The Handbook of Cuffless Blood Pressure Monitoring: A Practical Guide for Clinicians, Researchers, and Engineers [1st ed. 2019] 978-3-030-24700-3, 978-3-030-24701-0

Table of contents :
Front Matter ....Pages i-xii
Cuffless Blood Pressure Monitoring and the Advent of a New Era in Medicine and Society (Alberto Avolio, Fatemeh Shirbani, Isabella Tan, Mark Butlin)....Pages 1-7
Clinical Relevance of Continuous and Cuffless Blood Pressure Monitoring (Gianfranco Parati)....Pages 9-13
A Historical Journey on the Physiology of Blood Pressure Monitoring (Audrey Adji, Michael F. O’Rourke)....Pages 15-30
The Definition and Architecture of Cuffless Blood Pressure Monitors (Josep Solà)....Pages 31-42
Pulse Arrival Time Techniques (Marshal S. Dhillon, Matthew J. Banet)....Pages 43-59
Pulse Wave Velocity Techniques (Jim Li)....Pages 61-73
Pulse Decomposition Analysis Techniques (Martin C. Baruch)....Pages 75-105
Pulse Wave Analysis Techniques (Martin Proença, Philippe Renevey, Fabian Braun, Guillaume Bonnier, Ricard Delgado-Gonzalo, Alia Lemkaddem et al.)....Pages 107-137
Machine Learning Techniques (Xiaorong Ding)....Pages 139-162
Initialization of Pulse Transit Time-Based Blood Pressure Monitors (Ramakrishna Mukkamala, Jin-Oh Hahn)....Pages 163-190
Key Regulatory Aspects and the Importance of Consensus Standards in Bringing Devices to Market (Carole C. Carey)....Pages 191-202
Design of Clinical Trials to Validate Cuffless Blood Pressure Monitors (Willem J. Verberk)....Pages 203-224
Cuffless Blood Pressure Monitoring: The Future for the Evaluation and Management of Hypertension (George S. Stergiou)....Pages 225-230
Back Matter ....Pages 231-239

Citation preview

Josep Solà · Ricard Delgado-Gonzalo Editors

The Handbook of Cuffless Blood Pressure Monitoring A Practical Guide for Clinicians, Researchers, and Engineers

The Handbook of Cuffless Blood Pressure Monitoring

Josep Solà  •  Ricard Delgado-Gonzalo Editors

The Handbook of Cuffless Blood Pressure Monitoring A Practical Guide for Clinicians, Researchers, and Engineers

Editors Josep Solà Aktiia SA Neuchâtel, Switzerland

Ricard Delgado-Gonzalo Swiss Center for Electronics and Microtechnology (CSEM, Centre Suisse d’Electronique et de Microtechnique) Neuchâtel, Switzerland

ISBN 978-3-030-24700-3    ISBN 978-3-030-24701-0 (eBook) https://doi.org/10.1007/978-3-030-24701-0 © Springer Nature Switzerland AG 2019 This work is subject to copyright. All rights are reserved by the Publisher, whether the whole or part of the material is concerned, specifically the rights of translation, reprinting, reuse of illustrations, recitation, broadcasting, reproduction on microfilms or in any other physical way, and transmission or information storage and retrieval, electronic adaptation, computer software, or by similar or dissimilar methodology now known or hereafter developed. The use of general descriptive names, registered names, trademarks, service marks, etc. in this publication does not imply, even in the absence of a specific statement, that such names are exempt from the relevant protective laws and regulations and therefore free for general use. The publisher, the authors, and the editors are safe to assume that the advice and information in this book are believed to be true and accurate at the date of publication. Neither the publisher nor the authors or the editors give a warranty, express or implied, with respect to the material contained herein or for any errors or omissions that may have been made. The publisher remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. This Springer imprint is published by the registered company Springer Nature Switzerland AG The registered company address is: Gewerbestrasse 11, 6330 Cham, Switzerland

Preface

Let us start with a typical scenario from clinical practice today. During an annual physical exam: 143/97, Mr. Vailet, I have the feeling you are developing hypertension. The day after, at Mr. Vailet’s home: 134/88 Two days after: 129/84 No further measurements until the day before the next physical exam, at Mr. Vailet’s home: 148/95 Back to the physician’s office: Mr. Vailet, where are the weekly blood pressure readings I asked you to do? Measuring arterial blood pressure is a fundamental action when checking a patient’s overall health. Nowadays, physicians and patients alike still rely on a century-­old technology that requires the inflation of a cuff around the arm. This leads to intermittent, sparse, or nonexistent monitoring of a patient’s health status over long periods of time. As a result, millions of individuals are being wrongly treated, underdiagnosed, or simply not diagnosed at all. For more than a decade, an assembly of scientists, physicians, researchers, and engineers have been working on the development and validation of novel technologies to overcome the burden of cuff-based blood pressure measurements. This book is a tribute to this collective effort. In 2017, we started crafting “The Handbook of Cuffless Blood Pressure Monitoring: A Practical Guide for Clinicians, Researchers, and Engineers” that you have now in your hands. From the beginning, we aimed at creating the first v

vi

Preface

comprehensive publication providing an overview of the emerging field of cuffless blood pressure monitoring. We gathered the most knowledgeable authors around the world who could summarize the basics, the medical context, the potential, and the technical challenges of cuffless blood pressure monitoring. This work is now done. Dear reader, if you are a researcher, clinician, engineer, journalist, investor, or student who wants to get into the field of cuffless blood pressure, we are convinced you will enjoy diving into these chapters. Neuchâtel, Switzerland  Josep Solà   Ricard Delgado-Gonzalo

Acronyms

AAMI Association for the Advancement of Medical Instrumentation ABPM Ambulatory blood pressure monitoring AI/AIx Augmentation index ANSI American National Standards Institute AP Augmented pressure APG Acceleration plethysmogram BCG Ballistocardiography BHS British Hypertension Society BP Blood pressure CT Computed tomography DBP/DIA Diastolic blood pressure DIA Diastolic blood pressure DL Deep learning DNN Deep neural network DPTI Diastolic pressure-time index DT Diastolic time ECG Electrocardiography EHS European Home Systems EMAT Electromechanical activation time ESH European Society of Hypertension ESP End-systolic pressure FDA Food and Drug Administration FF Form factor FQP Four-quadrant plot GCP Good clinical practice GTF Generalized transfer function HR Heart rate ICG Impedance cardiography IDE Investigational device exemption IEEE Institute of Electrical and Electronics Engineers IPG Impedance plethysmography vii

viii

IR Infrared ISO International Organization for Standardization IVCT Isovolumic contraction time LED Light-emitting diode LSTM Long short-term memory LVET Left ventricular ejection time MAD Mean absolute differences MAP Mean arterial pressure MAPD Mean absolute percentage difference MDBP Mean diastolic blood pressure ML Machine learning MSBP Mean systolic blood pressure NIBP Noninvasive blood pressure NN Neural network NPMA N-point moving average PAT Pulse arrival time PCG Phonocardiography PDA Pulse decomposition analysis PEP Pre-ejection period PMA Premarket approval PP Pulse pressure PPG Photoplethysmography PTT Pulse transit time PWA Pulse wave analysis PWD Pulse wave decomposition PWTT Pulse wave transit time PWV Pulse wave velocity RI Reflection index RMSE Root mean square error RNN Recurrent neural network SBP Systolic blood pressure SCG Seismocardiography SD Standard deviation SEVR Subendocardial viability ratio SI Stiffness index SPTI Systolic pressure-time index SVM Support vector machine SVR Support vector regression SYS Systolic blood pressure

Acronyms

Contents

  1 Cuffless Blood Pressure Monitoring and the Advent of a New Era in Medicine and Society����������������������������������������������������������    1 Alberto Avolio, Fatemeh Shirbani, Isabella Tan, and Mark Butlin   2 Clinical Relevance of Continuous and Cuffless Blood Pressure Monitoring ��������������������������������������������������������������������    9 Gianfranco Parati   3 A Historical Journey on the Physiology of Blood Pressure Monitoring��������������������������������������������������������������������������������   15 Audrey Adji and Michael F. O’Rourke   4 The Definition and Architecture of Cuffless Blood Pressure Monitors������������������������������������������������������������������������������������   31 Josep Solà   5 Pulse Arrival Time Techniques ��������������������������������������������������������������   43 Marshal S. Dhillon and Matthew J. Banet   6 Pulse Wave Velocity Techniques��������������������������������������������������������������   61 Jim Li   7 Pulse Decomposition Analysis Techniques��������������������������������������������   75 Martin C. Baruch   8 Pulse Wave Analysis Techniques������������������������������������������������������������  107 Martin Proença, Philippe Renevey, Fabian Braun, Guillaume Bonnier, Ricard Delgado-­Gonzalo, Alia Lemkaddem, Christophe Verjus, Damien Ferrario, and Mathieu Lemay   9

Machine Learning Techniques����������������������������������������������������������������  139 Xiaorong Ding

ix

x

Contents

10 Initialization of Pulse Transit Time-Based Blood Pressure Monitors������������������������������������������������������������������������������������  163 Ramakrishna Mukkamala and Jin-Oh Hahn 11 Key Regulatory Aspects and the Importance of Consensus Standards in Bringing Devices to Market��������������������������  191 Carole C. Carey 12 Design of Clinical Trials to Validate Cuffless Blood Pressure Monitors ������������������������������������������������������������������������  203 Willem J. Verberk 13 Cuffless Blood Pressure Monitoring: The Future for the Evaluation and Management of Hypertension ������������������������  225 George S. Stergiou Index������������������������������������������������������������������������������������������������������������������  231

Contributors

Audrey Adji  St Vincent’s Clinical School, University of New South Wales, Sydney, NSW, Australia Department of Biomedical Sciences, Faculty of Medicine and Health Sciences, Macquarie University, Sydney, NSW, Australia Alberto  Avolio  Department of Biomedical Sciences, Faculty of Medicine and Health Sciences, Macquarie University, Sydney, NSW, Australia Matthew J. Banet  toSense, Inc., San Diego, CA, USA Martin C. Baruch  Caretaker Medical LLC, Charlottesville, VA, USA Guillaume Bonnier  Swiss Center for Electronics and Microtechnology (CSEM, Centre Suisse d’Electronique et de Microtechnique), Neuchâtel, Switzerland Fabian Braun  Swiss Center for Electronics and Microtechnology (CSEM, Centre Suisse d’Electronique et de Microtechnique), Neuchâtel, Switzerland Mark Butlin  Department of Biomedical Sciences, Faculty of Medicine and Health Sciences, Macquarie University, Sydney, NSW, Australia Carole C. Carey  C3-Carey Consultants, LLC, Fulton, MD, USA Ricard  Delgado-Gonzalo  Swiss Center for Electronics and Microtechnology (CSEM, Centre Suisse d’Electronique et de Microtechnique), Neuchâtel, Switzerland Marshal S. Dhillon  toSense, Inc., San Diego, CA, USA Xiaorong Ding  Institute of Biomedical Engineering, Department of Engineering Science, University of Oxford, Oxford, UK Damien  Ferrario  Swiss Center for Electronics and Microtechnology (CSEM, Centre Suisse d’Electronique et de Microtechnique), Neuchâtel, Switzerland

xi

xii

Contributors

Jin-Oh  Hahn  Department of Mechanical Engineering, University of Maryland, College Park, MD, USA Mathieu  Lemay  Swiss Center for Electronics and Microtechnology (CSEM, Centre Suisse d’Electronique et de Microtechnique), Neuchâtel, Switzerland Alia  Lemkaddem  Swiss Center for Electronics and Microtechnology (CSEM, Centre Suisse d’Electronique et de Microtechnique), Neuchâtel, Switzerland Jim Li  Global Medical Affairs, Omron Healthcare, Inc., Lake Forest, IL, USA Ramakrishna Mukkamala  Department of Electrical and Computer Engineering, Michigan State University, East Lansing, MI, USA Michael  F.  O’Rourke  St Vincent’s Clinical School, University of New South Wales, Sydney, NSW, Australia Faculty of Medicine, University of New South Wales, Sydney, NSW, Australia Gianfranco Parati  Department of Medicine and Surgery, University of Milano-­ Bicocca, Milan, Italy Department of Cardiovascular, Neural and Metabolic Sciences, Istituto Auxologico Italiano, IRCCS, San Luca Hospital, Milan, Italy Martin  Proença  Swiss Center for Electronics and Microtechnology (CSEM, Centre Suisse d’Electronique et de Microtechnique), Neuchâtel, Switzerland Philippe  Renevey  Swiss Center for Electronics and Microtechnology (CSEM, Centre Suisse d’Electronique et de Microtechnique), Neuchâtel, Switzerland Fatemeh Shirbani  Department of Biomedical Sciences, Faculty of Medicine and Health Sciences, Macquarie University, Sydney, NSW, Australia Josep Solà  Aktiia SA, Neuchâtel, Switzerland George S. Stergiou  Hypertension Center STRIDE-7, National and Kapodistrian University of Athens, Third Department of Medicine, Athens, Greece Isabella Tan  Department of Biomedical Sciences, Faculty of Medicine and Health Sciences, Macquarie University, Sydney, NSW, Australia Willem J. Verberk  Microlife AG, Widnau, Switzerland CARIM School for Cardiovascular Diseases, Maastricht University, Maastricht, The Netherlands Christophe  Verjus  Swiss Center for Electronics and Microtechnology (CSEM, Centre Suisse d’Electronique et de Microtechnique), Neuchâtel, Switzerland

Chapter 1

Cuffless Blood Pressure Monitoring and the Advent of a New Era in Medicine and Society Alberto Avolio, Fatemeh Shirbani, Isabella Tan, and Mark Butlin

Abstract  Measurement of arterial blood pressure (BP) by the brachial cuff sphygmomanometer has been a cornerstone of modern medicine, and notwithstanding its limitations of intermittent BP monitoring, the cuff sphygmomanometer has not been surpassed by any other noninvasive methodology. However, advances in sensor technology for arterial pulse detection have paved the way for the potential development of devices for cuffless measurement of BP, with the prospect of continuous monitoring. Just as the cuff sphygmomanometer could be considered as having provided a significant step in establishing elevated BP as a major factor of cardiovascular risk since the turn of the twentieth century, cuffless BP monitoring might be considered a disruptive technology for continuous BP monitoring in healthy individuals during daily living and could become an integral component of modern digital health platforms in the twenty-first century. This volume of the Handbook of Cuffless Blood Pressure Monitoring addresses the significant challenges faced by this methodology. The underlying physiology and hemodynamic principles involved in the generation of the arterial pressure pulse are complemented by the clinical relevance of the novel cuffless methodologies in relation to the conventional cuff sphygmomanometer. The underlying pressure dependency of arterial properties, which is at the base of the cuffless technique employing pulse transit measurement as a surrogate measure of BP, is addressed in association with instrumentation, measurement techniques, calibration procedures, device validation and regulatory requirements. This book provides a timely and comprehensive platform on how to approach the critical question of whether cuffless is the future of BP monitoring. Keywords  Arterial stiffness · Pulse wave velocity · Pulse transit time · Sphygmomanometer · Blood pressure · Cuffless techniques · Peripheral pulse · Arterial hemodynamics · Big data

A. Avolio (*) · F. Shirbani · I. Tan · M. Butlin Department of Biomedical Sciences, Faculty of Medicine and Health Sciences, Macquarie University, Sydney, NSW, Australia e-mail: [email protected] © Springer Nature Switzerland AG 2019 J. Solà, R. Delgado-Gonzalo (eds.), The Handbook of Cuffless Blood Pressure Monitoring, https://doi.org/10.1007/978-3-030-24701-0_1

1

2

A. Avolio et al.

Introduction There is a notable coincidence between the accepted dates marking the publication of the first clinical use of the cuff sphygmomanometer by Scipione Riva-Rocci (15 December 1896) in Turin, Italy and the first projection of cinema to a paying audience by brothers Auguste and Louis Lumière (28 December 1895) in Paris, France. In addition to the remarkable proximity of the official dates, only 1 year apart, the two milestone events have a common origin through the work of French physiologist Étiene-Jules Marey in his studies of the circulation of blood. His evolving attempts to obtain quantitative representation of the force of the arterial pulse as detected and registered by graphic means paved the way for the invention of the sphygmograph, This led to the construction of devices for achieving time-lapsed photography to capture movement, culminating in series of ground-breaking publications and in the coining of the term chronophotography, a technique used to study movement in flight [1] and which influenced the seminal flight experiment of the Wright brothers in 1903 at Kitty Hawk, marking the birth of modern aviation. Although the arterial pulse has been used in medical diagnoses and described qualitatively since antiquity, the use of the sphygmograph enabled the first ever quantitative registration of any physiological parameter—a cuffless measure of arterial pressure, and long before the registration of the electrocardiogram. It was the efforts of many scientists and inventors such as Marey in the search for mechanical devices capable of producing an image of the varying morphology of the arterial pulse, coupled with advances in photography, that enabled the evolution and convergence of sphygmography and chronophotography. The one resulting in the advent and clinical use of the cuff sphygmomanometer by the Italian physician Riva-Rocci and the other in the promotion of cinema by French scientists and entrepreneurs such as the Lumière brothers in the late nineteenth century. With the dawn of the twentieth century, technological advances in the form of the cuff sphygmomanometer and cinema produced respective fundamental changes in the way medicine was practiced and in the way the moving image was used for enhanced communication and enrichment of the human experience. Both have had an immeasurable impact on human health and culture in society.

 uffless Monitoring of Blood Pressure: A Disruptive Step C in Health Management The above advances were not made in isolation, but were integral component parts of a movement of progressive development of ideas and concepts that revolutionized scientific thinking and technological applications in the eighteenth and nineteenth centuries. History is rich with iconic milestones such as James Watt’s development of the steam engine (cc 1780s) that was a major driving force for the industrial revolution. Perhaps one of the most significant fundamental unifying

1  Cuffless Blood Pressure Monitoring and the Advent of a New Era in Medicine…

3

concepts in human history, the description of electromagnetic radiation by James Maxwell in 1865, has extended the application of human endeavor beyond the Newtonian view of the world to Einstein’s concepts of relativity and to quantum mechanics. In modern parlance, and given the significant impact these developments have made on society as a whole, they might be described as highly “disruptive” in comparison to what came before. And it is precisely the notion of “disruption” that is at the center of the transition from blood pressure being measured intermittently as a clinical parameter by the cuff sphygmomanometer to blood pressure being monitored continuously by cuffless devices. Just as the intellectual and scientific pathways that led to the development of the cuff sphygmomanometer were intertwined with advances and pitfalls in other avenues of scientific and technological endeavor in the nineteenth century, the development of cuffless blood pressure technology will tread similar ground in the twenty-first century. The difference being that the disruptive elements in the cuffless journey are supported and boosted by the immense power of all that is digital—the Internet, global positioning systems, satellite communication, cloud computing, miniaturization of computer chips and electronic circuits, intelligent algorithms, developments in sensor technology supported by intelligent materials. The difference will also extend to the use of the devices. Whereas the cuff sphygmomanometer, virtually unchanged since its inception, has been essentially used for detection of markers associated with pathology (hypertension), cuffless monitoring of blood pressure will be incorporated in wearable devices, thus enabling collection of data in normal daily living of individuals in society, in the form of “big data.” It is yet not known in what way this form of data collection will modify the practice of medicine and how it will affect the understanding of blood pressure profiles with daily habits or lead to phenotypic classifications of normal individuals in the context of genetic analysis and precision medicine. However, there is a precedent. When brachial cuff blood pressure values were systematically collected using pen and paper by life insurance companies in the early twentieth century [2], it led to the understanding of how asymptomatic high blood pressure can be one of the most powerful factors of cardiovascular risk.

Underlying Principles of Cuffless Methodology An important overarching concept underlying cuffless measurement of blood pressure is the fundamental relationship between transmural pressure and mechanical properties of the arterial wall which influence wave propagation phenomena [3]. This is the pressure dependency of the material stiffness of all blood vessels. This property is present in all species with pressurized circulatory systems and is a fundamental evolutionary property of arterial design [4]. Since, in any physical system, wave propagation is determined by the bulk modulus of the material, the speed of any disturbance (due to cardiac contraction and ejection) that travels along the arterial wall is directly related to vessel stiffness. And given the essential relationship of

4

A. Avolio et al.

arterial stiffness and distending pressure, any measure of the velocity of the travelling pulse wave would be a measure of arterial pressure. This is an important consideration because pulse wave velocity (PWV) is directly related to blood pressure and so an accurate measurement of PWV should theoretically deliver a measure of blood pressure, provided that the relationship of PWV and blood pressure of the specific arterial path length is known. Changes in blood pressure would then be registered as changes in PWV, or more specifically, for a fixed distance, as changes in pulse transit time (PTT) [3]. Hence, on theoretical considerations, cuffless measurement of blood pressure using PTT could be considered to be a more robust measure of intra-arterial pressure than that obtained by using indirect surrogate signals such as appearance of a distal pulse (palpatory method), Korotkoff sounds (auscultatory method) or features of the envelope of the oscillation of cuff pressure during cuff deflation (oscillometric method).

The Handbook of Cuffless Blood Pressure Monitoring This volume of The Handbook of Cuffless Blood Pressure Monitoring, the first of its kind, promises to be a guide in the modern twenty-first century journey of novel developments of methods and technology for monitoring of blood pressure, methodologies and techniques which go beyond the traditional cuff sphygmomanometer. It is a timely endeavor, and its publication will be an important milestone in this new field. It is both important and essential. Wearable, mainly wrist-worn devices are being produced that purport to measure blood pressure. Being connected to the Internet, they transmit data to databanks that log blood pressure values for individuals during daily activities. Being consumer devices, there is generally little or no regulatory requirement for the blood pressure measurement, other than device safety. With increasing use of these devices, it is conceivable that blood pressure data will be mined from big data sources and potentially “useful” knowledge will be extrapolated. However, there is no guarantee that the information from such data has any reliable relation to continuous blood pressure or any potential physiological significance. This is because of the inherent complexity in translating the noninvasive cuffless surrogate signal to a physiological arterial pressure. That is, the reliability of the basic methodology of calibration [5, 6]. This is in contrast to the cuff sphygmomanometer, where the cuff pressure is actually measured with high precision, and specific levels of cuff pressure are associated with arterial phenomena, such as appearance and disappearance of Korotkoff sounds associated with systolic and diastolic pressure, although to a varying degree of association and correlation. The chapters in this book are comprehensive and cover a wide range of topics, all of which are highly relevant to the scientific, technological, industrial, and clinical aspects of the field of cuffless and continuous measurement of arterial blood pressure. The material presented in all the chapters offers a robust platform on which to launch this new field and which builds on the large amount of published work based on basic concepts of the circulation involved in the genesis of the pressure pulse,

1  Cuffless Blood Pressure Monitoring and the Advent of a New Era in Medicine…

5

sensor technology, device development and fabrication, signal processing, deep learning strategies, modelling, calibration procedures, in vivo, in vitro, and in silico experimentation, and methods of validation. This book addresses critical aspects relevant to the monitoring of blood pressure using cuffless techniques in 12 chapters. The following 11 chapters can be divided into three broad topic sections: Topic Section 1 (Chaps. 2 and 3) describes the underlying physiology and hemodynamic principles involved in the generation of the arterial pressure pulse and the clinical relevance of the novel cuffless methodologies in relation to the conventional cuff sphygmomanometer. The important distinction between the new and old techniques for blood pressure monitoring is that the cuff sphygmomanometer is usually associated with the detection, treatment and management of hypertension, thus a pathological condition. Novel, cuffless techniques, embedded in wearable devices, will be associated with monitoring continuous changes in blood pressure predominantly in normal individuals during daily activities. This will undoubtedly have an impact on the meaning of such things as blood pressure thresholds that separate the “normal” state from the “pathological” state. This suggests significant “disruption” in the actual understanding of what constitutes “hypertension” in association with a plethora of other health data obtained by wearable devices and interpreted by data analytics. Topic Section 2 (Chaps. 4–9) describes the areas of cuffless devices involving instrument components, measurement techniques and calibration procedures. This Topic Section is critical to the fundamental understanding of the essential components of what constitutes a cuffless device for monitoring of arterial blood pressure. The basic requirement is the detection of the arterial pulse, with sensor modality being key for reliable pulse detection. An important distinction between sensors used for cuffless devices is that they do not necessity need to provide a quantitative measure of the force of the arterial pulse, but rather a fiducial signal which is in synchrony with the beating heart. The measurement of arterial pressure is based in temporal characteristics, so there is no requirement for absolute grading of measured signals. This enhances the capacity of available sensors for signal detection and the scope for instrumentation to be adapted to wearable devices. However, while sensors may be sufficiently robust for signal detection, the conversion of a measured time delay (irrespective of the degree of accuracy) to a physiologically relevant blood pressure remains one of the most challenging aspects of this field. To date, calibration issues remain unresolved, particularly in terms of stability, frequency of performing calibrations, procedures involved in obtaining adequate ranges of blood pressure and the differences in relationships between pulse transit time and arterial pressure for different arterial sites, or even at different levels of blood pressure. The limitations related to inconsistent physiological relationships can be addressed using data-driven procedures involving machine learning techniques [7]. However, it is not yet known how different databases can be sufficiently regulated to produce reliable training platforms for a broad range of algorithms applied to machine learning approaches for blood pressure monitoring.

6

A. Avolio et al.

Topic Section 3 (Chaps. 10–12) addresses the important area of device validation in the context of regulatory requirements and standards. To a certain extent, there is a conceptual overlap between “cuffless” and “continuous” techniques for blood pressure monitoring which is not adequately addressed by the current standards. Hence, there is a need for clear and definite guidance for evaluation from regulatory authorities so as to provide sufficient certainty for the medical devices industry. Finally, with appropriate validation, future projections are made on the expectations of the use of cuffless devices for diagnosis and management of hypertension. However, this will need to be in the context of the potential disruption to the concept of hypertension and blood pressure thresholds. All chapters are comprehensive and authoritative. They cover a broad range of relevant topics that enable and enhance understanding of the field of cuffless blood pressure monitoring at a considerable depth. While there will be some necessary and unavoidable repetition of some sections in different chapters, this is a positive aspect as it highlights their relative importance and contributes to placing this new field in the historical, technological and social continuum.

Is “Cuffless” the Future of Blood Pressure Monitoring? The comprehensive nature of this volume brings out the advances as well as the formidable challenges facing the journey of device development for reliable and continuous monitoring of blood pressure using cuffless techniques. The “state of the art” suggests that some milestones have already been achieved. Sensor technology for pulse detection is highly evolved and robust. Advances in component miniaturization and complex chip design have enabled the explosion of wearable devices incorporating measurement and processing of physiological signals, providing information on metrics that can guide decisions on health management. However, while some of these metrics are highly reliable (e.g., heart rate, blood oxygen levels, and many others) and have been available for some time, accurate, reliable continuous, cuffless monitoring of blood pressure has presented insurmountable challenges. There have been many patent submissions, start-up companies, and scientific publications, but to date there is no device that is universally accepted by the wider community beyond research laboratories and company boardrooms. This book will make a significant contribution to providing an informed answer to this important question.

References 1. Marey E-J. Le vol des insectes étudié par la chronophotographie. La Nat. 1892;20(1):135–8. 2. Fischer JW. The diagnostic value of the sphygmomanometer in examinations for life insurance. JAMA. 1914;63:3.

1  Cuffless Blood Pressure Monitoring and the Advent of a New Era in Medicine…

7

3. Mukkamala R, Hahn JO, Inan OT, Mestha LK, Kim CS, Töreyin H, Kyal S. Toward ubiquitous blood pressure monitoring via pulse transit time: theory and practice. IEEE Trans Biomed Eng. 2015;62:1879–901. 4. Shadwick RE. Mechanical design in arteries. J Exp Biol. 1999;202:3305–13. 5. Butlin M, Shirbani F, Barin E, Tan I, Spronck B, Avolio AP.  Cuffless estimation of blood pressure: importance of variability in blood pressure dependence of arterial stiffness across individuals and measurement sites. IEEE Trans Biomed Eng. 2018;65:2377–83. 6. Ding X, Zhang Y, Tsang HK. Impact of heart disease and calibration interval on accuracy of pulse transit time-based blood pressure estimation. Physiol Meas. 2016;37:227–37. 7. Khalid SG, Zhang J, Chen F, Zheng D.  Blood pressure estimation using photoplethysmography only: comparison between different machine learning approaches. J  Healthc Eng. 2018;2018:1548647.

Chapter 2

Clinical Relevance of Continuous and Cuffless Blood Pressure Monitoring Gianfranco Parati

Abstract  The assessment of dynamic features of blood pressure, which represent a response of cardiovascular control mechanisms to environmental stimulations and to daily life challenges, not only offers important insights into cardiovascular regulation patterns but also carries clinically relevant information. Availability of tools for continuous blood pressure monitoring represents a key step to make such an assessment possible, but its implementation in daily practice requires noninvasive, simple and minimally intrusive methods. These methods are expected to overcome the well-known limitations characterizing the conventional approach to blood pressure measurement based on discontinuous blood pressure readings obtained through repeated arm cuff inflations. In such a perspective, techniques able to provide continuous blood pressure monitoring without the need of a cuff inflation would be welcome. Keywords  Continuous blood pressure monitoring · Blood pressure variability · Arm cuff inflation · Cuffless blood pressure measurement technology

Introduction Blood pressure (BP) is one of the most dynamic physiologic variables among those routinely measured in clinical practice, consisting of a series of pulse waves continuously changing in terms of both frequency and amplitude. BP is indeed characterized by continuous and significant changes occurring over different time windows, with beat-by-beat oscillations being intertwined in a complex manner with G. Parati (*) Department of Medicine and Surgery, University of Milano-Bicocca, Milan, Italy Department of Cardiovascular, Neural and Metabolic Sciences, Istituto Auxologico Italiano, IRCCS, San Luca Hospital, Milan, Italy e-mail: [email protected] © Springer Nature Switzerland AG 2019 J. Solà, R. Delgado-Gonzalo (eds.), The Handbook of Cuffless Blood Pressure Monitoring, https://doi.org/10.1007/978-3-030-24701-0_2

9

10

G. Parati

fluctuations occurring from minute to minute and from hour to hour, at the time of the different behaviors which occur over the day and night, Further complexity is added by additional and more long-lasting fluctuations which may be observed over different days, over different seasons, and at the time of consecutive visits by a physician even in a time window of several years [1]. In physiological conditions these BP variations largely represent a response of cardiovascular control mechanisms to environmental stimulations and to daily life challenges, aimed at maintaining the so-called cardiovascular “homeostasis,” being additionally influenced by food, physical exercise, or sleep patterns. Moreover, sustained increases in BP variability may also reflect intrinsic alterations in cardiovascular regulatory mechanisms or the effect of underlying pathological conditions responsible for a dysregulation of neural and humoral factors involved in modulating cardiac and vascular functions. The interest in the assessment of these BP variations comes from the evidence that they may have important clinical significance and prognostic implications, as demonstrated by a series of experimental, clinical, and population studies. Indeed, enhanced BP fluctuations have been associated with a higher risk of subclinical organ damage, cardiovascular events, and cardiovascular and all-cause mortality independently and on top of what determined by elevated average BP values [2–4]. Thus, theoretically there is no doubt that the most accurate and detailed approach to the assessment of BP behavior in daily life would benefit from the possibility of performing continuous beat by beat recording of BP over 24 h. However, such continuous recordings are not easy to implement in daily practice, where discontinuous BP measurements obtained through repeated arm cuff inflations still represent the method routinely applied. In fact, BP assessment in daily practice is currently based on two main methods. The first is the auscultatory method, based on the occlusion arm cuff-based technique introduced in 1896 by Scipione Riva-Rocci, coupled with use of Korotkoff sounds. The second is the oscillometric method, which makes use of the application of repeated automated cuff inflations to directly measure mean BP and then systolic and diastolic BP values are estimated through proprietary algorithms. It is by means of these measurements that in year 2019 we are still currently managing hypertensive and cardiovascular patients in daily practice. Although these approaches have been shown to carry important clinical diagnostic and prognostic information when applied either in a clinic environment or in daily life conditions out of a clinic setting, they are not free from major limitations, which include the inherent inaccuracy of the techniques used, the inability to account for the dynamic nature of blood pressure and for its fluctuations over time and the difficulty in obtained undisturbed readings that might reflect subjects’ actual BP patterns both during wake and sleep. In fact, the arm cuff inflation is often itself responsible for difficulties, as it may be responsible for discomfort and sometimes even pain in patients in whom high levels of air pressure might be required for a proper BP measurement. These problems are in particular relevant when focussing on nighttime BP measurement, that

2  Clinical Relevance of Continuous and Cuffless Blood Pressure Monitoring

11

is, on the measurements of BP obtained during sleep which have been shown to carry highly relevant clinical information. Nocturnal cuff inflations may indeed significantly disturb sleep of the subjects in whom they are performed, and may ­importantly interfere with the BP levels they are aimed at assessing. This is exemplified by the fact that conventional 24 h ambulatory BP monitoring, which is based on discontinuous and frequent BP measurements (typically every 15–30 min), in spite of its acknowledged clinical value is poorly accepted by many patients due to discomfort associated with repeated cuff inflations, which during the night may significantly worsen the quality of sleep [5]. Even if in many patients this does not affect BP levels [6], in some, an increase in nocturnal BP can be induced, leading to an artifactual reduction in BP dipping pattern with a loss of its prognostic significance, as reported previously for untreated hypertensive patients with significant sleep deprivation during 24 h BP monitoring [7]. These methodological problems with currently available techniques for BP assessment make it difficult to clarify a number of still pending clinical issues, such as the definition of the actual link between obstructive sleep apnea and hypertension; the understanding of why heart attacks and stroke mostly happen during sleep, especially in early morning; the definition of whether periodic leg movements and snoring might actually cause a BP increase; and the demonstration of the real ability of continuous positive air pressure (CPAP) ventilation to reduce BP levels in hypertensive patients with obstructive sleep apnea. To avoid such inconveniences and to get rid of the related clinical difficulties, techniques able to provide continuous BP monitoring without need of a cuff inflation would be required. Currently, a few techniques for continuous BP recordings without cuff inflation are indeed available, but are rarely employed for difficulties in their applicability. The first of such techniques, available since many years, is the possibility of intra-­ arterial beat-by-beat BP recordings, which have allowed to make the first important progress in understanding BP variability phenomena and the dynamics of cardiovascular control mechanisms [8–10]. This method however is based on invasive recordings, and even if it was extremely useful in former experimental studies, it cannot obviously be proposed for a clinical application. Another approach to noninvasive continuous cuffless beat by beat BP recordings is based on the volume-clamp method described by Jan Penaz and implemented in the Finapres© and Portapres© devices by Karel Wesseling. This approach has turned out to be useful in research [11] but has not found an application in clinical practice due to methodological difficulties in its daily use, to problems in calibration of the BP values provided and to its high cost. Some other recent approaches have been proposed making use of mobile health technologies. Indeed, several smartphones apps have been developed to measure BP. However, validation studies for most of these smartphone-based BP measurement techniques have not been conducted. To date, mobile health BP monitors have shown poor accuracy compared with oscillometric readings [12–15]. Other approaches are based on assessment of pulse transit time, which can derive BP levels from pulse transit time (PTT) measurement on the basis of a stretch–strain relationship model, calibrated with BP from a single initial conventional BP mea-

12

G. Parati

surement [16–18]; or on pulse wave analysis, and are currently under investigations and development [19]. There is increasing interest towards thee modern approaches, because the cuffless and thus less interfering nature of the devices may render them particularly useful in patients who poorly tolerate traditional ambulatory BP monitoring and whenever undisturbed assessment of sleep BP is of importance, for example, in patients with sleep disordered breathing. There is therefore a strong need for research and development of reliable and accurate continuous cuffless blood pressure measuring technologies, able to face the yet unmet need of accurate and at the same time noninterfering repeated BP measurements. The present book offers an interesting and up-to-date insight into the progress made by technology in this stimulating and highly dynamic field.

References 1. Parati G, Ochoa JE, Lombardi C, Bilo G. Assessment and management of blood-pressure variability. Nat Rev Cardiol. 2013;10(3):143–55. 2. Parati G, Pomidossi G, Albini F, Malaspina D, Mancia G. Relationship of 24-hour blood pressure mean and variability to severity of target-organ damage in hypertension. J  Hypertens. 1987;5(1):93–8. 3. Stevens SL, Wood S, Koshiaris C, Law K, Glasziou P, Stevens RJ, et al. Blood pressure variability and cardiovascular disease: systematic review and meta-analysis. BMJ. 2016;354:i4098. 4. Parati G, Ochoa JE, Bilo G. Blood pressure variability, cardiovascular risk, and risk for renal disease progression. Curr Hypertens Rep. 2012;14(5):421–31. 5. O’Brien E, Parati G, Stergiou G, Asmar R, Beilin L, Bilo G, et  al. European Society of Hypertension position paper on ambulatory blood pressure monitoring. J  Hypertens. 2013;31(9):1731–68. 6. Parati G, Pomidossi G, Casadei R, Malaspina D, Colombo A, Ravogli A, Mancia G. Ambulatory blood pressure monitoring does not interfere with the haemodynamic effects of sleep. J Hypertens Suppl. 1985;3:S107–9. 7. Verdecchia P, Angeli F, Borgioni C, Gattobigio R, Reboldi G.  Ambulatory blood pressure and cardiovascular outcome in relation to perceived sleep deprivation. Hypertension. 2007;49:777–83. 8. Mancia G, Parati G, Pomidossi G, Casadei R, Di Rienzo M, Zanchetti A. Arterial baroreflexes and blood pressure and heart rate variabilities in humans. Hypertension. 1986;8(2):147–53. 9. Parati G, Saul JP, Di Rienzo M, Mancia G.  Spectral analysis of blood pressure and heart rate variability in evaluating cardiovascular regulation. A critical appraisal. Hypertension. 1995;25(6):1276–86. 10. Mancia G, Ferrari A, Gregorini L, Parati G, Pomidossi G, Bertinieri G, Grassi G, Di Rienzo M, Pedotti A, Zanchetti A.  Blood pressure and heart rate variabilities in normotensive and hypertensive human beings. Circ Res. 1983;53:96–104. 11. Parati G, Casadei R, Groppelli A, Di Rienzo M, Mancia G. Comparison of finger and intra-­ arterial blood pressure monitoring at rest and during laboratory testing. Hypertension. 1989;13:647–55. 12. Muntner P, et al. Measurement of blood pressure in humans: a Scientific Statement From the American Heart Association. Hypertension. 2019;73(5):e35–66.

2  Clinical Relevance of Continuous and Cuffless Blood Pressure Monitoring

13

13. Kumar N, Khunger M, Gupta A, Garg N. A content analysis of smartphone based applications for hypertension management. J Am Soc Hypertens. 2015;9:130–6. https://doi.org/10.1016/j. jash.2014.12.001. 14. Cortez NG, Cohen IG, Kesselheim AS. FDA regulation of mobile health technologies. N Engl J Med. 2014;371:372–9. https://doi.org/10.1056/NEJMhle1403384. 15. Bruining N, Caiani E, Chronaki C, Guzik P, van der Velde E, Task Force of the e-­Cardiology Working. Acquisition and analysis of cardiovascular signals on smartphones: potential, pitfalls and perspectives: by the Task Force of the e-Cardiology Working Group of European Society of Cardiology. Eur J  Prev Cardiol. 2014;21(2 Suppl):4–13. https://doi. org/10.1177/2047487314552604. 16. Schmalgemeier H, Bitter T, Bartsch S, Bullert K, Fischbach T, Eckert S, et al. Pulse transit time: validation of blood pressure measurement under positive airway pressure ventilation. Sleep Breath. 2012;16:1105–12. 17. Gesche H, Grosskurth D, Kuchler G, Patzak A.  Continuous blood pressure measurement by using the pulse transit time: comparison to a cuff-based method. Eur J  Appl Physiol. 2012;112:309–15. 18. Bilo G, Zorzi C, Oghoa Munera JE, Torlasco C, Giuli V, Parati G.  Validation of the Somnotouch™ NIBP non-invasive continuous blood pressure monitor according to the European Society of Hypertension International Protocol revision 2010. Blood Press Monit. 2015;20(5):291–4. 19. Baruch MC, Warburton DE, Bredin SS, Cote A, Gerdt DW, Adkins CM. Pulse decomposition analysis of the digital arterial pulse during hemorrhage simulation. Nonlinear Biomed Phys. 2011;5(1):1.

Chapter 3

A Historical Journey on the Physiology of Blood Pressure Monitoring Audrey Adji and Michael F. O’Rourke

Abstract  The arterial pulse has been the most basic sign of life for centuries. The radial pulse palpation has been pictured in the crest of the Royal Academy College of Physicians since 1628. The history of the arterial pulse entails the discovery of pulse, blood pressure and/or flow, and their measurements. This chapter begins with a review the description of the pulse and the related discoveries of pulse and blood pressure and/or flow since the ancient period until the late 1970s where the concept of haemodynamics and importance of pressure and flow pulsatility as well as methods to analyse the pulse in both time and frequency domains gained wider acceptance. Human aging is associated with an increase in blood pressure, particularly systolic and pulse pressures, and this is attributable to the loss of distensibility of the human aorta of which its function is to cushion pulsation from the ejecting heart. Stiffening of the major elastic arteries due to aging will cause the speed of the travelling pulse to be higher, and the reflected pulse wave from periphery to occur earlier, therefore will increase the amplitude of pressure. To understand how arterial haemodynamics is altered by the ageing process and cardiovascular disease is vital and this involves accurate measurement of central (or aortic) pressure. Finally, the chapter briefly considers the demand and technology to develop cuffless blood pressure measuring devices. This development could allow a device that can measure blood pressure accurately, with ease, comfortably and continuously. Keywords  Systolic pressure · Diastolic pressure · Mean pressure · Pulse pressure · Pressure pulse wave · Sphygmocardiography · Aging · Hypertension · Arterial stiffness

A. Adji (*) St Vincent’s Clinical School, University of New South Wales, Sydney, NSW, Australia Department of Biomedical Sciences, Faculty of Medicine and Health Sciences, Macquarie University, Sydney, NSW, Australia M. F. O’Rourke St Vincent’s Clinical School, University of New South Wales, Sydney, NSW, Australia Faculty of Medicine, University of New South Wales, Sydney, NSW, Australia © Springer Nature Switzerland AG 2019 J. Solà, R. Delgado-Gonzalo (eds.), The Handbook of Cuffless Blood Pressure Monitoring, https://doi.org/10.1007/978-3-030-24701-0_3

15

16

A. Adji and M. F. O’Rourke

From Pulse Wave to Systolic and Diastolic Pressures Many clinicians and researchers think of blood pressure only in terms of the brachial arterial cuff systolic pressure (SP) and diastolic pressure (DP) derived therefrom. But this is just a convenient way of getting two numbers for the top and bottom of the pressure wave in the brachial artery. Though usually now generated from mathematical algorithms in an office device, the vast amount of blood pressure data obtained in the past was from the auscultatory method where SP represents the first Korotkov sounds heard as the cuff pressure is reduced from above, and DP the cuff pressure at which Korotkov sounds are no longer heard. These are usually rounded off to the nearest 2 or 5 mmHg, so that SP measured at 120  mmHg may be from 110 to 130  mmHg, DP from 70 to 90  mmHg, and Pulse Pressure (PP  =  SP − DP) within an even wider range. For electronic devices, one still has the issue of respiratory and cardiac variability with respiration alone which is easily up to 5 mmHg in sinus rhythm with respiration, or 15 mmHg in atrial fibrillation. We are told by James Mackenzie (Father of Cardiology in the English-speaking world), to strike an average SP or DP, and by the American Heart Association guidelines [1, 2] to do the same—but there is likely to be bias (“experience”) in writing two numbers between hearing the K sounds, and writing in the medical record or entering two numbers into a computer. Clinicians (doctors and nurses) are aware of these practical issues, but non-medical persons (including scientists and patients) often are not. One needs take time and care to measure blood pressure, and taking as many recordings as possible. The heart is a periodic fluid pump; therefore it must generate a pulsatile blood flow pattern. The pressure generated by intermittent flow is determined by both the cardiac performance in ejecting blood out of the heart and the properties of the arterial system. The arterial system acts as a conduit to maintain continuous delivery of oxygenated blood to the tissues, and as a cushion to dampen the pressure and flow pulsatility and convert this oscillation into a near continuous flow at the capillaries. With each beat of the heart, a pulse wave of blood is expelled from the heart into the circulation. These waves can be felt as pulsations in arteries close to the skin surface, such as those in the wrist and the neck.

The Ancient Times The arterial pulse felt at the wrist has been used by physicians from ancient times to the modern epoch, for diagnosing illnesses. Throughout centuries, the arterial pulse was considered the most basic sign of life and was considered to contain an abundance of information on the health or disease of a patient. Graphic recordings of the pulse waveform at the wrist were first made in the late nineteenth century and were quickly embraced by physicians as a useful clinical tool, along with the stethoscope. The pulse waveform was recognised as providing information on elevated arterial pressure and on effects of arterial stiffening with age. Interest in the pulse lapsed

3  A Historical Journey on the Physiology of Blood Pressure Monitoring

17

with introduction of the cuff sphygmomanometer solely to measure peak and trough of the arterial pressure wave. Presentation these top and bottom numbers provides a veneer of science, which the method of measuring cuff blood pressure does not deserve. Ironically, as shown in modern SphygmoCor reports (Figs. 3.1 and 3.2), variation in waveforms values (2–3% in figures shown) is far less than variation in values of SP and DP taken with a cuff sphygmomanometer.

Fig. 3.1  An example of a young male with spurious systolic hypertension. His brachial SP was taken as 154 mmHg; hence, he would be diagnosed as hypertensive based on this number alone. His aortic SP, however, was calculated as 123 mmHg, which was normal. The elevated brachial SP is due to distortion in pulse wave transmission from the heart to upper limb, associated with young elastic arteries, resulting in sharp narrow systolic peak recorded in the radial pulse

18

A. Adji and M. F. O’Rourke

Fig. 3.2  An example of an elderly female with true isolated systolic hypertension. Her brachial SP was taken as 220 mmHg, and her aortic SP, was calculated as 209 mmHg, appropriately diagnosed as hypertensives. The elevated brachial SP is due to stiffening of the central elastic arteries, where the pulse wave transmitted faster from the heart to the periphery, resulting in the reflected pulse wave superimposed in the ejected pulse wave earlier, thus higher aortic SP

3  A Historical Journey on the Physiology of Blood Pressure Monitoring

19

The Mercury Manometer Stephen Hales performed the first measurement of blood pressure [3]. As described in his book “Statical Essays: Containing Haemastatics” in 1733, Hales recorded damped systolic arterial pressure using a tall glass tube in a horse from the height of a blood column and determined the effects of exsanguination on arterial pressure (Fig. 3.3). He reported that the flowing blood exerts a pressure on the blood vessel’s wall while the blood circulation obeys hydrostatic laws [4]. Jean Poiseuille (1799– 1869) advanced Hales’ method of measuring blood pressure by replacing the long glass tube with a U-tube mercury manometer (called haemodynamometer), published in his doctoral dissertation in 1828. Poiseuille calibrated his manometer to

Fig. 3.3  An attempt to measure blood pressure in a horse. An artist’s depiction of Hales’ experiment to measure blood pressure in a horse. From [3]

20

A. Adji and M. F. O’Rourke

record arterial pressure level in mmHg, and was able to calculate the force of the heart beat by observing the oscillation of mercury with his manometer [5]. He also found that there is no significant difference in mean pressure between central and tiny peripheral arteries, observed fluctuations in pressure with each heartbeat, and demonstrated that arterial pressure was maintained in smaller arteries [6]. Further modification of the Poiseuille manometer was made by Julius Hérisson of Paris in 1831/1833 by sealing the end of the mercury containing glass tube with a thin membrane which rested on the artery, thus enabled him to obtain crude reading of blood pressure from intact arteries [4]. In his book of 1835, Hérisson stated that he had no knowledge of Poiseuille’s manometer at the time he developed his, and further claimed his manometer was not similar to Poiseuille’s. Utilising Hérisson’s device, the pulse can be studied by its force, regularity and rhythm.

The Kymograph and the Sphygmograph In 1847, Carl Ludwig (1816–1895) made another modification to Poiseuille’s manometer by adding a float, thus invented the kymograph (or “wave writer”) to depict arterial pressure [3, 4], hence set out modern methods of graphic representation of the pulse [5]. Ludwig’s invention started from his interest on physiological relation between respiration rate and blood pressure. About the same time as Ludwig, Karl van Vierordt in 1855 suggested that an indirect, non-invasive technique to measure pulse might be generated from the counter pressure required to cause arterial pulsation to cease [3]. Vierordt’s instrument was large, had two heavily weighted levers which converted the unequal rise and fall of the pulse into a symmetrical curve [5]. Etienne Jules Marey (1830–1904), frustrated by the bulk of Vierordt’s instrument, developed a simpler and more accurate device to illustrate and measure the arterial pulse [5], by combining a sphygmograph and a kymograph [3] (Fig. 3.4). Marey’s device was introduced in the Lancet in 1860 with a notice “It may be doubted whether these instruments, though very ingenious, will ever prove actually useful in practice” [5]. Marey did the first formal study on arterial pulse waves to demonstrate the difference in the pulse of elderly and young adults (Fig.  3.5). Marey’s sphygmograph made apparent the asymmetrical nature of pulse waves and their variation under different physiological and pathological condition [5]. Marey wrote a chapter on arterial blood velocity in his popular textbook of medicine “Le circulation du sang à l’état physiologique et dans des maladies (The Circulation of Blood in the Physiological State and in Diseases)” in 1881. While Marey’s sphygmograph was accepted widely for recording and studying the pulse, it was still too complicated to use routinely in practice, hence the sphygmograph went through further modifications in England. John Burden­ Sanderson, as described in his “Handbook of the Sphygmograph” in 1867–1868, together with Francis Anstie, modified Marey’s sphygmograph by fixing the pressure to be 300  g at the centre button and adding a screw to raise and lower the spring’s free end by a measured amount at the distal end of the device.

3  A Historical Journey on the Physiology of Blood Pressure Monitoring

21

Fig. 3.4  Von Basch sphygmomanometer and stand, invented circa 1881. Von Basch’s sphygmograph incorporated Marey’s sphygmograph tambour—see also Fig. 3.5. From [3]

Independent from the development of sphygmography, the first truly accurate estimation of human blood pressure was done by Faivre in 1856 [3]. He connected an artery to a mercury manometer during a surgical operation and was able to obtain direct readings. Faivre found brachial artery pressure to be between 115 and 120 mmHg [3]. Non-invasive blood pressure measurement was pioneered by von Basch in 1880, where he used an inflatable rubber bag filled with water, then the edges of the bag were tightly drawn up around the neck of a mercury manometer bulb to record pressure indirectly from the radial artery [3]. In 1875, William Henry Broadbent delivered a series of lectures on the pulse, which was published later in 1890 in his book “The Pulse”. Broadbent advocated careful and systematic evaluation of the pulse, started with frequency and regularity of beats, then the force or strength, finally the character (i.e. rise, duration and fall) of the pulse [7]. He tested his clinical skills against pulse waveforms recorded by sphygmography, and encouraged his students and examinees to do the same. (He was chief censor (examiner) for the Royal College of Physicians of London for many years.) Frederick Mahomed (1849–1884), one of Broadbent’s students, made further modification to Marey’s sphygmograph and published his work in 1872. His modification included placing over the centre of the spring a vertical eccentric turned by a graduated screw, as well as replaced the side wires with ivory bars [8]. Mahomed recorded the pulse graphically, thus permitting the pulse to be studied as a wave [9].

22

A. Adji and M. F. O’Rourke

Fig. 3.5  Marey sphygmograph and the recorded pulse. Marey’s sphygmograph and his study on arterial pulse, demonstrating the difference in the pulse of elderly and young adults. From this sphygmogram, it is apparent the asymmetrical nature of pulse wave and their variation under different physiological and pathological condition. From [3]

The use of sphygmogram to detect elevated pressure was further emphasised by William Osler, where senile or physiological arteriosclerosis and high blood pressure were diagnosed on the basis of pulse wave contour—from high and prolonged pressure in late systole, “the tidal wave is prolonged and too much sustained”, and “the pulse (of high tension) is slow in its ascent, enduring, subsides slowly, and in the interval between the beats, the vessel remains full and firm”. He stressed that the

3  A Historical Journey on the Physiology of Blood Pressure Monitoring

23

(radial) “pulse of high tension” is similar to the pulse in arteriosclerosis but that a test could be used to distinguish between the two (which are often combined) “(if) when the radial (artery) is compressed with the index finger, the artery can be felt beyond the point of compressions, its walls are sclerosed” [10]. Robert Ellis Dudgeon (1820–1904) invented a new compact sphygmograph, which he exhibited in 1881 and published in 1882 [8]. Dudgeon made significant improvements to Marey’s and Mahomed’s sphygmograph [8]. Dudgeon’s ­sphygmograph was later used by James Mackenzie, founder of Cardiology as a specialist discipline in the English-speaking world. Sir James Mackenzie (1853–1925) is closely associated with non-invasive investigation of normal and pathological cardiovascular phenomena [8]. He was one of the first observers to be interested in the regularity or otherwise of the pulse, also laid the foundation of modern concept of heart failure, which was being studied by Frank in Germany and by Starling in England at this time [8]. While the introduction of the sphygmomanometer into clinical medicine in the early 1900s was accepted by some practitioners as a valuable aid for diagnosis, the “British Medical Journal” in 1905 stated in relation to use of clinical tools that “we pauperize our senses and weaken clinical acuity” [3].

The Sphygmomanometer The next development in blood pressure measurement was the introduction of blood pressure measurement by palpation. Scipione Riva-Rocci in 1896 published two papers titled “Un Nuovo Sfigmomanometro” in Gazetta Medica di Torino. His sphygmomanometer was based on established principles of Vierordt and further improvement by Marey and von Basch. Advantages of the Riva-Rocci device were the ease of its application, rapid action, precision and harmlessness [3, 11]. The Riva-Rocci technique involved compression of the arm around its whole circumference by inflating a rubber bag with air. Using the palpation of the radial artery, systolic blood pressure can be determined [3]. One flaw of this technique was that the cuff was narrow; thus the reading of pressure can be inaccurate [3]. Later, von Recklinghausen in 1901 fixed this defect by substituting the armband by a wider one [3]. Further advancement in pressure measurement was made by Nikolai Korotkoff, where in 1905, he reported that by placing a stethoscope over the brachial artery below the air-pressure cuff, tapping sounds—the sounds of the column of blood— could be heard as the cuff was deflated and blood flowed through the artery [11]. Korotkoff concluded that a perfectly constricted artery under normal conditions does not emit any sounds, and he introduced the auscultatory method to measure maximum and minimum level of blood pressure [11]. This auscultatory method was described in the Imperial Military Medical Academy in St Petersburg in 1905. Popularity of the sphygmomanometric cuff arose from its use in the Life Insurance industry which utilised its value of identifying apparently normal people who had

24

A. Adji and M. F. O’Rourke

markedly reduced age expectancy. In 1906, Dr. J. W. Fisher, medical director of the Northwestern Mutual Life Insurance Co., initiated the inclusion of a blood pressure reading in every routine examination by their examiners [12] (Fig. 3.6). By 1918 most insurance companies were measuring cuff blood pressures. Reliance on blood pressure measured from the brachial artery began since then, and cuff-based sphygmomanometry is widely used to date because it is easy to perform, despite merely describing the pressure as its peak (systolic) and nadir (diastolic) value.

Clinical Practice Today To a great extent pressure metrics have pervaded medicine, since virtually every diagnostic clinic visit is preceded by the ritual of taking the blood pressure. The conventional office cuff pressure measurement remains the “gold standard” for screening, diagnosis and assessment of arterial pressure level. Unfortunately, measurement of cuff pressure sometimes is performed “in the sloppiest manner” [13, 14]. These are strong words, but were made by respected, and usually reserved, clinical scientists. Cuff blood pressure is an ever-changing hemodynamic index; it varies with time, body position, heartbeat and many more physiological factors. Preoccupation with the sphygmomanometric “numbers” persists in the diagnosis and management of elevated blood pressure. At the turn of this (twenty-first) century, attention was transferred from diastolic to systolic pressure as the major index of risk in adult hypertension [15, 16] and to the importance of large arteries both in determining SP and PP and as the target of damage when pressure is high. This has been aided by new methods for determining and interpreting pulsatile arterial pressure, flow and diameter non-invasively [10].

Hypertension as a Risk Factor in Cardiovascular Disease Blood pressure will increase with age, even in apparently healthy individuals; it is acknowledged as a feature of human aging [17, 18] and has been demonstrated repeatedly including by the longitudinal study of Framingham [17]. In the past, risk evaluation was based on brachial DP, led by a misguided therapeutic proposition of James Orr, the literary executor of James MacKenzie following his death [19]. During 1910s, the use of sphygmomanometers gained popularity, especially within life insurance companies, who were able to relate blood pressure level to mortality outcomes [20] (Fig. 3.6). It was not until the 1960s when the investigators from the Framingham Heart Study began to recognise high brachial SP as an important cardiovascular risk factor [21]. This recognition was then followed by two major reports from the Veteran Affairs Cooperative Study on Antihypertensive Agents that lowering blood pressure resulted in significant reduction in major adverse events, including stroke and heart failure [22, 23]. The Framingham Heart Study [24] went

3  A Historical Journey on the Physiology of Blood Pressure Monitoring

25

Fig. 3.6  Inclusion of blood pressure measurement for insurance examination as published in JAMA 1914 by Dr. Fisher. From [12]

26

A. Adji and M. F. O’Rourke

on to conclude the importance of systolic hypertension in the middle-aged and elderly subjects, as opposed to DP.  Many other clinical scientists continue these epidemiological studies in association with the US National Heart, Lung and Blood Institute and American Heart Association [25]. The significance of SP was finally confirmed with the publication of the Systolic Hypertension in the Elderly Project (SHEP) [26]. Subsequently, the (brachial) SP and PP emerged as the (modifiable) key risk factor in cardiovascular disease following longitudinal studies of Framingham Heart Study [16, 27]. They found that coronary heart disease risk is more related to pulsatile stress due to large artery stiffness during systole, thus emphasised the consequence of isolated systolic hypertension and the importance of SP and/or PP level in adults. Franklin et al. went further and concluded that the two measures of pressure are superior to any single pressure component in predicting cardiovascular risk because these pressures assessed both arterial stiffness and resistance [28]. The latest publication from the Framingham Heart Study cohort again emphasised the importance of pulse pressure as the pulsatile load in increasing brachial blood pressure with age, at least for those over the age of 60 [17]. Increased arterial stiffness due to age is the main determinant of high blood pressure. Elevated blood pressure increases the load on the heart and stress on the arteries, hence assessment of arterial function should incorporate monitoring of blood pressure levels to gauge the ill-effect of arterial stiffness. It is now widely accepted that stiffening of the major (central) arteries with ageing affects the propagation of pressure pulse along the arterial tree, increases blood pressure, changes the ejection pattern of the heart, and eventually causes cardiac failure. Progressive stiffening of the large elastic arteries is attributable to loss of their distensibility with increasing age, characterised as faster travelling of the pulse generated by the heart (measurable through “aortic” pulse wave velocity) and higher central systolic pressure. Stiffening of the central arteries has been recognised as a cause of adverse cardiovascular outcomes, including a higher amplitude of central PP and an increase in the transmission of pulsatile flow into the microcirculation [2]—published as a Scientific Statement by the American Heart Association, and by the European Societies of Cardiology and Hypertension [29]. Most of the non-invasive studies published on central aortic pressure and related indices have shown clear superiority of these in determining clinical end points. Central aortic SP and aortic PP are increasingly important because these are not affected by amplification of pressure between aorta and upper limb (Figs. 3.1 and 3.2). They are more relevant to left ventricular load than SP and PP measured in the upper limb. SP varies among arterial segments due to the phenomenon of arterial pulse amplification [30]. Therefore, the common practice of using conventional brachial cuff sphygmomanometric measurements as pressure load imposed on the heart and on central circulation (including coronary and cerebrovascular arteries) may be erroneous [30]. Indices derived from the aortic pressure wave, including pressure pulse amplification to the upper limb, its augmentation from peripheral wave reflection, its integral during the periods of systole and diastole would provide more information on vascular load, cardiac function, and ventricular–vascular interaction.

3  A Historical Journey on the Physiology of Blood Pressure Monitoring

27

Management of elevated high blood pressure has evolved since cuff blood pressure was introduced clinically 120 years ago. The conventional office blood pressure recording remains the “gold standard” for screening, diagnosis and management of elevated blood pressure [29, 31]. At present, out-of-office blood pressure measurements are increasingly recommended to confirm the diagnosis of elevated blood pressure and to gauge the effectiveness of therapy.

The Need for Cuffless Blood Pressure Measuring Devices There is now a demand and a sufficiently advanced technology to develop cuffless blood pressure measuring devices. This development could allow for a device that can measure accurately, with ease, comfort and continuity. Current devices use modified tonometry or volume clamp/plethysmography of pulse wave transit time/ velocity to measure pressure. A cuffless blood pressure would utilise optical sensors, similar to fitness trackers. These wrist and finger sensors use beat-to-beat variability to compute systolic and diastolic readings by mathematical modelling. The benefits of these sensors include the ability to monitor blood pressure continuously and avoid sleep-disrupting cuff inflations when recordings are performed in the evening. These devices also will particularly be useful in the elderly or person with limited mobility. However, the accuracy of these devices is still questioned. Despite validation, measurements can vary as much as 20 mmHg when compared to blood pressure measured using standard brachial cuff, especially those with elevated blood pressure [32, 33]. Mobile or mHealth is increasingly suggested as a solution to poor blood pressure awareness and control. At present, most adults now own smartphones equipped with cameras and light/motion sensors. Concerns remain with this type of technology, where data safety is uncertain, and the information provided may be misleading or inaccurate [34]. With improvements in sensor technology, we have the ability to acquire high-­ fidelity pulse waveform signals from extremities to obtain various physiological parameters. These have led to development of several physiological parameters that correlate with blood pressure [35]. Pulse transit time (the amount of time from left ventricular contraction to pulse waveform acquisition at the distal site) is being looked at as an independent variable to predict blood pressure [36]. Central, not brachial, pulse wave velocity has long been known to be a predictor of vascular stiffness, and better than the brachial blood pressure [37]. Pulse tonometry (counter pressure to measure arterial distension) and vascular unloading (measuring counter pressure required to maintain constant finger blood volume) are two other techniques being attempted to get continuous non-invasive cuffless measurements. The challenge faced by most of these technologies has been calibration of local pressure measurement site. With the present technology, it is possible to implement these pulse-based systems that accurately predict the trends of blood pressure instead of measuring pressure values itself. Considerable fluctuations in this trend can be used

28

A. Adji and M. F. O’Rourke

as a warning signal for users to monitor their blood pressure and continuous monitoring of this variation can be helpful in clinical management of certain cases. With the advancing technology, trends will continue in development of cuffless blood pressure monitoring. Devices that can monitor blood pressure continuously and provide additional and comprehensive hemodynamic information will allow us to establish better correlations with clinical outcomes and open up new possibilities for cardiovascular and renal disease risk prediction. The importance of this initiative is well worth investigating.

References 1. Pickering TG, Hall JE, Appel LJ, Falkner BE, Graves J, Hill MN, Jones DW, Kurtz T, Sheps SG, Roccella EJ. Subcommittee of Professional and Public Education of the American Heart Association Council on High Blood Pressure Research. Recommendations for blood pressure measurement in humans and experimental animals: part 1: blood pressure measurement in humans: a statement for professionals from the Subcommittee of Professional and Public Education of the American Heart Association Council on High Blood Pressure Research. Hypertension. 2005;45:142–61. 2. Townsend RR, Wilkinson IB, Schiffrin EL, Avolio AP, Chirinos JA, Cockcroft JR, Heffernan KS, Lakatta EG, McEniery CM, Mitchell GF, Najjar SS, Nichols WW, Urbina EM, Weber T.  American Heart Association Council on Hypertension. Recommendations for improving and standardizing vascular research on arterial stiffness: a Scientific Statement From the American Heart Association. Hypertension. 2015;66:698–722. 3. Booth J. A short history of blood pressure measurement. Proc R Soc Med. 1977;70:793–9. 4. Wakerlin GE.  From Bright toward light: the story of hypertension research. Circ Res. 1962;11:131–6. 5. Lawrence C. Physiological apparatus in the Wellcome museum: 1. The Marey sphygmograph. Med Hist. 1978;22:196–200. 6. O’Rourke MF. Frederick Akbar Mahomed. Hypertension. 1992;19:212–7. 7. Fye WB. William Henry Broadbent. Clin Cardiol. 1990;13:62–4. 8. Lawrence C. Physiological apparatus in the Wellcome museum: 2. The Dudgeon sphygmograph and its descendants. Med Hist. 1979;23:96–101. 9. Ghasemzadeh N, Zafari AM. A brief journey into the history of the arterial pulse. Cardiol Res Pract. 2011;2011:164832. 10. Nichols WW, O’Rourke MF, Vlachopoulos C.  McDonald’s blood flow in arteries. 6th ed. London: Arnold Hodder; 2011. 11. Lewis WH Jr. The evolution of clinical sphygmomanometry. Bull N Y Acad Med. 1941;17:871–81. 12. Fisher JW. The diagnostic value of the sphygmomanometer in examinations for life insurance. JAMA. 1914;58:1752–4. 13. Messerli FH, White WB, Staessen JA.  If only cardiologists did properly measure blood pressure. Blood pressure recordings in daily practice and clinical trials. J Am Coll Cardiol. 2002;40:2201–3. 14. Kaplan NM. Commentary on the sixth report of the Joint National Committee (JNC-6). Am J Hypertens. 1998;11:134–6. 15. Franklin SS, Gustin W, Wong ND, Larson MG, Weber MA, Kannel WB, Levy D. Hemodynamic patterns of age-related changes in blood pressure: the Framingham Heart Study. Circulation. 1997;96:308–15. 16. Staessen JA, Gasowski J, Wang JG, Thijs L, Den Hond E, Boissel JP, Coope J, Ekbom T, Gueyffier F, Liu L, Kerlikowske K, Pocock S, Fagard RH.  Risks of untreated and treated

3  A Historical Journey on the Physiology of Blood Pressure Monitoring

29

isolated systolic hypertension in the elderly: meta-analysis of outcome trials. Lancet. 2000;355:865–72. Erratum in: Lancet. 2001;357:724. 17. Cheng S, Xanthakis V, Sullivan LM, Vasan RS.  Blood pressure tracking over the adult life course: patterns and correlates in the Framingham Heart Study. Hypertension. 2012;60:1393–9. 18. Gurven M, Blackwell AD, Rodríguez DE, Stieglitz J, Kaplan H. Does blood pressure inevitably rise with age?: longitudinal evidence among Forager-Horticulturalists. Hypertension. 2012;60:25–33. 19. O’Rourke MF.  From theory into practice: arterial haemodynamics in clinical hypertension. J Hypertens. 2002;20:1901–15. 20. Postel-Vinay P. A century of arterial hypertension 1896–1996. New Jersey: Wiley; 1996. 21. Kannel WB, Dawber TR, Kagan A, Revotskie N, Stokes IJ. Factors of risk in the development of coronary heart disease—six-year follow-up experience: the Framingham Study. Ann Intern Med. 1961;55:33–50. 22. Veterans Affairs Cooperative Study Group. Effects of treatment on morbidity in hypertension: results in patients with diastolic blood pressures averaging 115 through 129 mmHg. JAMA. 1967;202:1028–34. 23. Veterans Affairs Cooperative Study Group. Effects morbidity of treatment on in hypertension: II. Results in patients with diastolic blood pressure averaging 90 through 114 mmHg. JAMA. 1970;213:1143–52. 24. Kannel WB, Gordon T, Schwartz MJ. Systolic versus diastolic blood pressure and risk of coronary heart disease: the Framingham study. Am J Cardiol. 1971;27:335–46. 25. Greenland P, Peterson ED, Gaziano JM. Progress against cardiovascular disease: putting the pieces together. JAMA. 2014;312:1979–80. 26. Systolic Hypertension in Elderly Program Trial. Prevention of stroke by antihypertensive drug treatment in older persons with isolated systolic hypertension: final results of the systolic hypertension in the elderly program (SHEP). JAMA. 1991;265:3255–64. 27. Franklin SS, Khan SA, Wong ND, Larson MG, Levy D. Is pulse pressure useful in predicting risk for coronary heart disease?: the Framingham Heart Study. Circulation. 1999;100:354–60. 28. Franklin SS, Lopez VA, Wong ND, Mitchell GF, Larson MG, Vasan RS, Levy D. Single versus combined blood pressure components and risk for cardiovascular disease: the Framingham Heart Study. Circulation. 2009;119:243–50. 29. Williams B, Mancia G, Spiering W, Agabiti Rosei E, Azizi M, Burnier M, Clement DL, Coca A, de Simone G, Dominiczak A, Kahan T, Mahfoud F, Redon J, Ruilope L, Zanchetti A, Kerins M, Kjeldsen SE, Kreutz R, Laurent S, Lip GYH, McManus R, Narkiewicz K, Ruschitzka F, Schmieder RE, Shlyakhto E, Tsioufis C, Aboyans V, Desormais I; ESC Scientific Document Group. 2018 ESC/ESH Guidelines for the management of arterial hypertension. Eur Heart J 2018;39:3021–3104. 30. Roman MJ, Devereux RB. Association of central and peripheral blood pressures with intermediate cardiovascular phenotypes. Hypertension. 2014;63:1148–53. 31. Whelton PK, Carey RM, Aronow WS, Casey DE Jr, Collins KJ, Dennison Himmelfarb C, DePalma SM, Gidding S, Jamerson KA, Jones DW, MacLaughlin EJ, Muntner P, Ovbiagele B, Smith SC Jr, Spencer CC, Stafford RS, Taler SJ, Thomas RJ, Williams KA Sr, Williamson JD, Wright JT Jr. 2017 ACC/AHA/AAPA/ABC/ACPM/AGS/APhA/ASH /ASPC/NMA/PCNA Guideline for the prevention, detection, evaluation, and management of high blood pressure in adults: a Report of the American College of Cardiology/American Heart Association Task Force on Clinical Practice Guidelines. Hypertension. 2018;71:e13–e115. 32. Guggiari C, Bula C, Iglesias K, Waeber B. Measurement with an automated oscillometric wrist device with position sensor leads to lower values than measurements obtained with an automated oscillometric arm device from the same manufacturer in elderly persons. Blood Press Monit. 2014;19:32–7. 33. Pereira T, Correia C, Cardoso J. Novel methods for pulse wave velocity measurement. J Med Biol Eng. 2015;35:555–65. 34. Goldberg DM, Levy PD. New approaches to evaluating and monitoring blood pressure. Curr Hypertens Rep. 2016;18:49.

30

A. Adji and M. F. O’Rourke

35. Pandit JA, Batlle D. Snapshot hemodynamics and clinical outcomes in hypertension: precision in the measurements is key. Hypertension. 2016;67:270–1. 36. Mukkamala R, Hahn JO, Inan OT, Mestha LK, Kim CS, Töreyin H, Kyal S. Toward ubiquitous blood pressure monitoring via pulse transit time: theory and practice. IEEE Trans Biomed Eng. 2015;62:1879–901. 37. Sharma M, Barbosa K, Ho V, Griggs D, Ghirmai T, Krishnan SK, Hsiai TK, Chiao JC, Cao H. Cuff-less and continuous blood pressure monitoring: a methodological review. Technologies. 2017;5:21.

Chapter 4

The Definition and Architecture of Cuffless Blood Pressure Monitors Josep Solà

Abstract  This chapter introduces a systematic framework to navigate through the universe of cuffless blood pressure technologies. The chapter is organized in three parts. Initially, a glossary is provided to anchor the discussion and formalize the terms involved in cuffless blood pressure monitoring. Then, a classification of existing blood pressure monitoring technologies is proposed together with a formal definition of a cuffless blood pressure monitor. Finally, an generic system architecture is presented organizing all the elements that any cuffless blood pressure monitor shall contain. Keywords  Basic glossary of cuffless blood pressure monitoring · Definition of a cuffless blood pressure monitor · Architecture of a cuffless blood pressure monitor · Transducer layer · Processing layer · Initialization layer · Pulsatility sensor · Pulsatility-based algorithm · Initialization algorithm · Calibration parameters

Scope After covering the clinical and physiological aspects of cuffless blood pressure technologies, this chapter introduces the generic architecture of a cuffless blood pressure monitor. This architecture is intended to establish a systematic framework that will help the reader to navigate through the jungle of cuffless blood pressure technologies, and that will support the comparison of legacy and novel approaches. The chapter is organized as follows: initially a basic glossary to understand cuffless blood pressure technologies is provided. Then, a definition of cuffless blood pressure monitors is proposed, followed by the generic architecture of the elements that any cuffless blood pressure monitor shall present. Finally, illustrative examples of implementation of cuffless monitors are briefly enumerated.

J. Solà (*) Aktiia SA, Neuchâtel, Switzerland e-mail: [email protected] © Springer Nature Switzerland AG 2019 J. Solà, R. Delgado-Gonzalo (eds.), The Handbook of Cuffless Blood Pressure Monitoring, https://doi.org/10.1007/978-3-030-24701-0_4

31

32

J. Solà

Terms and Definitions Over the last several decades, the community of clinicians, scientists, engineers, manufacturers and users of cuffless blood pressure devices has incrementally introduced a list of terms in the field that describe the different units and functions of a cuffless blood pressure monitor. The goal of this section is to summarize the most relevant items in a compiled list of physiological terms (Table 4.1) and technological terms (Table 4.2) adopted in the field of blood pressure monitoring. The sorting of terms is intentionally not alphabetical, with the goal of incrementally educating the reader.

Table 4.1  Physiological terms adopted in the field of cuffless blood pressure monitoring Term Arterial-Blood Pressure (BP)

Systolic Blood Pressure (SBP) Diastolic Blood Pressure (DBP) Pulse Pressure (PP) Central artery Peripheral artery Vasomotion Muscular artery Elastic artery Perfusion Perfusion Index (PI) (Arterial) Pressure wave Forward pressure wave Backward pressure wave

Definition The pressure within an artery, commonly expressed in mmHg or kPa. Because the blood pressure continuously evolves during a cardiac cycle it is typically summarized in two values: the maximum value (Systolic), and the minimum value (Diastolic) The maximum blood pressure value during a cardiac cycle The minimum blood pressure value during a cardiac cycle The difference between systolic and diastolic blood pressure values during a cardiac cycle An artery close to the heart, typically the ascending aorta and the descending aorta An artery further (downstream) away from the heart. The brachial artery is a peripheral artery The change of tone of an artery created by the contraction/relaxation of its smooth muscles An artery whose wall contains layers of smooth muscles, and that is affected by vasomotion An artery whose wall contains few smooth muscles and that is not severely affected by vasomotion. A central artery is an elastic artery The delivery of (arterial) blood into an organ or tissue A quantified measurement of pulsatility, typically used in pulse oximetry A pressure impulse generated at the heart during the opening of the aortic valve that propagates along the entire arterial tree dilating the walls of each artery. An arterial pressure wave does not indicate the movement of blood An arterial pressure wave traveling from the heart towards the periphery

An arterial pressure wave traveling from the periphery towards the heart. A backward pressure wave is generated when a forward pressure wave is reflected at a given reflection site (Arterial) Pressure The time series of the superposition of the forward and backward pressure waveform waves at a given location of the arterial tree during one cardiac cycle Mean Arterial Blood The average of the arterial blood pressure waveform during a cardiac Pressure (MAP) cycle (continued)

4  The Definition and Architecture of Cuffless Blood Pressure Monitors

33

Table 4.1 (continued) Term Pulsatility

Definition The ability of an artery (or arterial bed) to pulsate, that is, to change its characteristics (e.g., diameter) at the arrival of an arterial pressure wave Pulsatility waveform The time series of the pulsatility of an artery (or arterial bed) during one cardiac cycle Blood velocity The velocity at which blood circulates within an artery. While blood velocity in central arteries is in the range of 0.2–0.4 m/s, in peripheral arteries it is decreased Pulse Wave Velocity The velocity at which an arterial pressure wave propagates along the (PWV) walls of the arterial tree. While PWV in a central artery is in the range of 5–15 m/s, in peripheral arteries it is higher Central pulse wave The pulse wave velocity of a central artery velocity Peripheral pulse The pulse wave velocity of a peripheral artery or a peripheral segment of wave velocity the arterial tree Local pulse wave The pulse wave velocity measured punctually on a given arterial site, or velocity along a homogenous segment of the arterial tree (along one single artery) Regional pulse wave The pulse wave velocity measured along a heterogeneous segment of the velocity arterial tree (along a concatenation of arteries) Pulse Transit Time The time that an arterial pressure wave requires to propagate along the (PTT) walls of a given segment of the arterial tree Pulse Arrival Time The time at which an arterial pressure wave arrives at a certain point of (PAT) the arterial tree. Typically, PAT measurements consider as initial time reference the R-Wave of an electro-cardiogram Pre-ejection period The time lapse between the R-wave of an electro-cardiogram and the (PEP) opening of the aortic valve Table 4.2  Technological terms adopted in the field of cuffless blood pressure monitoring Term Blood Pressure Monitor (BPM) Invasive blood pressure monitor Noninvasive blood pressure monitor Arterial-occlusion blood pressure monitor Cuff-based blood pressure monitor Full occlusion blood pressure monitor Semi occlusion No arterial-occlusion blood pressure monitor

Definition A device to determine blood pressure A blood pressure monitor that requires the puncture of an artery in order to directly measure intra-arterial pressure A blood pressure monitor that does not require the puncture of an artery A noninvasive blood pressure monitor that requires the partial or complete occlusion of an artery An arterial-occlusion blood pressure monitor that requires the inflation of a cuff around a limb An arterial-occlusion blood pressure monitor whose pressure exceeds the systolic blood pressure leading to complete arterial occlusion An arterial-occlusion blood pressure monitor whose pressure does not exceed the systolic blood pressure A blood pressure monitor that does not require the occlusion of an artery (continued)

34

J. Solà

Table 4.2 (continued) Term Cuffless blood pressure monitor Continuous blood pressure monitor Beat-to-beat blood pressure monitor Intermittent blood pressure monitor Manual blood pressure monitor

Definition A no arterial-occlusion blood pressure monitor that does not require the inflation of a cuff around a limb A blood pressure monitor that determines the entire pressure waveform

A blood pressure monitor that determines one systolic and a diastolic value at each cardiac cycle A blood pressure monitor that determines one systolic and one diastolic value over a series of cardiac cycles A blood pressure monitor that requires the intervention of a skilled operator in order to perform a measurement. The most common manual blood pressure technique used in practice is auscultation Automated blood A blood pressure monitor that does not require the intervention of a pressure monitor skilled operator in order to perform a measurement Arterial line (A-line) An invasive blood pressure monitoring method that requires insertion of a catheter into an artery via an arterial puncture Auscultation A noninvasive, cuff-based, intermittent manual determination of blood pressure that requires the expert operation of a cuff and a stethoscope. During the manual deflation of the cuff an operator identifies the presence Korotkoff sounds using the stethoscope put on top of the measured artery Oscillometry A noninvasive, cuff-based, intermittent automated determination of blood pressure that uses an electronic-driven pneumatic cuff and requires no particular intervention from an operator. Oscillometric devices are typically presented in the form of upper-arm or wrist cuffs Office Blood Pressure A manual or automated blood pressure measurement performed at a physician’s office or in clinics. Auscultation and oscillometric monitors Measurement are typically used (OBPM) Home Blood Pressure An automated blood pressure measurement performed by a patient at home under controlled conditions. The measurement is typically Measurement triggered by the patient at rest. Oscillometric monitors are typically used (HBPM) An automated blood pressure measurement performed on a patient Ambulatory Blood under uncontrolled conditions. The measurement is typically Pressure automatically triggered at 15–30 min intervals during the day and the Measurement night without any intervention from the patient. Oscillometric monitors (ABPM) are typically used Volume clamp A noninvasive, cuff-based, continuous automated determination of blood pressure that relies on the vascular unloading technique. Volume-clamp monitors perform an optical measurement of arterial volume at a phalanx while continuously applying a counter pressure by means of a servo-controlled finger-cuff Tonometry A noninvasive continuous automated determination of blood pressure that consists on the application of a force to a superficial artery in order flatten its walls and to measure its pulsatility. Tonometric monitors rely on different sensing technologies to quantify force and arterial displacement Pulse wave analysis The analysis of a pressure waveform in order to extract cardiovascular (PWA) information. PWA can also be referred to as pulse wave decomposition (PWD) or pulse contour analysis (PCA) (continued)

4  The Definition and Architecture of Cuffless Blood Pressure Monitors

35

Table 4.2 (continued) Term Initialization

Calibration function/ parameters Reinitialization

Definition The process of determining subject- or condition-specific parameters that are required to determine a blood pressure value in a blood pressure monitor. Initialization is typically required by cuffless blood pressure monitors, and involves the use of a cuff-based monitor The set of subject- or condition-specific parameters that are required to initialize the blood pressure monitor The process of updating the calibration function/parameters of a blood pressure monitor

Defining the Perimeter of Cuffless Blood Pressure Monitors While reading the existing literature and commercial claims around cuffless monitoring, it can be unclear what a cuffless blood pressure monitor actually is. The goal of this section is to provide a formal definition of a cuffless blood pressure monitor, and to determine its perimeter within the large catalog of existing blood pressure technologies. The arena of the most relevant blood pressure technologies in use today is categorized by the illustration in Fig. 4.1. The proposed classification starts, at its root, by splitting blood pressure monitors between invasive and noninvasive monitors. Invasive Blood Pressure monitors require the puncture of an artery (by means of an arterial catheter) and are out of scope of this book. The reader can find more information in [1]. Focusing on the family of noninvasive blood pressure monitors (monitors that do not require an arterial puncture), the proposed classification further splits between monitors that measure a physical pressure on the body and monitors that estimate blood pressure without applying any pressure to the body. Arterial-occlusion Blood Pressure monitors typically require the use of a cuff (or an actuator) that applies a certain pressure to the body in order to perform a measurement. These monitors can further be differentiated between those that create a prolonged supra-systolic occlusion of an artery (including manual auscultation techniques, and automated oscillometric techniques) and those that create only partial or an intermittent occlusion of an artery (including volume-clamp techniques, and tonometric techniques). These monitors are also out of scope of the current book. The reader can find more information in [1]. The perimeter of cuffless blood pressure monitors commences at the second branch of noninvasive monitors: a cuffless blood pressure monitor is defined as a device or a technology that noninvasively determines the blood pressure of an individual without creating any arterial occlusion. Furthermore, within the vast family of cuffless monitors one can differentiate between two categories:

Non-Invasive

Pulse Wave Analysis

Pulse Wave Velocity

Semiocclusion

Full-occlusion

Cuffless Blood Pressure monitors

No arterial occlusion

Arterial occlusion

A-line

Machine Learning

Feature-based

Local

Regional

Tonometry

Volume-clamp

Oscillometry

Auscultation

Peripheral Pulse Transit Time

Central Pulse Transit Time

Pulse Transit Time

Pulse Arrival Time

Fig. 4.1  Visual classification of the most relevant blood pressure technologies, with a proposed delimitation of the perimeter of Cuffless Blood Pressure monitors

Blood Pressure monitors

Invasive

36 J. Solà

4  The Definition and Architecture of Cuffless Blood Pressure Monitors

37

• Monitors based on the principle of pulse wave velocity (PWV), measuring local or regional PWV values by means of at least two pulsatility sensors and/or additional cardiovascular sensors. This family includes pulse arrival time (PAT) techniques, covered by Chap. 5 of this book, and pulse transit time (PTT) techniques, covered by Chap. 6 of this book. • Monitors based on the principle of pulse wave analysis (PWA), performing the analysis of pressure waveform by means of one single pulsatility sensor. This family includes Feature-based techniques, covered by Chap. 7 of this book, and machine learning-based techniques, covered by Chap. 8 of this book.

The Architecture of a Cuffless Blood Pressure Monitor Following the formal identification of what attributes define a cuffless blood pressure monitor, a generic architecture of the components of a cuffless solution is further proposed. The main characteristic of a cuffless blood pressure monitor is that it creates no arterial occlusion during the process of determining a blood pressure value. In practical terms, this means that a cuffless monitor does not apply any force to the body. The monitor simply extracts information from the user’s skin surface that is further processed to determine a blood pressure value. As explained later, most cuffless monitors happen to rely on information related to arterial pulsatility (the change of diameter of an artery or arterial bed occurring during a cardiac cycle). In some cases, a cuffless monitor might still apply some pressure to the skin surface in order to obtain reliable readings. However, this exerted pressure is related to the need of accurately measuring pulsatility signals, rather than related to the need of creating a physical interaction between a sensor and the arterial bed underlying the skin surface. The lack of applied force to the body leverages advantages and creates limitations. On one side, the obvious advantages of a cuffless monitor compared to an arterial-occlusion monitor are an increase in user comfort and the possibility of performing longterm continuous blood pressure monitoring. On the other side, the main disadvantage is that because of the lack of mechanical interaction between the monitor and the underlying arteries, the implemented sensor can only retrieve information related to the pulsatility of an artery, but cannot collect its actual pressure values. In practice, while arterial-occlusion monitors rely on pressure transducers to determine a physical blood pressure quantity, cuffless monitors rely on pulsatility sensors that have no access to physical pressure quantities. Consequently, while arterial-occlusion monitors are capable of outputting absolute blood pressure readings that are expressed in pressure units (kPa or mmHg), cuffless monitors are limited to output blood pressure readings that are expressed in non-pressure units (see Table 4.3). In order to prepare the introduction of a generic architecture for cuffless blood pressure monitors, Figs. 4.2 and 4.3 start by illustrating the typical architecture of invasive and arterial-occlusion monitors. The reader will note that both approaches rely on an actual pressure transducer in order extract a pressure waveform. For the

38

J. Solà

Table 4.3  Sensing principles and output units of different blood pressure monitoring technologies Blood pressure monitoring technology Invasive Noninvasive with arterial occlusion Noninvasive without arterial occlusion (cuffless)

Hydraulic pressure in catheter

Body

Sensing principle Pressure transducer Pressure transducer Pulsatility sensor

Pressure transducer

Output Blood pressure in mmHg Blood pressure in mmHg Blood pressure in non-pressure units, expressed in mmHg after an initialization procedure

Pressure waveform

Transducer layer

Peak detection algorithm Processing layer

Systolic BP Diastolic BP Output

Invasive blood pressure monitor

Fig. 4.2  Typical architecture of an invasive blood pressure monitor that relies on the puncture of an artery to measure a hydraulic pressure in a catheter. A pressure transducer generates than a pressure waveform that is further processed by an algorithm to generate an absolute blood pressure determination expressed in mmHg

Pneumatic pressure in cuff

Pressure actuator

Pressure transducer Body

Pressure command

Pressure waveform

Transducer layer

Pressurebased algorithm

Processing layer

Systolic BP Diastolic BP

Output

Non-invasive arterial occlusion blood pressure monitor

Fig. 4.3  Typical architecture of an arterial occlusion blood pressure monitor that relies on a pneumatic cuff (or another actuator) to generate an occlusion to the body and to measure a pressure. A pressure transducer generates then a pressure waveform that is further processed by an algorithm to generate an absolute blood pressure determination expressed in mmHg

4  The Definition and Architecture of Cuffless Blood Pressure Monitors

39

invasive monitors, the pressure transducer captures its signals directly from the hydraulic pressure that a catheter transmits from the interior of the artery (arterial puncture) towards the transducer. For the arterial-occlusion monitors, the pressure transducer captures its signals from the pneumatic pressure that a cuff transmits from the periphery of the artery towards the transducer (Fig.  4.3). In the case of tonometry blood pressure monitors, the pneumatic cuff is replaced by a pressure (or displacement) sensor, and the pressure transducer requires additional initialization information to translate force or displacement into pressure (typically an initialization value provided by an oscillometric blood pressure monitor). In both Figs. 4.2 and 4.3, the pressure waveform is further processed by an algorithm that determines absolute Systolic and Diastolic Blood Pressure values. The main feature of these approaches is that the pressure waveform is already expressed in pressure units, being either mmHg or kPa. As described previously, the common characteristic of all cuffless monitors is that they rely on pulsatility sensors instead of pressure transducers, and therefore the output of their processing layer is a blood pressure determination expressed in arbitrary units (not in mmHg). Because of this fundamental limitation, cuffless blood pressure monitors have historically introduced an additional processing layer that transforms the non-pressure (uncalibrated) blood pressure values into absolute blood pressure readings, by a so-called initialization procedure. With this information in hand, a generic architecture for a cuffless blood pressure monitor is introduced in Fig. 4.4. A detailed description of the different layers of the architecture is provided in the following section. Non-initialized BP Pulsatility energy

Pulsatility sensor

Pulsatility waveform

Pulsatilitybased algorithm Additional cardio-synchronous information

Body

Transducer layer

Processing layer

Initialization algorithm

Systolic BP Diastolic BP

Calibration Parameters

Initialization layer

Output

Non-invasive cuffless blood pressure monitor

Fig. 4.4  Generic architecture of a blood pressure monitor that requires no arterial occlusion, a so-called cuffless blood pressure monitor. The monitor first relies on a pulsatility sensor (transducer layer) that generates a pulsatility waveform expressed in arbitrary units. The pulsatility waveform contains information on the change in diameter of the artery occurring during at least one cardiac cycle. By analyzing the pulsatility waveform, a pulsatility-based algorithm (processing layer) generates uncalibrated determinations of blood pressure. Finally, an initialization algorithm translates the uncalibrated blood pressure values into absolute blood pressure determinations expressed in mmHg (initialization layer)

40

J. Solà

 xamples of Implementation of Cuffless Blood Pressure E Monitors A more detailed description of the generic architecture of Fig. 4.4 is provided in this section, including practical examples on what technologies are known to implement each layer of the architecture: 1. Transducer layer: At a given body location, and for each cardiac cycle, the arrival of an arterial pressure wave generates a change in diameter of the arterial bed (including arteries and arterioles underlying the skin surface) that further induces a general displacement of the surrounding materials and anatomical structures. This so-called pulsatility energy can be captured from the skin surface at most body locations by a large variety of sensor modalities, as depicted in Table 4.4. The common feature of any pulsatility sensor is thus that it generates an electrical signal that codes a pulsatility waveform: a waveform which contour describes the change of diameter of the underlying arterial bed. 2. Processing layer: This second layer of the architecture analyzes the pulsatility waveform generated by the transducer layer in order to generate a non-initialized blood pressure determination. Because the pulsatility waveform is generated by the change in diameter of the arterial bed, and because the change in diameter is generated by the arrival of a pressure wave at the arterial bed, pulsatility signals are known to contain relevant information on the pressure waveform in time, amplitude, and frequency domains. This information has been shown to be useful to generate estimates of blood pressure, and/or estimates of changes of blood pressure. Accordingly, Table 4.5 summarizes the most common strategies in use to process pulsatility waveforms in order to extract blood pressure-related information. Note that some pulsatility-based algorithms require the acquisition of additional cardio-synchronous signals in order to perform denoising tasks, or in order to introduce into the calculation information on triggering events, such as Table 4.4  Examples of sensing modalities implemented in the transducer layer of cuffless blood pressure monitors Sensor technology Ballistocardiography (BCG) Electrical impedance tomography (EIT) Impedance cardiography (ICG) Implantable photoplethysmography Phonocardiography (PCG) Photoplethysmography (PPG) Radar Seismocardiography (SCG) Tonometry Ultrasound (US) Video plethysmography (VPG)

Described in this book In Chaps. 5, 6, and 10 Not covered, see Sola [2] In Chaps. 5, 6, and 10 Not covered, see Theodor [3] In Chap. 5 In Chaps. 5, 6, and 8–10 Not covered, see Solberg [4] In Chap. 5 In Chaps. 5, 6, and 8 In Chap. 10 In Chaps. 6 and 8

4  The Definition and Architecture of Cuffless Blood Pressure Monitors Table 4.5  Examples of pulsatility-based algorithms and additional cardio-­ synchronous signals implemented in the processing layer of cuffless blood pressure monitors

Table 4.6  Examples of algorithms and parameters implemented in the initialization layer of cuffless blood pressure monitors

Pulsatility-based algorithms Regional PAT Regional PTT Local central PTT Local peripheral PTT Feature-based PWA Machine Learning-based PWA Additional cardio-synchronous information Electrocardiography (ECG) Impedance cardiography (ICG) Phonocardiography (PCG)

Initialization algorithm Parametric algorithm based on theoretical models Parametric algorithm based on empirical models Calibration parameters Person-specific Population-based Hybrid parameters

41

Described in this book In Chaps. 5, 9, and 10 In Chaps. 6, 9, and 10 In Chaps. 7 and 8 In Chaps. 6 and 9 In Chaps. 7, 8, and 10 In Chaps. 6–9 Described in this book In Chaps. 5–7, 9, and 10 In Chaps. 5, 6, and 10 In Chaps. 5 and 6

Described in this book In Chaps. 5, 6, and 8–10

In Chaps. 5, 6, and 8–10

Described in this book In Chaps. 6 and 8–10 In Chaps. 6 and 8–10 In Chaps. 6 and 8–10

the opening of the aortic valve. The output of the processing layer of a cuffless blood pressure monitor is typically a quantity expressed in non-pressure units such as milliseconds, millivolts or Hz that correlates to blood pressure. In some use-cases this output might be directly exploited to monitor blood pressure trends, for instance to perform assessments on blood pressure variability. However, a further processing layer is required to generate determinations of systolic and diastolic blood pressure expressed in mmHg or kPa. 3 . Initialization layer: This final layer of a cuffless architecture is intended to transform the non-initialized blood pressure determinations (expressed in non-­ pressure units) into systolic and diastolic blood pressure values (expressed in mmHg or kPa) that can be interpreted according to established clinical practices. The additional information required to perform such initialization and the periodicity of required initialization procedures largely vary according to the implemented transducer and processing layers. Table  4.6 summarizes typical implementations of this final layer.

42

J. Solà

Introduction to the Technical Chapters of this Book The presented glossary and architecture support the technical chapters of this book. In the following pages, the reader will find a review of the basics and the most recent advances in the cuffless techniques of pulse arrival time (Chap. 5), pulse wave velocity (Chap. 6), pulse wave decomposition (Chap. 7), pulse wave analysis (Chap. 8), and machine learning developments (Chap. 9). Further, this book presents a detailed analysis of existing initialization techniques (Chap. 10) and the current worldwide regulatory framework (Chap. 11). Finally, relevant aspects of clinical trials for cuffless blood pressure monitors are presented (Chap. 12), concluded by the expected next steps in the validation and implementation of cuffless technologies in the field (Chap. 13). Acknowledgements  The author wants to specially thank Bastien Di Marco and Elisa Oliver for the preparation of the infographics and the material of this chapter, Prof. Ramakrishna Mukkamala and Prof. George Stergiou for the inputs on the terms and definitions, and the entire Aktiia team for the revision work.

References 1. Geddes LA. Handbook of blood pressure measurement. Clifton: Humana Press; 1991. p. 1991. 2. Sola J.  Continuous non-invasive blood pressure estimation, PhD Thesis No. 20093. Zurich: ETHZ; 2011. 3. Theodor M.  Subcutaneous blood pressure monitoring with an implantable optical sensor. Biomed Microdevices. 2013;15(5):811–20. https://doi.org/10.1007 4. Solberg LE. Radar based central blood pressure estimation, PhD Thesis. Oslo: University of Oslo; 2015.

Chapter 5

Pulse Arrival Time Techniques Marshal S. Dhillon and Matthew J. Banet

Abstract  For over 90 years, researchers and clinicians have worked on systems to noninvasively measure pulse wave velocity, with the ultimate goal of measuring continuous, cuffless blood-pressure. The design of multisensor systems to make two or more measurements of the pulse wave along the human arterial tree has become the hallmark of these efforts. Pulse arrival time measurements are traditionally made by detecting a first fiducial point on the electrocardiogram (ECG) waveform, and second fiducial point on a distal pulsatile waveform, such as the photoplethysmogram (PPG) or impedance cardiogram (ICG). By looking at the time difference between these two fiducial points, an approximation for pulse wave velocity can be calculated. Some other challenges exist in transforming the pulse arrival times into absolute blood pressure, but these have been largely overcome, as evidenced by the clinical-grade systems that are commercially available in the marketplace today. Keywords  Pulse arrival time · Pulse transit time · Pulse wave velocity · Blood pressure · Continuous · Wearable · Cuffless · Noninvasive

Introduction Methods for measuring physiological parameters using pulse arrival time (PAT) and pulse transit time (PTT) have been researched and discussed for over 90 years. The majority of work done in this time period has focused on PAT as a basis for a continuous, noninvasive blood pressure measurement. Efforts in this area have greatly increased in the last 15 years, mostly because of technology advances (e.g., miniaturization, cost reduction of components, module consolidation, and wearable computing power) coupled with the unmet need for diagnosis and monitoring of hypertensive patients worldwide.

M. S. Dhillon · M. J. Banet (*) toSense, Inc., San Diego, CA, USA e-mail: [email protected] © Springer Nature Switzerland AG 2019 J. Solà, R. Delgado-Gonzalo (eds.), The Handbook of Cuffless Blood Pressure Monitoring, https://doi.org/10.1007/978-3-030-24701-0_5

43

44

M. S. Dhillon and M. J. Banet

Definitions PAT is generally defined as the time delay between the electrical activity of the heart and a peripheral pulse measured further down the arterial tree. Although the nomenclature is not consistent throughout the literature, PAT requires the measure of the electrical activity of the heart (typically using an electrocardiogram (ECG)) and some measure of a mechanical activity of the pulse wave (typically using a photoplethysmogram (PPG)). In contrast, PTT typically relies on two mechanical measurements of pulse wave activity, rather than one mechanical (e.g., PPG) and one electrical (e.g., ECG), as used in PAT measurements. Despite these differences, both techniques attempt to accurately calculate pulse wave velocity (PWV) within an individual patient, as this parameter has a strong relationship with blood pressure.

Sensing Elements To measure PAT, most methods utilize a fiducial from the QRS complex of the ECG waveform (Q-point or R-point), and the onset of a pulse in the PPG waveform, as shown in Fig. 5.1:

Fig. 5.1  ECG-PPG plot of PAT

5  Pulse Arrival Time Techniques

45

Transmissive and Reflective PPG Sensors PPG sensors use an optical emitter at a specific wavelength and one or more corresponding photodetectors to measure optical energy which is modulated by the underlying vasculature. PPG sensors can be classified into two separate measurement configurations: transmissive and reflective. Transmissive sensors have the emitter(s) on one side of the tissue to be interrogated, and the photodetector(s) on the opposite side of the tissue. In this configuration, the emitted light is transmitted through the tissue and modulated by the underlying vasculature; the modulated optical energy is then detected at the other side, as shown in Fig. 5.2. This configuration is common in pulse oximeters designed to measure fingers and ear lobes. Reflective sensors have the emitter(s) and the photodetector(s) on the same side of the tissue, usually on the same plane. In this configuration, the emitted light penetrates into the tissue, and some portion is modulated by the tissue and reflected back to the photodetector(s), as shown in Fig. 5.3. Reflective sensors are common for optical measurements on the forehead, arm, wrist, chest, and other areas where a transmissive sensor is impractical due to the length of the optical path through the tissue. Transmissive sensors provide a stronger pulsatile signal than reflective sensors, often on the order of 10× to 100×. With either configuration, the light sensed by the photodetector in a PPG sensor contains a mixture of static (DC) and dynamic (AC) components. The static component makes up the majority of the signal amplitude (>95% depending on the sensor and measurement site), and is modulated by skin, muscle, fat, non-pulsatile blood, and other tissue. The dynamic component is modulated of the heartbeat-induced volumetric change in the vasculature in the light path with each cardiac cycle, which yields a pulsatile signal [1].

Fig. 5.2 Transmissive optical sensor

Fig. 5.3  Reflective optical sensor

46

M. S. Dhillon and M. J. Banet

While many emitter wavelengths have been researched for systems that measure PAT and PTT, the most popular wavelengths are infrared (~940 nm), red (~660 nm), and green (~530 nm). Infrared (IR) and red are used in pulse oximetry sensors, and light sources that generate them are readily available for experimentation and system construction. Infrared is preferred over red due to its greater penetration depth, and lack of variation with changes in blood oxygen content. The penetration depth of the optical sensor is a subtle but important point, as greater penetration depth often yields PPG pulses that are representative of larger artery blood flow, as opposed to capillary or arteriole blood flow. Green emitters are used mainly in reflective-mode PPG sensors, where they have been shown to yield stronger pulsatile waveforms than an IR or red wavelength. This is due to the shorter penetration depth of the green wavelength, which provides a stronger measurement of pulsatile blood flow near the surface of the skin. While this is sufficient for measurements of pulse rate, the small artery pulsatile signals are subject to morphology and timing changes due to vasoconstriction and vasodilation (e.g., as a result of changes in skin temperature), and can introduce error into a PAT-based blood pressure measurement. Another consideration for PPG sensors is the applied pressure of the sensing optics onto the skin of the subject, as this can have a dramatic effect on the morphology of the measured pulse. Increasing the applied pressure, up to the level of the subject’s mean arterial blood pressure, can increase the measured pulse amplitude [2]. This is due to the principle of maximal vessel compliance during zero transmural pressure. However, this increase in amplitude can distort the timing information within each PPG pulse, as well as cause discomfort to the patient, so it should be approached with caution. In general, applied pressure is a bigger concern for transmissive PPG sensors, where a wrap or a spring-loaded housing can be used to impart large amounts of force on the tissue being measured.

Impedance Plethysmography and Impedance Cardiology Impedance plethysmography (IPG) and impedance cardiography (ICG) are other well-researched methodologies for measuring distal pulsatile waveforms in a PAT-­ measurement system. In this case, PAT is calculated as the time delay between the QRS complex of the ECG waveform, and the zero crossing (also referred to as “point B”) of the IPG/ICG waveform [3], as shown in Fig. 5.4. IPG/ICG is the central technology in many commercially available bioimpedance systems that noninvasively measure thoracic impedance (a proxy for body fluids), and hemodynamic properties such as cardiac output, stroke volume, and systemic vascular resistance. IPG/ICG systems traditionally use a multi-electrode system (as shown in Fig. 5.5) where a high-frequency, low-amperage electrical current is applied to the subject’s body through dedicated electrodes (impedance drive electrodes), and the modulated signal is measured through the remaining electrodes (impedance sense electrodes). These systems measure the impedance of the underlying tissue, and typically measure physiological activity at a greater depth than PPG sensors [4].

5  Pulse Arrival Time Techniques

47

Fig. 5.4  ECG-IPG plot of PAT

Fig. 5.5  IPG multi-electrode diagram

The applied electrical current is conducted by the various tissues and organs in the path of the impedance drive electrodes. Similar to PPG waveforms, the measured IPG/ICG waveform features a static (DC) component, and a dynamic (AC) component. The static component makes up the majority of the measured signal, and is affected by the amount of bone, muscle, fat, and other tissue in the measurement

48

M. S. Dhillon and M. J. Banet

path. The dynamic component is modulated by the pulsatile component of each cardiac cycle. Due to the low resistance of blood, the applied current is well conducted through the arterial vasculature between the impedance drive electrodes, and the resulting dynamic signal is a good indicator of the pulsatile blood volume. Since most IPG/ICG systems involve electrodes applied on the thorax, the dynamic IPG/ ICG component is also modulated by the respiration effort of the subject. The respiratory modulation is usually 10X-100X greater than the cardiac modulation of the IPG/ICG waveform, which can complicate the measurement of cardiac pulses. However, since the frequencies of interest for respiration and cardiac pulses have little overlap, they can be filtered out using analog or digital signal processing.

Other Sensing Technologies Other less common methods for measuring PAT include those using ballistocardiogram (BCG), seismocardiogram (SCG), and phonocardiogram (PCG) waveforms. BCG and SCG are both measurements of the vibration or recoil of the body due to the force of each heartbeat-induced pulse ejected from the left ventricle into the aorta. BCG is commonly measured using highly sensitive load sensors (e.g., load cells). These are present in passive devices like weight scales, chairs, and beds [5]. The SCG is commonly measured using accelerometers, typically placed on or near the sternum. Both measurements provide well-defined waveforms corresponding to each cardiac pulse, although their morphology is not typical of standard pulsatile waveforms, as shown in Fig. 5.6. The fiducial points are well described in literature and could provide useful for future research efforts.

Fig. 5.6  Typical BCG and SCG waveforms

5  Pulse Arrival Time Techniques

49

Fig. 5.7  S1/S2 heart sounds with ECG

PCG is a measurement of the acoustics of the cardiac cycle, and is essentially the output of a digital stethoscope. In a healthy heart, a PCG sensor measures the sounds of the mitral and tricuspid valves closing (S1) and the sounds of the pulmonary and aortic valves closing (S2). These are the classic “lub dub” sounds heard with each heartbeat through a stethoscope, as shown in Fig. 5.7. In other cases, the sounds S3 and S4, which are faint and slightly precede S1 and S2 respectively, can be measured with a PCG sensor. In some cases, the presence of these sounds can indicate cardiac abnormalities or congestive heart failure. A PCG sensor is also useful for identifying other sounds during the cardiac cycle which can indicate issues with heart valves (e.g., heart murmurs), or conditions like pericarditis. For PAT-based systems, the time delay between the fiducial of the ECG waveform and the S1/S2 markers can indicate how much of the cardiac cycle is spent in systoli and diastoli, and could potentially be used to identify changes in blood pressure over time.

Challenges of PAT-Based Blood Pressure Monitors Underlying Physiology Theoretical equations, such as Bramwell–Hill and Moens–Korteweg, describe the relationship between PWV (e.g., the speed at which a pulse wave leaves the left ventricle of the heart and travels down the arterial tree) to changes in blood pressure, with some simplified accounting for arterial compliance. PAT-based methods

50

M. S. Dhillon and M. J. Banet

attempt to utilize this relationship by providing a noninvasive, continuous measurement of PWV in order to calculate blood pressure. However, classic PAT, taken in isolation, presents some challenges to comprehensive blood pressure measurement, which are described below.

The Effect of Pre-ejection Period The goal of most PAT-based systems is to measure and calculate a value that represents the PWV for a given cardiac pulse, and to relate that measured value to the subject’s blood pressure. However, the common starting point of the time measurement—a fiducial marker on the ECG—is not usually indicative of the time that the mechanical pulse wave starts travelling from the left ventricle into the aorta. More specifically, there is a delay from the electrical depolarization of the heart’s ventricles (indicated by the ECG’s QRS complex), and the mechanical activation of the ventricular muscle. This period is known as the electromechanical activation time (EMAT). After EMAT is completed and the ventricular muscle is activated, there is another delay where the ventricles have started contraction but blood has not started to eject into the aorta. This time period is known as the isovolumic contraction time (IVCT). Together, the EMAT and IVCT are often referred to as the pre-ejection period (PEP). PAT measurements include both PEP and the PTT of the cardiac wave.

PAT = PEP + PTT 

(5.1)

PEP = EMAT + IVCT

(5.2)

Another issue with the electromechanical delay is that it can change for various reasons, often in a very short time period. Vasoactive drugs, hydration status, fluid overload, venous return, and a patient’s posture can all cause changes to the electromechanical delay, which can add a variable amount of error to measurement of PAT [6, 7]. The most effective solutions to eliminate this source of error are methods that directly measure the electromechanical delay or use additional sensors to directly measure the PTT.  Such sensors, for example, lack ECG-measuring systems, and instead utilize sensors that measure waveforms such as PPG, IPG/ICG, BCG, SCG, and PCG.

Motion Artifacts Another challenge for PAT-measuring systems (and indeed all wearable systems) is motion artifacts. The magnitude of motion artifacts is based upon the sensor’s location and the activities undertaken by a subject during measurement. While it may not be realistic to have a system that measures perfectly during vigorous exercise, a

5  Pulse Arrival Time Techniques

51

reasonable goal is for a PAT-measuring system is to operate accurately during normal motions undertaken by patients in an ambulatory environment: brushing teeth, walking, using the restroom, etc. There are two levels to dealing with motion artifact. The first is accurately identifying motion-corrupted waveforms and not including them in the calculation of blood pressure values. The second is using an active measure of the corrupting motion (e.g., with an accelerometer or gyroscope), and then using adaptive noise cancellation techniques to remove the motion artifact from the corrupted waveforms so that an accurate measurement can be made. The ECG waveform is relatively motion-tolerant, and thus the biggest challenge related to motion artifacts for most PAT-measuring systems is from the sensor measuring the distal waveform (e.g., the PPG waveform). For systems that utilize an optical sensor on the wrist or finger, being able to ignore and possibly read through periods of motion artifact is essential to avoiding erroneous blood pressure values being calculated. Systems that utilize sensors on the thorax, the neck, or the head for the distal waveform are less affected but are still well served by having a solution for ameliorating motion artifacts. For PAT-measuring systems that utilize a distal waveform sensor on the arm, the wrist, or fingers, hydrostatic pressure error is another challenge that needs to be addressed. When the sensor measuring the distal waveform is raised above the level of the heart, the actual blood pressure at the sensor site is lower than the blood pressure at heart level. Conversely, when the sensor is placed below the level of the heart, the blood pressure at the sensor site is higher than the blood pressure at heart level. Both of these changes in blood pressure are due to the hydrostatic effects of gravity on the blood in the arterial tree. The change in PAT due to hydrostatic effects is significant enough that efforts have been made to use this effect for calibration purposes [8]. For PAT-measuring systems where hydrostatic error is a concern, an objective measure of the position of the distal waveform sensor (e.g., using accelerometers) is recommended.

 ositioning PAT-Based Monitors Among Conventional P Measurements of Blood Pressure There are many existing methods for measuring blood pressure, namely because blood pressure is an important clinical parameter and useful in many different patient situations. The following section provides a brief overview of the existing methods, and the advantages and disadvantages compared to a PAT-based monitor.

Auscultation Auscultation is the classic and most widely accepted method for measuring blood pressure. In addition to its use in clinical settings, auscultation is also accepted as a reference method for validating blood pressure-monitoring technologies based on

52

M. S. Dhillon and M. J. Banet

Image 5.1  Auscultation example

PAT and PTT by regulatory agencies. It is based on the principle of occluding blood flow in an artery with a manual cuff (usually the brachial artery), and then listening for sounds indicating the return of turbulent blood flow, and laminar blood flow (called Korotkoff sounds) that correspond to systolic and diastolic pressure respectively [9]. Auscultation is accurate, clinically accepted, noninvasive and relatively easy to perform; however, its measurements are not continuous, and can be uncomfortable (Image 5.1).

Oscillometry Oscillometry is a widely accepted technology for measuring blood pressure that is automated and relatively simple to deploy as compared to auscultation. Like auscultation, it is based on the principle of occluding blood flow with an inflatable cuff. Unlike auscultation, it measures the pulses transduced through the inflatable cuff, and exploits the fact that the pulse amplitude is modulated by the difference between mean arterial pressure (MAP) and the applied pressure in the cuff. The pulse amplitudes increase when the applied pressure is between the patient’s diastolic blood pressure (DBP) and MAP. The pulse amplitudes decrease when the applied pressure is between MAP and systolic blood pressure (SBP). Variation of pulse amplitudes with applied pressure yields a quasi-symmetric waveform from which SBP and DBP can be determined using empirical formulas [10]. Oscillometric devices are noninvasive, clinically accepted, easy to use, and readily available for patients to

5  Pulse Arrival Time Techniques

53

Image 5.2  Oscillometry example

measure the blood pressure at home; however, like auscultation, oscillometric devices are not continuous, can be uncomfortable, and can be inaccurate if used improperly or not thoroughly validated (Image 5.2).

Invasive Measurements Intra-arterial catheters (sometimes called “arterial lines”) are the standard of care for critically ill patients where continuous blood pressure monitoring is necessary. Arterial catheters are also recognized as a reliable reference method for validating new technologies by regulatory agencies. This method utilizes a pressure transducer incorporated with an invasive catheter in contact (through fluid) with a pulsating artery (e.g., radial artery) to provide continuous, direct measurement of blood pressure. Arterial catheters are accurate and clinically accepted, and provide continuous measurement; however, they are invasive and require expert users to place and monitor the catheter site, and patients must be relatively stationary during monitoring (Image 5.3).

Volume Clamp The volume clamp method, also known as vascular unloading or the method of Penaz, provides a noninvasive, continuous measurement of blood pressure by using a combination of sensors. In these systems, a finger cuff connected to a high-speed

54

M. S. Dhillon and M. J. Banet

Image 5.3  Arterial line example

servo pump is used to cancel out volumetric change of blood flow in the fingers with each cardiac cycle. The lack of volumetric change is verified by a finger-mounted PPG sensor, which senses a waveform lacking an AC component. The pressure applied through the high-speed servo is essentially the inverse of the arterial blood pressure waveform at the finger. Using an oscillometric calibration and empirical models, a continuous blood pressure waveform representative of larger artery waveforms can be provided. Volume clamp methods are noninvasive and continuous; however, they require a large, complex measurement apparatus. The finger cuff can be uncomfortable for patients, and the system can be inaccurate if used improperly (Image 5.4).

Tonometry Arterial tonometry provides a noninvasive, continuous measurement of blood pressure by applying a sensitive transducer directly above a pulsating artery located above a rigid surface (e.g., bone). The transducer needs to be placed directly above the pulsating artery (usually the radial artery), and pressure needs to be applied to flatten the surface of the artery. Changes in arterial pressure with each cardiac cycle are measured by the transducer, and can be analyzed to yield blood pressure. Arterial tonometry is noninvasive and continuous; however, it is very sensitive to motion and requires precise placement of the transducer, making it difficult to implement (Image 5.5).

5  Pulse Arrival Time Techniques

55

Image 5.4  Volume clamp example

Image 5.5  Tonometry example

PAT-Based Measurements It is clear that while many methods exist for measuring blood pressure, each method features nonideal aspects that need to be considered depending on the use case for the blood pressure measurement. Furthermore, none of the existing methods for noninvasive blood pressure measurement meet the needs of the growing hypertensive population, both in the clinic and at home. An ideal solution for these environments would be accurate, wearable, passive, continuous, and easily accessible. PAT-measuring techniques have the potential to satisfy all of these goals, which is a reason so much effort has been invested in developing them.

56

M. S. Dhillon and M. J. Banet

Devices and History Pulse wave velocity systems, using both PAT and PTT, have been studied regularly for over 90 years. One of the earliest experiments involving a pulse wave velocity system was conducted by Hickson and McSwiney in 1924 [11]. In this experiment, two mechanical pulses were simultaneously recorded from the carotid and radial arteries, using hot-wire sphygmographs. In this setup, a thin filament of wire is heated to a very high temperature, using a battery or other energy source, and then connected to a galvanometer. The hot wire is then placed in a funnel or tube on the skin, above the artery of interest. Small perturbations of air in the funnel, caused by each peripheral pulse, cause a slight cooling in the hot wire and a commensurate current into the galvanometer. The authors calculated PWV using the two pulse signals, and found a slowing of the pulse wave as each subject raised their arm, due to hydrostatic pressure effects. Another seminal experiment was conducted Thomas in 1955 [12]. In this experiment, PAT was calculated using the time delay between an ECG waveform, and a crystal microphone placed on the dorsalis pedis artery as a pulse transducer. Both waveforms were recorded and pulse wave velocity was manually calculated from the time delay, and shown to track changes in diastolic pressure. In 1964, Weltman, Sullivan, and Bredon conducted an experiment which was one of the first to utilize a computerized PWV measurement system [13]. They used a standard ECG measurement setup, combined with a crystal microphone to transduce a peripheral pulse. The computerized system used adjustable triggers to detect the ECG R-peak and peripheral pulse, and then output an analog voltage that represented the time difference of the two signals. An image of the computerized system is shown below in Image 5.6. There are a handful of commercially available pulse arrival time systems as well. Nihon Kohden has a series of multiparameter vital sign monitors that also measures PAT to enhance their NIBP oscillometric measurement [14]. Using PAT measured from the R-peak of the ECG waveform to the onset of the PPG pulse as measured by a fingertip oximeter, the vital sign monitor can determine when more NIBP measurements are needed for a patient, rather than only inflating the NIBP cuff at predetermined intervals. These systems from Nihon Kohden have been approved by Japan’s PMDA and other regulatory agencies. Sotera Wireless markets a body-worn multiparameter vital sign monitor, ViSi Mobile, that features a measurement of continuous, noninvasive blood pressure (cNIBP) using PAT [15]. ViSi Mobile has sensors deployed on a patient’s chest, upper arm, wrist, and fingers. It measures PAT using the time delay between the ECG measured at the chest and a pulsatile PPG waveform measured at the base of the thumb or finger. In addition, the system contains a network of accelerometers and impedance-based sensing to mitigate some of the challenges faced by conventional PAT techniques. ViSi Mobile has been cleared by the US FDA and conforms to the applicable governing standard (ISO 81060-2) for blood pressure accuracy.

5  Pulse Arrival Time Techniques

57

Image 5.6  Computerized PWV system from 1964 [13]

Somnomedics has a body-worn multiparameter vital sign monitor (Somnotouch NIBP, somnomedics.eu) that is similar to the ViSi Mobile from Sotera Wireless. The Somnotouch NIBP system measures pulse arrival time using the time delay between the ECG and a pulsatile PPG waveform measured at the fingertip, using a conventional pulse oximetry probe [16]. The Somnotouch NIBP system has published results from a clinical trial conducted using the European Society of Hypertension International Protocol to quantify the accuracy of the blood pressure measurement. Biopac has a modular system that can measure ECG and PPG waveforms, and then process the waveforms with a proprietary software package that automatically calculates PAT [17]. Biopac also provides a variety of reflective and transmissive PPG sensors, IPG/ICG sensors, and accelerometers for multimodal data acquisition. The Biopac system is for research purposes only and not cleared for use in patient monitoring. There are also a group of handheld or wrist-worn devices that incorporate a combination of PPG and ECG sensors, and provide some measurement of absolute or relative blood pressure. Examples of these devices include the CheckMe Pro [18], Care Up [19], and the H2-BP [20]. These devices have been not widely accepted clinically or commercially due to the lack of supporting data for the accuracy of their blood pressure claims. While these devices are on the leading edge of accessibility and usability, their slow adoption indicates the need for objective clinical results to support blood pressure measurement accuracy for PAT-based systems.

58

M. S. Dhillon and M. J. Banet

Future Directions The future of devices that calculate blood pressure using PAT and PTT lies in the continued miniaturization and wearability of the systems, the accuracy of their blood pressure algorithms, and their increasing accessibility for home and ambulatory use. Major electrical component vendors are reducing the size, cost, and power consumption for many of the building blocks required for the sensing components required by pulse arrival time systems. In addition, there continues to be an emergence of specialized integrated circuits, combining multiple components into a single package for biomedical sensing applications, that will lead to low-cost, effective systems. As wearable systems are able to measure more features related to changes in blood pressure, artificial intelligence techniques are becoming a leading method for combining and distilling the information into an accurate blood pressure measurement. This shift requires more computation than traditional models that use only a handful of features, and increases in the computing power and memory of wearable systems is needed. Many popular consumer watches and wearables have already incorporated ECG and PPG sensors, making these technologies accessible to large amounts of consumers outside of the hospital. Research efforts have also shown promise for modifying the world’s most popular consumer device, the smartphone, into a simple and accurate blood pressure monitor [21]. The prevalence of these consumer sensors, and smartphone modifications, coupled with the technology improvements listed above, will usher in the next-generation of pulse arrival time systems. These next-­ generation systems will provide a new level of passive and accurate blood pressure monitoring for use in both home and clinic environments.

References 1. Mukkamala R, Hahn JO, et al. Toward ubiquitous blood pressure monitoring via pulse transit time: theory and practice. IEEE Trans Biomed Eng. 2015;62(8):1879–901. 2. Teng XF, Zhang YT. The effect of applied sensor contact force on pulse transit time. Physiol Meas. 2006;27:675. 3. Bang S, et  al. A pulse transit time measurement method based on electrocardiography and bioimpedance. In: 2009 IEEE Biomedical Circuits and Systems Conference. 2009. p. 153–6. 4. Patterson R. Fundamentals of impedance cardiography. IEEE Eng Med Biol Mag. 1989;8:35–8. 5. Inan OT, et  al. Robust ballistocardiogram acquisition for home monitoring. Physiol Meas. 2009;30:169. 6. Payne RA, et al. Pulse transit time measured from the ECG: an unreliable marker of beat-to-­ beat blood pressure. J Appl Physiol. 2006;100:136–41. 7. Zhang G, et al. Pulse arrival time is not an adequate surrogate for pulse transit time as a marker of blood pressure. J Appl Physiol. 2011;111:1681–6. 8. McCombie DB, et al. Motion based adaptive calibration of pulse transit time measurements to arterial blood pressure for an autonomous, wearable blood pressure monitor. Conf Proc IEEE Eng Med Biol Soc. 2008;2008:989–92.

5  Pulse Arrival Time Techniques

59

9. Perloff D, et  al. Human blood pressure determination by sphygmomanometry. Circulation. 1993;88:2460–70. 10. Alpert BS, et al. Oscillometric blood pressure: a review for clinicians. J Am Soc Hypertens. 2014;8:930–8. 11. Hickson SK, McSwiney BA. The effect of variation in blood pressure on pulse wave velocity in the brachial artery in man. J Physiol. 1924;59:217–20. 12. Thomas JG. A method for continuously indicating blood pressure. J Physiol. 1955;129(3):75–6P. 13. Weltman G, Sullivan G, Bredon D. The continuous measurement of arterial pulse wave velocity. Med Electron Biol Eng. 1964;2:145–54. 14. “Innovative Technologies”, Nihon Kohden. https://ae.nihonkohden.com/en/innovativetechnologies/ pwtt. 15. “Why Visi”, Sotera Wireless. https://www.soterawireless.com/why-visi/. 16. “Somnotouch NIBP”, Somnomedics. https://somnomedics.eu/products/cardiology/24h-blood pressure-24h-ecg/somnotouch-nibp/. 17. “Pulse Transit Time (PTT), Pulse Wave Velocity (PWV), and Pulse Wave Amplitude (PWA)”, Biopac. https://www.biopac.com/application/plethysmography/advanced-feature/ pulse-transit-time-ptt-and-relative-bp-measurement/. 18. “CheckMe Pro”, Viatom. https://www.viatomtech.com/checkme-pro. 19. “Features”, Care Up. http://care-up.com/. 20. “H2-BP”, H2 Care. http://h2care.com/18. 21. Chandrasekhar A, Kim CS, Naji M, Natarajan K, Hahn JO, Mukkamala R. Smartphone-based blood pressure monitoring via the oscillometric finger-pressing method. Sci Transl Med. 2018;10(431):eaap8674.

Chapter 6

Pulse Wave Velocity Techniques Jim Li

Abstract  Recently, pulse wave velocity (PWV), or its reciprocal pulse transit time (PTT), has been intensively investigated as a promising technique for continuous, cuffless, and noninvasive blood pressure (BP) monitoring. BP is mathematically derived through PTT, or the “time delay” in propagation of pressure waves in the vascular system, which can be easily derived from two pulse signals, including electrocardiography (ECG) and pulse plethysmography (PPG) signals, together with adequate calibration procedure. Practical steps in applying this method as well as mathematical models in estimating BP were reviewed; while limitations of this approach, such as the need for individual calibration and the need for a reasonably stable condition were discussed. The future of this technology can be potentially used in, but not limited to, continuous BP monitoring, BP change tracker, and trigger for absolute BP measurement. Furthermore, with machine learning, the initially extract surrogate cardiovascular indexes from physiological signals can be used to train and adapt to the model to further improve the accuracy of BP prediction. Keywords  Blood pressure · Pulse wave velocity · Pulse transit time · Pulse arrival time · Electrocardiogram · Pulse plethysmography · Wearable · Cuffless · Calibration · Machine learning

History Overview Inspired by the invasive intra-arterial continuous blood pressure measurement through clinical cannulation in 1949, researchers started exploring noninvasive alternative for continuous blood pressure monitoring. In 1963, arterial tonometry was used by Pressman et al. by applanating a superficial artery against a bone with an external transducer until the artery is flattened, where the tangential arterial wall tension no longer affects the vertical force measured [1]. Quickly, arterial tonometry J. Li (*) Global Medical Affairs, Omron Healthcare, Inc., Lake Forest, IL, USA e-mail: [email protected] © Springer Nature Switzerland AG 2019 J. Solà, R. Delgado-Gonzalo (eds.), The Handbook of Cuffless Blood Pressure Monitoring, https://doi.org/10.1007/978-3-030-24701-0_6

61

62

J. Li

was improved and widely used by the researchers in the measurement of pulse wave velocity (PWV) for evaluation of arterial stiffness and pulse wave analysis. The commonly applied sites include the radial artery, carotid artery, and femoral artery. The validity of measuring arterial BP depends on the applanation of the artery, and therefore encounters practical problems of sensor positioning, motion artifacts, calibration, etc. To enhance usability, tonometry sensor array was used to ensure at least one sensor among multiple ones is precisely positioned to capture good pulse wave signal. AtCor SphygmoCor (AtCor Medical, Sydney, Australia) and Form/VP-2000 and HEM-9000AI (Omron Healthcare Co., Ltd, Kyoto, Japan) are example of commercialized research tools that utilizing arterial tonometry. These tools were mainly for pulse wave analysis, arterial stiffness, and central BP measurement. Even though they are capable of providing continuous pulse waves for short period, they are not a true noninvasive continuous BP monitor. Another development in noninvasive continuous BP monitoring field is the use of volume clamp on the finger technique proposed by Peñáz from Czech in 1973 [2]. The system has an inflatable finger cuff with a built-in photoplethysmograph (PPG) sensor, and a closed loop servo system to apply a pulsating cuff pressure to the finger arteries that is precisely opposite to the intra-arterial pressure. In equilibrium, when the cuff pressure equals the arterial pressure, the difference between the intra-arterial pressure and external applied cuff pressure will become zero, which is called the set point. The key to volume clamp method is to reach the set point. On the basis of the volume clamp method, some research tools were commercialized, including Finapres (Finapres Medical Systems, Enschede, The Netherlands) and CNSystems (CNSystems Medizintechnik GmbH, Graz, Austria). Despite being noninvasive continuous blood pressure measurement techniques, both tonometry and volume clamp techniques are intrusive because both require the application of external pressure or force on cuff during the entire course of the monitoring, which leads to discomfort and motion artifacts. Another approach, which is based using PWV or its reciprocal, pulse transit time (PTT), can be an attractive alternative for measuring unobtrusive continuous BP. In principle, PWV depends on the property of the arterial wall, which varies with the arterial pressure. The usage of PTT can be dated back to 1959 when Weltman et  al. designed the PWV computer by utilizing the ECG and a pulse signal to define pulse transit time over a known arterial length [3]. Recently, PTT became a popular way of achieving cuffless continuous BP measurement through using wearable objects including watches, rings, shirts, eyeglasses, smartphones, and cameras, as well as daily objects such as sleeping cushions, chairs, and weighing scales. Most of these methods are still in the research stage. The few commercially available devices include Sotera ViSi Mobile (Sotera Wireless, San Diego, USA)  continuous, noninvasive BP (cNIBP) monitoring and SOMNOtouchTM-­ NIBP (SOMNOmedics GmbH, Randersacker, Germany), which are based on the PTT method. The ViSi’s cNIBP is determined on a beat-to-beat basis employing PTT and calibration with automatic noninvasive BP method.

6  Pulse Wave Velocity Techniques

63

Physiological Description PWV is usually assessed using the arrival time of a pressure wave propagating through the arterial tree in a certain distance between the proximal and distal arterial sites, in the form of PWV = L/PTT, where L is the distance between the proximal and distal sites. Because of the complexity of distance measurement, PWV can be indirectly approximated using PTT, which can be easily derived from two pulse signals, including electrocardiography (ECG) and pulse plethysmography (PPG) signals (Fig. 6.1). By using a calibration procedure, the measured PTT can be translated into arterial pressure by using an appropriate model. Pulse transit time refers to the time it takes a pulse wave to travel between two arterial sites. Like pulse wave velocity, PTT is a measure for arterial stiffness. When blood pressure increases, the vascular tone increases and the arterial wall becomes stiffer, causing the PTT to shorten. Conversely, when blood pressure falls, vascular tone decreases and the arterial wall becomes less stiffer, and PTT increases. Increased stiffness can be either structural—due to aging and atherosclerosis, or functional—due to higher sympathetic activity or elevated blood pressure [4]. But at the same time, the multiple factors besides blood pressure that influences vascular stiffness mechanisms itself make the technique based on PWV or PTT sensitive but not specific, particularly for peripheral/muscular arteries, and less prominent in central/elastic arteries. Such influence presents a limitation of PWV or PTT for blood pressure estimation in real life.

Fig. 6.1  Pulse wave, pulse transit time (PTT), and pulse arrival time (PAT)

64

J. Li

Another relevant concept and parameter is the pulse arrival time (PAT), as described in the previous chapter, which measures the time difference between the R-peak of ECG and a characteristic point of pulse plethysmography (PPG) waveform. PAT is the sum of PTT of the pressure wave and the pre-ejection period (PEP) delay:

PAT = PTT + PEP

(6.1)

PEP is the time needed to convert the electrical signal into a mechanical pumping force and isovolumetric contraction to open the aortic valve, and can be calculated by the delay between R-wave and impedance cardiography (ICG), as shown in Fig. 6.1. PEP is a delay that changes with various factors such as stress, physical activity, age and emotion [5]. The popularity of using PAT to estimate blood pressure is based on the simplicity of obtaining PAT by referencing ECG R wave, which is precise and easy to get. The disadvantage is the introducing the new variable of PEP, especially when PTT is small when the distal measurement point is close to the core body. Studies showed that PEP accounts for 7% of the RR interval to approximately 20% of PTT measured at the finger tips at rest [6]. The impact of PEP decreases with distance from the heart, but becomes more significant when heart rate lowered, and therefore should be subtracted out to obtain PTT. It is generally thought that PTT has higher correlation with SBP, DBP, and MAP than PAT, but with some studies stating PAT is a better indicator of SBP due to its dependency on both ventricular contraction and vascular function. Using PTT to effectively estimate blood pressure needs to be handled in a reasonably stable condition, where smooth muscle contraction is minimal and viscous effects are negligible. Such conditions are best met by measuring PTT through central arteries. It also needs to be individually calibrated to associate the PTT to the absolute blood pressure values. Furthermore, aging or diseases can potentially change the arterial elasticity characteristics; therefore, periodic recalibration is necessary to ensure the close correlation between PTT and blood pressure values during long-term BP monitoring.

Practical Approach The practical PTT-based cuffless BP monitoring involves a three-step approach: Step 1—obtain the proximal and distal arterial waveforms; Step 2—calculate PTT from the waveforms, either from foot-to-foot, or peak-to-­ peak of the waveforms; Step 3—calibration of PTT (in units of ms) to BP (in units of mmHg) There are various challenges along the way during implementation of the above steps. First challenge is to reliably obtain good quality arterial waveforms in a

6  Pulse Wave Velocity Techniques

65

r­ easonably convenient way. Generally speaking, the more convenient the waveform measurement is, the more noise contamination can be resulted. For example, when the proximal and distal sites are apart, the signal-to-noise ratio will be higher, but such design will lead to two different sensors, and compromise on the convenience. On the other hand, getting the two sites next to each other can make a great case on convenience, but the signal-to-noise ratio might be an issue given too small PTT between the two sites. Second challenge is to calculate the PTT with suboptimal waveforms obtained by applying algorithm to enhance the signal-to-noise ratio. It is necessary to strike a balance between waveform quality and convenience, and rely on signal processing algorithm to optimize the suboptimal waveforms. The third and the most difficult challenge is to choose the right calibration and appropriate calibration intervals. The calibration curve relating PTT to BP is dependent on quite a few factors, including individual specific characteristics such as body height, relevant artery cross-sectional area, and vascular tree compliance. Therefore, obtaining an individual specific calibration curve through multiple blood pressure readings which covers the target blood pressure range is optimal. However, constructing such a curve requires cuff BP measurements from the subject, and the calibration curve need to be updated at a rate faster than the arteriosclerotic process (e.g., up to a few years at a time). A final major challenge is the independent determination of both systolic BP and diastolic BP  at the same time. This challenge is important, as isolated systolic hypertension often occurs in the elderly. However, conventionally estimated PTT may only be a marker of diastolic BP.

Mathematical Modeling PTT based BP estimation relies on two basic equations for arterial wave propagation. The first is the relationship between the Young’s modulus (E) and the arterial pressure (P):

E = Eo e aP

(6.2)

where a is a parameter that is related to the vessel and Eo is the Young’s modulus for zero arterial pressure. The formula estimates the arterial pressure if a and Eo are adjusted for subject (age, gender, health condition, etc.) on the elasticity due to the change in the wall composition. The second is the Moens–Korteweg equation in which the elasticity of arteries determines the propagation speed, which is the pressure’s pulse wave velocity (PWV).

66

J. Li

PWV =

hE ρd

(6.3)

where h is the thickness in an elastic artery, d is the diameter, and ρ is the blood density. Combining Eqs. (6.2) and (6.3), the relationship between P and PWV can be derived by the Bramwell–Hill equation PWV = L / ∆t =

hEo e aP ρd

(6.4)

where L is the length the pulse wave passed through, and ∆t is PTT, or the time delay [7]: This equation indicates that an increase in pressure will lead to an increase in PWV and decrease in time delay. Various mathematical models were used to approximate the relationships between BP and the PTT including the following: 1. Logarithmic model—Rearranging the Bramwell–Hill equation, we will have the logarithmic relationship between BP and PTT in the form of [8]:



2  2  1   ρ dL P =  −  ln ( ∆t ) +   ln   a  a   hEo

  

Or in a simplified version as:

BP = A ∗ ln ( ∆t ) + B



(6.5)

Here, A and B are subject-specific constants and they can be obtained through a regression analysis between the reference BP and the corresponding PTT, or ∆t— the time delay [9]. The above mathematical relationship is the theoretical basis between PTT and BP. The more generalized relationship can be summarized as

BP = A ∗ f ( ∆t ) + B



(6.6)

where f(∆t) is a specific function of the PTT, or ∆t—the time delay. Among various approximation of BP from PTT, the following models were used besides the logarithmic model shown in (6.5): 2. Linear model—Assuming there is a negligible change in the arterial thickness and diameter with pressure variation, BP and the time delay can be linearly related [10]:

6  Pulse Wave Velocity Techniques



67

BP = A ∗ ∆t + B

(6.7)

BP = A / ∆t 2 + B

(6.8)

3. Inverse Square Model

where A and B are constants that are related to subject’s biometrics such as height and blood density [6] 4. Inverse Model—an inverse relationship between BP and PTT [11]

BP = A / ∆t + B

(6.9)

While PTT is the most important parameter for estimating BP, other factors, especially those that are related to cardiovascular activity can be added to enhance the robustness of the model. Specifically, heart rate (HR) represents the cardiac cycle and influences the heart’s preload and the cardiac output (CO), which positively impact BP as the pressure on the arterial walls. HR is proportional to the volume of blood ejected. HR and BP are regulated by the autonomic nervous system which has been found to be inversely related, depending on the baroreflex activity. HR is calculated from the RR interval in ECG signals and has been incorporated in several algorithms to estimate BP, contributing some improvement in accuracy.

BP = A ∗ f ( ∆t ) + B ∗ g ( HR ) + C



(6.10)

where A, B, and C are constants, and ∆t and HR are two variables [12]. In order to apply a mathematical model, one needs to vary the BP over a considerable range to obtain the curve that can relate PTT or PAT closely to BP. Models under different conditions, generally exercising, hydrostatic posture, Valsalva maneuver, and medication, have been commonly used [13–16].

Technical Details on Measurement Methods The most direct way is to measure the time delay between two arterial sites. In research, it is easy to handle such way, but in consumer oriented application, it is challenging, because the device needs to be simple and compact, which leads to short PTT due to the two testing sites being quite close to each other quite sensitive to error. Alternative way is to calculate the difference between two PATs at two different sites: for example, one is at the finger tip, but the other is at the wrist along the same arm. The approach cancels the same pre-ejection period PEP. The disadvantage of this approach is that the user needs to wear sensors at two different sites.

68

J. Li

The focus of the PPT measurement is on conveniently measuring waveforms, contact or noncontact, most of which are relevant waveforms instead of blood pressure waveforms. PPG, bioimpedance, electrocardiography (ECG), ballistocardiography  (BCG), and video plethysmography (VPG) are among the most popular methods. These are based on either optical or electrical principles. PPG method  Uses optical transmittance or reflectance to measure waveforms of proximal and distal blood volumes—the same technique as used in pulse oximeters to measure arterial oxygen saturation [17]. Transmission-mode and reflectance-­ mode PPG offer different advantages and limitations. The reflectance-mode PPG is less restrictive in measurement locations (including forehead, forearm, supraorbital artery, under the legs, and the wrist), greater signal amplitude and lower motion artifact [18]. The disadvantage includes lower signal-tonoise ratio, motion artifact, and variation due to relative distance between the light source and the sensor caused by the shorter green wavelength that is only able to penetrate the skin and reflects skin blood flow rather than deeper larger arteries. The transmission-mode PPG measurements are limited in locations: such as the earlobe, fingertip, and toe for maximizing the signal-to-noise ratio. Infrared (IR) and red optical wavelengths are widely used for transmission-mode PPG due to higher tissue penetration depth. IR is less sensitive to the oxygen content of hemoglobin and thus yields waveforms that are more stable over time. Transmission-­ mode PPG at the fingertip has been the most widely used method for obtaining the distal waveform. Bioimpedance method  Bioimpedance can be measured using skin surface electrode pairs placed on skin surface at both proximal and distal sites. Micro amp level electrical current with high frequency is applied onto the outer electrodes; and the resultant differential voltage is measured across the inner electrodes. The measured differential voltage has both an AC component for the pulsatile blood and the DC component for other stable tissue components (bone, fat, muscle, and interstitial fluid). For PTT measurements, bioimpedance plethysmography is used with four electrodes in two pairs, positioned locally along the same artery in the limb (such as the lower leg or forearm) to measure the differential voltage when blood is flowing from one set of electrodes to the other [19, 20]. ECG method  This method provides a time reference with ECG, an easy to measure and resistant to noise and artifact. The time delay between the ECG waveform and a distal arterial waveform is the pulse arrival time (PAT). The advantage is, ECG is easy to measure, but the disadvantage is, PAT does not have as tight relationship to BP as PTT, due to the pre-ejection period PEP [21]. Ballistocardiography (BCG) method  It measures the reaction forces of the body in response to cardiac ejection of blood into the aorta using daily common objects such as chairs, beds, weighing scales, and on-body accelerometers [22]. BCG provides a

6  Pulse Wave Velocity Techniques

69

proximal waveform, but it can be applied at a distal location, such as at the feet with a weighing scale or at the wrist with an accelerometer. Video plethysmography (VPG) method  A noncontact method uses camera such as that on a smartphone to measure arterial waveforms from the skin [23]. It uses ambient/external light serving as the excitation source. Both proximal and distal waveforms can be obtained from two sites, such as the face, finger, or hand. Infrared thermal imaging with a highly sensitive camera can be used to measure arterial waveforms from the skin, based on the principle that the measured skin temperature changes with pulsatile blood flow due to heat exchange between vessels and surrounding tissue. The actual measurement involves following measurement steps: (a) Sampling and filtering the proximal and distal waveforms. (b) Detecting the beats in the waveforms. (c) Detecting the feet or other features within the beats. (d) Calculating PTT as the time delay between the features. Handling artifacts in the waveforms (e.g., due to motion) is also crucial in practice yet is often not mentioned.

Calibration of PTT to BP PTT is a relative parameter that correlates with blood pressure. It needs calibration to have absolute value of blood pressure. Calibration can be done with automatic blood pressure monitors. Calibration process is to set up the necessary parameters for the model to establish one to one mapping relationship between the actual BP readings and the PTT measurements. There are several aspects of calibration: the first is how to calibrate, the second one is how frequent to calibrate, and lastly is if the calibration can be done by population instead of by individuals. The goal is to construct a calibration curve to map PTT measurement to absolute value of BP readings. The actual steps of calibration involve the following steps: 1. Define a mathematical model to relate PTT to BP, either using physical models or empirical regression models. Most of the physical models are based on the Moens–Korteweg and Bramwell–Hill equations with an assumed function to relate the elastic modulus or compliance to BP. 2. Measure multiple pairs of PTT and BP values from a subject during interventions that change the BP in a wide range. Commonly employed interventions to change BP include exercise (e.g., climbing steps, cycling on an ergometer), postural changes (i.e., seated, supine, standing), sustained handgrip, and the Valsalva maneuver. Large BP changes can be achieved using anesthesia induction, surgery, and ICU therapies but are limited to hospitalized subjects and are therefore not valid for chronic hypertension management.

70

J. Li

3. Estimate the parameters by fitting the model to the PTT-BP multiple measurements. The accuracy of the parameter estimation generally improves with the ratio of the number of data pairs to the number of parameters. Least squares regression has been commonly used for parameter estimation. Calibration Frequency  The calibration curve can be constructed either one-time at the beginning, or periodically throughout a period [24, 25]. For the studies require periodic calibration, the period between calibrations was within 2 h. Such frequent calibration is to account for body system changes such as vascular tone change or smooth muscle contraction. Shorter PTT is directly linked to the increased arterial stiffness, which is driven by either higher sympathetic nerve activity or elevated blood pressure, so the sympathetic nerve activity can also change the individual’s arterial stiffness “black box,” and therefore such change will require the calibration again. Besides individualized calibration, generalized calibration approach can also be utilized. The process is to use a population average value for one model parameter while estimating the other parameter from cuff BP measurement [26]. The advantage of this approach is to avoid BP perturbation. The disadvantage of using population averages is the less accurate BP values. To implement this method, the following steps would have to be performed: (a) collection of training data comprising pairs of PTT estimates and BP values during a set of BP varying interventions per subject from a vast number of diverse subjects; (b) estimation of the parameters of a calibration model for each subject; and (c) regression of these parameters on simple subject information. Collecting the necessary training data is a serious endeavor but may be the best way to popularize the PTT-based BP monitoring approach. There are challenges to have independent determination of systolic and diastolic BP, if these two BP values do not vary in the same direction (e.g., isolated systolic hypertension). This problem could be addressed by including additional simple covariates, such as heart rate, in the calibration model or by estimating multiple PTT values per beat via arterial modeling. Note that PTT estimates should correlate better with BP values than these covariates do to offer any real value. Limited by length and depth, this chapter is just aimed to serve as a general overview of key elements in PWV or PTT related cuffless continuous BP monitoring. Interested readers who want to study on mathematical models, calibration, and practical approaches should further reference to relevant literature in this field [27–29].

The Future of Technology With the clear advantage of this PWV or PTT based technology being less obstructive and therefore continuous, there are three potential applications for such technology to play meaningful roles in the real world:

6  Pulse Wave Velocity Techniques

71

1. Continuous BP monitoring—This is the most active field in which lots of startup companies or research interests reside. In order to provide BP readings on a continuous basis, the challenge is all around the calibration—specifically, how to improve calibration of PTT to BP accuracy, how to make calibration simple and easy, and how to make calibration less frequently are among the bottlenecks that could present as the biggest hurdle for PWV/PTT based technology become reality. One potential method is the universal calibration wherein the parameters of the model relating PTT to BP are determined simply from the subject’s age, gender, and other such information including cardiovascular risk factors. Another approach is to involve more of an individual’s biometrics parameters and use machine learning to improve the predictive power of future BP. 2. BP change tracker—This application will reflect the BP changes, without giving out the absolute BP value. Such BP changes are in mmHg unit, and such changes could be used to guide individuals to take appropriate measures. 3. Trigger for absolute BP measurement—In this application, the PTT device serves as a trigger point based on individualized setup. Once BP changes monitored by the PTT technology reach a preset value, the trigger will alert the individuals to take a BP measurement using traditional cuff-based devices. Even though the PWV or PTT based technology is not the main unit for BP measurement, such technology provide a quite user friendly setting with continuous protection against elevated BP level. Machine learning   Nowadays artificial intelligence through machine learning has been used in research for BP prediction and cuffless BP measurement. The general approach is to initially extract surrogate cardiovascular indexes from physiological signals, and then to use machine learning to train and adapt to the model, and finally to predict BP using the trained model. Key features could be extracted from physiological signals such as ECG, PPG, activity, or sleep using either time domain or frequency domain. A feature selection method was used to remove irrelevant or redundant features to avoid over fitting. Machine learning methods such as linear regression, neural network, and Bayesian network can be used to establish the BP prediction model. Chiang et al. used Random Forest with Feature Selection to enhance the performance of the BP prediction by filtering out unnecessary features from wearable sensors signals and past BP readings [30]. Xing et al. used fast Fourier transform on the PPG signal to extract the amplitude and frequency features to train an artificial neural network to estimate BP [31]. Kachuee et al. extracted multiple physiological parameters from ECG and PPG signals using multiple regression [32]. Future machine learning and math modeling—select optimal features that can best contribute to dynamic BP changes and then combine the physiological and mathematical modeling to predict continuous BP noninvasively and continuously. Considering the complexity of the cardiovascular system, selection of multiple indicators and an appropriate model is critical and requires full-system integration to ensure the accuracy of indirect measurements.

72

J. Li

References 1. Pressman G, et al. A transducer for the continuous external measurement of arterial blood pressure. IEEE Trans Biomed Electron. 1963;10:73–81. 2. Peñáz J.  Photoelectric measurement of blood pressure, volume and flow in the finger. In: Digest of the 10th International Conference on Medical and Biological Engineering. 1973. 3. Weltman G, et al. The continuous measurement of arterial pulse wave velocity. Med Electron Biol Eng. 1964;2:145–54. 4. Kortekaas MC, et al. Small intra-individual variability of the preejection period justifies the use of pulse transit time as approximation of the vascular transit. PLoS One. 2018;13(10):e0204105. 5. Peter L, et al. A review of methods for non-invasive and continuous blood pressure monitoring: pulse transit time method is promising? IRBM. 2014;35:271–82. 6. Fung P, et al. Continuous noninvasive blood pressure measurement by pulse transit time. Conf Proc IEEE Eng Med Biol Soc. 2004;1:738–41. 7. Nichols WW, et al. McDonald’s blood flow in arteries: theoretical, experimental and clinical principles, 3rd ed. J Cardiopulm Rehabil Prev. 1991;11:407. 8. Proença J, et al. Is pulse transit time a good indicator of blood pressure changes during short physical exercise in a young population? In: Proceedings of the 2010 Annual International Conference of the IEEE Engineering in Medicine and Biology (EMBC), Buenos Aires, Argentina. 31 Aug–4 Sept 2010. p. 598–601. 9. Geddes L, et al. Pulse transit time as an indicator of arterial blood pressure. Psychophysiology. 1981;18:71–4. 10. Wong MY, et al. An evaluation of the cuffless blood pressure estimation based on pulse transit time technique: a half year study on normotensive subjects. Cardiovasc Eng. 2009;9:32–8. 11. Masè M, et  al. Feasibility of cuff-free measurement of systolic and diastolic arterial blood pressure. J Electrocardiol. 2011;44:201–7. 12. Baek HJ, et al. Enhancing the estimation of blood pressure using pulse arrival time and two confounding factors. Physiol Meas. 2009;31:145–57. 13. Marcinkevics Z, et al. Relationship between arterial pressure and pulse wave velocity using photoplethysmography during the post-exercise recovery period. Acta Univ Latv Biol. 2009;753:59–68. 14. Longo A, et al. Posture changes and subfoveal choroidal blood flow. Investig Ophthalmol Vis Sci. 2004;45:546–51. 15. Parati G, et al. Comparison of finger and intra-arterial blood pressure monitoring at rest and during laboratory testing. Hypertension. 1989;13:647–55. 16. Steptoe A, et al. Pulse wave velocity and blood pressure change: calibration and applications. Psychophysiology. 1976;13:488–93. 17. Shelley KH. Photoplethysmography: beyond the calculation of arterial oxygen saturation and heart rate. Anesth Analg. 2007;105:S31–6. 18. Maeda Y, et  al. Relationship between measurement site and motion artifacts in wearable reflected photoplethysmography. J Med Syst. 2011;35:969–76. 19. Risacher F, et  al. Impedance plethysmography for the evaluation of pulse-wave velocity in limbs. Med Biol Eng Comput. 1993;31:318–22. 20. Critchley LA.  Impedance cardiography. The impact of new technology. Anaesthesia. 1998;53:677–84. 21. Geddes LA, et al. Pulse arrival time as a method of obtaining systolic and diastolic blood pressure indirectly. Med Biol Eng Comput. 1981;19:671–2. 22. Inan OT, et  al. Ballistocardiography and seismocardiography: a review of recent advances. IEEE J Biomed Health Inform. 2015;19(4):1414–27. 23. Yoshizawa M, et  al. A great impact of green video signals on tele-healthcare in daily life, especially for rural or disaster areas. Trans Jpn Soc Med Biol Eng. 2013;51:M-55. 24. Chen W, et al. Continuous estimation of systolic blood pressure using the pulse arrival time and intermittent calibration. Med Biol Eng Comput. 2000;38:569–74.

6  Pulse Wave Velocity Techniques

73

25. McCarthy BM, et al. An investigation of pulse transit time as a non-invasive blood pressure measurement method. J Phys Conf Ser. 2011;307:012060. 26. McCarthy BM, et al. An examination of calibration intervals required for accurately tracking blood pressure using pulse transit time algorithms. J Hum Hypertens. 2013;27:744–50. 27. Mukkamala R, Hahn JO, Inan OT, Mestha LK, Kim CS, Töreyin H, Kyal S. Toward ubiquitous blood pressure monitoring via pulse transit time: theory and practice. IEEE Trans Biomed Eng. 2015;62(8):1879–901. 28. Sharma M, Barbosa K, Ho V, Griggs D, Ghirmai T, Krishnan SK, Hsiai TK, Chiao JC, Cao H. Cuff-less and continuous blood pressure monitoring: a methodological review. Technologies. 2017;5:21. 29. Ding XR, et al. Continuous blood pressure measurement from invasive to unobtrusive: Celebration of 200th birth anniversary of Carl Ludwig. IEEE J Biomed Health Inform. 2016;20(6):1455–65. 30. Chiang PH, et  al. Personalized effect of health behavior on blood pressure: machine learning based prediction and recommendation. In: IEEE 20th International Conference on e-Health Networking, Applications and Services (Healthcom). 2018. https://doi.org/10.1109/ HealthCom.2018.8531109. 31. Xing X, et al. Optical blood pressure estimation with photoplethysmography and FFT-based neural networks. Biomed Opt Express. 2016;7:3007–20. 32. Kachuee M, et  al. Cuff-less high-accuracy calibration-free blood pressure estimation using pulse transit time. In: 2015 IEEE International Symposium on Circuits and Systems (ISCAS). 2015. p. 1006–9.

Chapter 7

Pulse Decomposition Analysis Techniques Martin C. Baruch

Abstract Pulse decomposition analysis (PDA) uses a pulse contour analysis approach to quantify hemodynamic parameters such as blood pressure and arterial tone changes. It is based on the concept that two central reflection sites are responsible for the shape of the pressure pulse envelope of the upper body. The two reflection sites, one located at the aortic juncture of thoracic and abdominal aortas, and the other at the iliac bifurcation, reflect the primary left ventricular ejection pulse to give rise to two reflected and two re-reflected component pulses. Within the pulse pressure envelope of each cardiac cycle these five component pulses arrive sequentially in the arterial periphery. Quantification of the temporal and amplitudinal behavior of the first three component pulses establishes a formalism that can be used to monitor certain hemodynamic states and their changes. The observational evidence and motivation for PDA are presented, as are pulse modeling approaches, practical implementation considerations and physiological confounders. Benchmark and clinical study comparisons are provided. The current status and outlook of the CareTaker physiological monitor, which utilizes PDA as its operational principle and has demonstrated compliance with several regulatory standards, are described. Keywords  Noninvasive · Continuous blood pressure · Arterial reflections · Finger cuff · CareTaker

Pulse Decomposition Analysis Pulse decomposition analysis (PDA) uses a pulse contour analysis approach to quantify hemodynamic parameters such as blood pressure and arterial tone changes. It is based on the concept that central reflection sites, as opposed to distal sites in the arterial periphery, are primarily responsible for the shape of the pressure pulse M. C. Baruch (*) Caretaker Medical LLC, Charlottesville, VA, USA e-mail: [email protected] © Springer Nature Switzerland AG 2019 J. Solà, R. Delgado-Gonzalo (eds.), The Handbook of Cuffless Blood Pressure Monitoring, https://doi.org/10.1007/978-3-030-24701-0_7

75

76

M. C. Baruch

envelope of the upper body. Specifically, PDA postulates that five individual component pulses give rise to the observed pulse shape. The first of these component pulses to arrive in the arterial periphery is the left ventricular ejection pulse which is then followed by reflections and re-reflections of the ejection pulse from two central arteries reflection sites. PDA further postulates that the quantification of the temporal and amplitudinal behavior of these component pulses gives rise to a formalism that can be used to monitor certain hemodynamic parameters and their changes. PDA has cleared several practical implementation stages and is the operational principle of the CareTaker physiological monitor, which has demonstrated compliance with several regulatory standards.

Underlying Considerations The existence and the physiological consequences of reflections in the arterial tree are now commonly accepted [1–4]. In this physiological model the arterial pressure pulse originates from the left ventricle and travels away from the heart through the arterial tree and is reflected at sites where the arterial tree branches or different diameter sections join, since these sites present an impedance mismatch to the propagating arterial pressure pulse. Clinical studies and theoretical modeling efforts have investigated various aspects of arterial pulse reflections, such as the “second systolic peak”, yet no clear model has been proposed specifying where exactly the reflections arise. For example, an asymmetric T-shaped model [5] has been proposed where the pulse originates at the T junction from the heart and the ends of the T represent generalized reflection sites of the lower body and the upper body. Similarly it has been proposed that the principal mechanism giving rise to reflections in the arterial tree are the various artery/arteriole interfaces throughout the body, since these sites, characterized by significant lumen changes and therefore impedance mismatches, will give rise to reflected pressure pulses that, counter propagating, will return from the arterial periphery.

Evidence of Central Reflection Sites In contrast to these models proposing distributed reflection sites, Pulse Decomposition Analysis (PDA) is based on the concept that distinct reflection sites dominate the shape of the arterial pressure pulse envelope, resulting in a superposition of distinct component pulses. Focusing on the structure of the digital or the radial arterial pressure pulse, because these are usually the clinically most relevant monitoring sites, the component pulses that are the temporal features of the arterial pulse envelope that follow the primary left ventricular ejection pulse are reflections arising from reflection sites in the core arteries, specifically the junction of thoracic

7  Pulse Decomposition Analysis Techniques

77

and abdominal aortas, and the interface between abdominal aorta and the common iliac arteries. Why is it reasonable to assume that there are distinct reflection sites in the arterial tree as opposed to the assumption that, as an alternative scenario, “the lower body” as a whole gives rise to reflections? The answer is twofold. One is that the features of the reflected wave are too distinct as to be the convolution of different reflections originating from different sites with different time delays and different amplitudes, which would tend to broaden out specific pulse features. The second answer is that the arrival times, determined by well-known arterial pulse propagation velocities, of the specific features of the radial pulse very much narrow the location possibilities of the reflection sites. Figure 7.1 presents radial pulse signatures collected from different individuals of different ages. Both traces exhibit pulse-like protrusions (black and red arrows) that have a time duration comparable to that of the primary pulse (blue arrows). Data that clarifies this point is presented in Fig. 7.2, which presents radial pulse data collected during a Valsalva episode. One consequence of Valsalva is the shortening of the cardiac ejection period. As a result, it is possible, in a comparatively young and elastic arterial tree, to see the complete separation of primary pulse and reflected pulse. Clearly the reflected pulse shows little to no broadening compared to the primary systolic peak, supporting the hypothesis that it originated at a distinct refection site. Figure 7.3 seeks to clarify this point further. While a distinct reflection site will give rise to a reflection bearing strong resemblance to the primary pulse, distributed and multitudinous reflection sites will give rise to a multitude of reflected pulses, arriving at different time delays and with different amplitudes.

Fig. 7.1  Examples of high-fidelity radial arterial pulse shapes. Top: 20 y. m. athlete. Bottom: 52 y. m. catheter laboratory patient. Note the pulse-like features, indicated by black and red arrows, following the primary ejection pulse (blue arrows), that have a similar temporal and amplitude profile as the primary pulse

78

M. C. Baruch

Fig. 7.2  Radial pulse during onset of Valsalva maneuver. Notice the vanishing of the second systolic pulse

Fig. 7.3  Qualitative comparison between reflected pulse (red on left resulting from distinct reflection sites (top) as compared to a reflected pulse resulting from distributed, amorphous reflection sites

The superposition of such a system of reflection sites would result in a featureless, broadened pulse. The presence of distinct pulse-like features in the radial signatures shown therefore suggests that, past the primary systolic peak, distinct reflection sites are responsible for the sequence of reflected pulses comprising the “diastolic wave.” While the presence of distinct pulse-like features in the radial pulse suggests the existence of specific and powerful reflection sites, their time of arrival relative to the primary pulse makes the argument significantly more concrete. Figure 7.4 presents an example of the radial pulse of a 44 year old male as well as the time intervals between its various component pulses. The first timing issue worth considering is to what degree the pulse features are influenced by the geometry of the arm, that is, could one of the pulse features observed be due to a reflection site in the arm? Arterial pulse velocities in the radial artery are on the order of 7–8 m/s. Since the pulse signal is collected at the wrist, the distance from that site to a site of a potential reflection, the interface between artery and arterioles at the wrist, is on the order of centimeters. Therefore, the reflection would return in a matter of a few milliseconds, as indicated in Fig. 7.4 by a blue line in the immediate vicinity of the primary pulse. Since all the reflected pulse features in the radial pulse appear at far greater time delays, as indicated in the figure, they have to originate elsewhere in the arterial tree.

79

7  Pulse Decomposition Analysis Techniques Second Systolic Peak Diastolic Peak

"Radial Reflection" (not visible) 115 ms

121 ms

172 ms 193 ms

168 ms

0.2 Seconds Fig. 7.4  Distinct pulse structure in the radial arterial pulse of a 44 y. male

Since arterial pulse propagation velocities are well known, it is possible to match time delays with potential reflection sites. Figure 7.5 presents a simplified sketch of the components of the aorta and the connecting arteries of the legs and the left arm. The sketch also lists typical arterial diameters as well as arterial pulse propagation velocities at the different sites as published in the medical literature [1]. Using approximate arterial distances and their respective velocities, the “second systolic” peak matches readily with the site labeled “reflection site I” while the third peak matches with “reflection site II.” Work by others supports these conclusions [2–4]. In 1985, Latham [6] performed a detailed experimental study to map out the shape of the pressure pulse in the different sections of the aorta using a specially designed catheter with spaced micro manometers (Fig. 7.6). His work demonstrated the existence of two major reflection sites to the distally traveling arterial pulse, one in the region of the renal arteries, the other beyond the bifurcation of the iliac arteries. At the location of the renal artery the diameter of the aorta, tapering continuously away from the heart, undergoes its greatest change at the juncture between the larger diameter thoracic aorta and the smaller diameter abdominal aorta. This discontinuity presents a significant impedance mismatch to the traveling pressure pulse, resulting in an appreciable part of its amplitude being reflected. Referring back to Fig.  7.2 and the Valsalva maneuver, the phenomenological explanation is that the maneuver reduces this reflection because, due the increasing pressure within the thoracic cavity, the diameter of the thoracic aorta decreases while the diameter of the abdominal aorta, which is outside the thoracic pressure cage, does not. The maneuver therefore alleviates the aortic diameter change at the renal arteries, reducing the impedance mismatch and thereby lowering the site’s reflection coefficient.

Fig. 7.5  Sketch of the aorta/arm complex arterial system and its effect on the arterial pressure pulse line shape observed at the radial/digital artery. Two reflection sites, one at the height of the renal arteries, the other one in the vicinity of the iliac bifurcation, give rise to the reflected pulses (gray) that trail the primary left ventricular ejection (black)

80 M. C. Baruch

7  Pulse Decomposition Analysis Techniques

81

Fig. 7.6  Arrangement of the catheter sensor positions in the aorta, with examples of pressure waveforms from patient C. Diameter and SD values refer to elastic tube model used to simulate observed effects. Reproduced with permission

Latham also found a second reflection site beyond the bifurcation of the iliac arteries, the contribution of which to arterial pulse reflections in the aorta was ascertained using manual femoral artery occlusion maneuvers. Other contributions to the tail end of the aortic pulse were attributed to diffuse arterial pulse reflections from the periphery. This, however, appears to be unlikely, given the distinct peak structure with a spacing comparable to that of the “second systolic” and the “diastolic” peak. Furthermore, the time delay of such diffuse reflections would extend up to 250 ms past the “diastolic” peak if they truly traversed the length of the legs. Indeed, other work by J. Kriz et al. [7] supports the hypothesis that the peaks visible past the “diastolic” peak are in fact due to re-reflections between the two reflection sites, a reasonable proposition given the strength of the sites’ reflection coefficients (10– 15% in the case of the renal arteries reflection site and up to 30% in the case of the iliac arteries reflection site [8]). The work by Kriz showed that it is possible to use force plate measurements as a noninvasive method to perform ballistocardiography, the body’s recoil due to the momentum generated by the heart’s activity, by displaying the motion of the heart muscle and the subsequent propagation of the pulse wave along the aorta and its branches. With subjects lying horizontally on a bed placed on a force plate they were able to identify the ground reaction forces arising from such center-of-mass

82

M. C. Baruch

altering events as the heart muscle contraction as well as the resulting ejection pressure pulse. The resolution of the apparatus was sufficient to clearly resolve events involving the redirection of momentum of the propagating arterial pulse, such the pulse’s traversal of the aortic arch, its partial reflection at the renal artery site, the iliac reflection well as the subsequent re-reflections of the reflected pulses. As an aside, in subjects with an aortic aneurism, the site of the arterial distention was clearly identifiable due to its effect on the neighboring “normal” reflection sites. The basic PDA model of the radial/digital arterial pressure pulse is therefore one of a convolution of the primary systolic peak, its single-pass reflections from the renal arteries and iliac arteries reflection sites, as well as their double-pass re-reflections.

Implementation Modeling of Pulse Reflections The existence of two distinct central pressure pulse reflection sites make it is possible to propose a simple model of the arterial paths the primary pulse and its reflections traverse and to compare its predictions with observations regarding the relative arrival times of the different component pulses. The model’s equations predict the time of arrival of each individual component pulse, subject to the total distance the pulse has traveled and the pressure-dependent pulse propagation velocity in each arterial segment. The different relevant arterial paths are denoted by xn, where x1 refers to the arm arterial path, while x2 and x3 refer to the thoracic and abdominal aorta, respectively. The variable tn refers to the time of arrival of the nth component pulse at the radial/digital arterial peripheral site. While in the case of the #1 pulse its arrival time, t1, is determined only by its travel along the arm complex arteries (x1 path), the arrival times for the #2 and #3 pulses take into account their initial travel as the primary ejection pressure pulse as well as, after impacting a reflection site, their subsequent return as a reflected pulse. As an example, the “second systolic” (#2) pulse traverses the thoracic aorta at systolic pressure, traverses it again as an R2 reflection after redirection at the renal arteries reflection site (indicated as R2 of pulse pressure plus diastolic pressure) and then enters the arm arteries where it loses another percentage of its amplitude due to the R1 reflection coefficient that incorporates artery segment transitions, such as the aortic/subclavian junction. The pressure dependence of the pulse propagation velocity is implemented using the Moens–Korteweg [9] equation relating pressure and velocity, ν =  √ ((hEeζP)/ (2ρα)). Its definitions are as follows: ν(P) is the velocity of the xth arterial pulse path at the pressure P indicated. E is the Young’s modulus, α is the artery’s diameter, h is the arterial wall thickness, ρ is the fluid density, ζ is the arterial compliance and P is the pressure. The Young’s modulus and the arterial compliance ζ are different for the different arterial segments.

7  Pulse Decomposition Analysis Techniques

83

Another significant feature of the model is that R2, the renal reflection coefficient, is dependent on pressure. The motivation for this is based on the following consideration. t1 = t2 = t3 =

x1 , vx1 , P11

(7.1)

x2 x2 x1 + + vx2 , P21 vx2 , P22 vx3 , P23

(7.2)

x3 x3 x2 x2 x1 + + + + vx2 , P31 vx3 , P32 vx3 , P33 vx2 , P34 vx1 , P35



P11 = Psyst - R1 PPulse



P21 = Psyst



P22 = Pdiast + R2 PPulse



P23 = Pdiast + R2 (1 - R1 ) PPulse



P31 = Psyst



P32 = Psyst - R2 PPulse



P33 = Pdiast + R3 (1 - R2 ) PPulse



P34 = Pdiast + R3 (1 - R2 ) (1 - R2 ) PPulse



P35 = Pdiast + R3 (1 - R2 ) (1 - R2 ) (1 - R1 ) PPulse

(7.3) (7.4)



(7.5)



(7.6) (7.7)



(7.8)



(7.9)



(7.10)



(7.11)



(7.12)

As discussed, the renal reflection (P2 pulse) originates at the junction between thoracic and abdominal aorta, a junction characterized by a significant change in arterial diameter. Since the thoracic aorta is the softest artery in the body, as evidenced by the lowest pulse pressure propagation velocities (4–5 m/s) and much more extensible than the abdominal aorta, increasing pressure will enlarge the diameter mismatch, giving rise to a more pronounced renal reflection pulse amplitude while falling pressure will produce the opposite effect, an effect observed in manipulative experiments performed by Latham. The central insight then is that the amplitude of the renal reflection will increase relative to the amplitude of the primary systolic (P1 pulse) peak because, while both component pulses travel the arteries of the arm complex, and are therefore both subject to the pulse narrowing and heightening due to the taper and wall composition changes of the peripheral arteries, only the renal

84

M. C. Baruch

reflection will have sampled the pressure-induced aortic impedance mismatch changes. This provides the motivation for taking the ratio of the amplitudes of the #2 and the #1 pulse, which is the PDA parameter P2P1. These considerations are put in context given the different response characteristics of central versus peripheral arteries that have been reported and discussed by others. Specifically, the fact that central arterial elasticity is determined by BP and not smooth muscle contraction, in contrast to peripheral arteries, provides a physiological explanation for the effects that are quantified and utilized as part of the PDA formalism [10, 11]. Preliminary tests involving a fit middle-age male subject demonstrate that this comparatively simple model is able to adequately predict the arrival times of the three primary component pulses during a maneuver such as Valsalva. Figure 7.7a–c, which present predicted and measured time delay curves for the three primary

Delay TIme (Seconds)

0.250

Pulse #1

0.225 0.200 0.175 0.150

A

0.125

Delay Time (Seconds)

0.38

0

0.36

20

40

60

80

100

40

60

80

100

40

60

80

100

Pulse #2

0.34 0.32 0.30 0.28 0.26

B

Delay Time (Seconds)

0.24 0.56

0

0.54

20

Pulse #3

0.52 0.50 0.48 0.46 0.44 0.42 0.40

C

0.38 0.36

0

20

Time (Seconds)

Fig. 7.7 (a, b, c) Relative overlap of delay times of the three primary pulses measured (red) and obtained using the model (black) with the diastolic and systolic blood pressures obtained from the Colin-Pilot, clinical monitor of noninvasive continuous blood pressure

7  Pulse Decomposition Analysis Techniques

85

pulses, give a sense of the agreement between the two. The predicted delay time values were obtained by isolating the diastolic and systolic peak to peak blood pressure values obtained from a continuous radial artery tonometer (Colin Pilot) and inserting these values into the PDA model. The timing of the individual component pulses was obtained using the QRS complex of a simultaneously obtained ECG signal as a starting signal. The agreement of the range of delay time values is no surprise since correlations were used to relate the blood pressures measured with the Colin unit to the measured pulse delay times. Encouraging is the fact that the overall time evolution of the predicted and measured delay times agrees well. In order to arrive at the above results, the pressure/velocity response curve for each of the three primary pulses had to be quantified by correlating the systolic and diastolic blood pressures measured with the tonometer with the delay times of the three primary pulses. In addition, the BP response behavior of the renal reflection coefficient R2 has to be quantified. The resulting fitted functions are displayed in Fig. 7.8 for the velocity responses of the different arterial sections and in Fig. 7.9 for the pressure response of reflection coefficient. The difference in velocity response, and therefore time delay response, between the different pulses is significant. While the arm complex displays an exponential response, thoracic and abdominal velocities follow more linear relationships. The pulse propagation velocity of the abdominal aorta region exceeds that of the arm arteries, comparable to results published in the medical literature. Of course, this result holds for subjects with “elastic” arteries. Results are quite different for patients with increased arterial stiffness, where arm pulse propagation velocities can reach 15 m/s. 9

Arm complex Adominal aorta Thoracic aorta

Velocity (m/s)

8

7

6

5

4

100

150

200

Pressure (mmHg) Fig. 7.8  Arterial pulse velocity profiles as a function of pressure, based on fit to experimental results in Fig. 7.6

86

M. C. Baruch

Reflection coefficient (unitless)

0.55 0.50 0.45 0.40 0.35 0.30 0.25 0.20 0.15 75

100

125

150

175

Pressure (mmHg) Fig. 7.9  Pressure response of R2, reflection coefficient of the junction of thoracic/abdominal aortic sections at the height of the renal arteries

The behavior of the three pulses is summarized in Fig. 7.10. The figure, which for graphical clarity inverts dependent and independent variables, summarizes the arrival time response, along the abscissa, of the individual component pulses as a function of varying arterial blood pressure and the correspondingly varying pulse propagation velocity, along the ordinate. Specifically, as blood pressure rises, so does the pulse propagation velocity. However, while in the diastolic regime the velocity increase is approximately linear with a linear pressure increase, the systolic regime is characterized by an exponential velocity increase response [12]. Commensurate with increasing blood pressure and increasing propagation velocity is a decreasing arrival time in the arterial periphery. Individual component pulses sample different sections of the velocity response curve depending on their pressure amplitude. Since the response curve is nonlinear, the different component pressure amplitudes give rise to different velocity variations between the different component pulses, that is, not only will the pulse envelope accelerate as blood pressure varies, but its components will do so relative to each other, changing the envelope of the process. Specifically, while the #1 pulse samples the top of the systolic pressure regime throughout its travel along the arterial tree to the radial pulse site, the #2 and #3 pulses do so only on the initial traversal of sections of the aorta, with a much greater part of their propagation time in significantly lower blood pressure ranges. As a result the exponential pressure/velocity relationship that governs their travel as outward bound primary pulses is masked by the linear pressure/velocity relationship that governs their travel as reflected pulses. More importantly, differential changes in travel time between the different pulses can be resolved because of the

7  Pulse Decomposition Analysis Techniques

87

Fig. 7.10  Effect of small pressure variations on the propagation velocity of the three primary component pulses. While the #1 pulse samples the nonlinear section of the response curve, the #2 and #3 pulse responses are essentially linear while the traverse the arterial system as reflections

different functional forms and gains of the velocity curves that govern the propagation of the different component pulses. While the timing considerations outlined above lend more qualitative credence to the approach, their relevance in the context of obtaining hemodynamic information through pulse wave analysis is somewhat limited because the external timer start, which in the above experiments was the ECG’s QRS peak, is usually not available. That leaves only relative timing determinations between the component pulses of a given pulse envelope which yield significantly less information because, as the Valsalva example above made clear, the component pulses display similar delay time evolutions, making their differential determination more difficult. In addition, detection of particularly the renal reflection can be challenging because of its highly dynamic amplitude response to blood pressure changes and the fact that it is the component pulse most prone to be obscured by the smoothing arterial pulse shape changes associated with stiffening arterial walls, a point that is discussed more in-­ depth in the following section.

Modeling of Pulses Just as is the case in the timing considerations above, a comparatively simple model can be used to generate the peripheral arterial pressure pulse envelopes of the upper body that are encountered clinically. Specifically, the triple overlap of a generalized asymmetric exponential function of the following form,

88

M. C. Baruch

1

1A*

1+ e 1+ e



w1 ö æ ç - x - xc + ÷ 2 ø è w3

w1 ö æ ç - x - xc + ÷ 2 ø è w2

(7.13)

can generate pulse envelopes that bear close resemblance to pulse shapes ranging from those recorded on young athletes to shapes associated with the arterial wall compliance and cardiac timing changes associated with more advanced age (Figs. 7.11 and 7.12, below which the parameters to generate the simulated curves are provided.). The individual component pulse in each case is modeled using an asymmetric line shape that is characterized by a fast onset and a significantly slower decay associated with peripheral resistance. Remarkably close pressure pulse envelope representations can be generated by simply adjusting the amplitudes and delays of the otherwise identical component pulses. While this type of modeling further supports the underlying PDA hypothesis and could potentially be useful in the short-term analysis of patient data with the goal of characterizing arterial wall health, it is impractical for the implementation of hemodynamic monitoring on a continuous, beat by beat basis. Since the procedure is a 0.6 0.4 0.2 0 0

2

10

100

200

300

400

500

600

700

100

200

300

400

500

600

700

-4

0 -2 -4 0

Fig. 7.11  Simulation of an arterial pressure pulse of a younger person with flexible arteries using Eq. (7.13) (top graph). Second derivative of the envelope is presented in the lower graph. Front rise = 0.4; backend = 1.5; backend3 = 2.3; delay1 = −2.0; ampl1 = 1.6; delay2 = −4.4; ampl2 = 0.3; delay3 = −7.0; ampl3 = 0.55; w1 = 0.5

7  Pulse Decomposition Analysis Techniques

89

0.8 0.6 0.4 0.2 0

0

2

100

10

200

300

400

500

600

700

-4

0 -2 -4 0

100

200

300

400

500

600

700

Fig. 7.12  Simulation of an arterial pressure pulse of an older person. Note the decreased delay times and enhanced renal reflection. Front rise = 0.4; backend = 1.5; backend3 = 2.3; delay1 = −2.0; ampl1 = 1.6; delay2 = −3.4; ampl2 = 1.1; delay3 = −4.9; ampl3 = 0.5; w1 = 0.5

multiparameter fit, the likelihood for an optimization algorithm to arrive at a nonsensical but optimized line shape is very high, and the computational load will become very significant as heart rates increase. The approach can, however, be used to guide a real-time pulse analysis approach where the aim is to examine sections of the pulse envelope or its derivatives and to track their evolution as hemodynamic changes such as that in blood pressure and heart rate occur. Other groups have performed similar simulations, with three and more underlying component pulses in the context of, for example, extracting information about cardiovascular function [13]. In what follows the principal physiological confounders affecting pulse analysis are discussed before giving an overview of the actual implementation issues.

Physiological Confounders A real-time pulse analysis approach has to be able to accommodate the different pulse shapes encountered across a wide patient spectrum. Long-term pulse shape changes that are age- and disease-related and arise as cardiac function and arterial wall structures are altered have been studied and modeled extensively, for example via the augmentation index [14]. Less well understood are the shorter-term effects

90

M. C. Baruch

that can modify the arterial pulse envelope significantly in a timeframe of minutes and severely compromise a previous blood pressure calibration of pulse parameters. Based on our research the principal short-term physiological confounders are arterial stiffness, heart rate and left ventricular ejection time (LVET) changes, as well as peripheral resistance changes. These confounders will be discussed next. Arterial Stiffness Considerations An observational fact that distinguishes arterial pressure pulses associated with stiff arteries from those associated with flexible arteries is the degree of features, or, in the context of PDA, the degree to which the component pulses are resolvable in the pressure pulse envelope. While pressure pulses recorded on flexible arteries have visually readily assessed distinct features, those recorded from stiffer arteries show fewer or more rounded features, or none at all. See Figs. 7.13, 7.14 and 7.15. While the above examples are due to long-term changes in the arterial wall structure, equally significant pulse shape changes due to, for example, vasodilation can be observed on much shorter time frames. Figures 7.16 and 7.17 display examples of vasodilation in different context which, in both cases, significantly modify the feature profile of the original pulse. Fig. 7.13  Young athlete: Location of P1 is indicated by the vertical line at about 100. The location of P2 is indicated by the two short vertical lines bracketing 200. For this individual, P1 and P2 are clearly resolved

Fig. 7.14  53 y. m. cath lab patient: P1 is indicated by the vertical line at 130, while P2 is indicated by the two short vertical lines bracketing 200. P2, following closely behind P1, has merged with P1 as an additive shoulder

7  Pulse Decomposition Analysis Techniques

91

Fig. 7.15  67 y m. pancreaticoduodenectomy surgery patient. P1 and P2 have essentially merged. Even the incisura in front of P3, indicated by the vertical line at 310, is essentially indistinguishable

10

6

4

4

P1 P2

P3

2 0

-2

0

0.2

0.4

4

P1 P2

3

Value

Value

4

10

0.6

0.8

1

1.2

1.4

Time (s)

P3

2 1 0

-1 0

0.2

0.4

0.6

1

0.8

1.2

Time (s)

Fig. 7.16  Digital pulse of a 58 y. m. prior to (left) and 30 s after ingestion of red wine. Note redistribution of amplitudes as well as the more pronounced inversion

10

6

4

P1

P2 P3

Value

Value

4 2 0

-2

0

0.2

0.4

P1

10000

Time (s)

0.6

0.8

1

P2

P3

5000 0 -5000

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

Time (s)

Fig. 7.17  Digital pulse of a 34 y. f. Cesarean section patient prior to (left) and after (right) administration of spinal anesthesia. Aside from the redistribution of component pulse amplitudes the three primary individual component pulses are now resolved (note one inversion on the left, two on the right) as the entire pulse envelope has lengthened with a decrease of arterial pulse propagation velocities (note the time shift to the right of P3), resolving the second systolic peak P2 (renal reflection) that was essentially indistinguishable before

Part of the PDA framework is an arterial stiffness parameter (AS) which quantifies the spectral content of the arterial pressure pulse that is due to the component pulses. The featuredness spectral content in turn is related to arterial stiffness as it is the mechanical filtering of the arterial wall that determines the extent to which the structure of the component pulses is resolved noninvasively. As an aside, because the featuredness spectral content of the pulse envelope is driven significantly by the resolution of the region of overlap of P2 and P3, respectively the renal and the iliac reflection pulses, the AS factor incorporates the pulse region of a parameter intro-

92

M. C. Baruch 35 30

a

25 20 15 10 5 0 –5

d P3

–10 –15

P2

–20

P1

–25 –30

0

100

200

300

400

500

600

Fig. 7.18  Calculation of AS factor in the second derivative and the d/a ratio introduced by Takazawa

duced by Takazawa that he was able to link to arterial stiffness on the evidence of vasodilator/pressor studies as well as demographic characteristics as part of an epidemiological study [15]. Similarly, preliminary validation tests indicate that the AS parameter tracks expected trends after the introduction of vasoactive agents as well as age-related population trends. The AS parameter is calculated using the numerical integral over the shaded area of the second derivative of the arterial pressure pulse profile (Fig. 7.18). The second derivative analysis approach provides better resolution in the identification of the component pulses, that is, the left ventricular ejection pulse (P1), the renal reflection pulse (P2) and the iliac reflection pulse (P3), in the pulse envelope. Also indicated in the figure are the constituents of the d/a ratio that was introduced by Takazawa, who labeled all of the inversions of the second derivative and then performed correlation studies between different ratios of these inversions and clinically relevant parameters. Both the AS integral and the d/a ratio resolve the region of the pulse envelope most affected by arterial stiffness changes, the overlap region of the P2 and P3. Since these reflection pulses are significantly weaker in pressure amplitude than the primary ejection pulse, the mechanical filtering of the arterial wall will have the largest effect here, obscuring the overlap as the wall stiffens and resolving if it is in a more compliant state. The benefit of integration is robustness, because often the identification of distinct inversions is difficult or impossible due to high heart rates or the smoothing effect of extremely stiff arteries. As Figs. 7.16 and 7.17 suggest, arterial stiffness changes are significant in the context of pulse analysis for the purpose of tracking blood pressure. In the case of

7  Pulse Decomposition Analysis Techniques

93

PDA the differential blood pressure response of central and peripheral arteries, which the ratio P2P1 represents, will be affected as the changed mechanical filtering of the pulse effects the amplitude and temporal distribution of the component pulses. Specifically, an increase in arterial stiffness will tend to increase the ratio while a decrease will have the opposite effect, with the possibility of overestimating a parallel trend or masking an opposite trend in blood pressure. Heart Rate Considerations Heart rate changes will affect primarily the timing aspects of pulse analysis, as the width of the component pulses changes with LVET, correspondingly shifting associated fiduciary markers. This is particularly a problem as heart rates approach and exceed approximately 120 bpm because, in addition to the changes in LVET, the average stress state of the arterial wall increases since the time duration between cardiac cycles is too short for the wall to completely relax, causing pulse propagation velocities to rise [16]. The overall effect is that the features being tracked on the pulse envelope will shift temporally relative to each other, with time intervals narrowing for rising heart rates and extending in the opposite case. In the case of stationary search windows, the need for shifting them in response to changed heart rates clearly has to be taken into account. In the case where discrete fiduciary points are tracked, such as for example certain inversions in the derivatives of the pulse envelope, the problem is more complex. In this case inversions points that were clearly resolvable at lower heart rates will merge with other features or lose profile amplitude, impeding detection. Peripheral Resistance Considerations A discussion of the effect of peripheral resistance on the pressure pulse envelope requires a brief clarification since the effect is sometimes invoked as being the cause of arterial pressure reflections [17]. More current views propose that it is the peripheral resistance that sustains pressure once the left ventricular ejection as subsided [18]. Consequently, it will determine the rate at which stored pressure “bleeds” into the arterial periphery, determining the decay time on the falling side of arterial pressure pulse. The above Fig. 7.19 presents an example of the evolution of the arterial pulse of a 35 male subject prior to, during, and shortly after a 2 min 100 W workout. While the changes due to heart rate/LVET/cardiac output in the front section of the pulse are obvious, the collapse of the reflected pulses on the falling side of the pulse envelope are also clearly evident as the corresponding reflection sites diminish to accommodate the increased blood flow associated with the increased peripheral muscle oxygen requirements. The interesting aspect, considering peripheral resistance, is the final slope at the tail end of the pulse envelope. While at rest the slope is very

94

M. C. Baruch

resting

15 seconds past end of workout

5 min past

0.25 Seconds/div Fig. 7.19  Evolution of the arterial pressure pulse of a 35 m from rest (left), toward the end of a 2 min 100 W workout, 15 s into recovery, and 5 min into recovery. Peripheral resistance changes in the tail end of the pulse are visible. Vertical red lines indicate, from left to right, the temporal positions of P1, P2, P3, respectively left ventricular ejection, renal reflection, iliac reflection

shallow, suggesting increased peripheral resistance, the slopes at the end of the workout and just after are much steeper, suggesting a steep pressure drop at the end of diastole. Five minutes later, while clearly not fully recovered as indicated by the still elevated heart rate, the slope is again dropping. Similar effects were observed in experiments specifically designed to alter peripheral resistance pharmacologically with the simultaneous introduction of multiple agents [19]. From the point of view of pulse analysis, extracting blood pressure during such hemodynamic changes is one of the most challenging tasks.

Final Considerations on Implementation This section has provided a brief overview of the challenges any blood pressure pulse analysis approach faces. They will affect the detectability of most fiduciary points on the pulse profile, which will vary or vanish entirely under certain circumstances. This applies to detection on the actual pressure pulse profile as well as any of its derivatives, which, while sometimes able to enhance detection of curvature inversions, can also obscure it because they are a subject to both amplitude as well as slope changes. As a result, for example, an initially low-amplitude inversion in the pulse envelope occurring over a short time frame will be amplified in its derivatives, facilitating detection, while being diminished as the amplitude grows but the slope change of the inversion decreases. As a result, it can in this case be easier to detect the fiduciary point in the original pulse profile. An example is P3, the iliac reflection, in the left and right panels of Fig. 7.17 (teal vertical line in rear section of pulse in both cases). While in the right panel the peak of P3 is easily identified, in

7  Pulse Decomposition Analysis Techniques

95

the left panel a derivative has to be used in the detection. Clearly, different detection approaches have to be used depending on circumstances. Problems arise connecting the detection of the same fiduciary point detected in different differentiation states because the use of derivatives, and the associated need for low-pass filtering or smoothing, introduces time shifts in the same detected peak depending on which differentiation state was used. These time shifts will exhibit as amplitude noise if the threshold condition for choosing one differentiation state versus another is crossed repeatedly. In the case of the detection of P2 there are a number of specific difficulties. Due to its high dynamic amplitude its temporal behavior is also dynamic, that is, it accelerates and decelerates relative to the P3 and P1 component pulses, as indicated in Fig.  7.10. At the same time P1 undergoes the most significant velocity changes, modulating the detectability of P2 significantly as it is more or less embedded in the tail decay of P1 (see component pulse simulations in Fig. 7.11). In our experience the most robust approaches involve using detection windows, involving calculating the average amplitude of a window extending, for example, 40 to 140 ms past P1, to track the P2 evolution, as opposed to attempting to detect the P2 time location directly. Another important consideration concerns the details of the response curve of P2/P1. While in mid-range pressures (90–140 mmHg) the response curve is linear, in the hypotensive and hypertensive range nonlinear corrections have to be implemented based on clinical data. Compensation for the confounder effects described above is another important topic. In the case of AS the challenge is at least twofold; on the one hand the AS score has to be stable if it is to be used in any compensatory calculation. Since it involves an integration over a derivative, care has to be taken that the line shapes are stable, that is, noise-contaminated line shapes are either rejected and/or adequately filtered. On the other hand is the actual compensation calculation. Dividing by an AS factor has proven most effective in mitigating the effects due to changing arterial stiffness. Compensation for heart rate changes can be implemented by narrowing and shifting search windows as component pulses narrow. Care has to be taken that heart rate inputs are filtered and adjustments are very gradual to avoid introducing additional noise.

Benchmarks and Clinical Evidence Completed and ongoing clinical studies [20–22], both published and internal, have sought to validate the PDA model and demonstrate accuracy. These efforts are ongoing to further enhance and refine the approach.

96

M. C. Baruch

CT Systole

165

CT Systole

160 155 150 145 140

0

500

Time (Seconds)

1000

Fig. 7.20  Blood pressure results for a 25 y. m. to the pressor. The immersion period is indicated by the red vertical lines. The subject’s digital pulse shape prior to and during application of the pressor is shown in the adjacent figures

Ice Stimulus Experiments As part of these experiments, subjects were monitored using the CareTaker™ (CT), which is the hardware platform on which the PDA formalism has been implemented before, during and after a 1-min ice–water immersion episode, an intervention that brings about a temporary increase in blood pressure on the order of 15–25 mmHg in the majority of subjects. Figure 7.20 presents an example of the systole response during and after immersion while Figs. 7.21 and 7.22 give examples of the pulse shape change prior to and during application of the pressor. Figure 7.23 displays the blood pressure response as well as the associated pulse shapes of a different subject. While the direction and magnitude of the responses in both cases agree with expectations, there is a difference in the time response to the stimulus between the two subjects that has been observed in other subjects also. The origin of the time delay in some subjects is unclear. Given that peripheral and central blood pressure can track differently at least temporarily—the differential treatment of central vs. peripheral hypertension (CAFE [23] study) is an example—it is conceivable that the ice immersion stimulus applied in the arterial periphery could elicit a delayed or, at least initially, modulated response in the core arteries and therefore in the CT’s response characteristics. Studies involving the simultaneous monitoring of both central and peripheral pressure will be required to further investigate this observation.

Valsalva Experiments A different set of experiments involved the investigation of the CareTaker™ response to the Valsalva maneuver,

7  Pulse Decomposition Analysis Techniques

97

Fig. 7.21  Pulse shape prior to application of pressor

Fig. 7.22  Pulse shape during peak of response

There are four main phases in the Valsalva maneuver [24]. In phase I, there is a transient rise in BP due to increased intrathoracic and intra-abdominal pressure causing mechanical compression of the aorta. In the early part of phase II, reduced preload and reduced stroke volume lead to a fall in cardiac output. Total peripheral resistance then increases, reversing the fall in BP to the point where, in some subjects, mean arterial pressure (MAP) can be at resting MAP level or above at the beginning of phase III. Phase III lasts a few seconds during which time BP falls due to a sudden decrease in intra-thoracic pressure. As part of phase IV, venous return and cardiac output return to normal while peripheral resistance remains high, ­resulting in an overshoot of BP. Figure 7.24 displays results for a 57 y. m. performing a Valsalva by maintaining a pressure in excess of 40 mmHg for 20 s on the distal side of an orifice. All four phases are resolved with expected relative amplitudes.

98

M. C. Baruch

Blood Pressure (mmHg)

180 160

1 Second

1 Second

140

1 Second

1 Second

CareTaker Systole

120

CareTaker Diastole Omron Systole

100

Omron Diastole

1 Second

80 16:28

16:30

16:32

16:34

16:36

Time (hh:mm:ss)

P2P1 parameter (unitless)

Fig. 7.23  Blood pressure results for a 37 y. m. with very strong response to the pressor. The immersion period is indicated by the shaded rectangle. The subject’s digital pulse shape changed significantly as a result of the immersion episode and the associated temporary blood pressure increase

0.35

0.30

0.25

0.20 100

125

150

175

Time (Seconds)

200

225

Fig. 7.24  The four phases recorded in the P2P1 PDA parameter during a Valsalva maneuver

Clinical Comparisons Central Arterial Line Comparisons In these experiments, performed at the Catheterization Laboratory at the University of Virginia Medical Center, the aortic blood pressures of 63 patients undergoing cardiac catheterization were monitored using central line catheters while the

7  Pulse Decomposition Analysis Techniques

99

Fig. 7.25  Overlap of central systolic pressure (red) obtained from catheter signal and P2P1 ratio obtained from PDA analysis of noninvasively obtained arterial signal (black) for patient 38

CareTaker™ system collected pulse line shapes at the proximal phalange of the pollex and an automatic cuff determined brachial blood pressure. While the patient rested in a supine position, the catheter was inserted into the femoral artery and advanced toward the heart through the aorta. As part of the study the catheter was positioned in the aorta at the height of the renal arteries for 90 s under fluoroscopy while the catheter signal was recorded. The CareTaker™ system recorded data throughout the preparation period as well as the 90 s overlap window. Both data streams were time synchronized by matching the recording computer’s time as closely as possible to the laboratory’s central time and matching the beat-to-beat inter-beat interval variability, whose randomness provides a unique time stamped signature. PDA parameters were then extracted, beat by beat, from the noninvasively collected CareTaker™ data and converted to systolic and diastolic blood pressures for comparison with systolic and diastolic blood pressures obtained directly from the catheter data tracings. Figure  7.25 displays an example of the overlap of the PDA pulse parameter P2P1 and the systolic blood pressure recorded by the central catheter, while Figs.  7.26 and 7.27 display the overall correlation comparisons of the study. Peripheral Arterial Line Comparisons As part of a study comparing blood pressures measured with the CareTaker™ and peripheral arterial line at Cooper Hospital, data from 24 adult patients requiring hemodynamic monitoring during major open abdominal surgery were analyzed. Patients were not excluded due to other medical conditions. Measurements were

100

M. C. Baruch

Fig. 7.26  Overall correlation of systolic blood pressures obtained through conversion of PDA parameters from noninvasively obtained arterial pulse signal, and central systole

Fig. 7.27  Overall correlation of diastolic blood pressures obtained through conversion parameters from noninvasively obtained arterial pulse signal, and central diastole

7  Pulse Decomposition Analysis Techniques

101

obtained during general anesthesia in these patients starting with induction. The induction point was chosen because the blood pressure fluctuations and variability typically found during this period provided an opportunity to compare tracking accuracy under baseline and induced controlled dynamic conditions. Figure 7.28 presents an example of an overlap lasting almost 3 h. A total of 3870 comparative data points was obtained from the A-line and CT device for the 30 min time window comparison. For the data set collected during the entire procedure, 58,701 comparative data points were obtained, spanning approximately 114.5  h. Across the 24 subjects, the percentage mean of excluded data was 2.8% (SD: 4.0, range: 0–12.7%) while the median was 1.0%. The correlation between the a-line and the CT device for MAP, systolic, and diastolic were 0.92, 0.86, and 0.91, respectively (p  20% in less than 1 min, were strongly correlated (r = 0.94) and accurately tracked (concordance rate of 100%) when compared with invasively assessed changes. The amplitude of the changes was also accurately estimated, as assessed via Bland– Altman analysis, with low biases (1.5%) and standard deviations (11%) for all three pressures, and >90% of percentage errors falling below the threshold of 20%. This novel approach, which may be used to trigger oscillometric measurements, may help to decrease the duration and clinical consequences of hypotensive events in patients with no arterial line monitoring

Applanation Tonometry Since the 1970s, several studies analyzed pulse waveforms obtained by applanation tonometry and showed similarity with pressure pulse recorded intra-arterially by catheterization [14, 15]. Since then, multiple techniques have been developed to initialize and process the peripheral tonometry waveforms to extract different features related to central blood pressure. Different reviews [27, 40, 41, 84] investigated—through the results of multiple clinical studies (see Table  8.2)—the correlation between PWA features with central blood pressure and clinical outcome. The impact of the device and technique used, as well as the initialization methodology on the accuracy of the results, were reported. Miyashita [27] aimed at providing basic knowledge and information on different techniques used to estimate central blood pressure using applanation tonometry. In this context, he focused his analysis on four different techniques, namely, the generalized transfer functions (GTF) used in the SphygmoCor® device (AtCor Medical

8  Pulse Wave Analysis Techniques

129

Pty Ltd., Australia), the late or second systolic shoulder of the peripheral pressure wave (SBP2) used in the OMRON HEM-9000AI device (OMRON Healthcare, Japan), the N-point moving average (NPMA) used in the BPro® device with the A-PULSE CASP® software (HealthSTATS International Pte Ltd., Singapore), and the simple substitution method used with any carotid tonometric device, when the carotid waveform is used as a direct surrogate of aortic waveform. Although carotid tonometry has been shown to provide waveforms similar to central pressure waveform and only requires initialization without specific transformation, it is not considered as a standard method for clinics [1]. Furthermore, the possible activation of baroreceptors in the carotid artery may affect the measurements. Concerning the radial tonometry techniques, Miyashita’s review concluded that the NPMA technique accuracy has shown not to be superior to the GTF, because individualized optimization is practically impossible with this technique, as it is with a GTF. GTF and SBP2 techniques appeared even in terms of accuracy but only if central and peripheral pressure initializations are unified. Finally, the review revealed the impact of reproducibility of the measurement and suggested to use automated arterial tonometry devices for more accurate results. Cheng et al. [40] who performed a meta-analysis on 22 studies (857 subjects, 1167 measurements), also aimed at assessing the accuracy of the GTF and SBP2 methods to estimate central blood pressure, although they mainly focused on the impact of initialization. In particular, they compared invasive and cuff-based initializations, and concluded that current tonometry-based central blood pressure estimation methods are acceptable when initialized with invasive aortic pressure, but showed evident errors when initialized with brachial sphygmomanometer. Narayan et al. [84] analyzed 164 studies using the different techniques and initialization methods mentioned before (GTF, SBP2, NPMA), but also categorized some of these studies by disease states (hypertension, coronary disease, renal impairment, diabetes) and analyzed their impact on central blood pressure accuracy estimation. Their conclusions validated the previous results on the better accuracy of invasive initialization, and did not show significant differences in central blood pressure estimation between the different groups of patients. In parallel with the previous studies that assessed the accuracy of central blood pressure from a methodological point of view, Cheng et al. [41] evaluated the prognostic potential of PWA based on two independent studies on large populations: the Kinmen study including 1272 individuals with right carotid artery pulse waveforms acquired with a tonometer, and the Cardiovascular Disease Risk Factors Two-­ Township Study including 2221 individuals with central aortic pressure waveform acquired with the SphygmoCor® device. Their findings—based on the correlation between biomarkers derived from the pulse waveform and cardiovascular mortality—suggested that PWA could predict long-term cardiovascular risk.

130

M. Proença et al.

Photoplethysmography The use of PPG waveforms to estimate indicators of central blood pressure or arterial stiffness has been developed using multiple approaches (see Table 8.3). One of the major techniques considered to estimate BP-related features based on the flow-­ pressure relationship is the second derivative of the photoplethysmogram, or acceleration plethysmogram (APG, as previously introduced in section “Photoplethysmography-Specific Features”). Millasseau et  al. [21] reviewed different studies that related the APG features to blood pressure, arterial properties and aging. These are based on the pioneer work developed by investigators in Japan which distinguishes five different waves in the APG signal and analyses the ratios of their amplitudes (see Fig. 8.9b). All ratios (b/a, c/a, d/a, and e/a) were found to correlate with age [37, 42, 46] when applied to large population groups, leading to the definition of a vascular aging index as (b-c-d-e)/a, hypothesized to be useful for screening arteriosclerosis. An alternative definition of this index was proposed by Baek [43] as (b-e)/a when the c- and d-waves are missing. The negative ratio -d/a showed to correlate with the aortic augmentation index and was found to be a potentially useful way of evaluating acute effects of vasoactive agents [36, 37]. The ratio d/a however showed only a mild correlation with blood pressure, and the correlation was even less significant with the b/a ratio [42, 44–46, 81]. The b/a ratio was however found to be a useful index of atherosclerosis and altered arterial distensibility [47]. It was further found that b/a and c/a discriminate independently between subjects with essential hypertension and healthy controls [48]. Based on previous studies, Millasseau et al. [21] concluded that PPG waveform analysis provides a rapid and simple means of assessing vascular tone and arterial stiffness, but warned on the importance of employing appropriate signal conditioning since inappropriate filtering can distort the signal, especially when using second derivative techniques. They also insisted on the fact that the measurement has to be performed under standardized conditions (e.g. subjects have to be at rest when measured). Based on a similar mathematical approach than the APG five-wave method, a recent technique [49] also showed potential for blood pressure monitoring. The recent studies performed with this new technique assessed its accuracy in different contexts—including during induction of general anaesthesia—and showed promising results for continuous blood pressure monitoring and detection of acute blood pressure changes [50–52].

The Future of the Technology Two different cuffless technologies aiming at the analysis of the pulse waveform have been presented in this chapter. The first is based on the local measurement of peripheral arterial pressure (applanation tonometry) and the second is based on the measurement of the variation of the local blood volumes in the tissues using optical

8  Pulse Wave Analysis Techniques

131

measurements (photoplethysmography). The objectives of these two techniques are the estimation of central values based on peripheral measurements by means of PWA techniques. This section presents the directions where we think the research on PWA is heading towards and discusses the associated challenges that will be faced. As any prediction of the future, the discussed developments are certainly approximate. However, the presented material represents a likely scenario that opens a window into the future of this technology. The main challenges in applanation tonometry are the inaccuracies involved with initializations based on oscillometric cuffs [40, 84], and the dependency of the observed signals to the operator [85]. While the problem of initialization remains difficult to solve, new systems and methods need to be developed to minimize the dependency on the operator. Such systems would make possible and facilitate the comparison of results across different studies resulting from different measurement campaigns. Although tonometer embedded in wristwatchlike supports exist for measurements at the radial artery, they remain intrinsically limited by the difficulty in reliably measuring pressure waveforms given the small size of the location of interest. In contrast, the positioning of PPG probes at the fingertip is easy, independent on the operator, and reproducible. However, the analysis of pulse waveform based on PPG measurements is more challenging due to the facts that it is a more indirect observation of the central hemodynamic parameters and that the underlying mechanisms generating the observed signal are only partially understood. There are a certain number of challenges that have to be taken care off in order to obtain relevant and repeatable results. The first challenge concerns the sensors. Optical measurements provide information about the variations in absorption of the blood volumes that are present in the tissues along the optical light path. In order to obtain usable signals, the system has to provide an ad hoc contact between the sensor and the skin. Adequate optical signal quality requires that a pressure is applied by the sensor onto the skin surface. The blood volumes that are in the vicinity of the sensor are affected by the applied pressure (especially the venous component) and thus affects the observed signal [19]. The challenge is therefore to design sensors such that these signal modifications are reproducible, or to develop methods that are able to adapt to the signal variability induced by pressure changes. The second challenge is related to the fact that the measurement site is located at the periphery of the arterial tree and will therefore be affected by changes such as vasomotion and the control of the autonomic nervous system. New methods have to be developed to consider their impacts on the sensor signals in order to permit the estimation of central hemodynamic features. Finally, the last identified challenge is the inter-­subject variability: the observed PPG signals are impacted by the morphology and physiology of the subject. The challenge consists in the identification the impact of these inter-subject differences on the observed signal and to develop methods aiming in either adapting the analysis methods to the specificities of the subject or to discard them in order to recover hidden central hemodynamic characteristics such as central blood pressure.

132

M. Proença et al.

The second direction for the development of PWA concerns the development of unobtrusive systems allowing the extraction of relevant information about the cardiovascular system. The estimated physiological parameters will surely exhibit a degraded accuracy when compared to those of reference systems. Nonetheless, they will offer the tremendous advantage of permitting a long-term acquisition of hemodynamic features in an unobtrusive manner. A certain number of challenges remain open considering the application of this technology for long-term noninvasive measurements. These challenges can be clustered into three main categories: (1) the measurement system, (2) the exogenous components in the acquired signals and (3) the validation of long-term measurements. The first challenge, the measurement system, consists in the development of a system that is able to operate in an “out-of-­the-lab” environment. This means that such system has to be unobtrusive, easy to use, and has to obtain measurements with a sufficient quality to apply PWA algorithms. The development of such system will require an important amount of work in the design of ergonomic solutions. The second challenge consists in the determination, the identification and the separation of endogenous components (that are related to the function of the cardiovascular system) from exogenous components. The endogenous components are related directly to the central hemodynamic features (e.g. diastolic and systolic blood pressure, augmentation index) that PWA aims to identify. The exogenous components are of two types. The first is related to the non-central measurement location, whose signals will be subject to variations due to physiological and physical factors such as the control of the autonomic nervous system, the influence of gravity, or vasomotion. These variations will affect the local measurement and lead to erroneous estimation of the central values if their effects are not corrected for. The second type of exogenous components is due to other factors, such as motion, sensor pressure, that can affect the estimation of central values. Finally, the third challenge that has to be faced is the problem of the validation of the measurement in “out-of-the-lab” conditions. The development of a standalone system for long term measurements opens the ability to continuously measure hemodynamic parameters, but no reference system exists for the validation of such a system (the current reference for long-term measurement is the oscillometric cuff sphygmomanometer, limited by its accuracy, its incompatibility with continuous monitoring and its obtrusiveness, particularly at night).

Conclusion This chapter presents an overview of the techniques aiming at extracting relevant haemodynamic parameters from the analysis of pulse waves. Early developments were based on distally obtained pressure measurements (applanation tonometry) generally obtained at the radial location. These measurement techniques are available and routinely used in clinical settings. However, the field of research in this domain remains open and some challenges remain to be solved.

8  Pulse Wave Analysis Techniques

133

Recently two factors have opened new opportunities for research and development of new PWA solutions and products. The first factor is the recent development of PPG-based wearable systems that enable the acquisition of physiological signals in ambulatory conditions such as the daily life. The second factor is the recent transposition of the concepts highlighted in tonometry to optical measurements. The fusion of these two domains opens great opportunities for the measurement and the understanding of haemodynamic variations in the early detection and follow-up of pathological conditions.

References 1. Nichols W, O’Rourke M, Vlachopoulos C.  McDonald’s blood flow in arteries: theoretical, practical and clinical principles. Boca Raton, FL: CRC; 2011. 2. Marey EJ.  Physiologie médicale de la circulation du sang basée sur l'étude graphique. A. Delahaye; 1863. 3. Mahomed FA.  The physiology and clinical use of the sphygmograph. Med Times Gazette. 1872;1:62. 4. Wellcome Collection (image credit). 2018. https://wellcomecollection.org/works/x92jruvr. Accessed Oct 2018. 5. Riva-Rocci S. Un nuovo sfigmomanometro. Gazzetta medica di Torino. 1896;47:981–1001. 6. Shevchenko YL, Tsitlik JE. 90th anniversary of the development by Nikolai S. Korotkoff of the auscultatory method of measuring blood pressure. Circulation. 1996;94(2):116–8. 7. Salvi P. Pulse waves. How vascular hemodynamics affects blood pressure. Berlin: Springer; 2017. 8. Tiru B, Bloomstone JA, McGee WT. Radial artery cannulation: a review article. J Anesth Clin Res. 2012;3(5):1000209. 9. Mitchell GF. Arterial stiffness and wave reflection: biomarkers of cardiovascular risk. Artery Res. 2009;3(2):56–64. 10. Vlachopoulos C, Aznaouridis K, Stefanadis C.  Prediction of cardiovascular events and all-­ cause mortality with arterial stiffness: a systematic review and meta-analysis. J  Am Coll Cardiol. 2010;55(13):1318–27. 11. McEniery CM, Cockcroft JR, Roman MJ, et al. Central blood pressure: current evidence and clinical importance. Eur Heart J. 2014;35(26):1719–25. 12. Roman MJ, Devereux RB, Kizer JR, et al. Central pressure more strongly relates to vascular disease and outcome than does brachial pressure: the Strong Heart Study. Hypertension. 2007;50(1):197–203. 13. Drzewiecki GM, Melbin J, Noordergraaf A.  Arterial tonometry: review and analysis. J Biomech. 1983;16(2):141–52. 14. Chen CH, Ting CT, Nussbacher A, et al. Validation of carotid artery tonometry as a means of estimating augmentation index of ascending aortic pressure. Hypertension. 1996;27(2):168–75. 15. Kelly R. Noninvasive carotid pressure wave registration as an indicator of ascending aortic pressure. J Vasc Med Biol. 1989;1:241–7. 16. Matthys K, Verdonck P.  Development and modelling of arterial applanation tonometry: a review. Technol Health Care. 2002;10(1):65–76. 17. Pauca AL, O’Rourke MF, Kon ND. Prospective evaluation of a method for estimating ascending aortic pressure from the radial artery pressure waveform. Hypertension. 2001;38(4):932–7. 18. Salvi P, Lio G, Labat C, et al. Validation of a new non-invasive portable tonometer for determining arterial pressure wave and pulse wave velocity: the PulsePen device. J  Hypertens. 2004;22(12):2285–93.

134

M. Proença et al.

19. Reisner A, Shaltis PA, McCombie D, Asada HH. Utility of the photoplethysmogram in circulatory monitoring. Anesthesiology. 2008;108(5):950–8. 20. Millasseau SC, Guigui FG, Kelly RP, et  al. Noninvasive assessment of the digital volume pulse: comparison with the peripheral pressure pulse. Hypertension. 2000;36(6):952–6. 21. Millasseau SC, Ritter JM, Takazawa K, Chowienczyk PJ. Contour analysis of the photoplethysmographic pulse measured at the finger. J Hypertens. 2006;24(8):1449–56. 22. Allen J.  Photoplethysmography and its application in clinical physiological measurement. Physiol Meas. 2007;28(3):R1. 23. Sazonov E, Neuman MR, editors. Wearable sensors: fundamentals, implementation and applications. Amsterdam: Elsevier; 2014. 24. Westerhof N, Stergiopulos N, Noble MI.  Snapshots of hemodynamics: an aid for clinical research and graduate education. Cham: Springer Science & Business Media; 2010. 25. Murgo JP, Westerhof N, Giolma JP, ltobelli SA. Aortic input impedance in normal man: relationship to pressure wave forms. Circulation. 1980;62(1):105–16. 26. Karamanoglu MMAR, O’Rourke MF, Avolio AP, Kelly RP.  An analysis of the relationship between central aortic and peripheral upper limb pressure waves in man. Eur Heart J. 1993;14(2):160–7. 27. Miyashita H.  Clinical assessment of central blood pressure. Curr Hypertens Rev. 2012;8(2):80–90. 28. London G, Guerin A, Pannier B, Marchais S, Benetos A, Safar M. Increased systolic pressure in chronic uremia. Role of arterial wave reflections. Hypertension. 1992;20(1):10–9. 29. Hausvater A, Giannone T, Sandoval YHG, et al. The association between preeclampsia and arterial stiffness. J Hypertens. 2012;30(1):17–33. 30. Khalil A, Jauniaux E, Harrington K. Antihypertensive therapy and central hemodynamics in women with hypertensive disorders in pregnancy. Obstet Gynecol. 2009;113(3):646–54. 31. Elgendi M.  On the analysis of fingertip photoplethysmogram signals. Curr Cardiol Rev. 2012;8(1):14–25. 32. Chua CP, Heneghan C. Continuous blood pressure monitoring using ECG and finger photoplethysmogram. In: Engineering in Medicine and Biology Society, 2006. EMBS’06. 28th Annual International Conference of the IEEE; 2006. p. 5117–5120. 33. Padilla JM, Berjano EJ, Saiz J, et  al. Assessment of relationships between blood pressure, pulse wave velocity and digital volume pulse. In: Computers in cardiology, 2006. Lyon: IEEE; 2006. p. 893–6. 34. Wang L, Pickwell-MacPherson E, Liang YP, Zhang YT. Noninvasive cardiac output estimation using a novel photoplethysmogram index. In: Proceedings of 31st IEEE EMBS, 2009. p. 1746–1749. 35. Millasseau SC, Kelly RP, Ritter JM, Chowienczyk PJ. Determination of age-related increases in large artery stiffness by digital pulse contour analysis. Clin Sci. 2002;103(4):371–7. 36. Takazawa K. Clinical usefulness of the second derivative of a plethysmogram (acceleration plethysmogram). J Cardiol. 1993;23:207–17. 37. Takazawa KTN, Fujita M, Matsuoka O, et  al. Assessment of vasocative agents and vascular aging by the second derivative of photoplethysmogram waveform. Hypertension. 1998;32:365–70. 38. Mukkamala R, Hahn JO. Toward ubiquitous blood pressure monitoring via pulse transit time: predictions on maximum calibration period and acceptable error limits. IEEE Trans Biomed Eng. 2018;65(6):1410–20. 39. Mukkamala R, Hahn JO, Inan OT, et  al. Toward ubiquitous blood pressure monitoring via pulse transit time: theory and practice. IEEE Trans Biomed Eng. 2015;62(8):1879–901. 40. Cheng H-M, Lang D, Tufanaru C, Pearson A.  Measurement accuracy of non-invasively obtained central blood pressure by applanation tonometry: a systematic review and meta-­ analysis. Int J Cardiol. 2013;167(5):1867–76. 41. Cheng HM, Chuang SY, Wang JJ, et  al. Prognostic significance of mechanical biomarkers derived from pulse wave analysis for predicting long-term cardiovascular mortality in two population-based cohorts. Int J Cardiol. 2016;215:388–95.

8  Pulse Wave Analysis Techniques

135

42. Bortolotto LA, Blacher J, Kondo T, et al. Assessment of vascular aging and atherosclerosis in hypertensive subjects: second derivative of photoplethysmogram versus pulse wave velocity. Am J Hypertens. 2000;13:165–71. 43. Baek HJ, Kim JS, Kim YS, et al.. Second derivative of photoplethysmography for estimating vascular aging. In: Information technology applications in biomedicine, 2007. ITAB 2007. 6th International Special Topic Conference on IEEE; 2007. p. 70–72. 44. Hashimoto J, Chonan K, Aoki Y, et al. Pulse wave velocity and the second derivative of the finger photoplethysmogram in treated hypertensive patients: their relationship and associating factors. J Hypertens. 2002;20(12):2415–22. 45. Hashimoto J, Watabe D, Kimura A, et al. Determinants of the second derivative of the finger photoplethysmogram and brachial-ankle pulse-wave velocity: the Ohasama study. Am J Hypertens. 2005;18(4):477–85. 46. Takada H, Washino K, Harrell JS, Iwata H. Acceleration plethysmography to evaluate aging effect in cardiovascular system. Using new criteria of four wave patterns. Med Prog Technol. 1996;21(4):205–10. 47. Imanaga I, Hara H, Koyanagi S, Tanaka K. Correlation between wave components of the second derivative of plethysmogram and arterial distensibility. Jpn Heart J. 1998;39:775–84. 48. Simek J, Wichterle D, Melenovsky V, et al. Second derivative of the finger arterial pressure waveform: an insight into dynamics of the peripheral arterial pressure pulse. Physiol Res. 2005;54(5):505. 49. Proença M, Solà J, Lemay M, Verjus C. Method, apparatus and computer program for determining a blood pressure value. WO 2016 138965 A1, 9th of September 2016; 2016. 50. Ghamri Y, Proença M, Hofmann G, et al. Using pulse oximetry waveform analysis to track changes in blood pressure during anesthesia induction. 2019. Manuscript in preparation. 51. Solà J, Proença M, Braun F, et al. Continuous non-invasive monitoring of blood pressure in the operating room: a cuffless optical technology at the fingertip. Curr Direct Biomed Eng. 2016;2(1):267–71. 52. Solà J, Proença M, Schoettker P, et al. Blood pressure monitoring using a smartphone camera: performance of the OBPM technology. In: IEEE Conference on Biomedical and Health Informatics; 2018. 53. Pauca AL, Kon ND, O’Rourke MF.  The second peak of the radial artery pressure wave represents aortic systolic pressure in hypertensive and elderly patients. Br J  Anaesth. 2004;92(5):651–7. 54. Soderstrom S, Nyberg G, O'Rourke MF, et al. Can a clinically useful aortic pressure wave be derived from a radial pressure wave? Br J Anaesth. 2002;88:481–8. 55. Davies JI, Band MM, Pringle S, et  al. Peripheral blood pressure measurement is as good as applanation tonometry at predicting ascending aortic blood pressure. J  Hypertens. 2003;21(3):571–6. 56. Sharman JE, Lim R, Qasem AM, et al. Validation of a generalized transfer function to noninvasively derive central blood pressure during exercise. Hypertension. 2006;47(6):1203–8. 57. Chen CH, Nevo E, Fetics B, et al. Estimation of central aortic pressure waveform by mathematical transformation of radial tonometry pressure: validation of generalized transfer function. Circulation. 1997;95(7):1827–36. 58. Hope SA, Tay DB, Meredith IT, Cameron JD. Use of arterial transfer functions for the derivation of aortic waveform characteristics. J Hypertens. 2003;21(7):1299–305. 59. Hope SA, Meredith IT, Tay D, Cameron JD. Generalizability’ of a radial-aortic transfer function for the derivation of central aortic waveform parameters. J Hypertens. 2007;25(9):1812–20. 60. Rajani R, Chowienczyk P, Redwood S, et al. The noninvasive estimation of central aortic blood pressure in patients with aortic stenosis. J Hypertens. 2008;26(12):2381–8. 61. Hickson SS, Butlin M, Mir FA, et al. The accuracy of central SBP determined from the second systolic peak of the peripheral pressure waveform. J Hypertens. 2009;27(9):1784–8. 62. Hope SA, Meredith IT, Cameron JD.  Effect of non-invasive calibration of radial waveforms on error in transfer-function-derived central aortic waveform characteristics. Clin Sci. 2004;107(2):205–11.

136

M. Proença et al.

63. Smulyan H, Siddiqui DS, Carlson RJ, et al. Clinical utility of aortic pulses and pressures calculated from applanated radial-artery pulses. Hypertension. 2003;42:150–5. 64. Zuo JL, Li Y, Yan ZJ, et  al. Validation of the central blood pressure estimation by the SphygmoCor system in Chinese. Blood Press Monit. 2010;15(5):268–74. 65. Ding FH, Fan WX, Zhang RY, et al. Validation of the noninvasive assessment of central blood pressure by the SphygmoCor and Omron devices against the invasive catheter measurement. Am J Hypertens. 2011;24(12):1306–11. 66. Takazawa K, Kobayashi H, Shindo N, et  al. Relationship between radial and central arterial pulse wave and evaluation of central aortic pressure using the radial arterial pulse wave. Hypertens Res. 2007;30(3):219. 67. Williams B, Lacy PS, Yan P, et al. Development and validation of a novel method to derive central aortic systolic pressure from the radial pressure waveform using an N-point moving average method. J Am Coll Cardiol. 2011;57(8):951–61. 68. Garcia-Ortiz L, Recio-Rodríguez JI, Canales-Reina JJ, et al. Comparison of two measuring instruments, B-pro and SphygmoCor system as reference, to evaluate central systolic blood pressure and radial augmentation index. Hypertens Res. 2012;35(6):617. 69. Westerhof BE, Guelen I, Stok WJ, et al. Individualization of transfer function in estimation of central aortic pressure from the peripheral pulse is not required in patients at rest. J Appl Physiol. 2008;105(6):1858–63. 70. Shih YT, Cheng HM, Sung SH, et  al. Quantification of the calibration error in the transfer function-derived central aortic blood pressures. Am J Hypertens. 2011;24(12):1312–7. 71. Cheng HM, Wang KL, Chen YH, et al. Estimation of central systolic blood pressure using an oscillometric blood pressure monitor. Hypertens Res. 2010;33(6):592. 72. Van Bortel LM, Balkestein EJ, van der Heijden-Spek JJ, et al. Non-invasive assessment of local arterial pulse pressure: comparison of applanation tonometry and echo-tracking. J Hypertens. 2001;19:1037–44. 73. Karamanoglu MMAR, Feneley MP. Derivation of the ascending aortic-carotid pressure transfer function with an arterial model. Am J Phys Heart Circ Phys. 1996;271(6):H2399–404. 74. Kelly R, Fitchett D.  Noninvasive determination of aortic input impedance and external left ventricular power output: a validation and repeatability study of a new technique. J Am Coll Cardiol. 1992;20(4):952–63. 75. Otsuka T, Kawada T, Katsumata M, Ibuki C. Utility of second derivative of the finger photoplethysmogram for the estimation of the risk of coronary heart disease in the general population. Circ J. 2006;70(3):304–10. 76. Otsuka T, Kawada T, Katsumata M, et al. Independent determinants of second derivative of the finger photoplethysmogram among various cardiovascular risk factors in middle-aged men. Hypertens Res. 2007;30(12):1211. 77. Kawada T, Otsuka T.  Factor structure of indices of the second derivative of the finger Photoplethysmogram with metabolic components and other cardiovascular risk indicators. Diabetes Metab J. 2013;37(1):40. 78. Chan GS, Middleton PM, Celler BG, et al. Automatic detection of left ventricular ejection time from a finger photoplethysmographic pulse oximetry waveform: comparison with Doppler aortic measurement. Physiol Meas. 2007;28(4):439. 79. Tusman G, Acosta CM, Pulletz S, et al. Photoplethysmographic characterization of vascular tone mediated changes in arterial pressure: an observational study. J Clin Monit Comput. 2018. 80. Coutrot M, Joachim J, Dépret F, et al. Noninvasive continuous detection of arterial hypotension during induction of anaesthesia using a photoplethysmographic signal: proof of concept. Br J Anaesth. 2019;122(5):605–12. 81. Avolio A.  The finger volume pulse and assessment of arterial properties. J  Hypertens. 2002;20(12):2341–43. 82. Avolio AP, Butlin M, Walsh A. Arterial blood pressure measurement and pulse wave analysis—their role in enhancing cardiovascular assessment. Physiol Meas. 2010;31(1):R1.

8  Pulse Wave Analysis Techniques

137

83. Wilkinson IB, MacCallum H, Flint L, et al. The influence of heart rate on augmentation index and central arterial pressure in humans. J Physiol. 2000;525(1):263–70. 84. Narayan O, Casan J, Szarski M, et al. Estimation of central aortic blood pressure: a systematic meta-analysis of available techniques. J Hypertens. 2014;32(9):1727–40. 85. Hirata K, Kawakami M, O'Rourke MF.  Pulse wave analysis and pulse wave velocity.  Circ J. 2006;70(10):1231–9. 86. Cloud GC, Rajkumar C, Kooner J, et al. Estimation of central aortic pressure by SphygmoCor® requires intra-arterial peripheral pressures. Clin Sci. 2003;105(2):219–225. 87. Gunarathne A, Patel JV, Hughes EA, Lip GY. Measurement of stiffness index by digital volume pulse analysis technique: clinical utility in cardiovascular disease risk stratification. Am J Hypertens. 2008;21(8):866–72.

Chapter 9

Machine Learning Techniques Xiaorong Ding

Abstract  Driven by the exponential growth in the computational power and the increasing size of the collected data sets, there has been growing interest in using data-driven approaches based on machine learning techniques to resolve the problems and overcome the challenges that facing the area of cuffless blood pressure measurement. Compared with the theory-driven analytical approaches, the machine learning method is very promising with its ability to learn the function of the complex system if the model is trained well, and to address the latent affecting factors that cannot be considered in the analytical model. This chapter first addresses the motivation of employing data-driven method, then provides a brief introduction of machine learning  method for cuffless blood pressure estimation. It then presents some of the state-of-the-art examples and applications of such technology and finally discusses the outlook of its future development. Keywords  Machine learning · Cuffless blood pressure · Data-driven method · Analytical model · Pulse transit time · Blood pressure indicator · Learning model · Deep learning

Introduction Ever since 1980s, various research efforts have been devoted to the development of cuffless techniques for blood pressure (BP) measurement, which is to obtain the continuous BP reading indirectly and continuously without the traditional inflatable cuff. The classical way is to calibrate the indicators that can reveal the BP changes to the BP values, to get the indirect BP estimation with the measurement of the indicators. Those indicators are usually the features that are available from non-­ invasive cardiovascular signal recordings. The cardiac signals can be vital signs such as the electrocardiogram (ECG) and the photoplethysmogram (PPG). Pulse X. Ding (*) Institute of Biomedical Engineering, Department of Engineering Science, University of Oxford, Oxford, UK e-mail: [email protected] © Springer Nature Switzerland AG 2019 J. Solà, R. Delgado-Gonzalo (eds.), The Handbook of Cuffless Blood Pressure Monitoring, https://doi.org/10.1007/978-3-030-24701-0_9

139

140

X. Ding

transit time (PTT), for example, is one of the most common technical indicators and has been widely used for the estimation of the cuffless BP. To get the indirect BP from the indicator(s), there is a need to map the indicator(s) to the BP via calibration algorithm that can model the relationship between the indicator(s) and BP. In classical calibration, the PTT is projected to the BP values through explicit theoretical expressions based on the underlying physiological mechanism (e.g. the principle of pulse wave velocity propagation (PWV)). The readers may refer to Chap. 7 for more details of the PWV techniques. When there are more BP indicators being involved, and the interpretations of the relationship between these indicators and BP are not always straightforward, BP is regressed on these indicators for a set number of known sample subjects, and the regression models are used to estimate the BP values for an unknown subject from its measured indicators. However, there are too many variables relating to the BP changes, and the relationship between each variable and BP are too complicated to be expressed via simple physiologically based mathematic model. As data sets get bigger and computers become more powerful, the data-driven approaches based on machine learning techniques have been attempted to model the relationship between BP indicator variables and BP.  In this chapter, we will first address the motivation of employing data-driven method, following by the brief introduction of machine learning for cuffless BP estimation. Finally, we will illustrate some of the state-of-the-art examples and applications of such technology and present the outlook of its future development.

Motivation of Using Data-Driven Method Model-Based Method Most PTT-based cuffless BP measurement approaches are based on the principle of PWV recording through the Moens–Korteweg (M-K) equation [1]: PWV =

Eh ρD

(9.1)

which correlates PWV with the modulus of elasticity of the artery E, the thickness of the arterial wall h, the diameter of the artery D and the blood density ρ. Since the changes of ρ are negligible, ρ can be assumed to be constant. Then, E is the major factor that PWV relies on, among the other three variables (i.e., E, h, and D). Further according to Hughes et  al. [2], the elastic modulus E is exponentially correlated with the mean distending pressure P as given by Eq. (9.2):

9  Machine Learning Techniques

141

E = E 0 eγ P



(9.2)

where E0 is the zero-pressure modulus, and γ a constant typically between 0.016 and 0.018. Both E0 and γ are site and individual dependent. The combination of Eqs. (9.1) and (9.2) thus gives the relationship between PWV and arterial BP as given by Eq. (9.3): PWV =

hE0 eγ P . ρD

(9.3)

As PWV is reciprocally related with PTT as given by Eq. (9.4): PWV =



L PTT

(9.4)

where L represents the transit distance of the pressure waveform traveling between two arterial sites, the arterial pressure P can be derived accordingly with Eq. (9.4) substituted in Eq. (9.3), as given by Eq. (9.5). P=

ρ L2 D  1  −2 ln PTT + ln  γ hE0 

(9.5)

Therefore, with initial calibration of PTT to BP, beat-to-beat BP can be theoretically derived from beat-to-beat PTT that can measured from each cycle of the cardiac pulse signals. The Relationship Between PTT and BP Since Weltman’s seminal work on PTT in 1964 [3], a considerable number of studies have focused on the application of PTT in psychophysiological research in 1970s, as PTT is an indirect continuous measure of BP variation. At that period, the non-invasive continuous recording of arterial BP was not yet available due to many unsolved physiological and technical problems [4]. During 1970s–1980s, there have been active research efforts on the relationship between PTT and BP, aiming to investigate whether PTT is feasible to evaluate BP changes [3–22]. Since 2000, the study on PTT for BP estimation has attracted more and more attention [23]. The fundamental of the applications of PTT for BP measurement is its relationship with BP.  The methods and results of the studies about PTT-BP relationship are briefly summarized in Table 9.1. Those studies have investigated the relationship between PTT and BP with experimental stressors or clinical interventions to elicit changes in BP. At the early stage of PTT study in 1980s, some investigators pointed out to use arterial PWV as a continuous measure of BP changes, and they carried out experiment to assess the relationship between PTT and mean BP (MBP). Gribbin et  al. [6] measured

142

X. Ding

Table 9.1  Summary of the studies about the relationship between PTT and BP Correlation coefficient SBP DBP MBP – – 0.92– 0.99 Rest, cold pressor, pornographic Non-invasive −0.85 −0.30 – movie, unsignalled shock Inter-arterial −0.76 −0.44 – Sphygmomanometer −0.49 −0.03 – Unsignalled shock, Stroop color–word interference test, isometric handgrip Rest, paced respiration, and Intra-arterial −0.62 −0.32 −0.48 mental arithmetic Rest Sphygmomanometer −0.95 0.09 – Lower body negative pressure Intra-arterial 0.80 – – Bicycle test Sphygmomanometer >0.80 – – Intra-arterial −0.62 −0.14 −0.28 Drug administration (glyceryl trinitrate, angiotensin II, norepinephrine, salbutamol) Exercise Sphygmomanometer −0.92 −0.38 –

Reference N Condition [6] 26 Externally applied pressures [8] [11]

14 7 94

[12]

4

[17] [18] [19] [20]

44 14 18 20

[22]

41

BP measurement Intra-arterial

brachial-­to-­radial PWV in humans and correlated it with MBP changes. The PWV was found to change linearly with mean pressure. Similar study by Steptoe et al. [7] reported that the changes in PTT was dependent on arterial pressure changes which were caused by manoeuvres including mental arithmetic and isometric exercise. The PTT that calculated from ECG and peripheral pulse, was linearly correlated with mean arterial pressure, with correlation coefficients varying between −0.91 and −0.98 for five subjects. After that, there have been studies examining the correlation of PTT with systolic BP (SBP) and diastolic BP (DBP). As can be seen from Table 9.1, PTT had stronger correlation with SBP than DBP for almost all the studies [8, 9, 11–13, 17]. In addition, although the correlations between PTT and BP depend upon, how PTT and BP were measured, the method of recording and BP changes induction method, and whether the subjects displayed differences in BP reactivity, the results of those studies reveal that: although most of the studies indicate the changes in PTT can reliably track the variations in BP, which support the usage of PTT as an indirect measure of BP change, majority of the studies show that the correlation coefficients between PTT and BP vary from individual to individual, which means that in any one subject an absolute change in arterial pressure is not exclusively due to change in PTT.

9  Machine Learning Techniques

143

Modelling Methods of PTT-Based BP Estimation Based on the theoretical relationship between PTT and BP as mentioned earlier, as well as their experimental or empirical relationship, a wide variety of research studies have been conducted to model PTT with BP to achieve BP estimation. Table 9.2 summarizes the studies about PTT-based cuffless BP measurement in the recent two decades. The findings of Table 9.2 indicate that the PTT-BP models can be roughly classified into three types: (1) physiological models derived based on M-K equation or B-H equation, (2) linear or nonlinear regression models, which account for more than 70% of these studies, and (3) empirical model with parameters identified from a population. The studies of the relationship between BP-signatures and BP, as well as the models that are used to represent their relationships, demonstrate that the association of the BP-signature and BP is far more complicated than the simple liner or nonlinear regression model. Taking PTT for example, for different individuals it either has positive or negative, strong or weak correlation with BP.  Furthermore, most of the analytical models work on the premise of some assumptions. For example, the M-K equation assumes that the ratio of the wall thickness to the vessel radius and blood density keep constant, and the artery wall is presumed to be isotropic and there are volumetric changes between the artery wall and the pulse pressure. Obviously, these assumptions do not conform to the real situation. Furthermore, intrinsic or extrinsic factors like age, temperature, mental stress, and different behaviour pattern would also affect the BP changes. Theoretically, all the factors that may impact the variations of BP should be considered into the analytical model to achieve accurate BP estimate. But in practice, it is impossible to include all these elements with engineering implementation, such as the wearable systems. All these disadvantages of the analytical model drive the application of the data-driven learning method to build the function for the complicated relationship between the indicators and BP. The method will accordingly be elaborated in the following section.

Data-Driven Approaches Based on Machine Learning In a nutshell, machine learning is a subset of artificial intelligence in the field of computer science. It often uses statistical techniques to give computers the ability to “learn” with data, without being explicitly programmed [34]. Given the exponential growth in the computational power and the increasing size of the collected data sets, there has been growing interest in using big data analysis techniques to resolve the problems and overcome the challenges that facing the area of cuffless BP measurement [35]. The data-driven based machine learning method is thus very promising with its ability to learn the function of the complex system if the model is trained well, and to address the latent affecting factors that cannot be considered in the analytical model.

SBP = a1∗PTT + b1∗HR + c1 DBP = a2∗PTT + b2∗HR + c2

Wong (2009, China) [22]

Cattivelli (2009, US) [28]

2

SBP = a∗lnPTT + b SBP = a∗L/PTT + b SBP = a∗(L/PTT)2 + b SBP = a1∗PTT + b1 DBP = a2∗PTT + b2

 PTT0  SBP = DBP + PP0 ⋅    PTT 

2 PTT0 1  PTT0  DBP = MBP0 + ln − PP0 ⋅   γ PTT 3  PTT 

a BP = +b PTT 2

Muehlsteff (2006, Germany) [19]

Poon (2005, China) [27]

Fung (2004, Canada) [26]

First author (year, country) 2.1.3. Method (model) Young (1995, US) SBP = a1/PTT + b1 [24] DBP = a2/PTT + b2 Chen (2000, Japan) [25] 2 SBP = SBP 0 − ( PTT − PTT0 ) γ PTT0

2

18 healthy

85 (39 hypertensive)

22 patients

0.6 ± 9.8

−0.08 ± 11.32

0.0 ± 2.9 Half year: 2.1 ± 7.3 −0.07 ± 4.96

0.9 ± 5.6

2.1.6. Accuracy (mmHg) SBP DBP −0.01 −0.37 (−14.9–14.8) (−29.0–28.2) RMSE: 3.70 ± 1.85

RMSE: 7.5 RMSE: 6.9 RMSE: 7.3 Oscillometric BP (exercise, 41 healthy (14 for 0.0 ± 5.3 half year)/oscillometric BP half year) Half year: 1.4 ± 10.2 MIMIC database (invasive 25 ICU patients −0.41 ± 7.77 BP, 1 h) / invasive BP

Cuff BP (physical test)/ cuff BP

Cuff BP (initial calibration)/average of auscultatory and oscillometric BP

Cuff sphygmomanometry (10 min)/cuff sphygmomanometry

2.1.4. Calibration (method/ interval)/reference method 2.1.5. Subjects Oscillometric BP (5 min)/ 35 patients invasive radial BP 20 patients Intermittent BP from invasive BP (5 min)/ invasive radial arterial BP

Table 9.2  Summary of the studies about PTT based cuffless BP measurement

144 X. Ding

Liu (2018) [33]

Huynh (2018) [32]

Ding (2016, China) [31]

Chen (2012, Singapore) [30]

Gesche (2011, Germany) [29]

kij

−

PWVd

kij

PWVs

PIR 0 PIR 2

2

MBP = HR∗(k1∗DRPPG TD + b1) PP = MBP∗(k2∗t/HP + b2) SBP = MBP + 2/3PP; DBP = MBP-1/3PP

 D  DBP = DBP0 + B ⋅   ln (1 + K ( Z max 0 − Z min ) )  PTT 

2

 D  SBP = DBP0 + B ⋅   ln (1 + K ( Z max 0 − Z max ) )  PTT 

DBP = DBP0 ⋅

PIR 0  PTT0  SBP = DBP0 ⋅ + PP0 ⋅   PIR  PTT 

DBP = bij e

SBP = bij e

−

BPPTT = P1 × PWV × e + P2 × PWV P4 − ( BPPTT ,cal − BPcal )

( p3×PWV )

Finapres BP

Oscillometric BP

Finapres BP

Intra-arterial BP (9 subjects)/intra-arterial BP

Sphygmomanometry (13 subjects, exercise)/ sphygmomanometry

20 healthy

15 healthy

27 healthy

35 healthy

63

SD: 2.85

RMSE: 8.47 ± 0.91

−0.37 ± 5.21

1.49 ± 6.51

SD: 10.10 95% CI: −19.8~19.8

SD: 1.75

RMSE: 5.02 ± 0.73

−0.08 ± 4.06

2.16 ± 6.23

9  Machine Learning Techniques 145

146

X. Ding

Compared with the theory-driven analytical approaches, there is no need for the machine learning method to understand the physiological process underlying the prediction and it is without priori assumption. For the data-driven approaches for the cuffless BP estimation, the data can signal that are informative of the BP changes, for example, the ECG, PPG, or any other biomedical signals that can be collected in a non-invasive and unobtrusive manner. The machine learning method or in particular the deep learning is able to extract meaningful features from these signals, and model the relationship between the learned features and BP with the objective to minimize the difference between the predicated value and the reference. The main feature and value of using the machine learning method for cuffless BP estimation lies in its learning ability. To achieve accurate estimation of cuffless BP with analytical model, it is often necessary to identify correlations between multiple feature inputs and external factors that are rapidly producing millions of data points. The analytical model is then built on past data and relying on the expert knowledge to establish a relationship between the variables. While the machine learning method can learn constantly from the data. It starts with the outcome variables, that is, BP, and then automatically looks for predictor variables and their interactions. When the target output is clear but the important input variables are unknown to make the decision, the machine learning becomes valuable by giving it the goals and then it learns from the data which factors are important in achieving that goal. In comparison to the static analytical model, the machine learning algorithms can constantly improve over time as more data is collected and assimilated.

 heory-Driven Analytical Model Versus the Data-Driven T Learning Model There are three major differences between the theory-driven analytical model and the data-driven learning model, in terms of generalizability, interpretability and reliability. First, the theory-driven method is generalizable, while the learning model is very hard to generalize beyond the original problem. The learning model is often achieved on a training dataset with the objective to solve a very specific problem. Usually for a new problem, new information and datasets that relate to that problem is required. Second, the analytical model is interpretable, but the machine learning model is usually inexplicable. One prominent feature of the learning method is that it is a black box method. The learning is driven by a well-defined objective function, without any prior assumptions and the understanding of the process. The analytical model, on the contrast, usually works under some hypothesis and requires the expert knowledge, which makes it understandable. Third, the analytical is light weight, while the learning model is computationally demanding. The major differences between these two types of methods are listed in Table 9.3.

9  Machine Learning Techniques

147

Table 9.3  Data-driven learning method versus model-based method 1 2 3 4 5

Theory-driven analytical model Generalizable Interpretable Hypothesis Need expert opinion Light weight

Data-driven learning model Hard to generalize beyond the original problem No understanding of the process No need for priori assumptions Learn from data Computationally demanding

 rief Introduction of Machine Learning for Cuffless BP B Prediction Short Historical Overview With the development of the machine learning algorithm, and its successful exploitation various areas such as speech recognition, natural language processing and computer vision, there are more and more efforts towards the field of cuffless BP estimation or prediction. The attempt of machine learning method for the arterial BP estimation started since 2000s. One of the earliest studies as far as we are concerned is the one in published 2005 by Kim and the colleagues [36]. In that study, three features such as the PTT, weight and arm length of each subject, were used as the inputs to a two hidden-­ layer artificial neural network for the estimation of SBP. It has achieved an estimation error of 4.53 ± 2.68 mmHg, which outperformed the method of multiple. There are very few studies that have used machine learning for the cuffless BP estimation until 2011 Monte-Moreno has proposed to use machine learning algorithm to estimate BP from a PPG waveform [37]. It is driven by the idea that there is functional relationship between the morphology of the PPG waveform and the BP. With the energy profile of the PPG signal being extracted as the inputs, several machine learning techniques have been tested, including ridge linear regression, a multilayer perceptron neural network, support vector machines and random forests (Fig. 9.1). The best performance was by means of the Random Forest Tree method, which has resulted a coefficient of determination between the reference and the prediction of 0.91 and 0.89 for SBP and DBP, respectively. The performance also was complying with the Grade B protocol of the BHS. Most of the machine learning based studies use features extracted from PPG signals [38–41], since the PPG signal is the blood volume change that caused mainly by the BP. Almost all the studies have been using machine learning to model the relationship between feature variables and the target variable—BP. Regarding the input features, the hand-designed features from the PPG signal are the most common type of inputs, including the time-domain profiles such as the amplitude, interval, intensity, and area, and the frequency-domain features like the power, peak frequency, and maximum amplitude. There are a few studies using the whole wave-

148

X. Ding

Fig. 9.1  Diagram of the system of machine learning based cuffless BP estimation [37]

form or the whole waveform-based features [37, 41], with most of these studies using both ECG and PPG signals [42–45]. In the early 2010s, the simple regression methods have been commonly used, such as the univariate or multivariate linear regression [44, 46] Later, other algorithms such as support vector regression [39, 47] and random forest [37, 42, 48] have been employed to develop the BP model. With the development of the machine learning algorithm, and its successful application in various areas such as speech recognition, natural language processing and computer vision, researchers began to develop models based on the advanced machine learning method such as deep learning method. For example, multiple neural perceptron and neural network have been employed to develop the nonlinear relationship between the predictors and BP [38, 49]. In the recent years, the long-short-term-memory (LSTM) architecture of a recurrent neural network (RNN) has been studied and achieved better performance than classical machine learning method, due to its advantage of accommodating the multiscale temporal dependency between the sequential raw signal values and the corresponding BP values [50–52]. However, there are few studies involving the use of the real sense of “deep learning” method for estimating cuffless BP. Deep learning often involves the representation learning of the features from raw data rather than using handcrafted features. Although one study mentioned using deep learning [53], it does not contain the representative learning.

How Does Machine Learning Work for Cuffless BP Estimation? While the analytical model-based method maps BP indicators to BP via a generative mathematical model, with some assumptions or hypothesis, the machine learning based regression method is able to predict the BP values by a complicated model

9  Machine Learning Techniques

149

that can be learnt from the data with specific learning algorithm. Simply speaking, the machine learning algorithm works to learn a target function (f) that best maps input variables (X) to an output variable (Y): Y = f(X). The general learning task is usually to make predictions in the future (Y) given new examples of input variables (X). This is also called predictive modelling or predictive analytics and our goal is to make the most accurate predictions possible. For the application of cuffless BP estimation, the handcrafted features, such as PTT and heart rate, or other features that are extracted from cardiac signals such as ECG and PPG signal are used as the input variable for the learning of the relationship between those input variables and the output—BP. More advanced methods, like deep learning, can also be employed to extract representative features from the raw signals. For example, autoencoder is one representative architecture for the abstract feature learning. Figure 9.2 illustrates the machine learning method for the cuffless BP measurement.

Classical Machine Learning Method for Cuffless BP Estimation The indirect estimation of BP from the features or signals is a regression problem by nature. In statistics, the problem of regression is that of learning a function that allows to estimate a certain quantity of interest, the dependent variable, from serval observed variables, known as covariates, features or independent variables. For the case of cuffless BP estimation, we are interested in estimating BP based on its indicators (e.g. the cardiac output, peripheral resistance). The function that models the

Fig. 9.2  Diagram to show the machine learning method for the cuffless BP measurement, where the boxes with red orange indicate the components that can learn from data

150

X. Ding

relationship between the response and the predictors is learnt from training data and can then be used to predict the response for new collected data. Traditional calibration of the PTT-BP model uses only one or two calibration points to get the regression coefficients, which would not be accurate due the reason we explained earlier. With a big training dataset, it is assumed that the coefficients of the regression model would be more accurate. Commonly used classical machine learning methods include linear regression, support vector machine regression, random forest, and adaptive regression, which will be elaborated as below. Linear and Polynomial Regression Linear regression is by far the most simple and popular example of a regression algorithm. The univariate linear regression is the simplest technique used to model the relationship between a single input independent variable (feature variable) X and an output dependent variable BP using a liner model [54]:

BP = β 0 + β1 X + ε

(9.6)

For the application of cuffless BP measurement, there are studies in the 2000s taken PTT as the sole feature variable and modelled the relationship between the PTT and BP with a linear model [55–60]. Usually two calibration values were measured to calibrate the model, rather than collecting a group of data to do the calibration. Since the relationship between PTT and BP is far more complicated than the linear model, the estimations with single linear model is not accurate. The general case is the multiple variable linear regression where a model is created for the relationship between multiple independent inputs variables X1, X2, …, Xn = X and an output dependent variable BP:

BP = β 0 + β T X + ε .

(9.7)

The model by nature is linear such that the output is a linear combination of the input variables. With the awareness that only PTT cannot track BP well, there are other indicators that can feature the BP changes, and there are studies using multiple features, such as PTT, heart rate (HR), and other BP indicators to estimate BP [61, 62]. The advantages of the linear regression are that it is very efficient and useful when the relationship to be modelled is not extremely complex and when there is not enough data, and it is very simple to understand which variable is the most valuable for the estimate or prediction of the output. However, it would not be effective at modelling highly complex non-linear relationship.

9  Machine Learning Techniques

151

Regression Trees and Random Forests Linear regression provides a global model of the variable or process to predict the output, where only one predictive formula is holding over the entire data-space. When the target output has more than one features which have complicated and nonlinear interaction with each other, to develop a sole global model can be very challenging. Even though it can be achieved successfully, it would be very difficult to understand. The alternative approach for nonlinear regression is to subdivide, or partition the space into small regions. In the divided small regions, the interactions are more manageable. Each of the small region is subdivided again until finally the space that can be fitted with simple models. The global model thus consists of two parts: the iterative partition, and a simple model for each element of the partition [54]. Regression trees use the tree to represent the iterative partition. Each of the terminal nodes (leaves) of the tree represents an element of the partition and has attached to it a simple model which works for that element only. A point x belongs to a leaf if x falls in the corresponding cell of the partition, and to figure out which cell we are in, we start at the root node of the tree, and ask a sequence of questions about the features. For classic regression trees, the model in each cell is just a constant estimate of Y. That is, suppose the points (xi, yi), (xj, yj), …, (xn, yn) are the samples belonging to the leaf node. Then our model for the leaf node is the sample mean of the dependent variable in all the leaf nodes: y=

1 n ∑ yi n i =1

Random forest is a collection of decision trees, with the input vector running through multiple decision trees. Ever since its introduction by Breiman in 2001 [63], it has become a very popular learning technique. The biggest advantage of the regression trees is that it is great at learning complex, high non-linear relationship. They usually can achieve high performance, better than simple linear and polynomial regression and often comparable with neural networks [37, 42]. In addition, it is very easy to interpret, with variables and variable values corresponding to nodes can be pinpointed, thus to understand the importance of the feature variables and the process. However, due the nature of training decision trees, they are very prone to overfitting, but this can be overcome by using proper tree pruning and larger random forest ensembles. Support Vector Regression Support vector regression (SVR) is a nonparametric technique, as it relies on the kernel function. In ε-SVR, the set of training data includes both the predictor variables and observed response values. The goal is to find a function f(x) that deviates from yn by a value no more than ε for each training point x, and to be as flat as possible (Fig. 9.3). The idea of SVR is closely related to that of support vector machine

152

X. Ding

Fig. 9.3  A schematic of support vector regression using ε-insensitive loss function [64]

(SVM) for classification. In SVM, a separating hyperplane is expected such that all points are at a certain distance from this plane. If there are points that are too close to the separating hyperplane, a penalty occurs. Similarly, in SVR, a function is desirable such that all points are within a certain distance from this function. Again, if points are outside this distance—the “ϵ-tube”, there will be a penalty.

Deep Learning Method for Cuffless BP Prediction Deep learning is a subfield of machine learning. It uses deep networks with many intermediate layers of artificial “neurons” between the input and the output, and, like the visual cortex, these artificial neurons learn a hierarchy of progressively more complex feature detectors. By learning feature detectors that are optimized for a specific task such as classification, deep learning can substantially outperform systems that rely on features supplied by domain experts or that are designed by hand [34]. It is about using multiple levels of representation and abstraction that help to make sense of data such as images, sound, and text. Over the past few years, deep learning has been a popular technique for most artificial intelligence type problems, overshadowing classical machine learning. The underlying reason for this is that deep learning has repeatedly demonstrated its superior performance on a wide variety of tasks including speech, natural language, and playing games. The properties and advantages of deep learning including (1) scales effectively with data: deep networks scale much better with more data than classical ML algorithms, as shown in Fig. 9.4. That is why usually to improve the accuracy with a deep network it is just better to use more data; (2) deep learning excels at modelling extremely complicated relationships between inputs and out-

9  Machine Learning Techniques

153

puts, as it can represent more complex features and to “learn” increasingly complex models for predictions. Neural networks which consists of more than three layers of neuros (including the input and output layer) are called as the deep neural networks (DNN). And training them is called deep learning. Neural network is a machine learning algorithm inspired from the working of human brain which enable a system to learn from some observational data. A simple neural network consists of an input layer, a single hidden layer and an output layer. Deep learning is a machine learning technique that performs learning in more than two hidden layers (Fig. 9.5). It is a DNN consisting multiple layers of nonlinear processing units (hidden layers). It performs feature extraction and transformation. Each successive layer of DNN uses the output from the previous layer as input. The deep learning method will perform better when there is more data, more complicated models, and more computation.

Fig. 9.4  The performance of deep learning versus classic machine learning in terms of data size

Fig. 9.5  A deep neural network architecture for imagine recognition [65]

154

X. Ding

Currently, there is few studies in cuffless BP estimation with the real sense of deep learning method. That is, to use the data-driven feature learning detectors to extract abstract representations of the raw input and mapping the learned features to the output. Instead, most of the current studies have still used the handcrafted features as the input, and attempted the deep neural networks, such as deep neural network, long short-term memory (LSTM) to model the relationship between the features and the output.

Summary of State-of-the-Art, Application Examples Review of Emerging Studies In the recent years, most of the studies have been employed the hand-designed features to indicate the BP changes, and attempted various machine learning algorithms, including deep learning architecture to predict BP. One representative study was conducted by Su et al., in which a deep recurrent neural network was proposed to predict long-term BP of multiple components [50]. Figure 9.6 shows the overview of the proposal model. The whole network was trained with backpropagation through time to miniaturize the mean squared error (MSE) of the difference between BP prediction and the ground truth of total N training samples:

Fig. 9.6  The Deep recurrent neural network that consists of the bottom layer bidirectional LSTM (green dashed box) and the LSTM layer with residual connections (orange dashed box), where X1, X2, …, Xt represent the extracted features, and y1, y2, …, yt the predicted BP time series [50]

9  Machine Learning Techniques



(

155

)

 { x1:T , y1:T }N =

1 N T ∑∑(zt − yt )2 + λθ 2 N i =1 t =1

(9.8)

where yt = [SBP, DBP, MBP] represents the ground truth, zt the corresponding prediction, θ2 and λ are the L2 regulation of model parameters and the corresponding coefficient, respectively. Validation of the proposal model on 84 health subjects with 7 features that are extracted from ECG and PPG signal showed that the 4-layer deep RNN has achieved the best performance for SBP and DBP with RMSE of 3.73 and 2.43 mmHg, which is supervisor to the methods that are using classic machine learning algorithms, such as decision tree and SVR. Further, it can obtain accurate estimation for multiple days BP, with one representative estimation illustrated in Fig. 9.7. Wang et al. used artificial neural network with 22 extracted features and one hidden layer to estimate SBP and DBP, which achieved the MAE of 4.02 ± 2.79 mmHg and 2.27 ± 1.82 mmHg for SBP and DBP, respectively [66]. Table 9.4 summarizes the studies in the recent 3  years that are using machine learning method for the development of BP models. It can be observed that the deep learning model with a bigger training dataset generally can obtain a better accuracy than the classical machine learning method [50, 52, 53, 67].

Discussions With the advancement of machine learning techniques, more and more researchers have attempted the machine learning method for cuffless BP measurement. However, given the limited amount of data that are used for training of the model, the performance of the model has not achieved to the level as expected. Taking the example of the deep RNN, it is designed mostly to learn features from raw signals, but for the task of BP estimation, the handcrafted features are still being used. In this case, the use of the deep neural network would not take effect for modelling the relationship between the input and the output. The factors determining how well the machine learning algorithm will perform depends on its ability to make the training error small, and to make the gap between training and test error small. These two factors correspond to the two central challenges in machine learning: underfitting and overfitting. Underfitting occurs when the model is not able to obtain a sufficiently low error values on the training set. Overfitting occurs when the gap between the training error and the test error is too large. This happens when the sample size for training is too small, which means the training dataset does not adequately represent the whole population. As the result, the trained model would only be accurate for the training set but not for the test dataset. It is expected to achieve the best accuracy for the task of BP estimation with the most appropriate machine learning method and with a big enough data set that covers the information for the target population.

Fig. 9.7  Comparison of the Deep RNN prediction and the reference value of one representative subject for multi-day continuous BP estimation: (a–d) represent the results of first day, second day, fourth day and sixth month after the initial measurement, respectively [50]

156 X. Ding

9  Machine Learning Techniques

157

Table 9.4  Summary of the recent studies using machine learning based method for cuffless BP estimation

Reference Input features Xing et al. Spectrum amplitude [68] and phase of PPG waveform Sun et al. [44]

PPG and ECG signals, PAT and 18 PPG features from PPG and ECG signals

Jain et al. [43]

32 parameters extracted from ECG and PPG

Duan et al. 11 out 56 features [39] from PPG signal

Machine learning algorithm Artificial neural network with one hidden layer Multiple linear regression

Sparse regression (to trim the redundant features)

Training and test 69 subjects

19 subjects reference: Volume-­ clamp method Leave-on-­ subject-out cross validation Training: 99 subjects Test: 10 subjects Reference: OMRON HBP1300 57 subjects

Support vector machine regression He et al. 18 features from ECG Random Forest One-hour [42] and PPG signals continuous BP: 1246 pairs DBP 1260 pairs SBP Shobitha 18 features extracted Relevance 26 subjects et al. [40] from PPG signal vector machine Miao et al. 14 features extracted Multiple linear 73 subjects [47] from ECG and PPG regression Support vector regression 22 subjects Lin et al. 19 PPG indicators Linear [46] and PTT regression method Su et al. [50]

7 features extracted from ECG and PPG signals

Four-layer deep 84 healthy RNN (LSTM) subjects

Performance SBP: 0.06 ± 7.08 mmHg DBP: 0.01 ± 4.66 mmHg

SBP: 0.43 ± 13.52 mmHg

SBP: MAD: 4.43 mmHg (SD: 4.90 mmHg) DBP: MAD: 2.46 mmHg (SD: 3.31 mmHg)

SBP: 4.77 ± 7.68 mmHg DBP: 3.67 ± 5.69 mmHg MBP: 3.85 ± 5.87 mmHg SBP: 8.29 ± 5.84 mmHg

SBP: Kappa score = 0.99 DBP: Kappa score = 0.99 SBP: −0.00 ± 3.10 mmHg DBP: −0.00 ± 2.20 mmHg Combination of PPG and PTT achieves a better performance than PTT-based method SBP: 3.73 mmHg (RMSE) DBP: 2.43 mmHg (RMSE) (continued)

158

X. Ding

Table 9.4 (continued)

Reference Ertugrul et al. [45]

Machine learning algorithm Input features ECG and PPG signals Extreme learning machine A sequence-to-­ sequence model: perceptron + LSTM Artificial neural network (one hidden layer)

Training and test UCI datasets

120 subjects

Radha et al. [51]

Activity features Heart rate variability PPG morphology features

Wang et al. [67]

Spectral and morphological features from PPG signal

Polinski et al. [69]

PTT, RR interval, and Single layer respiration signal recurrent neural network ECG and PPG LSTM 50 healthy subjects 85 subjects Eight hidden Waveform layer deep information, handcrafted features neural and personal features networks from ECG and PPG signals 441 subjects Whole base feature Adaptive from PPG boosting regression

Ghosh et al. [52] Wu et al. [53]

Mousavi et al. [41]

72 subjects: 70% training 15% validation 15% testing 21 subjects

Performance SBP: 6.93 mmHg (MAE) MBP: 8.86 mmHg (MAE) DBP: 19.43 mmHg (MAE) SBP: 5.65 mmHg (RMSE)

SBP: 4.02 ± 2.79 mmHg DBP: 2.27 ± 1.82 mmHg

SBP: 1.06 mmHg (MAE) DBP: 0.63 mmHg (MAE) SBP: 0.02 ± 4.8 mmHg DBP: 1.5 ± 3.7 mmHg SBP: 3.63 mmHg (MAD) DBP: 2.45 mmHg (MAD)

SBP: −0.05 ± 8.90 mmHg MBP: 0.07 ± 4.91 mmHg DBP: 0.19 ± 4.17 mmHg

The Future of the Technology Challenges of the machine learning method for cuffless BP estimation are that the training needs to cover all situations of all individuals to ensure the accuracy of the model. The general limit and challenges of using machine learning are summarized as below: First, the successful training of the model relies on the enough accurate labelled data. However, for BP measurement, even the reference BP values are sometimes not 100% accurate. Moreover, there should have enough data for the training. That is, to include all kinds of population, and for each subject the data should cover all situations of BP changes, to obtain the most generalizable model for the application of different types of population. Second, machine learning, specifically the deep learning is kind of black box technique. Clinicians, scientists, patients, and regulators would all prefer a simple

9  Machine Learning Techniques

159

explanation for how a neural net arrives at its prediction of a specific case. However, when a deep neural network is trained to make predictions on a big data set, it typically uses its layers of learned, nonlinear features to model a huge number of complicated but weak regularities in the data. It is usually impossible to interpret these features as their meaning depends on complex interactions with the learned uninterpretable features in other layers. Correspondingly, there is opportunity and space for future studies to make a breakthrough: (1) to construct standard database that can be used to validate various algorithms. Current studies are using different datasets that are collected by different designed clinical trials. Due to different subject characteristics and different signal acquisition systems, there is no way to compare the performance of different methods from different groups; (2) to use DNNs and learning algorithms, including recurrent, generative adversarial, reinforcement, representation, and transfer [35, 70], to learn abstract features from the raw signals or learn the true data distribution from the training set rather than with handcrafted features to capture the most relevant information for the accurate prediction of BP; (3) to explore new signal modalities that can be acquired with wearable/unobtrusive system.

References 1. Nichols W, O’Rourke M, Vlachopoulos C.  McDonald’s blood flow in arteries: theoretical, experimental and clinical principles. Boca Raton, FL: CRC; 2011. 2. Hughes DJ, Babbs CF, Geddes LA, Bourland JD. Measurements of Young’s modulus of elasticity of the canine aorta with ultrasound. Ultrason Imaging. 1979;1(4):356–67. 3. Weltman G, Sullivan G, Bredon D. The continuous measurement of arterial pulse wave velocity. Med Biol Eng Comput. 1964;2(2):145–54. 4. Obrist PA, Light KC, McCubbin JA, Hutcheson JS, Hoffer JL. Pulse transit time: relationship to blood pressure. Behav Res Methods Instrum. 1978;10(5):623–6. 5. Williams JGL, Williams B. Arterial pulse wave velocity as a psychophysiological measure, (in English). Psychosom Med. 1965;27(5):408–14. 6. Gribbin B, Steptoe A, Sleight P. Pulse wave velocity as a measure of blood pressure change, (in English). Psychophysiology. 1976;13(1):86–90. 7. Steptoe A, Smulyan H, Gribbin B. Pulse-wave velocity and blood-pressure change—calibration and applications, (in English). Psychophysiology. 1976;13(5):488–93. 8. Obrist PA, Light KC, McCubbin JA, Hutcheson J, Hoffer JL. Pulse transit time: relationship to blood pressure and myocardial performance. Psychophysiology. 1979;16(3):292–301. 9. Allen RA, Schneider JA, Davidson DM, Winchester MA, Taylor CB.  The covariation of blood pressure and pulse transit time in hypertensive patients, (in English). Psychophysiology. 1981;18(3):301–6. 10. Geddes LA, Voelz MH, Babbs CF, Bourland JD, Tacker WA. Pulse transit time as an indicator of arterial blood pressure,(in English). Psychophysiology. 1981;18(1):71–4. 11. Newlin DB. Relationships of pulse transmission times to pre-ejection period and blood pressure. Psychophysiology. 1981;18(3):316–21. 12. Lane JD, Greenstadt L, Shapiro D, Rubinstein E. Pulse transit time and blood pressure—an intensive analysis, (in English). Psychophysiology. 1983;20(1):45–9. 13. Pollak MH, Obrist PA. Aortic radial pulse transit time and ECG Q-wave to radial pulse wave interval as indexes of beat-by-beat blood pressure change, (in English). Psychophysiology. 1983;20(1):21–8.

160

X. Ding

14. Marie GV, Lo CR, Vanjones J, Johnston DW. The relationship between arterial blood pressure and pulse transit time during dynamic and static exercise, (in English). Psychophysiology. 1984;21(5):521–7. 15. Sawada Y, Yamakoshi K. A correlation analysis between pulse transit time and instantaneous blood pressure measured indirectly by the vascular unloading method, (in English). Biol Psychol. 1985;21(1):1–9. 16. Zong W, Moody G, Mark R. Effects of vasoactive drugs on the relationship between ECG-­ pulse wave delay time and arterial blood pressure in ICU patients. In: Computers in Cardiology, IEEE; 1998. p. 673–6. 17. Nitzan M, Khanokh B, Slovik Y. The difference in pulse transit time to the toe and finger measured by photoplethysmography (in English). Physiol Meas. 2002;23(1):85–93. 18. Ahlstrom C, Johansson A, Uhlin F, Länne T, Ask P. Noninvasive investigation of blood pressure changes using the pulse wave transit time: a novel approach in the monitoring of hemodialysis patients. J Artif Organs. 2005;8(3):192–7. 19. Muehlsteff J, Aubert X, Schuett M. Cuffless estimation of systolic blood pressure for short effort bicycle tests: the prominent role of the pre-ejection period. In: Conf Proc IEEE Eng Med Biol Soc, New York, USA, vol. 1, 2006. p. 5088–92. 20. Payne RA, Symeonides CN, Webb DJ, Maxwell SRJ. Pulse transit time measured from the ECG: an unreliable marker of beat-to-beat blood pressure, (in English). J  Appl Physiol. 2006;100(1):136–41. 21. Marcinkevics Z, Greve M, Aivars JI, Erts R, Zehtabi AH. Relationship between arterial pressure and pulse wave velocity using photoplethysmography during the post-exercise recovery period. Acta Univ Latviensis Biol. 2009;753:59–68. 22. Wong MYM, Poon CCY, Zhang Y-T. An evaluation of the cuffless blood pressure estimation based on pulse transit time technique: a half year study on normotensive subjects, (in English). Cardiovasc Eng. 2009;9(1):32–8. 23. Zheng YL, et al. Unobtrusive sensing and wearable devices for health informatics. IEEE Trans Biomed Eng. 2014;61(5):1538–54. 24. Young CC, Mark JB, White W, DeBree A, Vender JS, Fleming A. Clinical evaluation of continuous noninvasive blood pressure monitoring: accuracy and tracking capabilities. J Clin Monit. 1995;11(4):245–52. 25. Chen W, Kobayashi T, Ichikawa S, Takeuchi Y, Togawa T. Continuous estimation of systolic blood pressure using the pulse arrival time and intermittent calibration (in English). Med Biol Eng Comput. 2000;38(5):569–74. 26. Fung P, Dumont G, Ries C, Mott C, Ansermino M. Continuous noninvasive blood pressure measurement by pulse transit time. In: Conf Proc IEEE Eng Med Biol Soc, 2004, vol. 1, IEEE, 2004. p. 738–41. 27. Poon CCY, Zhang YT. Cuff-less and noninvasive measurements of arterial blood pressure by pulse transit time. In: Conf Proc IEEE Eng Med Biol Soc, Shanghai, China. IEEE; 2005. p. 5877–5880. 28. Cattivelli FS, Garudadri H.  Noninvasive cuffless estimation of blood pressure from pulse arrival time and heart rate with adaptive calibration. In: Sixth international workshop on wearable and implantable body sensor networks (BSN 2009): IEEE; 2009. p. 114–9. 29. Gesche H, Grosskurth D, Küchler G, Patzak A.  Continuous blood pressure measurement by using the pulse transit time: comparison to a cuff-based method. Eur J  Appl Physiol. 2012;112(1):309–15. 30. Chen Y, Wen C, Tao G, Bi M. Continuous and noninvasive measurement of systolic and diastolic blood pressure by one mathematical model with the same model parameters and two separate pulse wave velocities. Ann Biomed Eng. 2012;40(4):871–82. 31. Ding XR, Zhang YT, Liu J, Dai WX, Tsang HK. Continuous cuffless blood pressure estimation using pulse transit time and photoplethysmogram intensity ratio. IEEE Trans Biomed Eng. 2016;63(5):964–72. 32. Huynh TH, Jafari R, Chung WY. Noninvasive cuffless blood pressure estimation using pulse transit time and impedance plethysmography. IEEE Trans Biomed Eng. 2018.

9  Machine Learning Techniques

161

33. Liu J, Yan B, Zhang Y, Ding XR, Peng S, Zhao N. Multi-wavelength photoplethysmography enabling continuous blood pressure measurement with compact wearable electronics. IEEE Trans Biomed Eng. 2018. 34. LeCun Y, Bengio Y, Hinton G. Deep learning. Nature. 2015;521(7553):436. 35. Goodfellow I, Bengio Y, Courville A, Bengio Y. Deep learning. Cambridge: MIT Press; 2016. 36. Kim JY, Cho BH, Im SM, Jeon MJ, Kim IY, Kim SI. Comparative study on artificial neural network with multiple regressions for continuous estimation of blood pressure. In: Engineering in Medicine and Biology Society, 2005. IEEE-EMBS 2005. 27th Annual International Conference of the IEEE; 2005. p. 6942–5. 37. Monte-Moreno E. Non-invasive estimate of blood glucose and blood pressure from a photoplethysmograph by means of machine learning techniques. Artif Intell Med. 2011;53(2):127–38. 38. Ruiz-Rodríguez JC, et al. Innovative continuous non-invasive cuffless blood pressure monitoring based on photoplethysmography technology. Intensive Care Med. 2013;39(9):1618–25. 39. Duan K, Qian Z, Atef M, Wang G. A feature exploration methodology for learning based cuffless blood pressure measurement using photoplethysmography. In: Engineering in medicine and biology society (EMBC), 2016 IEEE 38th annual international conference of the IEEE; 2016. p. 6385–8. 40. Shobitha S, Amita P, Krupa BN, Beng GK.  Cuffless blood pressure prediction from PPG using relevance vector machine. In: Electrical, Electronics, Communication, Computer, and Optimization Techniques (ICEECCOT), 2017 International Conference on. IEEE; 2017. p. 75–8. 41. Mousavi SS, Firouzmand M, Charmi M, Hemmati M, Moghadam M, Ghorbani Y. Blood pressure estimation from appropriate and inappropriate PPG signals using a whole-based method. Biomed Signal Process Control. 2019;47:196–206. 42. He R, Huang Z-P, Ji L-Y, Wu J-K, Li H, Zhang Z-Q.  Beat-to-beat ambulatory blood pressure estimation based on random forest. In: Wearable and implantable body sensor networks (BSN), 2016 IEEE 13th international conference on: IEEE; 2016. p. 194–8. 43. Jain M, Kumar N, Deb S, Majumdar A. A sparse regression based approach for cuff-less blood pressure measurement. In: Acoustics, speech and signal processing (ICASSP), 2016 IEEE international conference on: IEEE; 2016. p. 789–93. 44. Sun S, Bezemer R, Long X, Muehlsteff J, Aarts R. Systolic blood pressure estimation using PPG and ECG during physical exercise. Physiol Meas. 2016;37(12):2154. 45. Ertuğrul ÖF, Sezgin N.  A non-invasive time-frequency based approach to estimate cuffless arterial blood pressure 2. Methods. 2018;2:18. 46. Lin W-H, Wang H, Samuel OW, Liu G, Huang Z, Li G.  New photoplethysmogram indicators for improving cuffless and continuous blood pressure estimation accuracy. Physiol Meas. 2018;39(2):025005. 47. Miao F, et al. A novel continuous blood pressure estimation approach based on data mining techniques. IEEE J Biomed Health Inform. 2017;21:1730–40. 48. Bose SSN, Kandaswamy S. Sparse representation of photoplethysmogram using K-SVD for cuffless estimation of arterial blood pressure. In: Advanced computing and communication systems (ICACCS), 2017 4th international conference on: IEEE; 2017. p. 1–5. 49. Pan J, Zhang Y. Improved Blood Pressure Estimation Using Photoplethysmography Based on Ensemble Method. In: Pervasive Systems, Algorithms and Networks & 2017 11th International Conference on Frontier of Computer Science and Technology & 2017 Third International Symposium of Creative Computing (ISPAN-FCST-ISCC), 2017 14th International Symposium on. IEEE; 2017. p. 105–11. 50. Su P, Ding X, Zhang Y, Li Y, Zhao N. Predicting blood pressure with deep bidirectional LSTM network, arXiv preprint. In: 2018 IEEE EMBS International Conference on Biomedical & Health Informatics (BHI), Las Vegas, Nevada, USA, 2018. p. 323–8. 51. Radha M, et al. Wrist-worn blood pressure tracking in healthy free-living individuals using neural networks. arXiv preprint arXiv:1805.09121, 2018. 52. Ghosh S, Banerjee A, Ray N, Wood PW, Boulanger P, Padwal R. Using accelerometric and gyroscopic data to improve blood pressure prediction from pulse transit time using recurrent neural network. In: 2018 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP). IEEE; 2018. p. 935–9.

162

X. Ding

53. Wu D, Xu L, Zhang R, Zhang H, Ren L, Zhang Y-T. Continuous cuff-less blood pressure estimation based on combined information using deep learning approach. J Med Imaging Health Inform. 2018;8(6):1290–9. 54. Regression Trees. http://www.stat.cmu.edu/~cshalizi/350-2006/lecture-10.pdf. 55. Chan K, Hung K, Zhang Y.  Noninvasive and cuffless measurements of blood pressure for telemedicine. In: Engineering in Medicine and Biology Society, 2001. Proceedings of the 23rd Annual International Conference of the IEEE, vol. 4. IEEE; 2001. p. 3592–3. 56. Payne R, Symeonides C, Webb D, Maxwell S. Pulse transit time measured from the ECG: an unreliable marker of beat-to-beat blood pressure. J Appl Physiol. 2006;100(1):136–41. 57. Douniama C, Sauter C, Couronne R. Blood pressure tracking capabilities of pulse transit times in different arterial segments: a clinical evaluation. In: Computers in Cardiology, 2009. IEEE; 2009. p. 201–4. 58. Wong MY-M, Poon CC-Y, Zhang Y-T. An evaluation of the cuffless blood pressure estimation based on pulse transit time technique: a half year study on normotensive subjects. Cardiovasc Eng. 2009;9(1):32–8. 59. Choi Y, Zhang Q, Ko S. Noninvasive cuffless blood pressure estimation using pulse transit time and Hilbert–Huang transform. Comput Electr Eng. 2013;39(1):103–11. 60. Heravi MY, Khalilzadeh M, Joharinia S. Continuous and cuffless blood pressure monitoring based on ECG and SpO2 signals ByUsing Microsoft visual C sharp. J  Biomed Phys Eng. 2014;4(1):27. 61. Kim JS, Kim KK, Baek HJ, Park KS. Effect of confounding factors on blood pressure estimation using pulse arrival time. Physiol Meas. 2008;29(5):615. 62. Ye S-y, Kim G-R, Jung D-K, Baik S, Jeon G. Estimation of systolic and diastolic pressure using the pulse transit time. World Acad Sci Eng Technol. 2010;67:726–31. 63. Breiman L. Random forests. Mach Learn. 2001;45(1):5–32. 64. Deka PC. Support vector machine applications in the field of hydrology: a review. Appl Soft Comput. 2014;19:372–86. 65. Are deep neural nets “Software 2.0”?. 2018. 66. Wang G, Atef M, Lian Y. Towards a continuous non-invasive cuffless blood pressure monitoring system using PPG: systems and circuits review. IEEE Circuits Syst Mag. 2018;18(3):6–26. 67. Wang L, Zhou W, Xing Y, Zhou X. A novel neural network model for blood pressure estimation using photoplethesmography without electrocardiogram. J Healthc Eng. 2018;2018 68. Xing X, Sun M. Optical blood pressure estimation with photoplethysmography and FFT-based neural networks. Biomed Opt Express. 2016;7(8):3007–20. 69. Poliñski A, Czuszyñski K, Kocejko T. Blood pressure estimation based on blood flow, ECG and respiratory signals using recurrent neural networks. In: 2018 11th International Conference on Human System Interaction (HSI): IEEE; 2018. p. 86–92. 70. Yu K-H, Beam AL, Kohane IS.  Artificial intelligence in healthcare. Nat Biomed Eng. 2018;2(10):719.

Chapter 10

Initialization of Pulse Transit Time-Based Blood Pressure Monitors Ramakrishna Mukkamala and Jin-Oh Hahn

Abstract  Pulse transit time (PTT)-based monitors can potentially permit cuffless blood pressure (BP) measurements, but initialization of the monitors so that the measured time delay may be mathematically converted to absolute BP is difficult. The goal of this chapter is to facilitate the achievement of reliable and practical initialization of PTT-based BP monitors. We present parametric models to relate PTT to BP; explain three different methods to determine the model parameters for a patient via cuff BP measurements that range from more accurate to more convenient; describe models that incorporate readily available covariates in addition to PTT and parameter determination methods for these enhanced models; discuss the required time period for reinitialization of the monitors with cuff BP measurements; and make recommendations and provide our outlook on such monitors. Keywords  Aging · Arterial stiffness · Blood pressure monitoring · Calibration · Cuff-less blood pressure · Hypertension · Initialization · Mobile health · Pulse transit time · Pulse wave velocity

Introduction Blood pressure (BP) monitors based on pulse transit time (PTT, the time delay for the pressure wave to travel between two arterial sites) are being widely pursued [1]. One reason is that PTT is inversely correlated with BP in a person due to fundamental properties of pulse wave propagation in nonlinear arteries. Hence, PTT may offer a physics-based principle for BP measurement. The other reason is that PTT can be simply measured as the relative timing between proximal and distal arterial R. Mukkamala (*) Department of Electrical and Computer Engineering, Michigan State University, East Lansing, MI, USA e-mail: [email protected] J.-O. Hahn Department of Mechanical Engineering, University of Maryland, College Park, MD, USA e-mail: [email protected] © Springer Nature Switzerland AG 2019 J. Solà, R. Delgado-Gonzalo (eds.), The Handbook of Cuffless Blood Pressure Monitoring, https://doi.org/10.1007/978-3-030-24701-0_10

163

164

R. Mukkamala and J.-O. Hahn

waveforms. Hence, PTT does not involve variable force application for its measurement and may thereby permit passive, cuff-less BP monitoring. However, initialization of PTT-based BP monitors so that the PTT measurements in units of ms may be mapped to BP measurements in units of mmHg is a major challenge. To explain, BP (P) and PTT (τ) over a measured length (l) may be approximately related via a simple model as follows:

l P = K1 + K 2 . t

(10.1)

The slope K1 (>0) and intercept K2 (0) indicates the degree of the elastic modulus dependency on BP [2]. Another useful formula is the Wesseling model for the compliance of the ex vivo human aorta as follows:



é1 1 æ P - P0 ö ù A ( P ) = Amax ê + tan -1 ç ÷ú êë 2 p è P1 ø úû C (P) =



dA = dP

Amax

, é æ P - P ö2 ù 0 p P1 ê1 + ç ÷ ú êë è P1 ø úû

(10.6) (10.7)

where Amax is the maximum cross-sectional area at infinite BP; P0 is the BP at which the compliance is maximal; and P1 reflects the BP range for which the compliance is relatively constant [3]. The Wesseling model also includes a relationship between its three parameters and age/gender as follows:



ì4.12 females Amax = í î5.62 males

(10.8) (10.9)



ì72 - 0.89age females P0 = í î76 - 0.89age males



P1 = 57 - 0.44age,

(10.10)

where Amax, Pj, and age are in units of cm2, mmHg, and years. Because of differences between cardiovascular and chronological aging [4], these formulas may be thought of as representing an average person for each age and gender. As indicated in Fig.  10.2a, the overall Wesseling model indicates that the arterial cross-sectional area becomes less dependent on BP with aging, while the compliance becomes more dependent on BP with aging. Equations (10.8)–(10.10) also indicate that the compliance is larger in males. The second mechanism may also be quantified via mathematical modeling. Fung’s hyper-elastic model of the arterial wall and a cylindrical tube geometry without assumption of a thin wall was recently employed to derive Ma’s formulas for A(P) and C(P) [5]. These model-based formulas have both similarities and differences with the experimentally based Eqs. (10.6) and (10.7). Due to the different quantitative formulations of the two underlying mechanisms, multiple, theoretical models arise to relate PTT to BP. One popular model, which may be obtained by substituting Eq. (10.5) into Eq. (10.4) [6], is a logarithmic relationship as follows:



æt ö P = K1 ln ç ÷ + K 2 , èlø

(10.11)

168

R. Mukkamala and J.-O. Hahn (A)

5.5 5

0.06

175

0.05

150

0.04

125

(B)

3.5

2

A [cm 2]

4

3 2.5

30 years old 70 years old

2 1.5 25

50

75

100 125 P [mmHg]

150

175

P [mmHg]

C [cm /mmHg]

4.5

0.03

100

0.02

75

0.01

50

0 25

50

75

100 125 P [mmHg]

150

175

25

0

5

10 l/ [m/s]

15

20

Fig. 10.2  A parametric model relating PTT to BP may be defined with the Wesseling model of the aortic cross-sectional area (A)-BP relationship [3] integrated into the Bramwell–Hill equation (see æ dA ö Eq. 10.2). (a) The model A − P and C ç = ÷ - P relationships for a young and old male (see è dP ø l Eqs. 10.6-10.10). (b) The corresponding model P relationships defined by two parameters t (see Eq. 10.14). The model indicates that the shape of the PTT-BP relationship changes with age

2 1 æ 2r r ö and K 2 = ln ç ÷ and are thus person- and time-specific parama a è E0 h ø eters. Another popular model, which may arise by substituting Eqs. (10.6) and (10.7) into Eq. (10.2) and then assuming large P (i.e., (P>>P0 + P1) [1], is a line relationship in PWV as follows: where K1 = -

where K1 =

P = K1

l + K2 , t

(10.12)

2 r P1 and K2  =  P0 and  are thus person- and time-specific paramep +2

ters. Ma’s A(P) and C(P) formulas were substituted into Eq. (10.2), and a reduced model was derived as a quadratic relationship in PWV [5] as follows: 2



ælö P = K1 ç ÷ + K 2 , èt ø

(10.13)

where K1 and K2 are functions of the material and geometric properties of the artery and are thus person- and time-specific parameters. Equation (10.13) may also arise by linearizing Eq. (10.5) and then substituting the linear model into Eq. (10.4) [7]. A more general model, which arises by substituting Eqs. (10.6) and (10.7) into Eq. (10.2) without any assumptions on P [8], is as follows:

10  Initialization of Pulse Transit Time-Based Blood Pressure Monitors

t = l

2819.7 æ æ P - P ö2 ö æ 1 1 æ P - P0 0 ÷ ç + tan -1 ç p P1 ç 1 + ç ÷ ç ç è P1 ø ÷ è 2 p è P1 è ø

169

, öö ÷ ÷÷ øø

(10.14)

where τ is in units of ms, l is in units of m, and P0 and P1 are person- and time-­ specific parameters. Using the age-dependent P0 and P1 formulas in Eqs. (10.9) and (10.10) and as illustrated in Fig. 10.2b, the model indicates that the shape of the PTT-BP relationship may be age-dependent and becomes nearly a line relationship in PWV in the elderly. (Note that at very low BP of a younger person, the PTT-BP relationship may not be one-to-one due to the dominance of elastin, which is a linear elastic fiber, over collagen in these conditions [1].) While all of the models here are theoretically based, none may be strictly correct. In particular, Eq. (10.4) was derived assuming constant E, whereas Eq. (10.2) was derived assuming small changes in A.

Empirical Models Empirical models are mainly based on PTT-BP data fitting considerations. A popular model is a line relating PTT to BP as follows:



P = K1

t + K2 , l

(10.15)

where K1 and K2 are person- and time-specific parameters. Nonlinear, yet parsimonious, functions have also been used to relate PTT to BP. Examples of such models are as follows:

P = K1 x 2 + K 2 x + K 3

(10.16)



P = K1 x p + K 2

(10.17)



P = K1e K2 x ,

(10.18)

t l or , Kj are person- and time-specific parameters, and p is either a fixed l t value or likewise varies with the person and time [9, 10]. A more unique model relates PTT to pulse pressure (PP = systolic BP – diastolic BP) rather than an absolute level of BP as follows: where x =

2



ælö PP = K ç ÷ , èt ø

(10.19)

170

R. Mukkamala and J.-O. Hahn

where K is a person- and time-specific parameter [11]. Empirical models have also been proposed based on initialization considerations. One such model was specifically designed to prevent non-physiologic BP (i.e., negative or too high BP) and is as follows: P=

K1 ö æt ç l - K2 ÷ è ø

2

+ K3 ,

(10.20)



where Kj are person- and time-specific parameters [12]. While this design may be practical for initialization, note that physiologically correct limiting behavior is for PTT to be finite at zero BP and approach zero as BP approaches infinity. See [1] for a list of publications on the aforementioned and other PTT-BP models.

Comparison of Models as a Practical Calibration Curve Form A model relating PTT to BP may constitute a practical calibration curve form, if it fits actual PTT-BP data over a wide BP range using a minimal number of parameters and does not output unrealistic BP. Several studies have compared PTT-BP models in terms of fitting simultaneous measurements of PTT and BP over a range of BP. These studies did not include single-parameter models, because such models do not generally fit PTT-BP data (see, e.g., plots in [10, 13, 14]). Also note that Eq. (10.19), a one-parameter model relating PWV squared to PP, may not hold in general, as K depends on the difference in the arterial cross-sectional areas at systole and diastole (see Eq. (10.34) below) and may thus vary considerably within a person [15–17]. The comparative studies in humans have generally concluded that various models fit the data similarly and increasing the number of model parameters hardly improves the data fitting. The reason is that the data were obtained over too limited a BP range due to ethical considerations. However, comparative studies in animals wherein BP may be perturbed widely do indicate that Eq. (10.12), a line model in PWV, fits PTT-BP data significantly better than Eq. (10.15), a line model in PTT [18, 19]. A recent animal study indicated that Eq. (10.18), an exponential model with t x = , fits PTT-BP data well over a 100 mmHg range of BP and best amongst varil ous two-parameter models [10]. While two parameters appear to be the minimum number needed for a model to fit data, models with more parameters such as Eq. (10.20) can prevent non-physiologic BP. However, instead of adding more model parameters, unrealistic BP could be averted by invoking the known BP range (see, e.g., vast BP statistical data for different genders, ethnicities, and age groups [20]). So, if a two-parameter model yielded BP outside this range, then the PTT-based monitor could either not output BP or output the maximum BP for an unusually low PTT or the minimum BP for an unusually high PTT. The former option makes more sense when the unrealistic BP is due to gross error in PTT, whereas the latter option

10  Initialization of Pulse Transit Time-Based Blood Pressure Monitors

171

might be reasonable when the unrealistic BP is due to a physiologic change in the model parameters. Since the cause of the unrealistic BP would be unknown, it may be safest to not output BP. If no BP outputs persist, then reinitialization of the monitor may be indicated.

Model Parameter Determination There are three methods to determine the multiple parameters of a PTT-BP model and thereby initialize a PTT-based BP monitor for a person (see Fig.  10.1). A person-­specific method involves measuring cuff BP and PTT during interventions that perturb BP in the person to determine all model parameters. A population-based method involves using basic information about the person along with a training dataset comprising cuff BP and PTT measurements from a cohort of different subjects to determine all model parameters. A hybrid method involves measuring cuff BP and PTT in the person to determine one model parameter and using the person’s basic information and a similar training dataset to determine the remaining parameters. The three methods represent different trade-offs between accuracy and convenience.

Person-Specific Methods Figure 10.3 illustrates the person-specific method for determining the PTT-BP model parameters. This method generally involves: (1) employing one or more interventions to perturb BP in the person; (2) measuring cuff BP and PTT during the baseline period and each intervention; and (3) fitting the model to the multiple PTT-BP data pairs to determine all parameters. By yielding a calibration curve that is specific to the person at the time of measurement, the method is most accurate but least convenient. Table 10.1 summarizes interventions that have been employed to perturb BP in order to initialize and/or assess a PTT-based monitor. These interventions generally represent different trade-offs between effect and convenience. Slow breathing [21] and postural changes (e.g., from supine to upright [22]) are relatively convenient but change BP little. Cold pressor (i.e., immersing a limb in ice water) [23] and different types of exercise (e.g., cycling [24], sustained handgrip [25], and mental arithmetic [26]) increase BP significantly but are less convenient. On the other hand, cycling and similar exercises could be incorporated in a person’s daily life, and BP may also fall below pre-exercise level sometime after the exercise [27]. The Valsalva maneuver (i.e., exhaling with nose and mouth closed) [28] is fairly unique in that it decreases BP appreciably but is inconvenient in that a non-standard, continuous BP monitor such as a finger cuff, volume-clamp device [29, 30] is needed to detect the momentary BP reduction.

172

R. Mukkamala and J.-O. Hahn

One intervention that is both significant and relatively convenient is a hydrostatic maneuver in which the vertical height (h) of the effective BP measurement site of a PTT-based monitor with respect to the heart is varied. As shown in Fig. 10.4, due to the weight of the blood column, such a maneuver will cause the BP at the measurement site (i.e., “local BP”) to change by ρgh, where g is gravity. A change in h of

Fig. 10.3  The person-specific method for determining the PTT-BP model parameters involves measuring cuff BP and PTT in the person during interventions that perturb BP and then optimally fitting the model to the measured BP-PTT pairs to determine all parameters Table 10.1  Blood pressure (BP) interventions Intervention Slow breathing [21] Supine to standing [22] Cold pressor [23] Exercise (e.g., cycling) [24, 27] Sustained handgrip [25] Mental arithmetic [26] Valsalva maneuver [28]

BP effect [mmHg] 0.15 Hz) power of HRV reflects augmented parasympathetic nervous function, while increases in the ratio of the low freæ LF ö quency(0.04–0.15  Hz) power to high frequency power of HRV ç ÷ indicates è HF ø higher sympatho-vagal balance [52]; and pulse rate variability correlates with HRV [53] and can be obtained when an ECG waveform is unavailable. Beat-to-beat PTT variability may reflect beat-to-beat BP variability, which is also a result of autonomic nervous control. An exemplary model relating peripheral PTT detected at the level of diastole and an autonomic nervous index to BP may be as follows:



LF ö l æ + K3 , Pd = ç K1 + K 2 HF ÷ø t è

(10.37)

where K2  6 months

tions in PEP and/or smooth muscle contraction rather than 1 month of aging. In the other study [61], finger PAT and additional waveform features were measured in 10 healthy subjects and mapped to BP using a hybrid method. The hybrid calibration curve was then tested 1 day, 3 days, and 6 months later. The BP errors increased significantly after just 1  day but then remained stable over the 6-month period. These results likewise suggest that the calibration curve changed with PEP and/or smooth muscle contraction but not 6 months of aging. Note that these data pertain to measurements of PTT through peripheral arteries and that aging impacts peripheral PTT less than aortic PTT [1]. Based on the aforesaid studies, it may be reasonable to perform cuff initializations every few months. However, further studies are surely needed to arrive at a conclusion.

Conclusions In summary, PTT-based BP monitors have gained popularity, because they can potentially permit cuffless and passive measurement of BP in a principled way. However, initialization of these monitors so that PTT in units of ms can be mapped to BP in units of mmHg is a serious challenge, as the PTT-BP relationship is defined by at least two person- and time-specific parameters. Parametric models inspired by theory or data are available to relate PTT to BP, with some constituting a practical calibration curve form. There are three methods for determining the model

186

R. Mukkamala and J.-O. Hahn

parameters that reflect different trade-offs between accuracy and convenience. A person-­specific method determines all of the parameters from multiple cuff BP measurements during BP interventions in the person and is thus most accurate but least convenient. A population-based method determines all parameters from only basic information about the person in conjunction with a training dataset that includes cuff BP measurements from other people and is thus most convenient but least accurate. A hybrid method determines one parameter from a single cuff BP measurement in the person and the remaining parameter(s) from basic person information plus a similar training dataset and thus balances accuracy and convenience. Additional features of the waveforms obtained to detect PTT such as peak-to-peak amplitudes and autonomic nervous system indices may be incorporated in the model to improve accuracy. Limited evidence seems to suggest that the model parameters must be updated every several months to account for changes due to cardiovascular aging. Hence, PTT-based monitors may provide ultra-convenient BP measurements only in the periods between cuff initializations. Our recommendations for initializing PTT-based BP monitors are as follows. We currently advocate for Eq. (10.14) to serve as the calibration curve form, because it can represent a linear in PWV relationship and change shape with aging using only two parameters. In the event that this or another model yields unrealistic BP, we further suggest that the monitor not output BP. This conservative approach assumes that the PTT input is fraught with error. For the person-specific method to determine the model parameters, we presently endorse exercise or a hydrostatic maneuver as BP interventions. Exercise may be convenient (or necessary) for many and can raise and then lower BP with respect to pre-exercise levels. Hand lowering and raising and, to a lesser extent, postural changes are simple and can increase and decrease local BP by a substantial and graded amount. Moreover, these interventions induce BP changes that can be captured with a standard automatic cuff. For the population-­ based and hybrid methods to determine the model parameters, it may possibly be worthwhile to form a rich training dataset to maximize the chances for success by applying a battery of interventions (cold pressor to increase BP via vasoconstriction, mental arithmetic to increase BP via cardiac output, Valsalva maneuver to decrease BP, and perhaps others) and a finger cuff, volume-clamp device to capture all of the BP changes in large number of subjects with diverse BP. Such data could be collected as part of a large, multicenter study and made available to the entire research community for analysis. In addition, we believe that it is essential to include a waveform amplitude feature with PTT to track PP and thereby independently measure systolic BP and diastolic BP. For the person-specific method, we advise to only add this one feature to maintain tractability of the parameter determination. Finally, we recommend reinitializing PTT-based BP monitors more frequently than what may be necessary such as every week or month to ensure maximal accuracy. A key advantage of a PTT-based BP monitor is that it can yield many measurements over time. These measurements can be averaged to not only mitigate the BP prediction error but also eliminate intra-person BP variability. In this way, the moni-

10  Initialization of Pulse Transit Time-Based Blood Pressure Monitors

187

tor may be useful for hypertension screening despite errors in individual measurements that exceed the AAMI bias and precision limits [8]. Nevertheless, in our opinion, it may still be difficult to map PTT to BP with enough accuracy under all scenarios. Hence, it may be prudent to focus on niche areas such as nighttime and 24-h ambulatory BP monitoring, which are clinically important [64, 65] and for which cuff usage is especially problematic. For example, PTT-based monitors could be initialized in a person using cuff BP measurements before going to bed and after getting up to potentially yield sufficient accuracy, while entirely avoiding disruptive cuff inflations, during sleep [44]. In conclusion, PTT-based BP monitors may be promising, but substantial research is still needed to initialize the monitors so as to yield absolute BP measurements that are clinically useful. Our intent for this chapter is to serve as a foundation for embarking on this research. Acknowledgements  This work was supported by the National Institutes of Health under Grant EB-018818.

References 1. Mukkamala R, Hahn J-O, Inan O, Mestha L, Kim C-S, Toreyin H, Kyal S. Towards ubiquitous blood pressure monitoring via pulse transit time: theory and practice. IEEE Trans Biomed Eng. 2015;62:1879–901. 2. Hughes DJ, Babbs CF, Geddes LA, Bourland JD. Measurements of Young’s modulus of elasticity of the canine aorta with ultrasound. Ultrason Imaging. 1979;1:356–67. 3. Wesseling KH, Jansen JRC. Computation of aortic flow from pressure in humans using a nonlinear, three-element model. J Appl Physiol. 1993;74:2566–73. 4. Tanaka H, Dinenno FA, Monahan KD, Clevenger CM, DeSouza CA, Seals DR. Aging, habitual exercise, and dynamic arterial compliance. Circulation. 2000;102:1270–5. 5. Ma Y, Choi J, Hourlier-Fargette A, Xue Y, Chung HU, Lee JY, Wang X, Xie Z, Kang D, Wang H, Han S, Kang S-K, Kang Y, Yu X, Slepian MJ, Raj MS, Model JB, Feng X, Ghaffari R, Rogers JA, Huang Y.  Relation between blood pressure and pulse wave velocity for human arteries. Proc Natl Acad Sci. 2018;115:11144–9. 6. Pruett JD, Bourland JD, Geddes LA.  Measurement of pulse-wave velocity using a beat-­ sampling technique. Ann Biomed Eng. 1988;16:341–7. 7. Muehlsteff J, Aubert XL, Schuett M. Cuffless estimation of systolic blood pressure for short effort bicycle tests: the prominent role of the pre-ejection period. In: Annual International Conference of the IEEE Engineering in Medicine and Biology—Proceedings; 2006. p. 5088– 92. https://doi.org/10.1109/IEMBS.2006.260275. 8. Mukkamala R, Hahn J-O. Toward ubiquitous blood pressure monitoring via pulse transit time: predictions on maximum calibration period and acceptable error limits. IEEE Trans Biomed Eng. 2018;65:1410–20. 9. Thomas SS, Nathan V, Zong C, Soundarapandian K, Shi X, Jafari R. BioWatch: a noninvasive wrist-based blood pressure monitor that incorporates training techniques for posture and subject variability. IEEE J Biomed Heal Inform. 2016;20:1291–300. 10. Gao M, Olivier NB, Mukkamala R. Comparison of noninvasive pulse transit time estimates as markers of blood pressure using invasive pulse transit time measurements as a reference. Physiol Rep. 2016;4:e12768.

188

R. Mukkamala and J.-O. Hahn

11. Ding X, Zhang Y, Liu J, Dai W, Tsang HK.  Continuous cuffless blood pressure estimation using pulse transit time and photoplethysmogram intensity ratio. IEEE Trans Biomed Eng. 2016;63:964–72. 12. Wibmer T, Doering K, Kropf-Sanchen C, Rüdiger S, Blanta I, Stoiber KM, Rottbauer W, Schumann C.  Pulse transit time and blood pressure during cardiopulmonary exercise tests. Physiol Res. 2014;63:287–96. 13. Carroll JD, Shroff S, Wirth P, Halsted M, Rajfer S. Arterial mechanical properties in dilated cardiomyopathy. J Clin Invest. 1991;87:1002–9. 14. Butlin M, Shirbani F, Barin E, Tan I, Spronck B, Avolio AP.  Cuffless estimation of blood pressure: importance of variability in blood pressure dependence of arterial stiffness across individuals and measurement sites. IEEE Trans Biomed Eng. 2018;65:2377–83. 15. Seo J, Pietrangelo SJ, Lee H-S, Sodini CG.  Noninvasive arterial blood pressure waveform monitoring using two-element ultrasound system. IEEE Trans Ultrason Ferroelectr Freq Control. 2015;62:776–84. 16. Beulen BW, Bijnens N, Koutsouridis GG, Brands PJ, Rutten MC, van de Vosse FN. Toward noninvasive blood pressure assessment in arteries by using ultrasound. Ultrasound Med Biol. 2011;37:788–97. 17. Vappou J, Luo J, Okajima K, Di Tullio M, Konofagou EE.  Non-invasive measurement of local pulse pressure by pulse wave-based ultrasound manometry (PWUM). Physiol Meas. 2011;32:1653–62. 18. Geddes L, Voelz M.  Pulse transit time as an indicator of arterial blood pressure. Psychophysiology. 1981;18:71–5. 19. Zhang G, Gao M, Xu D, Olivier NB, Mukkamala R. Pulse arrival time is not an adequate surrogate for pulse transit time as a marker of blood pressure. J Appl Physiol. 2011;111:1681–6. 20. National Center for Health Statistics, Drizd T, Dannenberg AL, Engel A. Blood pressure levels in persons 18–74 years of age in 1976–80, and trends in blood pressure from 1960 to 1980 in the United States; 1986. 21. Joseph CN, Porta C, Casucci G, Casiraghi N, Maffeis M, Rossi M, Bernardi L. Slow breathing improves arterial baroreflex sensitivity and decreases blood pressure in essential hypertension. Hypertension. 2005;46:714–8. 22. Currens JH. A comparison of the blood pressure in the lying and standing positions: a study of five hundred men and five hundred women. Am Heart J. 1948;35:646–54. 23. Hines EA Jr, Brown GE. The cold pressor test for measuring the reactibility of the blood pressure: data concerning 571 normal and hypertensive subjects. Am Heart J. 1936;11:1–9. 24. Froelicher VF, Myers J. Exercise and the heart. Philadelphia, PA: Saunders; 2006. 25. Petrofsky JS, Lind AR. Aging, isometric strength and endurance, and cardiovascular responses to static effort. J Appl Physiol. 1975;38:91–5. 26. Al’Absi M, Bongard S, Buchanan T, Pincomb GA, Licinio J, Lovallo WR.  Cardiovascular and neuroendocrine adjustment to public speaking and mental arithmetic stressors. Psychophysiology. 1997;34:266–75. 27. Kenney MJ, Seals DR. Postexercise hypotension. Key features, mechanisms, and clinical significance. Hypertension. 1993;22:653–64. 28. Parati G, Casadei R, Groppelli A, Di Rienzo M, Mancia G. Comparison of finger and intra-­ arterial blood pressure monitoring at rest and during laboratory testing. Hypertension. 1989;13:647–55. 29. Imholz BP, Wieling W, van Montfrans GA, Wesseling KH. Fifteen years experience with finger arterial pressure monitoring: assessment of the technology. Cardiovasc Res. 1998;38:605–16. 30. Wesseling KH, De Wit B, van der Hoeven GMA, van Goudoever J, Settels JJ. Physical, calibrating finger vascular physiology for finapres. Homeostasis. 1995;36:67–82. 31. McCombie DB, Reisner AT, Asada HH. Motion based adaptive calibration of pulse transit time measurements to arterial blood pressure for an autonomous, wearable blood pressure monitor. In: 2008 30th Annual International Conference of the IEEE Engineering in Medicine and Biology Society; 2008. p. 989–92. https://doi.org/10.1109/IEMBS.2008.4649321.

10  Initialization of Pulse Transit Time-Based Blood Pressure Monitors

189

32. Wang Y, Liu Z, Ma S. Cuff-less blood pressure measurement from dual-channel photoplethysmographic signals via peripheral pulse transit time with singular spectrum analysis. Physiol Meas. 2018;39:025010. 33. Rosner B, Polk BF. Predictive values of routine blood pressure measurements in screening for hypertension. Am J Epidemiol. 1983;117:429–42. 34. Gavish B, Ben-dov IZ, Bursztyn M. Linear relationship between systolic and diastolic blood pressure monitored over 24 h: assessment and correlates. J Hypertens. 2008;26:199–209. 35. Master AM, Lasser RP. The relationship of pulse pressure and diastolic pressure to systolic pressure in healthy subjects 20–94 years of age. Am Heart J. 1965;70:163–71. 36. Keesman KJ.  System identification—an introduction. Berlin: Springer; 2011. https://doi. org/10.1007/978-0-85729-522-4. 37. Ljung L.  System identification: theory for the user. Upper Saddle River, NJ: Prentice Hall; 1999. 38. Wiard RM, Inan OT, Argyres B, Etemadi M, Kovacs GTA, Giovangrandi L. Automatic detection of motion artifacts in the ballistocardiogram measured on a modified bathroom scale. Med Biol Eng Comput. 2011;49:213–20. 39. Gao M, Cheng H-M, Sung S-H, Chen C-H, Olivier NB, Mukkamala R. Estimation of pulse transit time as a function of blood pressure using a nonlinear arterial tube-load model. IEEE Trans Biomed Eng. 2017;64:1524–34. 40. ISO 81060-2:2018 Non-invasive sphygmomanometers—part 2: clinical validation of automated manometer type. 2018. https://www.iso.org/standard/73339.html 41. Chen Y, Wen C, Tao G, Bi M, Li G. Continuous and noninvasive blood pressure measurement: a novel modeling methodology of the relationship between blood pressure and pulse wave velocity. Ann Biomed Eng. 2009;37:2222–33. 42. Watanabe N, Bando YK, Kawachi T, Yamakita H, Futatsuyama K, Honda Y, Yasui H, Nishimura K, Kamihara T, Okumura T, Ishii H, Kondo T, Murohara T. Development and validation of a novel cuff-less blood pressure monitoring device. JACC Basic Transl Sci. 2017;2:631–42. 43. Zheng Y-L, Yan BP, Zhang Y-T, Poon CCY. An armband wearable device for overnight and cuff—less blood pressure measurement. IEEE Trans Biomed Eng. 2014;61:2179–86. 44. Zheng Y, Poon CCY, Yan BP, Lau JYW. Pulse arrival time based cuff-less and 24-h wearable blood pressure monitoring and its diagnostic value in hypertension. J Med Syst. 2016;40:195. 45. Gesche H, Grosskurth D, Küchler G, Patzak A.  Continuous blood pressure measurement by using the pulse transit time: comparison to a cuff-based method. Eur J  Appl Physiol. 2012;112:309–15. 46. SOMNOtouch™ NIBP. https://somnomedics.eu/products/cardiology/24h-bloodpressure24h-ecg/somnotouch-nibp/ 47. Kim C-S, Ober SL, McMurtry MS, Finegan BA, Inan OT, Mukkamala R, Hahn J-O. Ballistocardiogram: mechanism and potential for unobtrusive cardiovascular health monitoring. Sci Rep. 2016;6:31297. 48. Wiens AD, Etemadi M, Roy S, Klein L, Inan OT. Toward continuous, noninvasive assessment of ventricular function and hemodynamics: wearable ballistocardiography. IEEE J  Biomed Heal Inform. 2015;19:1435–42. 49. Kim C-S, Carek AM, Inan OT, Mukkamala R, Hahn J-O. Ballistocardiogram-based approach to cuffless blood pressure monitoring: proof of concept and potential challenges. IEEE Trans Biomed Eng. 2018;65:2384–91. 50. Huynh TH, Jafari R, Chung WY. Noninvasive cuffless blood pressure estimation using pulse transit time and impedance plethysmography. IEEE Trans Biomed Eng. 2019;66:967–76. https://doi.org/10.1109/TBME.2018.2865751. 51. Ding X, Yan BP, Zhang YT, Liu J, Zhao N, Tsang HK. Pulse transit time based continuous cuffless blood pressure estimation: a new extension and a comprehensive evaluation. Sci Rep. 2017;7:11554.

190

R. Mukkamala and J.-O. Hahn

52. Task Force of the European Society of Cardiology and the North American Society of Pacing and Electrophysiology. Heart rate variability: standards of measurement, physiological interpretation and clinical use. Circulation. 1996;93:1043–65. 53. Schäfer A, Vagedes J. How accurate is pulse rate variability as an estimate of heart rate variability?: a review on studies comparing photoplethysmographic technology with an electrocardiogram. Int J Cardiol. 2013;166:15–29. 54. Zhang Q, Zhou D, Zeng X. Highly wearable cuff-less blood pressure and heart rate monitoring with single-arm electrocardiogram and photoplethysmogram signals. Biomed Eng Online. 2017;16:1–20. 55. Yoon YZ, Kang JM, Kwon Y, Park S, Noh S, Kim Y, Park J, Hwang SW. Cuff-less blood pressure estimation using pulse waveform analysis and pulse arrival time. IEEE J Biomed Heal Inform. 2018;22:1068–74. 56. Harris WS, Schoenfeld CD, Weissler AM. Effects of adrenergic receptor activation and blockade on the systolic preejection period, heart rate, and arterial pressure in man. J Clin Invest. 1967;46:1704–14. 57. Chen Y, Shi S, Liu YK, Huang SL, Ma T. Cuffless blood-pressure estimation method using a heart-rate variability-derived parameter. Physiol Meas. 2018;39:095002. 58. Ma HT.  A blood pressure monitoring method for stroke management. Biomed Res Int. 2014;2014:571623. 59. Yousefian P, Shin S, Mousavi A, Kim CS, Mukkamala R, Jang DG, Ko BH, Lee J, Kwon UK, Kim YH, Hahn JO. Data mining investigation of the association between a limb ballistocardiogram and blood pressure. Physiol Meas. 2018;39:075009. 60. Kachuee M, Member S, Kiani MM, Member S. Cuffless blood pressure estimation algorithms for continuous health-care monitoring. IEEE Trans Biomed Eng. 2017;64:859–69. 61. Miao F, Fu N, Zhang YT, Ding XR, Hong X, He Q, Li Y.  A novel continuous blood pressure estimation approach based on data mining techniques. IEEE J  Biomed Heal Inform. 2017;21:1730–40. 62. Lin W-H, Wang H, Samuel OW, Liu G, Huang Z, Li G.  New photoplethysmogram indicators for improving cuffless and continuous blood pressure estimation accuracy. Physiol Meas. 2018;39:025005. 63. Ding X, Zhang Y, Tsang HK. Impact of heart disease and calibration interval on accuracy of pulse transit time–based blood pressure estimation. Physiol Meas. 2016;37:227–37. 64. Pickering TG, Shimbo D, Haas D.  Ambulatory blood-pressure monitoring. N Engl J  Med. 2006;354:2368–74. 65. Boggia J, Li Y, Thijs L, Hansen T, Kikuya M, Bjorklund-Bodegard K, Richart T, Ohkubo T, Kuznetsova T, Torp-Pedersen C, Lind L, Ibsen H, Imai Y, Wang J, Sandoya E, O’Brien E, Staessen J. Prognostic accuracy of day vs. night ambulatory blood pressure: a cohort study. Lancet. 2007;370:1219–29.

Chapter 11

Key Regulatory Aspects and the Importance of Consensus Standards in Bringing Devices to Market Carole C. Carey

Abstract Software and hardware innovations, smart devices, digitization in healthcare, IoT (Internet of Things), and other emerging device technologies are driving the medical device industry as one of the fastest growing markets around the globe. Medical devices carry some measure of risk that can potentially cause problems or result in harm impacting patient safety significantly. The challenge manufacturers face is obtaining government marketing authorization for the proposed new medical device before it can be legally sold in its respective jurisdiction. This chapter examines key regulatory aspects and the importance of the use of standards and their increasing importance as a regulatory tool in medical device regulation. Five countries or regions with advanced medical device regulatory systems were explored: the USA, European Union (EU), Canada, Australia, and Japan. Keywords  Medical devices · Regulatory framework · Device classification · Risks and controls · International consensus standards · Harmonization · Marketing authorization

Background The twenty-first century has brought a proliferation of innovative, unobtrusive, and wearable personal devices both for clinical and consumer applications. The rising trend is possible with the development of novel sensors, new materials, advances in medicine, science, and biomedical engineering. Emerging technologies in nanotechnology, information technology, medical robotics, artificial intelligence, and neurotechnologies in brain–computer interfaces—just to name a few, are impacting C. C. Carey (*) C3-Carey Consultants, LLC, Fulton, MD, USA e-mail: [email protected] © Springer Nature Switzerland AG 2019 J. Solà, R. Delgado-Gonzalo (eds.), The Handbook of Cuffless Blood Pressure Monitoring, https://doi.org/10.1007/978-3-030-24701-0_11

191

192

C. C. Carey

healthcare and personalized medicine. Wearable technology has the potential to improve the management of various diseases with better clinical outcomes, reduced costs (more affordable), and faster access to medical devices that can provide diagnostic, monitoring, and therapeutic solutions. The medical devices industry is a highly regulated sector, particularly in jurisdictions in different parts of the world with established regulatory systems. All devices carry a certain degree of varying risk (low, moderate, and high) that can cause harm to the human body and result in adverse event(s). Examples of devices range from thermometers, stethoscopes to multi-function physiological monitoring systems to ventricular bypass (assist) devices. Regulatory bodies share a common mission, to promote and protect the public health of its citizens. Risks and benefits must be carefully weighed to determine safety, effectiveness, performance, and quality before they can be sold and distributed in the marketplace. The aim is to maximize benefit and minimize risk.

Introduction This chapter focuses on five case studies in different parts of the world with well-­ established regulatory systems: the USA, European Union (EU), Canada, Australia, and Japan. It provides an overview of the regulatory frameworks and key aspects to consider before placing a medical device in legal distribution. In all five different jurisdictions, a risk-based classification scheme is used to guide regulators in the evaluation of risks associated with a medical device. After the classification of the device is determined, appropriate regulatory assessments and controls are applied. The higher the risks to the patient and/or user, the greater is the level of control. It is worthwhile mentioning that smaller countries are working together to harmonize their medical device regulations in order to accelerate the development of economic growth in the region, promote stability, and facilitate collaboration. Such is the case with the ASEAN (Association of Southeast Asian Nations) member states. The ten-member states are Brunei Darussalam, Cambodia, Indonesia, Laos, Malaysia, Myanmar, Philippines, Singapore, Thailand, and Vietnam. They developed the ASEAN Medical Device Directive (AMDD) [1]. The Directive uses a common format to simplify the registration process and application requirements for manufacturers. Although implementation is still in process and some countries are more developed than others, the cooperation allows for all the member states provide a clearer regulatory pathway to market for manufacturers; thus, allowing medical devices to reach the intended population sooner to benefit patients.

11  Key Regulatory Aspects and the Importance of Consensus Standards in Bringing…

193

Medical Device Definition What is a medical device? In very broad terms, a device or equipment intended for medical purposes is a medical device. The World Health Organization (WHO) briefly described a medical device as, An article, instrument, apparatus or machine that is used in the prevention, diagnosis or treatment of illness or disease, or for detecting, measuring, restoring, correcting or modifying the structure or function of the body for some health purpose. Typically, the purpose of a medical device is not achieved by pharmacological, immunological or metabolic means.

This definition is an abridgement from the Global Harmonization Task Force (GHTF) final document that was published by the Study Group 1 on May 12, 2016. [2] GHTF no longer exists. However, their work is continuing under the International Medical Device Regulators Forum (IMDRF) where the GHTF archived documents are kept for preservation. While regulatory authorities across different jurisdictions have very similar definition (in content) of a medical device, it should be recognized that some products may not be considered medical devices in some jurisdictions. Devices in the gray area zone may fall in different risk classification categories (borderline between two different classes). In order to verify if a product meets the definition of a medical device and its regulatory classification, manufacturers should consult with the appropriate regulatory authority responsible for regulating medical devices in their jurisdiction. The US Food and Drug Administration (FDA) regulatory definition of a medical device can be found in Section 201(h) of the Food, Drug and Cosmetic Act. Health Canada (HC) defines the term medical devices in the Food and Drugs Act, while Australia’s Therapeutic Goods Administration (TGA) is stated in 41 BD of the Therapeutic Goods Act 1989 (the Act). In the Government of Japan’s Ministry of Health, Labour and Welfare (MHLW), it is identified in the Pharmaceuticals and Medical Devices Act (PMD Act). The European Union medical device definition is described in Article 1 of the Council Directive 93/42/EEC.

Regulatory Framework and Medical Device Classification Each country or region recognizes the necessity to classify a medical device based on its intended use and the level of risk posed to the patient and/or the user (Risk-­ based Classification Scheme). There are several considerations when ensuring the safety of a device, such as device features and characteristics, design complexity, invasive or noninvasive, and local or systemic use. The environmental conditions of use and whether it is radiation-emitting are other factors to consider as well as the potential harm if misused. Medical devices can be as simple as a tongue depressor to a wearable monitoring electrocardiograph (ECG) watch, an infusion pump, a robotic surgery device—to more complex, high risk device that is life sustaining and life supporting such as a cardiac implantable pacemaker/defibrillator.

194

C. C. Carey

The USA The Center for Devices and Radiological Health (CDRH) is one of several Centers within FDA responsible for regulating firms who manufacture, repackage, relabel, and/or import medical devices sold in the USA. Additionally, CDRH also regulates medical and non-medical radiation-emitting electronic products such as lasers, X-ray systems, ultrasound equipment, microwave ovens, and color televisions. If a product is both a medical device and an emitter, it must meet the regulations pertaining to medical device as well as the requirements for radiation-emitting product. Based on the laws set forth in the Food Drug and Cosmetic Act, most of FDA’s medical device and radiation-emitting product regulations are detailed in Title 21 Code of Federal Regulations (CFR) Parts 800-1299 [1]. The FDA recognizes three levels of classification based on their risks and regulatory controls necessary to provide a reasonable assurance of safety and effectiveness. The 1976 Medical Device Amendment Act created a risk-based classification system and classified all medical devices into three classes, Class I, II, and III [2]. Device classification also considers the intended use of the device and the indications for use. For example, an external defibrillator is intended to restore a normal heartbeat by sending an electric pulse or shock to the heart with an irregular and abnormal heart beat (arrhythmia). It is indicated for use on victims of cardiac arrest who are unconscious, not breathing and without circulation (without a pulse). In Table 11.1 below are the three different classes in the FDA device classification system, including the fact that a device may also fall under the exempt status. An exempt device means the FDA will not require a premarketing application. However, it is not exempt from the General Controls requirements. Most Class I devices are Exempt. Some Class II devices may have also been exempted by regulation. In this case, it can be viewed as three classes with four categories. General Controls are basic requirements necessary for all devices irrespective of their classification, unless exempted by regulation (example, a toothbrush). Devices must not be adulterated or misbranded. Other general controls also include adherence to Good Manufacturing Practices (GMP)/Quality System Regulations (QS Regulations), Premarket notification (or 510(k)) unless exempt, Records and Reports (adverse event reporting), device tracking, and UDI (unique device identification).

Table 11.1  US medical device classification Classification Class I, class II exempt Class I Class II Class III

Risk level Low Low to moderate risk Moderate to high risk High risk

Regulatory control/application General controls/exempt from 510(k) General controls/510(k) General controls, special controls/510(k) General controls, premarket approval (PMA

11  Key Regulatory Aspects and the Importance of Consensus Standards in Bringing…

195

The classification of the device will determine the path to market and the premarketing application required. In general, devices under Class I or Class II (example, blood pressure measuring devices) will require a premarket notification or 510(k) submission to the FDA. An FDA clearance letter will be needed before going to market. FDA Class III devices (example, automated external defibrillators)  will require a premarket approval (PMA) process. In this case, an FDA approval letter will be needed before going to market. Scientific reviews of premarketing applications are conducted by FDA. FDA has a Third-Party Review Program and accredits the Third Party to review some Class II devices.

European Union The EU does not have a “Food and Drugs Administration” but harmonizes their efforts into one law that can be applied throughout the European Union. In close cooperation with Member State’s Health Authorities, the European Commission has the task of regulating medical devices and harmonizing requirements and legislation adopted through a set of directives, collectively known as Medical Device Directive (MDD). Manufacturers need to be aware of the current changes in the EU legal regulatory framework for medical devices. Regulation (EU) 2017/745 [3] repeals the existing Council Directives, Medical Devices (93/42/EEC) and Active Implantable Medical Devices (90/385/EEC) and will officially replace them in 2020. EU Medical Device Regulation (MDR) 2017/745 was published on May 5, 2017 and came into force on May 25, 2017. Manufacturers with currently approved medical device have 3 years transition period to meet the new EU MDR requirements. The implementation date for the European Union Medical Device Regulation is May 22, 2020. The device classification structure in the EU also follows a risk-based system. However, determining classification and the path to market employs a different structure. As a general rule, to gain market access in Europe will require obtaining a CE (Conformité Européenne) marking approval. CE mark is the legal requirement to place a product in the EU.  With a CE mark, it signifies that a medical device meets all relevant essential requirements of the European Directives that outline the safety and performance requirements for medical devices in the European Union. The body within the government of the EU Member States which has the authority to ensure that the requirements of the CE marking directives are carried out in that particular member state is known as Competent Authority. The certification organization that conducts the conformity assessment against the relevant EU Directives (which the national Competent Authority of a member state designates to carry out the procedure and issue the CE Certificate) is referred to as Notified Body (NB). The conformity assessment normally involves an audit of the manufacturer’s quality system and a review of manufacturer’s technical documentation on the safety and performance of the device.

196

C. C. Carey

To decide what is required to CE mark a device, it is important to determine the classification first. The EU rule-based classification scheme for medical devices is set out in Annex IX of Directive 93/42/EEC within the MDD [4]. The European Medicines Agency (EMA) is also involved in the assessment of certain categories of medical devices. By following the rule-based system, each manufacturer could determine the classification of its own device as early as possible in device development; they could classify their own devices. Intended use determines the classification. The classification rules consider criteria such as duration of contact with the patient, degree of invasiveness, and the part of the body affected by the use of the device. In the initial process of probable identification of device class, the devices are segmented into broad categories: noninvasive, invasive, active, and special rules. The conformity assessment routes lead to other routes (applicable certain rules) within the directive. They are further segmented into classes as shown in Table 11.2. The classification categories are Class I (including Is (sterile) & Im (measuring function)), Class IIa, Class IIb and Class III, with Class III ranked as the highest risk. The higher the classification, the greater is the level of assessment. With exception of Class I, devices will generally require working with a Notified Body.

Canada Health Canada’s Medical Devices Bureau of the Therapeutic Products Directorate (TPD) is the government authority who reviews, evaluates, and monitors diagnostic and therapeutic medical devices—to assess their safety, effectiveness and quality before being commercialized in Canada. The regulatory and legal framework are detailed under the authority of the Food and Drugs Act. The medical device classification scheme is also based on risk and is rule-based similar to the EU system. There are four device classifications using a set of 16 rules found in the Canadian Medical Devices Regulations (CMDR) SOR/98-282 [5] There is also a published Guidance on the Risk-based Classification System for Non-In Vitro Diagnostic Devices [6]. Medical devices are classified into one of four classes, Class I, II, III, and IV where I represents the lowest risk and IV represents the highest risk (Table 11.3). In order to determine the appropriate classification of the device, the Classification Rules for Medical Devices as detailed in Schedule 1 of the Medical Device Regulations (Regulations) must be applied. As already mentioned, the rules were developed with strong similarities to the European Union’s Table 11.2  The EU medical device classification system (devices require CE mark)

Classification Class I, including Is and Im Class IIa Class IIb Class III

Risk level Low to medium risk Medium risk Medium to high risk High risk

11  Key Regulatory Aspects and the Importance of Consensus Standards in Bringing… Table 11.3 Canada’s medical device classification system

Classification Class I Class II Class III Class IV

Risk level Lowest risk Low to medium risk Medium to high Highest risk

197

Type of licence Establishment licence Medical device licence Medical device licence Medical device licence

Council Directive 93/42/EEC. It cannot be assumed, however, that a medical device classified in one class according to the European Union’s classification system will be classified in the same class based on the Canadian classification system or as in other jurisdictions. Be aware that many Class I products in the USA may be assessed as Class II products in Canada. Some USA Class II products could also fall into Class III Canadian classification system. With the exception of Class I devices, a Medical Device Licence is required for Class II, III, and IV products before they can be sold in Canada. If the information submitted in the Medical Device Licence Application meets the requirements of the Medical Device Regulations, a License is issued. For Class I devices, they are monitored through Establishment Licences. An Establishment Licence allows importers, distributors and manufacturers to operate in Canada if they do not sell their medical devices through a licensed importer or distributor.

Australia At the time of this writing, the Australian Regulatory Guidelines for Medical Devices (ARGMD) Version 1.1 (May 11, 2011) is currently under review.1 The Therapeutic Goods Administration (TGA) under the Australian government’s Department of Health and Ageing oversees medical device regulation. A major regulation is the Therapeutic Goods (Medical Devices) Regulations 2002 [7]. Australian TGA’s requirements are determined by the medical device intended purpose and its classification category, similar to the risk-based management paradigm in the USA, the EU, Canada, and Japan. It also adds separate fifth classification for active implantable medical devices. A manufacturer determines classification of their medical device using Schedule 2 of the Therapeutic Goods (Medical Devices) Regulations 2002. The risk-based approach is matched to the level of regulation with the risks presented by particular therapeutic goods. Table  11.4 lists the five main classification for medical devices depending on the level of risk they pose, Class I, Class IIa, Class IIb, Class III and Active Implantable Medical Devices (AIMD). Every medical device is assessed against a set of Essential Principles (conformity assessment)—such as, compliance with quality, safety and performance ­principles as well as compliance with regulatory controls for manufacturing pro The information provided here is based on information available online (accessed: May 1, 2019) https://www.tga.gov.au/publication/australian-regulatory-guidelines-medical-devices-argmd 1

198

C. C. Carey

Table 11.4  Australia’s medical device classification system Classification Class I Class I—supplied sterile Class I—with measuring function Class IIa Class IIb Class III Active implantable medical devices

Risk level Lowest Low to medium risk

Medium to high risk High risk High risk

cesses. For very low risk devices, these products can be self-certified. For all other devices, conformity assessment can be conducted by TGA or a recognized independent body, including notified bodies in Europe or US FDA-accredited bodies. TGA can still choose to audit any device (application audit) as an added layer of assurance that a device meets Australia’s regulatory requirement prior to marketing. Although there is close alignment in the conformity assessment procedures between the EU and Australia, it should be recognized that differences between Australian Essential Principles and EU Essential Requirements exist. The upside is that with a European CE marking, the TGA’s approval process will be easier. Australia recognizes CE marking. Declarations of Conformity to the Australian Regulations are also required in order to register with TGA. Market authorization is by inclusion in the Australian Register of Therapeutic Goods (ARTG). ARTG is an electronic, searchable registration system that holds information about a product name, device classification, ‘sponsor’ and manufacturer details. Every ARTG entry belongs to a Sponsor. A Sponsor is a person or company who is responsible for applying and for supplying a medical device in Australia, as well as for maintaining the ARTG. All classes of devices must submit a Medical Device Application (Intended purpose, Classification, and GMDN code2) through TGA’s eBS (eBusiness Services) portal system. If TGA approves the application, an ARTG listing number will be issued (Certificate of Inclusion). The listing will be in the ARTG database on the TGA website and can begin marketing the device in Australia. More recently, TGA published a new guidance [8] on the use of Market Authorization evidence from regulatory bodies by Australian market applicants to abridge the TGA conformity assessment process, allowing speedier commercialization in the country. The regulatory bodies include the US Food and Drug Administration, Health Canada, Japan’s Pharmaceutical and Medical Devices Agency (PMDA), and Medical Device Single Audit Program (MDSAP) audit organizations.

 The sponsor obtains the classification and GMDN (Global Medical Device Nomenclature) codes from the manufacturer. GMDN is a list of internationally agreed generic names/descriptors used to identify all medical device products and managed by the GMDN Agency. 2

11  Key Regulatory Aspects and the Importance of Consensus Standards in Bringing…

199

Japan In Japan, the responsibility to regulate the sale and distribution of medical devices to protect and promote the health of its citizens is performed by a close cooperation and collaboration between the government and an independent administrative agency. The Pharmaceutical Safety and Environmental Health Bureau (PSEHB) under the Ministry of Health, Labor, and Welfare (MHLW)3 is in charge of developing regulatory policies, issuing Ministerial Ordinances, and implementing safety standards. PSEHB is the final judge on approval decisions. On the other hand, the Pharmaceutical Medical Devices Agency (PMDA) has been providing the technical functions, e.g., approval reviews of medical device applications, consultations concerning clinical trials, GMP inspections, monitoring of adverse events reports and other administrative functions. The functional integration of MHLW and PMDA allows them to handle a wider range of activities from clinical studies to approval reviews to post-market measures throughout the device lifecycle. Japan has also established the Registered Certification Bodies (RCB) third party program. RCBs are certified by PMDA and can perform technical reviews for certain designated products. The law and regulations that govern MHLW’s authority to regulate medical devices is written in the Pharmaceutical and Medical Devices Act (PMD Act). In November 2014, it replaced the Japanese Pharmaceutical Affairs Law (PAL). PAL was renamed as the Law for Ensuring Quality, Efficacy, and Safety of Drugs and Medical Devices (commonly known as the Pharmaceutical and Medical Devices Act [9]. Medical Device Registration is required before a medical device is placed in the market.4 Japan has a Marketing Authorization Holder (MAH) system in which only local companies in Japan with a valid MAH license may import and sell medical products into the Japanese market. It is a mechanism to ensure that all imported medical products have a local company in Japan that can take full regulatory responsibility for the imported medical products. Thus, all foreign medical device companies who want to sell products in Japan must first be approved as certified non-Japanese manufacturers. Then, a Marketing  Authorization  Holder (MAH)  or a designated Table 11.5  Japan’s medical device classification system Classification Class I (general medical device) Class II (controlled medical device) Class III (specially controlled device) Class IV (specially controlled device)

Risk level Extremely low risk Low risk Medium/high risk High risk

Registration type Notification (self declaration) Certification or approval Certification or approval Approval (PMDA)

 Organization Chart, MHLW https://www.mhlw.go.jp/english/org/detail/dl/organigram.pdf  Be aware that the term Registration carry a different meaning in the US system. It is associated with Establishment Registration and Device Listing; which is done only after the device has obtained FDA market clearance or approval. 3 4

200

C. C. Carey

­ arketing authorization holder (DMAH) is appointed to manage their product regm istrations and interact with Japan’s regulatory authorities. Regarding device classification, similar to the other jurisdictions discussed above, the Japanese regulatory framework uses a risk-based classification system, Class I, II, III and IV (Table 11.5) which aligns with the principles outlined in the ‘Principles of Medical Devices Classification’ Study Group 1, GHTF Final Document, June 27, 2006.5 For a Class I general medical device, it does not require an approval process; only a notification is required and the manufacturer does a self-­ declaration. A Class II controlled device can be designated for RCB review or reviewed by PMDA. A Class III and IV specially controlled device must undergo PMDA and MHLW review for approval. However, under the new PMD Act some Class III devices may be designated for RCB review and certification.

Consensus Standards Good technical standards are built on globally accepted principles of consensus, openness, balance, and transparency. The recognition and use of consensus-driven, harmonized standards play a prominent and significant role in the regulation of medical devices around the world. Standards should address important public health issues. Whether the work was developed and consolidated at the national or international level, the use of established standards provides regulatory authorities an additional tool to carry out its mission—to promote and protect public health. Recognized published standards are used to satisfy certain premarket requirements to validate device conformance to safety, quality and effectiveness or to demonstrate compliance with Essential Principles. Some commonly used international standards by medical device manufacturers are: • • • •

ISO 13485—Quality management systems ISO 14971—Application of risk management to medical devices ISO 10993—Biological evaluation of medical devices ISO 60601—Series of technical standards for the safety and performance of medical electrical equipment

Regulators have published guidance documents on the appropriate use of consensus standards, how industry can use “Declarations of Conformity” and when additional data is needed. For example, the US FDA’s final guidance, Appropriate Use of Voluntary Consensus Standards in Premarket Submissions for Medical Devices, Guidance for Industry and Food and Drug Administration Staff, was issued September 14, 2018.6 The FDA also has a searchable Recognized Consensus Standards data-

 GHTF archived documents are accessible at the IMDRF site, http://www.imdrf.org/ghtf/ghtfarchives-sg1.asp 6  https://www.fda.gov/media/71983/download 5

11  Key Regulatory Aspects and the Importance of Consensus Standards in Bringing…

201

base. These are standards recognized by the FDA, either wholly or in part.7 Health Canada’s document, Guidance Document: Recognition and Use of Standards under the Medical Devices Regulations, was effective on September 11, 2006 and published by the authority of the Ministry of Health.8 Japan’s PMDA outlines the Certification Criteria, Approval Criteria and Review Guideline for medical devices in which they identify the use of technical standards.9 In the EU, harmonized standards on medical devices are referred to in Article 8 of Regulation (EU/MDR) 2017/745. A group of experts representing competent authorities of EU countries was established as the Medical Device Coordination Group (the MDCG). They participate in the development of proposals for standardization and provide assistance to implement the regulation.10 Manufacturers should always check the government websites for updated versions and information related to standards and conformity assessments program and policies.

Conclusion Around the world, it is apparent that regulatory authorities in different jurisdictions have something in common. As described in this chapter, they all apply the principles of risk-based regulatory paradigm in the management and supervision of medical devices. The methodology  is grounded in risk and provides assurance that products are safe, effective, and perform as they are intended prior to their sale and distribution to the broader population. While this chapter focused on device classification and key aspects of regulations in bringing new devices in the marketplace, it is important to recognize that the same risk-based concepts are applied across the device life cycle from design to premarket to postmarket stages in order to accomplish the mission of promoting and protecting public health. It is important for manufacturers to understand the differences and similarities in the structure, regulatory approaches and regulatory pathways to market and leverage harmonized guidelines and policies. In all cases, the use of established industry consensus standards and adherence to conformity assessments can meet certain aspects of the regulatory requirements. Standards should promote and not hinder innovation. However, they must also be periodically reviewed to stay in-step with technological advances and real-world experience.

 https://www.accessdata.fda.gov/scripts/cdrh/cfdocs/cfStandards/search.cfm  https://www.canada.ca/content/dam/hc-sc/migration/hc-sc/dhp-mps/alt_formats/hpfb-dgpsa/pdf/ md-im/md_gd_standards_im_ld_normes-eng.pdf 9  http://www.std.pmda.go.jp/stdDB/index_en.html 10  https://ec.europa.eu/growth/sectors/medical-devices/regulatory-framework_en 7 8

202

C. C. Carey

References 1. ASEAN Medical Device Directive. 2015 [Online]. https://asean.org/wp-content/ uploads/2016/06/22.-September-2015-ASEAN-Medical-Device Directive.pdf. 2. Study Group 1 of the Global Harmonization Task Force (GHTF). Definition of the terms ‘Medical Device’ and ‘In Vitro Diagnostic (IVD) Medical Device’. GHTF Final Document (revision of GHTF/SG1/N29:2005). 2012. [Online]. http://www.imdrf.org/docs/ghtf/final/sg1/ technical-docs/ghtf-sg1-n071-2012-definition-of-terms-120516.pdf. 3. U.S.  Code of Federal Regulations (CFR) [Online]. https://www.accessdata.fda.gov/scripts/ cdrh/cfdocs/cfcfr/CFRSearch.cfm?CFRPartFrom=800&CFRPartTo=1299. 4. Public Law 94-295 94th Congress, Medical Device Amendments of 1976, amend the Federal, Food, Drug, and Cosmetic Act, 28 May 1976. 5. Legislative Act, Regulation (EU) 2017/745 of the European Parliament and of the Council of 5 April 2017. 6. European Commission. Guidance document, classification of medical devices. Annex IX of Council Directive 93/42/EEC, MEDDEV 2.4/1 rev.9, June 2010. 7. Canadian Medical Devices Regulations (CMDR) SOR/98-282, last amended March 4, 2019. Current to March 27, 2019. 8. Minister or Health, Government of Canada. Guidance on the risk-based classification system for non-In Vitro Diagnostic Devices (non-IVDDs). Effective date, June 12, 2015. 9. Therapeutic Goods (Medical Device Regulations 2002. Compilation No. 39), latest compilation date is December 1, 2018. 10. Australian Therapeutics Goods Administration Health Safety Regulation. Use of market authorisation evidence from comparable overseas regulators/assessment bodies for medical devices (including IVDs). Ver 1.1, Nov 2018. 11. Regulatory Information Task Force, Japanese Pharmaceutical Manufacturers Association. Pharmaceutical administration and regulations in Japan; 2017.

Chapter 12

Design of Clinical Trials to Validate Cuffless Blood Pressure Monitors Willem J. Verberk

Abstract  Currently the market is flooded with cuffless blood pressure monitors. However, these devices are not always as accurate as consumers and physicians might expect. Therefore, the clinical accuracy of these blood pressure monitors should be verified using a suitable standard. As the current standards for blood pressure monitors are designed for cuff-based devices these are not suitable for verifying the accuracy of monitors without cuffs due to essential differences between blood pressure monitors with and without cuffs. For example, almost all cuffless monitors must be calibrated, may be used as wearables during activity and may measure blood pressure at body parts other than upper-arm and wrist. For this reason, there is an urgent need for a new standard to test cuffless monitors that covers these different aspects. This chapter describes characteristics of cuffless monitors, highlights the differences between cuff-based and cuffless monitors, and proposes methods for testing. Additionally, attention is paid to the selection of participants in relation to good clinical practice and to statistical reporting of the figures. Finally, some published clinical studies to the accuracy of cuffless blood pressure monitors are discussed. Keywords  Cuffless blood pressure monitor · Cuff · Sphygmomanometer · Blood pressure measurement · Validation protocol · Standards · Continuous blood pressure measurement · Intermittent blood pressure measurement · Good clinical practice

W. J. Verberk, Ph.D (*) Microlife AG, Widnau, Switzerland CARIM School for Cardiovascular Diseases, Maastricht University, Maastricht, The Netherlands e-mail: [email protected] © Springer Nature Switzerland AG 2019 J. Solà, R. Delgado-Gonzalo (eds.), The Handbook of Cuffless Blood Pressure Monitoring, https://doi.org/10.1007/978-3-030-24701-0_12

203

204

W. J. Verberk

Introduction Blood pressure measurement is the most performed medical procedure. Usually it is done by an upper-arm cuff inflating to supra-systolic pressure after which blood pressure is determined by mean of Korotkoff sounds (auscultatory technique) or the oscillometric technique. However, the cuff has numerous disadvantages. Patients consider the cuff as inconvenient or causing pain, particularly among those with high blood pressure. With frequent measurements such as 24-h blood pressure measurement it can lead to skin irritation, bruising and sleep disturbance. In addition, cuff pressure increases both systolic and diastolic blood pressure by approximately 7 and 5 mmHg, respectively [1]. Another serious challenge in clinical practice, can be the finding of a cuff that fits correctly for some patients. This is not only related to the size of the arm but also to deviating arm shapes (e.g., conical shape) and composition of tissue (fat, muscles) can lead to incorrect blood pressure measurement [2]. Finally, cuff blood pressure measurement on the upper arm only allows intermittent blood pressure measurement and is therefore not capable of measuring sudden changes in blood pressure which makes it unsuitable for use in acute care. The above disadvantages related to upper-arm cuffs have encouraged many manufacturers and researchers to develop noninvasive (continuous or intermittent) cuffless blood pressure monitors. Thus far, most of these cuffless monitors estimate blood pressure transcutaneously using the principal of pulse transit time (PTT) or pulse arrival time (PAT) [3]. This is based on the time it takes for a pulse wave to travel through the artery from one place to another at a fixed distance apart. When knowing the time delay and the distance between the two pulsatile arterial hemodynamic related signals the pulse wave velocity (PWV) can be calculated [4]. With the PWV the blood pressure value can be determined. Cuff-less devices have high potential because the sensors can be easily implemented in so-called wearables (watch, shirt, patch, etc.) that makes it suitable during activity (e.g., 24-h ambulatory blood pressure measurement and fitness purpose). In addition, the technique can be used in combination with, or implemented in smartphones which enables easy blood pressure measurement to a wide population. This has all led and will lead to numerous cuff-less blood pressure monitors entering the market. Obviously, not all these devices are as accurate as consumers and physicians might expect. Because the devices are relatively new and the process of making standards is time consuming, there is currently no internationally accepted standard to verify the accuracy of these devices. However, a standard is urgently needed for both clinicians and consumers who are overloaded with advertisements for “quick and easy blood pressure measurement.” In 2014, The Institute of Electrical and Electronics Engineers (IEEE) presented a standard entitled “IEEE Standard for Wearable, Cuff-less Blood Pressure Measuring Devices” [5]. However, this standard does not seem to cover all aspects of cuffless blood pressure measurement, especially regarding validation of continuous cuffless blood pressure monitors. Currently, the ANSI/AAMI/ISO Joint Working Group is developing a new standard (ISO/CD 81060-3) for clinical investigation of

12  Design of Clinical Trials to Validate Cuffless Blood Pressure Monitors

205

continuous automated measurement with noninvasive sphygmomanometers. This chapter is aimed at highlighting certain aspects that need to be considered for designing a clinical trial for validation of cuff-less blood pressure monitors.

 uff or Cuffless Devices and Continuous or Intermittent C Devices Blood pressure monitors can be distinguished according to two different classifications; cuff versus cuff-less devices and continuous compared to intermittent devices as explained in Table 12.1. Although, the name “cuff-based monitors” seems self-­ explaining, there is sometimes confusion about the definition of a cuff. A cuff is a part of the blood pressure monitor that is wrapped around a limb (thus not necessarily an arm) of the subject. It has a bladder that can be inflated to partially or fully occlude the artery after which it is deflated. A cuffless blood pressure monitor generally works with a sensor (electrocardiographic, photoplethysmographic, and/or phonocardiographic) that allows the assessment of the pulsation of arterioles (heart beat) at different skin depths. The fact that a cuff-less blood pressure monitor enables continuous blood pressure measurement does not necessarily mean that it is a continuous blood pressure monitor. A continuous noninvasive automated sphygmomanometer is, according to the definition, “a device able to estimate blood pressure from the pulse wave of each heart cycle without arterial puncture.” However, the device does not need to display the blood pressure values per heartbeat (continuously) but may display the blood pressure value averaged from multiple heart beats instead. This means that cuffless monitors may provide intermittent blood pressure values whereas on the other side cuff-based blood pressure monitors exist that measure blood pressure continuously. For instance, a device that measures finger arterial pressure continuously using a finger cuff and infrared plethysmograph [6]. Table 12.1  Cuff or cuffless devices versus continuous or intermittent blood pressure monitors Cuff-­ based

Cuff-­ less

Continuousa Determines and presents blood pressure per heartbeat using a cuff For example, by using a finger cuff and infrared plethysmograph measurements Determines and presents blood pressure per heartbeat without using a cuff For example, by mean of pulse wave velocity measurements

Intermittent Determines and presents blood pressure over multiple heartbeats For example, by using automated oscillometric measurements Determines blood pressure per heart beat (continuous) but presents blood pressure values averaged over multiple heartbeats For example, by mean of pulse wave velocity measurements

Suitable for verifying sudden changes in blood pressure

a

206

W. J. Verberk

 equired Tests for Validation of Cuffless Blood Pressure R Monitors Current ANSI/AAMI/ISO standards are designed for testing the accuracy of automated cuff-based blood pressure monitors. However, cuffless devices essentially differ from cuff-based automated oscillometric sphygmomanometers in design and often also in intended purpose so current standards are not suitable to test cuffless blood pressure monitors (Table 12.2). The following tests are required for the validation of a cuffless blood pressure monitor;

Static Test Automated oscillometric devices provide intermittent blood pressure values in a static condition. The validation protocol, therefore, is a relatively simple and static test verifying the sphygmomanometer under testing against a reference upper-arm cuff or invasive blood pressure monitor in a similar condition and environment [7, 8]. Reference and test blood pressure measurement can be done either sequentially or simultaneously. The accuracy of intermittent cuffless blood pressure monitors, designed for static measurement, can be tested almost in a similar way. However, an important difference with cuff-based upper-arm devices is that almost all cuffless blood pressure monitors need to be calibrated first. This process of determining subject specific parameters is needed for estimating blood pressure with the cuffless monitor. It affects overall performance of the device and therefore needs to be covered in the new standards.

Calibration Most cuffless monitors estimate blood pressure from PTT obtained PWV using a mathematical formula [9]. As the relationship of PTT to blood pressure (the calibration curve) depends on multiple characteristics such as arterial stiffness and body

Table 12.2  Standards for noninvasive blood pressure monitors ANSI/AAMI/ Standard ISO 81060-1 Device to Non-automated be tested (aneroid and mercury blood pressure monitors) Standard under development

a

ANSI/AAMI/ ISO 81060-2, BHS, IP Intermittent automated blood pressure monitors

ANSI/AAMI/ISO 81060-3a continuous blood pressure monitors (can also provide intermittent blood pressure values)

IEEE cuffless wearable Wearable cuffless blood pressure monitors (seems not suitable for devices that show beat to beat blood pressure values)

12  Design of Clinical Trials to Validate Cuffless Blood Pressure Monitors

207

length it is highly variable from person to person. However, also within a person this relationship fluctuates due to, for example, changes in heart rate, vasomotion, and hydrostatic pressure [10]. Because many arterial and human factors affect the relationship between PWV and blood pressure, calibration of the cuffless device seems indispensable. Some investigators claim that there is no need for calibration when using elastic arteries to measure blood pressure such as the aorta [11], but this needs further investigation. Until then, calibration (or initialization) with a cuff-based upper arm or invasive blood pressure monitor must be done for obtaining subject specific parameters that are needed for estimating blood pressure using a cuffless monitor.

Stability Test Currently, some cuffless blood pressure devices have been validated according to the standards of ANSI/AAMI/ISO 81060-2:2013 [12] or the International Protocol of the ESH 2010 [13]. For these validation studies the device under testing was calibrated shortly before the test measurement was taken. A cuffless device generally appears to be accurate for only a short time after calibration because there is no time for damping or drifting and thus the blood pressure is almost similar to the moment of calibration. However, the accuracy of the cuffless device generally deteriorates with blood pressure changes and/or with time after the initial calibration so that the accuracy over time must be investigated. Obviously, cuffless blood pressure measurement is useless if it requires calibration with an upper-arm cuff monitor before every cuffless measurement. Therefore, the manufacturer must disclose the maximum timeframe in which the device is accurate before it needs recalibration. This means a “stability test” must be done to verify accuracy over a longer period in which recalibration is not needed. The stability of the displayed blood pressure values must be investigated during the whole maximum period specified by the manufacturer. The IEEE standard suggests performing two test measurements within two calibration points; directly after initial calibration and before the next calibration is required. However, when performing only an initial and end-testing the device might not be accurate in-between these tests. Therefore, devices should be tested daily or even hourly depending on its disclosed accuracy period. If the disclosed timeframe covers a long-time period, a higher testfrequency in the beginning followed by less frequent testing towards the end should be recommended. For instance, if the disclosed period is 1 month, twice-daily measurements may be assessed for the first 3 days thereafter daily measurements for 10 days followed by weekly performances until the end.

Test with Blood Pressure Change Many cuffless devices are wearable and measure blood pressure beat by beat that allows tracking acute blood pressure changes over time. This is a major advantage because blood pressure variability is an important cardiovascular risk predictor

208

W. J. Verberk

[14]. Sometimes it is essential to verify acute changes in blood pressure such as in the operation theater or acute care. However, this advantage of cuffless monitors also adds an extra difficulty as its capability to track (quick) blood pressure changes must be tested for accuracy. Noninvasive Blood Pressure Change Test The IEEE standard considers only auscultatory blood pressure measurement as a reference device. Obviously, this method is unsuitable for testing quick blood pressure changes. The performance of three sequential auscultatory blood pressure measurements with 1-min interval time takes approximately 5 min. Nevertheless, the auscultatory method may be used to test slow blood pressure changes (e.g., over 15 min) in intermittent cuffless monitors. In practice this procedure might not be easy because the changed blood pressure value must be relatively stable for more than 5 min. Invasive Blood Pressure Change Test The capability for the cuffless device to follow quick blood pressure changes can only be verified by comparing to simultaneously performed invasive blood pressure measurement. Blood pressure changes to be verified must be significant (15– 30  mmHg) with a defined minimum number of both changes and patients. This means that if not enough blood pressure changes can be obtained, more patients must be included. It may be recommended to present the number of blood pressure changes in a histogram as suggested in the IEEE standards (Fig. 12.1).

Time Duration of Measurement Procedure For intermittent cuffless blood pressure monitors, it is important that the manufacturer discloses the number of heart beats from which the blood pressure value is averaged. Standards should provide a restriction to the time duration cuffless monitors may take for assessing blood pressure. This time should not extend the time it takes for an upper-arm cuff monitor to measure blood pressure (e.g., 30  s). The lowest number of cardiac cycles from which blood pressure is calculated should be used for the validation study. There are two reasons for this: First, it makes the validation procedure shorter and easier to perform, second the 30 s may be indicated as a sort of “worst case scenario.” If the short period of the device under testing appears accurate, one may expect that the average taken over a longer period with more heart beats is more accurate. The latter also indicates that the allowance of a larger time frame might make it too easy to pass the standards.

12  Design of Clinical Trials to Validate Cuffless Blood Pressure Monitors

209

Fig. 12.1  Histogram of blood pressure changes

For continuous cuffless blood pressure measurement validation, a clear time frame should be defined. It may seem logical to use approximately 5 min as this is the same time that is normally taken for intermittent blood pressure comparison. This will lead to approximately 300 measurements (5  ×  60 heart beats) to compare.

Sequential or Simultaneous Comparison The accuracy of intermittent cuffless blood pressure monitors can be verified with either sequential or simultaneous blood pressure measurement.

Sequential Sequential blood pressure measurement only seems plausible for auscultatory measurement reference of intermittent (limb) monitors and therefore the existing standard [8] may be followed; Triplicate test device measurements should follow reference measurements. At least 30 s should be allowed after reference measurement to avoid venous congestion, but not more than 60 s to reduce differences due to blood pressure variability. Differences are calculated from the test device measurement and the average of the adjacent two reference readings. Preferably, the same limb should be used to avoid lateral blood pressure differences.

210

W. J. Verberk

Simultaneous For simultaneous comparison using an auscultatory reference the test device should start with the cuff deflation of the reference sphygmomanometer. If both the reference blood pressure monitor and test device are (upper) arm devices and compared with simultaneous measurement, then the inter-arm blood pressure difference must be verified with simultaneous measurements [15]. If the difference exceeds a certain threshold (e.g., 5 mmHg) then the patient should be excluded from testing. It could also be decided to compensate for the differences. Then the reference and test devices should be changed between arms and the difference related to inter-arm difference can be subtracted [7]. For verifying a longer time-period the usefulness of compensating for inter-arm blood pressure difference seems not a good idea as reproducibility of (nonpathological) inter-arm difference in the long-term may be questioned [16]. For testing a cuffless intermittent device using invasive blood pressure measurement, the average blood pressure value should be calculated from the same number of heartbeats for both the reference and test method. Obviously, cuffless continuous blood pressure measurement must be validated with simultaneous invasive measurements.

Blood Pressure Measurement at Different Parts of the Body Wearable cuffless devices measure blood pressure at different sites of the body. Therefore, the reference measurement site should be disclosed by the manufacturer. As different sites of the body have different blood pressure values the test measurement site should be as close as possible to the reference site. In addition, new standards must define accepted reference sites, or state that any site may be used as a reference if appropriately measured. Current standards state that the same limb, the aorta and subclavian or femoral arteries are appropriate references for simultaneous validation [7].

Statistical Analysis and Reporting Published clinical studies to the accuracy of cuffless devices show a wide heterogeneity in design (sample size, number of measurements, reference device), statistical analysis and reporting of data. This makes it difficult to interpret the outcome. A universal protocol with proper design, described statistics and graphs can help to understand the outcome and makes it easy to compare studies. In this section, a list of important variables to report is provided.

12  Design of Clinical Trials to Validate Cuffless Blood Pressure Monitors

211

Population Sample Blood Pressure Ranges Most studies that investigated the accuracy of cuffless blood pressure measurement thus far, used a limited number of (healthy) patients who are within an average blood pressure range. However, Ding et al. showed that blood pressure from a normotensive group agreed better with the reference blood pressure than those of the hypertensive group [17]. This underscores the importance of selecting patients with a wide blood pressure range distribution as is required in current standards. Gender Distribution Almost all validation guidelines require an equal distribution for women and men (e.g., IEEE n  =  20/20; ANSI/AAMI/ISO  >  30%) despite the lack of convincing evidence that gender affects blood pressure measurement. However, gender appears to be of significant influence on vessel composition [18, 19]. Because most cuffless blood pressure monitors directly depend on PWV an equal distribution between men and women seems essential for clinical investigation of measurement accuracy. For invasive blood pressure measurement, it may be difficult to find a perfect distribution. Therefore, the current requirements of a least 30% of either gender in the total population seems appropriate. Sample Size and Measurement Frequency Sample size determination is essential in the design of clinical trials; less patients reduce power, but more patients may cause unnecessary inconvenience. The latter applies certainly for invasive blood pressure measurement but also for the duration of the procedure required for testing noninvasive monitors as this can take several days to weeks. Sample Size for Noninvasive Testing The ANSI/AAMI/ISO standard requires a minimum of 85 participants and a minimum of three paired tests (test and reference). This number was based on what was considered an acceptable difference between the average reference and test blood pressure value, which is 5 mmHg and an acceptable standard deviation of 8 mmHg. The statistical approach was a t-test with a 5% chance that a proper device was rejected (α = 0.05) and a 2% chance that an inaccurate device would be approved 2% (β = 0.02) [20]. See Formula 12.1.

212

W. J. Verberk

Formula 12.1 Subjects calculation for ANSI/AAMI/ISO standards: 2



  σ2 n = 2  z α + z1− β  2 .  1− 2  δ

(n = sample size Ζ = statistic for normal distribution σ = Sample standard deviation; 2 δ = mean difference) 2 (8 ) With the in the text described values this results in: 2 (1.96 + 2.06 ) = 83 2 ( 5) subjects The IEEE standard proposes 45 subjects based on an analysis by Yan et al. [21] who claim that the ANSI/AAMI/ISO sample size calculation is based on the erroneous assumption that the error rate of the 85 included subjects have a normal distribution. According to the authors, blood pressure variability is related to the height of blood pressure values which is not covered by the included three blood pressure ranges. Therefore, the authors consider a t-distribution (four degrees of freedom [t4]) to be a better model than the normal distribution on prediction of the errors for most blood pressure monitors [22]. Sample Size for Invasive Testing According to the guidelines for good clinical practice, one should not perform invasive blood pressure measurement for the sole purpose of a validation study [23]. Therefore, it may be difficult to find patients so that less patients (15 instead of 85 as in ANSI/AAMI/ISO) and less strict selection criteria should be allowed. For example, there are no age range requirements for intra-arterial verification (as in the first edition of the AAMI standard). This all leads to the fact that the range of observed blood pressures can be less evenly distributed. The lower number of patients can be partly compensated by increasing the measurement procedure frequency (e.g., 10 instead of 3 performances). This in total does not lead to the same measurement number as for noninvasive comparison (n = 85 × 3 measurements) but because invasive blood pressure measurement is performed simultaneously, differences because of blood pressure variability are low. Special Subject Population If a cuffless blood pressure monitor is intended for use in a special-subject population (e.g., children, pregnant women, end stage renal disease patients) a special validation is needed. As physiological characteristics (vessel composition, respiration rate, heart rate etc.) of these patients may influence the accuracy of the device. However, if the device is already validated in a general population there is no need to investigate the entire number of subjects again. Then an additional 35 (with

12  Design of Clinical Trials to Validate Cuffless Blood Pressure Monitors

213

auscultatory reference) or 5 subjects (with invasive reference) as recommended in current standards should suffice [7].

Performance Criteria Mean Difference and Standard Deviation The simplest and most commonly used method to present differences between reference and test blood pressure measurement is by calculation of the mean difference and standard deviation. The current standard for noninvasive validation of automated sphygmomanometers prescribes the mean value and standard deviation of the differences from the average of three test determinations and three reference determinations [7]. This standard has defined an acceptable difference from “the gold standard” (Mercury) of up to 5 mmHg. Due to the variation in blood pressure it is also agreed that a single deviation of more than 5 mmHg should not immediately lead to rejection of a device. Therefore, a standard deviation of 8 mmHg from at least 255 measurements is allowed. However, the criteria of 5 ± 8 mmHg could still lead to a high number of subjects that fall outside the acceptable rate. Therefore, the standard added a second criteria; the mean error of the average measurements from at least 85 subjects should be 5 mmHg with acceptable SD-values (from 4.79 to 6.95) that decrease as the mean errors increase. This schedule must ensure that at least 85% of all subjects fall within the 10-mmHg error range (and approximately 50% of subjects have errors that fall within the commonly accepted error of 5 mmHg). The acceptable differences of 5 ± 8 mmHg between the reference and test device were determined a few decades ago. It may be expected that at that time the quality of the automated blood pressure monitors were lower than is currently the case. Nowadays, it may be questioned if we should continue accepting this error or perhaps it is time to lower the acceptance threshold considering the method of validation. However, if a device is manually tested by two observers in sequential order, it might not be possible to decrease this threshold due to inter-observer bias and blood pressure variability. On the other hand, use of current developed techniques that support observers with their readings might decrease observer bias [24]. Corrected Standard Deviation for Repeated Measurements In continuous blood pressure measurement validation, many more readings are compared to the reference than with intermittent measurement. Whereas the latter requires 3 or 10 measurement procedures, simultaneous continuous measurement over a period of 5 min leads to approximately 300 measurement comparisons for one subject. This means that many pairs of measurements are made in relatively few patients, which needs to be considered in the statistical analysis. For this Bland and Altman suggested a corrected experimental standard deviation. This considers both

214

W. J. Verberk

the within and between subject variance. The measure of correction correlates to the ratio number of measurements per subjects to number of study participants [25]. Following this, the previously defined acceptable error of 5  mmHg and standard deviation of 8  mmHg might not be acceptable anymore. Therefore, the accepted differences may need to be reconsidered based on current clinical evidence. Error-Bands BHS and ESH Protocols Instead of using the mean and standard deviation some standards use so-called error bands. The mean error of the subjects from ANSI/AAMI/ISO is in line with the BHS-protocol that requires that 60% of the cumulative readings fall within 5 mmHg, 85% within 10 mmHg and 95% within 15 mmHg [26]. Later, the BHS protocol was replaced by the logistically easier ESH protocol that requires 33 instead of 85 subjects. The ESH protocol demands 65% within 5 mmHg, 81% within 10 mmHg and 96% within 15 mmHg [27]. Recently, it has been decided to discontinue the ESH protocol because of concerns regarding its statistical power [20] and to design a new joint protocol of ANSI/AAMI/ISO and ESH [28]. The new “universal protocol” mostly resembles the ANSI/AAMI/ISO protocol which indicates that mean and standard deviations is a preferred outcome above error-bands. Absolute Differences Mean absolute differences (MAD) and mean absolute percentage difference (MAPD) is suggested in the IEEE standard. Because differences may also relate to blood pressure levels, the accuracy is also calculated within the four different blood pressure classifications they propose. MAD at different blood pressure classification categories should be within 6 mmHg for systolic and diastolic measurement analyzed separately. According to the designers of the IEEE standard MAD has the advantage that the standard deviation is not needed as the absolute values also consider the spread and the range. As compared to standard deviation MAD does not put so much weight on outliers which relevance could be discussed (a coincidental error or important measure). Yan et al. showed that a MAD of approximately 5–6 is considered acceptable as this comes close to the criteria of ANSI/AAMI/ISO and BHS [29].

Statistical Figures Bland–Altman Plot and Histogram of Differences For comparing the agreement between two types of measurements the Bland– Altman plot (Fig. 12.2) is often used [30]. This plot, that shows the 95% limits of agreement between two methods, was introduced in 1986 to replace the, until then

12  Design of Clinical Trials to Validate Cuffless Blood Pressure Monitors

215

Fig. 12.2  Bland–Altman plot; BPref indicates blood pressure values of the reference technology; BPDUT blood pressure values of the Device Under Testing

most used outcome, correlation coefficient. The latter was considered misleading because a high correlation does not necessarily imply a good agreement between two methods. The Bland–Altman plot shows the difference of two paired measurements against the mean of these two measurements. It depends on the assumptions that the mean and standard deviation of the differences are constant throughout the range of measurements, and that these differences have an approximately normal distribution [31]. To check the latter a histogram of the differences is proposed as demonstrated in Fig. 12.3. If the histogram is skewed the Bland–Altman plot may lead to misinterpretation. The 95% limits of agreement are shown as the mean of the two values ±1.96 standard deviations. These limits are expected to cover the general difference between the method measurements for 95% of pairs. Four Quadrant Plot (for Blood Pressure Changes) To visualize how the device under testing behaves in blood pressure changes (∆PDUT) as compared to the blood pressure changes of the reference (∆PREF) device a Four-quadrant plot (FQP) may be helpful (Fig. 12.4). When both the studied technology and the reference technology indicate an increase in blood pressure, the data points will appear in the upper right quadrant of the FQP. Similarly, the lower left quadrant contains data points resulting from decreases in blood pressure for both the test and reference device. Therefore, the upper right and the lower left quadrants of the FQP represent concordant measurements of the studied and reference technology regarding direction of changes. The more dots in these areas the more concordant a device is. However, this does not necessarily mean that the device is

216

W. J. Verberk

Fig. 12.3  Histogram of differences between reference device and the device under testing

Fig. 12.4  Four quadrant plot of the continuous noninvasive sphygmomanometer versus the reference sphygmomanometer. The values on the X-axis refer to ∆P values of the reference technology (REF), whereas the y-axis refers to the ∆ values of the Device Under Testing (DUT)

12  Design of Clinical Trials to Validate Cuffless Blood Pressure Monitors

217

accurate. For accuracy the points should be located close to the 45° diagonal (Unity line). In addition, the acceptable maximum error could be defined in advance and presented as lines (PERRORmax) to either site of the 45° diagonal of the concordant quadrants between which the data should fall [32].

 xecute and Monitor the Measurement Campaign; Good E Clinical Practice For the performance of a validation study it is of essential importance that the standards for Good Clinical Practice (GCP) for the clinical investigation of medical devices for human subjects [23] are followed and the study is approved by the local Medical Ethical Committee. For this some items, in particular, apply to cuffless blood pressure device validation and therefore deserve extra attention.

Invasive Blood Pressure Regarding the performance of an invasive blood pressure validation the ANSI/ AAMI/ISO standard states that invasive blood pressure measurement may not be assessed just for the purpose of validating a blood pressure monitor [7]. The studies should be conducted on clinical patients in whom an intra-arterial line has already been placed for reasons other than sphygmomanometer verification. This could include patients undergoing carotid arteriography or cardiac catheterization, or patients undergoing clinical research studies approved by an institutional review board that involve intra-arterial blood pressure monitoring [33].

Duration of the Blood Pressure Measurement Procedure It should be predetermined how long a person will be measured, and this period should not be longer than strictly needed to ensure the convenience and/or safety of the study participant. Furthermore, all requirements according to GCP and the declaration of Helsinki should be respected, such as obtaining the informed consent from all participants and all adverse events must be reported.

Reference Device According to ANSI/AAMI/ISO standard a reference measurement can be performed auscultatory (both aneroid and mercury) and invasively [7, 33]. The IEEE standard only considers auscultatory blood pressure measurement with a mercury

218

W. J. Verberk

device as a reference [34]. Auscultatory reference measurement should be provided by two trained observers and measured simultaneously with one reference sphygmomanometer (using a “Y” connector). Invasive blood pressure measurement shall comply with the IEC standards [35], the maximum allowable error is ±2 mmHg and the intra-arterial transducer must be kept at the level of the heart.

Observer Measurement For the noninvasive auscultatory evaluation of the cuffless blood pressure measurement two well-trained observers are needed. Measurements must be determined using the first and fifth Korotkoff sound for systolic and diastolic blood pressure, respectively. All measurements should be recorded to the nearest 2 mmHg. If measurements from the two observers are no more than 4 mmHg apart, the mean value of the two is used as the reference measurement. Otherwise, the measurement should be taken again.

Test Environment A validation study should preferably be performed in a recognized center. The test environment room should have a good climate (room temperature) and should be quiet (no disturbing noises). Finally, the test device must be used according to the manufacturer’s instruction.

Phase Yes or No? Although the ANSI/AAMI/ISO standard does not suggest using phases/interim analyses, it deserves consideration from an ethical perspective. Most cuffless blood pressure monitors are at an initial stage of development and tested for the first time on real subjects. Having a phase 1 may prevent unnecessary inconvenience to the participants and wasting time and money. Phase 1 should be considered as an interim analysis, as is performed in many studies. If the device fails at that stage, it is useless to continue. For IEEE in Phase 1, only 20 subjects are recruited for which the MAD of the collected data must be within 7 mmHg for both systolic and diastolic pressure. If the MAD is higher than 7 mmHg further testing is not needed anymore. If lower, then Phase 2 can start in which another 25 subjects are recruited.

12  Design of Clinical Trials to Validate Cuffless Blood Pressure Monitors

219

 ublished Studies to Investigate the Accuracy of Cuffless P Blood Pressure Monitors Thus far, several studies have aimed to “validate” cuffless blood pressure monitors in real patients. However, these accuracy studies contain flaws, did not pass the tests of accuracy or used inappropriate statistics. Therefore, it seems that currently, none of the cuffless devices can be considered accurate for their intended purpose. Some examples are discussed below.

Reference Devices Used Most cuffless continuous blood pressure monitors were not tested against a proper reference measurement but against another noninvasive automated device that measured blood pressure continuously (e.g., a Finapres (Medical Systems, the Netherlands) device) [17, 36]. This device provides beat to beat measurements by means of finger-cuff blood pressure measurement [37]. The fact that a device has been validated against invasive blood pressure does not make it suitable as a reference. The device under testing may have an acceptable difference with the Finapres but not with invasive blood pressure measurement due to the difference between the latter two. Studies to the accuracy of intermittent cuffless blood pressure measurement often compared it against another cuff-based oscillometric blood pressure monitor [11, 38–41]. Although it is practically more convenient than auscultatory blood pressure measurement, it is not a reliable reference.

Static, Stability and Blood Pressure Change Tests The few studies that performed a validation study according to one of the existing validation protocols [12, 13] must be questioned for its usefulness in clinical practice as the devices were validated directly after calibration and no stability test was performed. Perhaps a minimal time delay of 1 h or so before starting the validation procedure may be suggested to exclude unstable cuffless devices. Multiple other studies also presented static tests only (in either supine [39, 41] or sitting position [40–42]). Dependent on the intended purpose of a wearable cuffless device this may not be enough. Obviously, it is easier to determine blood pressure accurately if the patient is not moving as there is no chance of movement artifacts and auscultatory reference blood pressure measurement during movement is nearly impossible. However, a cuffless wrist monitor that provides continuous measurement, may need to be tested in different arm positions and walking with a normal arm swing [43]. The IEEE standard recommends that blood pressure should be measured in “changing conditions.” In studies to the accuracy of cuffless monitors with changing blood

220

W. J. Verberk

pressure conditions several methods were used to induce blood pressure changes, such as: holding breath, mental stress test, pedalling while sitting on a chair [38] or on a bike [44], rope skipping [45], simple leg-stretching [46], the use of intra-­arterial nitroglycerin [47] and handgrip exercise [48]. Although, this all undoubtedly leads to significant blood pressure changes, it also causes large blood pressure fluctuations. It, therefore, remains uncertain if these changes can be accurately verified using an auscultatory blood pressure measurement that takes approximately 30 s. Schoot et al. [41] verified a cuffless monitor with a modified International Protocol [27]. The interesting part of this study was that the cuffless device was verified with the subjects in both supine and sitting position, which leads to different blood pressure values [49]. First, the calibration and test measurements were performed in supine position, thereafter subjects were also measured in sitting position. As may be expected, measurements taken in the “calibration-position” led to smaller errors than measurements obtained in the sitting position. However, in both positions the device failed to pass the defined criteria for accuracy.

Subjects Investigated Most of the currently performed studies to the accuracy of cuffless blood pressure monitors showed limitations regarding the population sample. Often the selected number of patients was (too) low and blood pressure ranges were not considered. That the latter is important for clinical testing was shown in the study of Ding et al. where the cuffless blood pressure monitor showed higher errors in the hypertensive than in normotensive subjects [17].

Statistical Outcomes of Performed Validations Studies Generally, studies to cuffless blood pressure measurement accuracy provided no or minor information of how the average blood pressure value of the cuffless device under testing was calculated, that is, the number of measured heart beats used and/ or the duration of measurement. An exception is the study of Ding et al. who compared 5024 beats from 33 subjects obtained in sitting and supine position [17]. The authors were also one of the few who compared the test methods over an extended calibration interval of 24 h. This showed that a longer duration after calibration led to higher errors. Obviously, new studies to cuffless blood pressure monitors follow, more or less, the existing standards for cuff-based blood pressure monitors resulting in mean and standard deviation and correlation coefficients as the preferred outcome measures. For graphic presentation, Bland–Altman and correlation plots were often used. Studies that performed different statistics used, for example, repeatability of measured variables over continuous cardiac cycles presented as the ratio of standard

12  Design of Clinical Trials to Validate Cuffless Blood Pressure Monitors

221

deviation to mean value of a number of cycles expressed in percentage [42]. The feasibility for hypertension screening (yes or no) was investigated using receiver operating characteristic curves [42]. A t-test was performed to determine significant difference between reference blood pressure device and the device under testing [50]. Another study presented the results following different standards: mean absolute difference (IEEE), mean error and standard deviation of the error (ANSI/ AAMI/ISO), and the cumulative percentage of readings falling within 5, 10, and 15 mmHg (BHS) [51]. Finally, Sola et al. investigated the capability of a cuffless monitor to track continuously blood pressure changes against invasive blood pressure measurement using the four-quadrant plot for presenting the results [52].

Calibration Free Cuffless Blood Pressure Measurement Matsumara et al. claimed to have developed a cuffless blood pressure measurement method that requires no calibration. The device works with a smartphone and a traditional finger photo plethysmograph [53]. According to the authors the calibration is not needed because the used method is based on heart rate and modified normalized pulse volume instead of pulse wave velocity. Because the device is meant as a replacement of the intermittent cuff-based monitor it might have been validated according to the current ANSI/AAMI/ISO standard, but the authors followed their own study design and expressed the outcome as a correlation coefficient larger than 0.7 compared to a standard brachial cuff sphygmomanometer.

Summary and Conclusion The quick development of cuffless blood pressure monitors and the published studies to their accuracy suggest an urgent need for a new widely accepted standard to guarantee the quality of these devices. The existing standards are insufficient, and the lack of new standards has led to large heterogeneity in clinical tests. A new standard should consider the following items in particular: 1. Details of how the cuffless device calculates blood pressure should be provided. 2. A stability test should be performed to verify the time at which a cuffless blood pressure monitor remains accurate after calibration (until recalibration). 3. Cuffless monitors intended for medical purpose must also be tested for their capability to track (quick) blood pressure changes. This means that a static test is not enough. 4. Continuous cuffless devices that are meant to track quick blood pressure changes can only be validated with simultaneous invasive blood pressure measurement. 5. Acceptable errors from the reference devices may need to be redefined considering more measurement comparisons and other circumstances (blood pressure changes).

222

W. J. Verberk

6. Regarding the population sample: a minimal number of patients is needed to guarantee enough power. The population needs to be balanced for gender, age and blood pressure level. 7. Results should be presented as mean and standard deviation and the results should be shown in Bland–Altman and/or 4-quadrant plots if the device is tested for tracking blood pressure changes. These plots can reveal correlation and agreement but also show the bias between the devices. 8. The standards for Good Clinical Practice should be followed. Unnecessary subject measurements should always be prevented. Until devices are appropriately tested for their intended purpose, considering the above points, continuous or intermittent cuffless blood pressure monitors should not be used for healthcare purposes.

References 1. Gazzola K, Honingh M, Truijen J, Zuliani G, Van Den Born BH. Effect of cuff inflation on blood pressure during self-measurement. J Hypertens. 2018;36(9):1798–802. 2. Palatini P, Benetti E, Fania C, Saladini F. Only troncoconical cuffs can provide accurate blood pressure measurements in people with severe obesity. J Hypertens. 2019;37(1):37–41. 3. Zhang G, Gao M, Xu D, Olivier NB, Mukkamala R.  Pulse arrival time is not an adequate surrogate for pulse transit time as a marker of blood pressure. J  Appl Physiol (1985). 2011;111(6):1681–6. 4. Bramwell JC, Hill AV.  The velocity of the pulse wave in man. Proc R Soc B Biol Sci. 1922;93(652):298–306. 5. IEEE standard for wearable cuffless blood pressure measuring devices. IEEE std 1708-2014; 2014. p. 1–38. 6. Chin KY, Panerai RB.  Comparative study of Finapres devices. Blood Press Monit. 2012;17(4):171–8. 7. Non-invasive sphygmomanometers — Part 2: clinical investigation of the intermittent automated measurement type. ISO/CD 81060-2:2013(E); 2013. 8. AAMI. Association for the Advancement of Medical Instrumentation; non-invasive sphygmomanometers -part 2: clinical investigation of automated measurement type, ANSI/AAMI/ISO 81060-2013; 2013. 9. Wang R, Jia W, Mao Z-H, Sclabassi RJ, Sun M. Cuff-free blood pressure estimation using pulse transit time and heart rate. In: International conference on signal processing proceedings international conference on signal processing 2014; 2014. p. 115–8. 10. Butlin M, Shirbani F, Barin E, Tan I, Spronck B, Avolio A. Cuffless estimation of blood pressure: importance of variability in blood pressure dependence of arterial stiffness across individuals and measurement sites. IEEE Trans Biomed Eng. 2018;65(11):2377–83. 11. Pm N, Joseph J, Karthik S, Sivaprakasam M, Chenniappan M.  Bi-modal arterial compliance probe for calibration-free cuffless blood pressure estimation. IEEE Trans Biomed Eng. 2018;65(11):2392–404. 12. Boubouchairopoulou N, Kollias A, Chiu B, Chen B, Lagou S, Anestis P, et al. A novel cuffless device for self-measurement of blood pressure: concept, performance and clinical validation. J Hum Hypertens. 2017;31(7):479–82. 13. Bilo G, Zorzi C, Ochoa Munera JE, Torlasco C, Giuli V, Parati G. Validation of the Somnotouch-­ NIBP noninvasive continuous blood pressure monitor according to the European Society of Hypertension International Protocol revision 2010. Blood Press Monit. 2015;20(5):291–4.

12  Design of Clinical Trials to Validate Cuffless Blood Pressure Monitors

223

14. Webb AJS, Mazzucco S, Li L, Rothwell PM.  Prognostic significance of blood pressure variability on beat-to-beat monitoring after transient ischemic attack and stroke. Stroke. 2018;49(1):62–7. 15. Verberk WJ, Kessels AG, Thien T. Blood pressure measurement method and inter-arm differences: a meta-analysis. Am J Hypertens. 2011;24(11):1201–8. 16. Grossman A, Weiss A, Beloosesky Y, Morag-Koren N, Green H, Grossman E.  Inter-arm blood pressure difference in hospitalized elderly patients--is it consistent? J Clin Hypertens (Greenwich). 2014;16(7):518–23. 17. Ding X, Yan BP, Zhang YT, Liu J, Zhao N, Tsang HK. Pulse transit time based continuous cuffless blood pressure estimation: a new extension and a comprehensive evaluation. Sci Rep. 2017;7(1):11554. 18. Borlotti A, Khir AW, Rietzschel ER, De Buyzere ML, Vermeersch S, Segers P. Noninvasive determination of local pulse wave velocity and wave intensity: changes with age and gender in the carotid and femoral arteries of healthy human. J Appl Physiol (1985). 2012;113(5):727–35. 19. Eikendal ALM, den Ruijter HM, Haaring C, Saam T, van der Geest RJ, Westenberg JJM, et al. Sex, body mass index, and blood pressure are related to aortic characteristics in healthy, young adults using magnetic resonance vessel wall imaging: the AMBITYON study. Magma (New York, NY). 2018;31(1):173–82. 20. Friedman BA, Alpert BS, Osborn D, Prisant LM, Quinn DE, Seller J.  Assessment of the validation of blood pressure monitors: a statistical reappraisal. Blood Press Monit. 2008;13(4):187–91. 21. Yan R.  Chapter 6: proposal for the evaluation of wearable cuff-less devices. Thesis. In: Evaluation of the wearable cuff-less blood pressure measuring devices; 2009. 22. Yan IR, Poon CC, Zhang YT. Evaluation scale to assess the accuracy of cuff-less blood pressure measuring devices. Blood Press Monit. 2009;14(6):257–67. 23. Clinical investigation of medical devices for human subjects  — good clinical practice ISO 14155:2011(E); 2011. 24. Alpert BS. The Accutension Stetho, an automated auscultatory device to validate automated sphygmomanometer readings in individual patients. J Hum Hypertens. 2018;32(6):455–9. 25. Bland JM, Altman DG. Agreement between methods of measurement with multiple observations per individual. J Biopharm Stat. 2007;17(4):571–82. 26. O’Brien E, Petrie J, Littler W, de Swiet M, Padfield PL, Altman DG, et al. An outline of the revised British Hypertension Society protocol for the evaluation of blood pressure measuring devices. J Hypertens. 1993;11(6):677–9. 27. O’Brien E, Atkins N, Stergiou G, Karpettas N, Parati G, Asmar R, et al. European Society of Hypertension International Protocol revision 2010 for the validation of blood pressure measuring devices in adults. Blood Press Monit. 2010;15(1):23–38. 28. Stergiou GS, Alpert B, Mieke S, Asmar R, Atkins N, Eckert S, et  al. A universal standard for the validation of blood pressure measuring devices: association for the Advancement of Medical Instrumentation/European Society of Hypertension/International Organization for Standardization (AAMI/ESH/ISO) Collaboration statement. Hypertension (Dallas, Tex: 1979). 2018;71(3):368–74. 29. Yan IR, Poon CC, Zhang YT.  A protocol design for evaluation of wearable cuff-less blood pressure measuring devices. In: Conference proceedings: annual international conference of the IEEE Engineering in Medicine and Biology Society. IEEE Engineering in Medicine and Biology Society annual conference 2009; 2009. p. 7045–7. 30. Bland JM, Altman DG. Statistical methods for assessing agreement between two methods of clinical measurement. Lancet (London, England). 1986;1(8476):307–10. 31. Bland JM, Altman DG.  Applying the right statistics: analyses of measurement studies. Ultrasound Obstet Gynecol. 2003;22(1):85–93. 32. Saugel B, Dueck R, Wagner JY.  Measurement of blood pressure. Best Pract Res Clin Anaesthesiol. 2014;28(4):309–22. 33. Association for the Advancement of Medical Instrumentation. American National Standard. ANSI/AAMI/ISO 81060-2:2013 non-invasive sphygmomanometers—part 2: clinical investigation of automated measurement type. Arlington: AAMI; 2013.

224

W. J. Verberk

34. Non-invasive sphygmomanometers—part 1: requirements and test methods for non-automated measurement type. ANSI/AAMI/ISO 81060-1:2007; 2007. 35. Medical electrical equipment—part 2-34: particular requirements for the basic safety and essential performance of invasive blood pressure monitoring equipment. IEC 60601-2-­ 34:2011; 2011. 36. Lin WH, Wang H, Samuel OW, Liu G, Huang Z, Li G.  New photoplethysmogram indicators for improving cuffless and continuous blood pressure estimation accuracy. Physiol Meas. 2018;39(2):025005. 37. Silke B, McAuley D.  Accuracy and precision of blood pressure determination with the Finapres: an overview using re-sampling statistics. J Hum Hypertens. 1998;12(6):403–9. 38. Tang Z, Tamura T, Sekine M, Huang M, Chen W, Yoshida M, et al. A chair-based unobtrusive cuffless blood pressure monitoring system based on pulse arrival time. IEEE J Biomed Health Inform. 2017;21(5):1194–205. 39. Shin H, Min SD. Feasibility study for the non-invasive blood pressure estimation based on ppg morphology: normotensive subject study. Biomed Eng Online. 2017;16(1):10. 40. Huynh T, Chung W-Y. Radial electrical impedance: a potential indicator for noninvasive cuffless blood pressure measurement. J Sens Sci Tech. 2017;26(4):239–44. 41. Schoot TS, Weenk M, van de Belt TH, Engelen LJ, van Goor H, Bredie SJ.  A new cuffless device for measuring blood pressure: a real-life validation study. J  Med Internet Res. 2016;18(5):e85. 42. Nabeel PM, Joseph J, Karthik S, Sivaprakasam M, Chenniappan M. Bi-modal arterial compliance probe for calibration-free cuffless blood pressure estimation. IEEE Trans Biomed Eng. 2018;65(11):2392–404. 43. Wang Y, Liu Z, Ma S. Cuff-less blood pressure measurement from dual-channel photoplethysmographic signals via peripheral pulse transit time with singular spectrum analysis. Physiol Meas. 2018;39(2):025010. 44. Zhang Q, Zhou D, Zeng X. Highly wearable cuff-less blood pressure and heart rate monitoring with single-arm electrocardiogram and photoplethysmogram signals. Biomed Eng Online. 2017;16(1):23. 45. Miao F, Fu N, Zhang YT, Ding XR, Hong X, He Q, et  al. A novel continuous blood pressure estimation approach based on data mining techniques. IEEE J  Biomed Health Inform. 2017;21(6):1730–40. 46. Watanabe N, Bando YK, Kawachi T, Yamakita H, Futatsuyama K, Honda Y, et al. Development and validation of a novel cuff-less blood pressure monitoring device. JACC Basic Trans Sci. 2017;2(6):631–42. 47. Xiao-Rong D, Yan BP, Yuan-Ting Z, Jing L, Peng S, Ni Z.  Coherence analysis of invasive blood pressure and its noninvasive indicators for improvement of cuffless measurement accuracy. Conf Proc IEEE Eng Med Biol Soc. 2017;2017:2255–8. 48. Kim S, Lee JD, Park JB, Jang S, Kim J, Lee SS. Evaluation of the accuracy of a new cuffless magnetoplethysmography blood pressure monitor in hypertensive patients. Pulse (Basel, Switzerland). 2018;6(1-2):9–18. 49. Cicolini G, Pizzi C, Palma E, Bucci M, Schioppa F, Mezzetti A, et al. Differences in blood pressure by body position (supine, Fowler’s, and sitting) in hypertensive subjects. Am J Hypertens. 2011;24(10):1073–9. 50. Krisai P, Vischer AS, Kilian L, Meienberg A, Mayr M, Burkard T. Accuracy of 24-hour ambulatory blood pressure monitoring by a novel cuffless device in clinical practice. Heart (British Cardiac Society). 2019;105(5):399–405 51. Solà et al, 2017 - Performance of Systolic Blood Pressure estimation from radial Pulse Arrival Time (PAT) in anesthetized patients Presented at EMBEC2017 - European Medical and Biological Engineering Conference 2017, Tampere (FI), 11-15 June 2017. 52. Solà J, Ghamri Y, Proença M, Braun F, Lemkaddem A, Pierrel N, et al. Tracking blood pressure changes in anesthetized patients: the optical blood pressure monitoring (oBPM) technology; Conference Paper 40th Annual International Conference of the IEEE Engineering in Medicine and Biology Society, At Honolulu, Honolulu, Hawaii, USA. July 17-21, 2018. 53. Matsumura K, Rolfe P, Toda S, Yamakoshi T. Cuffless blood pressure estimation using only a smartphone. Sci Rep. 2018;8(1):7298.

Chapter 13

Cuffless Blood Pressure Monitoring: The Future for the Evaluation and Management of Hypertension George S. Stergiou

Abstract  The upper arm-cuff blood pressure measurement using the manual auscultatory or automated oscillometric method remains the cornerstone for the diagnosis and management of hypertension. However, this method has major deficiencies which seriously undermine the accurate evaluation of blood pressure, including the fact that it estimates rather than measures blood pressure, differs from intra-arterial measurement, might be affected by cuff inflation and arterial occlusion, and assesses only intermittently and only static blood pressure, ignoring thereby the continuous dynamic blood pressure variability. The development of reliable continuous cuffless blood pressure measurement technology can provide a complete picture of the 24-h blood pressure profile and behaviour for long periods, which will reflect the true burden of blood pressure on the cardiovascular system. However, there are several technical research questions that need to be addressed regarding the accuracy of cuffless blood pressure measurement and its persistence with time and in conditions of acute and chronic changes in blood pressure and other hemodynamic parameters. There are also clinical research questions, including the optimal sampling method, reproducibility, clinical relevance and thresholds for hypertension diagnosis. When the abovementioned research questions for this novel method are adequately addressed, the conventional methods for office, ambulatory and home blood pressure measurement will become absolute and replaced by the evaluation of the complete evaluation of the true blood pressure profile. Keywords  Accuracy · Blood pressure measurement · Cuffless · Methodology · Validation

G. S. Stergiou (*) Hypertension Center STRIDE-7, National and Kapodistrian University of Athens, Third Department of Medicine, Athens, Greece e-mail: [email protected] © Springer Nature Switzerland AG 2019 J. Solà, R. Delgado-Gonzalo (eds.), The Handbook of Cuffless Blood Pressure Monitoring, https://doi.org/10.1007/978-3-030-24701-0_13

225

226

G. S. Stergiou

 Century of Clinical Use of Cuff-Based Blood Pressure A Measurement in Hypertension Management The history of blood pressure measurement evolved in 1733 when Reverend Stephen Hales conducted his famous experiment by placing a glass tube in the artery of horse and observed the blood to rise high due to blood pressure [1]. However, it was in 1896 that Scipione Riva-Rocci first estimated blood pressure using an inflatable cuff to encircle the upper arm aiming to occlude the brachial artery [1]. In 1905, Nicolai Sergeivich Korotkof who was surgeon at the Russian army founded the auscultatory measurement of systolic and diastolic blood pressure using a Riva-­Rocci upper arm cuff [1, 2]. The oscillometric cuff-based blood pressure measurement was introduced in 1876 by the French physiologist Etienne Marey [2, 3]. The upper arm-cuff blood pressure measurement has been the cornerstone for the evolution of clinical hypertension. In the last 100 years the evolution of clinical hypertension has been an excellent model for evidence-based medicine. A total of 61 prospective observational outcome studies including one million adults with 12.7 million person-years at risk and 56,000 cardiovascular deaths during follow-up showed that blood pressure is strongly and directly related to cardiovascular and total mortality, without no evidence of a threshold up to at least 115 mmHg systolic and 75 mmHg diastolic [4]. More importantly, a total of 122 interventional randomized outcome clinical trials including almost 350,000 subjects demonstrated the benefits of treatment induced blood pressure lowering using different antihypertensive drug classes in reducing the risk of fatal and nonfatal cardiovascular events [5]. This enormous database of mega-trials, which is the basis of the current management of hypertension, has been almost exclusively based on auscultatory upper-arm cuff blood pressure measurements. In the last 20 years the automated oscillometric blood pressure measurement method is progressively replacing the manual auscultatory method and has been increasingly used in hypertension outcome trials [6]. The oscillometric blood pressure measurement method aims to replicate the auscultatory method, which is taken as reference in empirical development of estimation algorithms [2].

Problems of Cuff-Based Blood Pressure Measurement Despite the indisputable value of the upper-arm cuff blood pressure measurement method in the identification and management of hypertension worldwide, this method has major deficiencies which seriously undermine the accurate evaluation of blood pressure. The main issues with the upper-arm cuff blood pressure measurement method (auscultatory and oscillometric) are listed below. (a) Provides an “estimation” of the blood pressure level rather a true measurement.

13  Cuffless Blood Pressure Monitoring: The Future for the Evaluation…

227

A true “measurement” can only be obtained directly using the invasive intra-­ arterial method. (b) Underestimates the systolic and overestimates the diastolic blood pressure. Picone et al. performed a meta-analysis of studies comparing intra-arterial brachial versus cuff-based brachial blood pressure and showed that the latter considerably underestimates intra-arterial systolic blood pressure (by about 6 mmHg) and overestimates diastolic blood pressure (by about 6 mmHg), leading thereby to a large underestimation of pulse pressure by about 12 mmHg [7]. From the scientific and the physiological point of view, the invasive intra-­ arterial blood pressure measurement is the meaningful and relevant one [8]. (c) Provides intermittent rather than continuous measurement of blood pressure. The current methodology applied for the evaluation of blood pressure in clinical practice can give only a very limited picture (snapshots) of the 24  h blood pressure profile. Thus, the widely used methods for blood pressure measurement in the office, at home and with 24 h ambulatory monitoring provide only 2–100 blood pressure readings out of 100,000 heart cycles per 24 h (each one generating a blood pressure value). This is a serious limitation of the cuff-­ based blood pressure measurement method as the dynamic variation of blood pressure during routine daily challenges cannot be demonstrated. (d) Is unable to evaluate beat-to-beat blood pressure variability. The dynamic variation of blood pressure during routine physical and mental activities certainly puts addition stress on the heart and vasculature independent of the average blood pressure [9, 10]. As mentioned above, the currently used conventional methods for the evaluation of blood pressure in the office, at home and with 24 h ambulatory monitoring obtain a very small sample of the 24 h beat to-beat profile and therefore provide very limited information on the burden of blood pressure variability. Thus, at present the management of hypertension is based on the average value of few blood pressure measurements and inevitably the impact of variability has been neglected [9, 10]. (e) Blood pressure is measured in static conditions only and fluctuations are not assessed. A major disadvantage of both the auscultatory and the oscillometric blood pressure measurement method is their inability to obtain a valid reading during body movement. Thus, the blood pressure reactivity and instability, and the episodic blood pressure rise during usual daily activities cannot be evaluated [9, 10]. (f) The blood pressure level may be affected by the cuff inflation. The cuff inflation and the resulting compression of the upper arm is sensed by the user and may by itself affect the blood pressure level during wakefulness and during sleep [11] giving thereby a distorted estimate of the true blood pressure. (g) The blood pressure level may be affected by the vascular occlusion. It is not clear whether and in what extent the cuff inflation induced arm compression which alters the circulation by occluding the brachial artery and vein affects the level of measured blood pressure.

228

G. S. Stergiou

 he Promise of Cuffless Blood Pressure Measurement T Technology Blood pressure is a continuous variable with dynamic characteristics of variability during routine physical and mental activities. Therefore, the crucial question for the medical engineer and the crucial need for the clinical doctor is how to quantify accurately the burden of the continuous beat-to beat blood pressure fluctuation on the cardiovascular system. The full picture of the dynamic blood pressure behaviour cannot be obtained by intermittent cuff-based blood pressure measurement as usually obtained in clinical practice. It can only be provided by continuous beat-to-beat blood pressure measurement which at present can be obtained only invasively (intra-arterial) and therefore is impractical for clinical use and unethical for clinical research (unless in case of medical indication for intra-arterial monitoring and for short-term only). Continuous beat-to-beat measurement of ambulatory blood pressure in the finger has been developed using the volume-clamp method, which has been shown to underestimate blood pressure and is restricted to non-invasive evaluation of beat-to-beat changes in research studies [12, 13]. The ultimate approach for blood pressure evaluation should be (1) non-invasive and cuffless, so that the individual does not sense the measurement procedure and therefore an unbiased evaluation is feasible in multiple repeated measurements, (2) continuous so as the full 24-h profile of blood pressure can be evaluated rather than a small sample of measurement and (3) unaffected of body movements so that the dynamic fluctuation of blood pressure during routine daily activities can be accurately recorded. The technologies for cuffless blood pressure measurement such those described in this book (see previous chapters) have the potential to satisfy all the requirements for optimal evaluation of the 24 h blood pressure profile. An accurate blood pressure measuring device that can be used for several days and record a large sample of blood pressure measurements without affecting the individual’s routine behaviour will provide a complete picture of the true blood pressure level and variability. Thus, for the first time in the history of clinical hypertension the true impact of the blood pressure on the cardiovascular system will be accurately quantified.

Technical Research Questions Further to the technical and research debates presented in the previous chapters of this book, several research questions arise regarding the accuracy of cuffless blood pressure measurement and its persistence in conditions of acute and chronic changes in blood pressure and other hemodynamic parameters. International standards which are specific for the evaluation of the accuracy of cuffless continuous blood pressure monitors should be urgently developed and agreed. • Which individual patient parameters need to be provided to improve the accuracy of cuffless blood pressure estimation. • Which reference blood pressure measuring device and method to be used for static calibration.

13  Cuffless Blood Pressure Monitoring: The Future for the Evaluation…

229

• Which calibration method can be accurately replicated in general practice. • Whether recalibration is necessary for retaining blood pressure measurement accuracy and at what conditions and time intervals. • Whether shortly after calibration accurate cuffless blood pressure measurement is obtained in static conditions. • Whether accurate cuffless blood pressure measurement is retained in conditions of acute change in blood pressure and heart rate (increase or decrease) and other hemodynamic parameters (physical or mental activity, emotional change, sleep, disease state, etc.). • Whether accurate cuffless blood pressure measurement is retained when the blood pressure level is changed (increased or decreased) in response to drug effect or other factors (disease state, body weight change, temperature change or others).

Clinical Research Questions Accurate non-invasive continuous cuffless blood pressure measurement has never been obtained in clinical practice. This method is not the same as intra-arterial continuous blood pressure monitoring which is performed in bedridden patients, neither is the same as 24 h ambulatory blood pressure monitoring which is intermittent, takes measurements only in static conditions and is affected by cuff inflation and arm compression. As a novel method for blood pressure measurement, several fundamental clinical research questions need to be addressed. • Reproducibility compared to 24  h cuff-based ambulatory blood pressure monitoring. • Relationship with indices of preclinical target organ damage and cardiovascular risk compared to 24 h cuff-based ambulatory blood pressure monitoring. • Optimal 24 h blood pressure sampling and number of monitoring days that give the highest reproducibility and correlation with indices of preclinical target organ damage. • Thresholds for hypertension diagnosis for 24 h, daytime and night-time average blood pressure values compared to conventional ambulatory blood pressure. • Optimal indices and thresholds for blood pressure variability.

Conclusion The development of a reliable continuous cuffless blood pressure measurement technology is expected to end the era of conventional blood pressure measurement and change the practice of hypertension management. This method promises to provide a complete picture of the 24 h blood pressure profile and behaviour for long periods, which will represent the true burden of blood pressure on the cardiovascular system. Thus, the conventional methods currently used for blood pressure evaluation in the office and at home, and for 24 h ambulatory blood pressure monitoring

230

G. S. Stergiou

will probably become obsolete. When the abovementioned research questions on the measurement accuracy and the clinical relevance of this novel method are adequately addressed, then the chapter on blood pressure measurement for hypertension diagnosis and management will have to be rewritten.

References 1. O’Brien E, Fitzerald D.  The history of indirect blood pressure measurement. In: O’Brien E, O’Malley K, editors. Handbook of hypertension, Vol. 14: blood pressure measurement. Amsterdam: Elsevier; 1991. p. 1–54. 2. Pickering TG. Blood pressure measurement. In: Ambulatory monitoring and blood pressure variability, Part 1. London: Science Press; 1990. p. 2.1–2.16. 3. Marey EJ. Pression et vitesse du sang. In: Masson G, editor. Physiologie experimentale, Vol. 2. Paris; 1876. p. 307. 4. Lewington S, Clarke R, Qizilbash N, Peto R, Collins R, Prospective Studies Collaboration. Age-specific relevance of usual blood pressure to vascular mortality: a meta-analysis of individual data for one million adults in 61 prospective studies. Lancet. 2002;360(9349):1903–13. 5. Thomopoulos C, Parati G, Zanchetti A. Effects of blood pressure-lowering treatment on cardiovascular outcomes and mortality: 14—effects of different classes of antihypertensive drugs in older and younger patients: overview and meta-analysis. J Hypertens. 2018;36(8):1637–47. 6. Giorgini P, Weder AB, Jackson EA, Brook RD. A review of blood pressure measurement protocols among hypertension trials: implications for “evidence-based” clinical practice. J Am Soc Hypertens. 2014;8:670–6. 7. Picone DS, Schultz MG, Otahal P, Aakhus S, Al-Jumaily AM, Black JA, Bos WJ, Chambers JB, Chen CH, Cheng HM, Cremer A, Davies JE, Dwyer N, Gould BA, Hughes AD, Lacy PS, Laugesen E, Liang F, Melamed R, Muecke S, Ohte N, Okada S, Omboni S, Ott C, Peng X, Pereira T, Pucci G, Rajani R, Roberts-Thomson P, Rossen NB, Sueta D, Sinha MD, Schmieder RE, Smulyan H, Srikanth VK, Stewart R, Stouffer GA, Takazawa K, Wang J, Westerhof BE, Weber F, Weber T, Williams B, Yamada H, Yamamoto E, Sharman JE.  Accuracy of cuff-measured blood pressure: systematic reviews and meta-analyses. J  Am Coll Cardiol. 2017;70(5):572–86. 8. Stergiou GS, Kollias A, Protogerou AD. Evidence on blood pressure measurement methodology and clinical implementation: research agenda for the 21st century. J  Am Coll Cardiol. 2017;70(5):587–9. 9. Parati G, Stergiou GS, Dolan E, Bilo G.  Blood pressure variability: clinical relevance and application. J Clin Hypertens (Greenwich). 2018;20(7):1133–7. 10. Rothwell PM. Limitations of the usual blood-pressure hypothesis and importance of variability, instability, and episodic hypertension. Lancet. 2010;375:938–48. 11. Sheshadri V, Tiwari AK, Nagappa M, Venkatraghavan L. Accuracy in blood pressure monitoring: the effect of noninvasive blood pressure cuff inflation on intra-arterial blood pressure values. Anesth Essays Res. 2017;11:169–73. 12. Imholz BP, Wieling W, van Montfrans GA, Wesseling KH. Fifteen years experience with finger arterial pressure monitoring: assessment of the technology. Cardiovasc Res. 1998;38:605–16. 13. O’Brien E, Parati G, Stergiou G, Asmar R, Beilin L, Bilo G, Clement D, de la Sierra A, de Leeuw P, Dolan E, Fagard R, Graves J, Head GA, Imai Y, Kario K, Lurbe E, Mallion JM, Mancia G, Mengden T, Myers M, Ogedegbe G, Ohkubo T, Omboni S, Palatini P, Redon J, Ruilope LM, Shennan A, Staessen JA, vanMontfrans G, Verdecchia P, Waeber B, Wang J, Zanchetti A, Zhang Y, European Society of Hypertension Working Group on Blood Pressure Monitoring. European Society of Hypertension position paper on ambulatory blood pressure monitoring. J Hypertens. 2013;31:1731–68.

Index

A Acceleration plethysmogram (APG), 118 Active implantable medical devices (AIMD), 197 Aging, 164, 167, 181, 183–186 Analytical models, 143, 146, 147 Android-based tablet, 102 ANSI/AAMI/ISO standard, 217, 221 Aortic PTT measurements, 181 Applanation tonometry, 109, 110, 120–124, 128–131 Architecture of a cuffless BPM advantages, 37 arterial-occlusion monitors, 37–39 characteristics, 37, 39 disadvantages, 37 initialization layer, 41 invasive BPM, 37, 38 limitation, 39 non-pressure units, 37 principles and output units, 37, 38 processing layer, 40, 41 pulsatility signals, 37 transducer layer, 40 user’s skin surface, 37 Arm cuff inflation, 10 Arterial catheters, 53, 109 Arterial hemodynamics, 5 Arterial length, 62 Arterial lines, 53 Arterial-occlusion BPM, 35 Arterial-occlusion monitors, 37–39 Arterial periphery, 76 Arterial pulse, 2, 16 Arterial reflections

aorta/arm complex arterial system, 79, 80 catheter sensor positions, 79, 81 digital/radial, 76 distinct pulse-like features, 78, 79 features, 77 noninvasive method, 81 propagation velocities, 77 pulse-like protrusions, 77 radial pulse signatures high-fidelity, 77 qualitative comparison, 77, 78 Valsalva maneuver, 77–79 second systolic/diastolic peak, 81 Arterial stiffness (AS), 4 aging, 26 cath lab patient, 90 coronary heart disease risk, 26 digital pulse, 91 featuredness, 91 flexible arteries, 90 identification, distinct inversions, 92 pancreaticoduodenectomy surgery, 91 parameter, 92 PDA framework, 91 pressure pulses, 90 pulse analysis, 92 second derivative analysis, 92 vasodilation, 90 wall structure, 90 young athlete, 90 Arterial system, 16 Arterial tonometry, 61 Artificial neural network, 155 ASEAN Medical Device Directive (AMDD), 192

© Springer Nature Switzerland AG 2019 J. Solà, R. Delgado-Gonzalo (eds.), The Handbook of Cuffless Blood Pressure Monitoring, https://doi.org/10.1007/978-3-030-24701-0

231

232 Association for the Advancement of Medical Instrumentation (AAMI), 177 Association of Southeast Asian Nations (ASEAN) member states, 192 Asymmetric T-shaped model, 76 Augmentation index (AIx), 112, 115 Auscultation, 51, 52 Auscultatory method, 10, 16, 23, 226 Australia’s medical device classification system, 197, 198 Australian Register of Therapeutic Goods (ARTG), 198 Australian Regulatory Guidelines for Medical Devices (ARGMD), 197 Automated oscillometric devices, 206 Autonomic nervous index, 182 B Ballistocardiography (BCG), 48, 81 measurement, 68 waveform, 180, 181 Basic glossary of cuffless BP monitoring, 32–35 Beat-by-beat oscillations, 9 Beat-to-beat BP variability, 227 B-H equation, 143 Big data, 3, 4 Bioimpedance method, 68 Biopac system, 57 Bland–Altman plot, 214 Blood pressure (BP) estimation, 139 horse, measurements, 19 indicators, 140, 148, 150 insurance examination, 24, 25 measurement, 139 pulse waveforms, 112 PWA, 118, 119 variability (see BP variability) Blood pressure monitoring (BPM) clinical practice, 24 cuffless (see Cuffless BPM) graphic recordings, 16 physiological terms, 32–33 technological terms, 32–35 BP change tracker, 71 BP measurement challenge, clinical practice, 204 disadvantages, 204 intermittent, 204 upper-arm cuff inflating, 204 wearable cuffless devices, 210

Index BP variability assessment, 10 cardiovascular regulatory mechanisms, 10, 11 physiological conditions, 10 Brachial-ankle pulse wave velocity, 117 Bramwell–Hill equation, 66, 166, 168 C Calibration, 62–65, 71, 206, 207 method, 229 parameters, 35, 41 PTT to BP, 69, 70 Calibration curve cardiovascular aging, 164 construction, 164 hybrid, 184, 185 parameters, 164 person-specific, 177 population-based, 177, 178 practical form, 170, 171 Canada’s medical device classification system, 196, 197 Cardiac output (CO), 67 Cardiac signals, 139, 149 CareTaker (CT), 76, 96, 99, 101–103 Carotid artery, 62 Center for Devices and Radiological Health (CDRH), 194 Central arterial line comparisons, 98–100 Chronophotography, 2 Clinical and consumer applications, 191 Clinical hypertension, 226, 228 Clinical studies, 115, 119–128 CNSystems, 62 Cold pressor, 171 Communication modalities, 102 Competent Authority, 195 Complex chip design, 6 Computer vision, 147, 148 Consumer devices, 4 Continuous BP, 84, 103, 205, 206, 213 Continuous BP monitoring, 71 arm cuff inflation, 10 auscultatory method, 10 beat-by-beat oscillations, 9, 10 clinical diagnostic, 10 clinical issues, 11 CPAP, 11 in daily practice, 10 invasive recordings, 11 noninvasive, 11 oscillometric method, 10

Index prognostic information, 10 techniques, 11 Continuous/intermittent vs. cuff/cuffless devices, 205 Continuous measurement, 43, 50, 52–56 Continuous noninvasive automated sphygmomanometer, 205 Continuous, noninvasive blood pressure (cNIBP), 56, 62 Continuous positive air pressure (CPAP), 11 Cuff definition, 205 inflation, 227 Cuff-based BP measurement hypertension management, 226 issues, 226, 227 Cuff-based upper-arm devices, 206 Cuff/cuffless vs. continuous/intermittent devices, 205 Cuffless BP, 164, 181–183 sensor, 205 machine learning (see Machine learning) validation (see Validation protocol) Cuffless BPM, 39 Cuffless devices, 3, 5, 6, 204, 206 Cuffless techniques monitoring, BP, 5, 6 wearable devices, 5 D Data-driven based machine learning method analytical model, 146 artificial intelligence, 143 big data analysis techniques, 143 deep learning, 146 feature and value, 146 M-K equation, 140 PTT and BP relationship, 141, 142 PTT-based BP estimation, 143–145 PWV recording, 140, 141 static analytical model, 146 vs. theory-driven analytical model, 146, 147 Data mining, 183 Deep learning, 148 advantages, 152 artificial neurons, 152 classification, 152 cuffless BP, 153, 154 DNN, 153, 154 vs. machine learning, 152, 153 neural networks, 153 properties, 152 Deep neural networks (DNNs), 153, 154, 159

233 Deep RNN, 154–156 Definition of a cuffless BPM arterial-occlusion, 35 classification, 35, 36 noninvasive, 35 perimeter, 35 PWA, 37 PWV, 37 Designated marketing authorization holder (DMAH), 199–200 Devices hemodynamic information, 28 mobile/mHealth, 27 optical sensors, 27 pulse tonometry, 27 sensor technology, 27 volume clamp/plethysmography, 27 wrist and finger sensors, 27 Device validation, 6 Diastolic blood pressure (DBP), 52, 142, 147, 155, 173 Diastolic pressure (DP), 16 Dynamic component (DC), 45, 48 E Electrical bioimpedance (EBI) waveforms, 180, 181 Electrocardiogram (ECG), 44, 46, 49–51, 56–58, 63, 139 measurement methods, 68 Electronic devices, 16 Empirical model, 143 EU medical device classification system, 195, 196 European Medicines Agency (EMA), 196 F Femoral artery, 62 Finapres, 62 Finger cuff, 102, 103 Food and Drugs Act, 193, 196 Four-quadrant plot (FQP), 215 Framingham Heart Study, 24, 26 Fung’s hyper-elastic model, 167 G Generalized transfer functions (GTF), 114, 128 Global Harmonization Task Force (GHTF), 193 Good clinical practice (GCP) clinical investigation, medical devices, 217 duration, BP measurement procedure, 217

234 Good clinical practice (GCP) (cont.) invasive BP, 217 observer measurement, 218 phases/interim analyses, 218 reference device, 217 test environment, 218 Good Manufacturing Practice (GMP), 194 Green emitters, 46 H Haemodynamometer, 19 Harmonization, 192, 195, 200, 201 Health Canada (HC), 193 Health management, 2, 3, 6 Heart rate (HR), 67, 115 PDA, 93 Hemodynamic parameters, 229 Homeostasis, 10 Hot-wire sphygmographs, 56 Hughes model, 166 Hybrid method, 179, 180 Hydrostatic maneuver, 172, 173 Hypertension, 187 ageing, 26 central aortic SP and PP, 17, 18, 26 conventional brachial cuff sphygmomanometric measurements, 26 cuff-based BP measurement, clinical use, 226 Framingham Heart Study, 24, 26 management, elevated BP, 27 risk evaluation, 24 SHEP, 26 stiffening, 26 I Ice stimulus experiments, 96–98 Impedance cardiography (ICG), 46–48 Impedance plethysmography (IPG), 46–48 Infrared (IR), 46, 68 Initialization algorithm, 39, 41 Initialization layer, 39, 41 Initialization procedure, 39 Initialization, PTT-based BPM calibration curve, 164 measurements, 164 parametric models, 164–170 properties, 163 reinitialization period, 183–185 waveform features accuracy, 183, 184 aortic PTT measurements, 181

Index autonomic nervous function and index, 182 BCG, 180, 181 data mining, 183 diastolic and systolic BP, 180 EBI, 180, 181 J-K amplitude, 180 PAT, 182–184 peak-to-peak amplitudes, 180, 181 PEP, 180 PP formula, 180 PPG, 180 proximal and distal arterial, 180 single cuff measurement, 181 smooth muscle contraction, 180 Valsalva maneuver, 183 Institute of Electrical and Electronics Engineers (IEEE), 204, 212, 214, 217, 219 Intermittent BP measurement, 204, 205 International consensus standards, 200, 201 International Medical Device Regulators Forum (IMDRF), 193 International standards, 228 Inter-subject BP variations, 177 Intervention-induced BP variations, 177 Intra-arterial catheters, 53 Intra-thoracic pressure, 97 Invasive BP, 217 Invasive BPM, 37, 38 Invasive measurements, PAT, 53, 54 Inverse model, 67 Inverse square model, 67 J Japan’s medical device classification system, 199, 200 J-K amplitude, 180 K Korotkoff sounds, 4, 16, 52 Kymograph, 20 L Laplace’s law, 166 Laplacian distributed data, 176 Least squares estimation, 176 Left ventricular ejection time (LVET), 90 Linear least squares regression, 174, 176 Linear model, 66 Linear/nonlinear regression models, 143 Linear regression, 150, 151

Index Logarithmic model, 66 Long-short-term-memory (LSTM), 148, 154 M Machine learning advancement, 155 artificial neural network, 155 black box technique, 158 cuffless BP analytical model-based method, 148 application, 149 calibration, 150 computer vision, 147, 148 deep learning, 152–154 features, 147 indirect estimation, 149 linear regression, 150 measurement, 149 natural language processing, 147, 148 polynomial regression, 150 PPG signals, 147, 148 random forests, 151 regression trees, 151 speech recognition, 147, 148 SVR, 151, 152 target function, 149 univariate/multivariate linear regression, 148 data-driven method (see Data-driven based machine learning method) deep RNN, 154–156 development, BP models, 155, 157–158 DNNs, 159 hand-designed features, 154 labelled data, 158 limit and challenges, 158 methods, 71 MSE, 154 signal modalities, 159 techniques, 5 Marey’s sphygmograph, 20, 22 Marketing Authorization Holder (MAH) system, 199 Mean absolute differences (MAD), 214 Mean absolute percentage difference (MAPD), 214 Mean arterial pressure (MAP), 52, 97 Mean BP (MBP), 141, 155 Mean pressure, 20 Mean squared error (MSE), 154 Measurement methods, PWV BCG, 68 bioimpedance, 68 calibration, 69, 70

235 disadvantages, 67 ECG, 68 PATs, 67 PPG, 68 PPT, 68 VPG, 69 The Medical Device Coordination Group (the MDCG), 201 Medical Device Directive (MDD), 195 Medical Device Regulations, 196, 197 Medical devices definition, 193 industry, 192 in legal distribution, 192 regulations, 192 Medical Device Single Audit Program (MDSAP), 198 Mercury manometer, 19, 20 Methodology, 227 Miniaturization, 6 Ministry of Health, Labor and Welfare (MHLW), 199 Mobile health BP monitors, 11 Mobile/mHealth, 27 Moens–Korteweg (M-K) equation, 65, 82, 140, 143, 166 Motion artifacts, 50, 51 Multi-electrode system, 46 N Natural language processing, 147, 148 Neural networks, 153 Nocturnal cuff inflations, 11 Noninvasive, 43, 46, 50, 52–56, 81, 84, 91, 99, 100, 102 Noninvasive cuffless surrogate signal, 4 Non-PTT modalities, 179 Notified Body (NB), 195, 196 Novel cuffless methodologies, 5 O Optical measurements, 131 Optical sensors, 27 Oscillometric method, 10 Oscillometry, 52, 53 P Parameter error, 175 Parametric models, PTT-BPM empirical models, 169, 170 hybrid method, 179, 180 person-specific methods, 171–177

236 Parametric models, PTT-BPM (cont.) population-based methods, 177–179 practical calibration curve form, 170, 171 theoretical models arterial vessel, 166 Bramwell–Hill equation, 166, 168 cardiovascular and chronological aging, 167 Fung’s hyper-elastic model, 167 Hughes’ model, elastic modulus, 166 Laplace’s law, 166 M–K equation, 166 person- and time-specific parameters, 168, 169 physiologic mechanisms, 166 PWV, 166 Wesseling model, 167, 168 Peak-to-peak amplitudes, 180, 181 Peak-to-peak time delay, 176 Periodic fluid pump, 16 Peripheral arterial line comparisons, 99, 101, 102 Peripheral resistance arterial pressure pulse, 93, 94 arterial pressure reflections, 93 heart rate/LVET/cardiac output, 93 implementation, 94, 95 left ventricular ejection, 93 Person-specific calibration curve, 177 Person-specific methods, 171–177 Pharmaceutical Affairs Law (PAL), 199 Pharmaceutical and Medical Devices Act (PMD Act), 193, 199 Pharmaceutical and Medical Devices Agency (PMDA), 198–200 Pharmaceutical Safety and Environmental Health Bureau (PSEHB), 199 Phases/interim analyses, 218 Phonocardiogram (PCG), 48, 49 Photoplethysmography (PPG), 110, 111, 117, 118, 125–128, 130, 131, 139, 147, 149, 155, 178–184 applied pressure, 46 DC, 45 ECG waveform, 44 green emitters, 46 IR, 46 photodetector, 45 pulse wave, 44 reflective sensors, 45 SC, 45 transmissive sensors, 45 zero transmural pressure, 46 Poiseuille manometer, 20 Polynomial regression, 150

Index Population-based methods, 177–179 Practical calibration curve form, 170, 171 Predictive analytics, 149 Predictive modelling, 149 Pre-ejection period (PEP), 50, 64, 174 Premarket approval (PMA) process, 195 Principles, cuffless methodology, 3, 4 Processing layer, 39–41 PTT-based BPM initialization (see Initialization, PTT-based BPM) Pulsatility-based algorithm, 39–41 Pulsatility energy, 40 Pulsatility sensors, 37, 39, 40 Pulse arrival time (PAT), 63, 204 auscultation, 51, 52 BCG, 48 Biopac, 57 BP measurement site, 173 cNIBP, 56 continuous, noninvasive BP measurement, 43 and cuff BP, 179 definitions, 44 ECG-PPG plot, 44 handheld/wrist-worn devices, 57 hot-wire sphygmographs, 56 invasive measurements, 53, 54 IPG/ICG, 46–48 limitation, 64 measurements, 44, 55 motion artifacts, 50, 51 NIBP, 56 oscillometry, 52, 53 PCG, 48, 49 PEP, 50, 174 PPG (see Photoplethysmogram (PPG)) PWV, 49, 56, 57 S1/S2 heart sounds, 49 SCG, 48 and smooth muscle contraction, 181 somnomedics, 57 Sotera Wireless markets, 56 time delay, 56 time difference, 64 tonometry, 54, 55 variability, 183 ViSi Mobile, 56 volume clamp method, 53–55 Pulse contour analysis, 75 Pulse decomposition analysis (PDA) central arterial line comparisons, 98–100 central reflection sites, 76–82 component, 76 hardware platform, 102–104

Index hemodynamic parameters, 75 HR, 93 ice stimulus experiments, 96–98 implementation AS, 90–93 modeling of pulses, 87–89 physiological confounders, 89, 90 pulse reflections, 82–87 peripheral arterial line comparisons, 99, 101, 102 peripheral resistance (see Peripheral resistance) physiological model, 76 principal mechanism, 76 radial/digital arterial pressure pulse, 82 second systolic peak, 76 Valsalva experiments, 96–98 Pulse plethysmography (PPG), 63, 68 Pulse pressure (PP), 16, 26 Pulse reflections arrival time, 82 arterial paths, 82 arterial pulse velocity profiles, 85 delay times, 84 model’s equations, 82 Moens–Korteweg equation, 82 preliminary tests, 84 pressure dependence, 82 pressure response, 85, 86 propagation velocity, 85–87 QRS complex, 85 renal reflection coefficient, 83, 84 second systolic, 82 timing considerations, 87 Valsalva, 84 Young’s modulus, 82 Pulse tonometry, 27 Pulse transit time (PTT), 4, 5, 11, 27, 204 advantages, 70 arterial length, 62 arterial sites, 63 arterial stiffness, 63 BP estimation, 143 and BP relationship, 141–142 calibration, 69, 70, 141 continued miniaturization, 58 description, 63 estimation, cuffless BP, 140 infrared, 46 limitation, 63, 64 mathematical models, 66, 67 mechanical measurements, 44 and PAT (see Pulse arrival time (PAT)) PEP, 50, 64

237 physiological mechanism, 140 practical approach, 64, 65 reciprocal/PWV, 62 regulatory agencies, 52 ViSi’s cNIBP, 62 wearable objects, 62 Pulse wave analysis (PWA), 37 advantages, 132 AIx, 112 amplification of pressure waveform, 113 aortic pressure, 112 applanation tonometry, 109, 110, 130, 131 arterial catheters, 109 BP, 112, 118, 119 challenges, 132 classification, 112, 113 clinical evidence applanation tonometry, 120–124, 128, 129 PPG, 125–128, 130 DBP, 112 development, 132 elastic and geometric properties, 111 endogenous components, 132 exogenous components, 132 frequently encountered features, 115–117 GTF, 114 history, 108 inter-subject variability, 131 measurement site, 131 measurement system, 132 morphological analysis, 107 morphological differences, 115 optical measurements, 131 peripheral pressure, 113 PPG, 110, 111, 117, 118, 131 SBP, 112 sphygmograph recording, 108 sphygmomanometer, 108 systolic pressure, 113 TFs, 113, 114 Pulse waveform, 16 Pulse wave velocity (PWV), 4, 37, 166, 168–170, 178, 180, 182, 186, 204 advantages, 70 and arterial BP, 141 BP change tracker, 71 brachial-to-radial, 142 Bramwell–Hill equation, 66 cardiac pulse, 50 computerized measurement system, 56, 57 continuous BP monitoring, 71 continuous measurement, 50 description, 63 ECG, 63

238 Pulse wave velocity (PWV) (cont.) hydrostatic pressure effects, 56 limitation, 63, 64 machine learning methods, 71 mathematical modeling, 65–67 measurement, 62 (see also Measurement methods, PWV) Moens–Korteweg equation, 65 PAT and PTT, 56 PPG, 63 predictor, vascular stiffness, 27 principle, 62, 140 recording, 140 set point, 62 time delay, 56 trigger for absolute BP measurement, 71 volume clamp method, 62 volume clamp/plethysmography, 27 Young’s modulus for zero arterial pressure, 65 Q Quadratic equation, 176 Quality System Regulations, 194 R Radial artery, 62 Random Forest Tree method, 147 Random forests, 151 Recurrent neural network (RNN), 148 Red optical wavelengths, 68 Reflectance-mode PPG, 68 Reflective PPG sensors, 45 Registered Certification Bodies (RCBs), 199 Regression trees, 151 Regulatory framework and device classification Australia, 197, 198 Canada, 196, 197 environmental conditions, 193 European Union (EU), 195, 196 Japan, 199, 200 USA, 194, 195 Regulatory systems, 192 Reinitialization period, 183–185 Ridge linear regression, 147 Risk-based classification system, 192, 193, 200 Risks and controls, 192, 194, 197 Riva-Rocci technique, 23 Riva-Rocci upper arm cuff, 226

Index S Second systolic peak, 76 Seismocardiogram (SCG), 48 Sensors, 5 Set point, 62 Single cuff measurement, 181 Slow breathing, 171 Smooth muscle contraction, 180 Somnomedics, 57 Somnotouch PTT-based BPM, 179 Sotera Wireless markets, 56 Speech recognition, 147, 148 Sphygmocardiography, 23, 24 Sphygmograph recording, 108 Sphygmography, 2 accurate estimation, BP, 21 asymmetrical, pulse waves, 20 clinical medicine, 23 modification, 21, 23 non-invasive BP measurement, 21, 23 recording, pulse, 20 Vierordt’s instrument, 20 Sphygmomanometer, 21, 23, 24, 108 clinical parameter, 3 clinical use, 2 cuff-based automated oscillometric, 206 detection, markers, 3 management, hypertension, 5 noninvasive, 205 static test, 206 technological advances, 2 Stability test, 207 Standard deviation, 176 Static component (SC), 45, 47 Stethoscopes, 192 Stretch–strain relationship model, 11 Sub-endocardial viability ratio (SEVR), 115 Support vector machine (SVM), 151–152 Support vector regression (SVR), 151, 152 Systemic BP, 173 Systolic BP (SBP), 52, 142, 147, 155, 173 Systolic Hypertension in the Elderly Project (SHEP), 26 Systolic pressure (SP), 16, 112 T Technology, cuffless BP measurement accuracy, 228, 229 accurate device, 228 approaches, 228 clinical practice, 229 continuous beat-to-beat measurement, 228 continuous variable, 228

Index Test environment, 218 Theory-driven analytical model vs. data-driven based machine learning method, 146, 147 Therapeutic Goods Administration (TGA), 193, 197, 198 Therapeutic Products Directorate (TPD), 196 Thermometers, 192 Time-domain profiles, 147 Tonometry, 54, 55 Tonometry sensor array, 62 Transducer layer, 39, 40 Transfer functions (TFs), 110, 113, 114, 121 Transmission-mode PPG measurements, 68 Transmissive PPG sensors, 45 U Univariate/multivariate linear regression, 148 Unpredictable error, 175 US Food and Drug Administration (FDA), 193 US medical device classification, 194, 195 V Validation protocol accuracy studies BP change tests, 219, 220 calibration, 221 reference device, 219 stability, 219, 220 static, 219, 220 statistical outcomes, 220, 221 subjects investigated, 220 ANSI/AAMI/ISO standards, 206 Bland–Altman plot, 214, 215 BP change test histogram, 209 invasive, 208 noninvasive, 208 calibration, 206 GCP (see Good clinical practice (GCP)) histogram, differences, 215, 216 mean difference and standard deviation correction, measurements, 213 error-bands BHS, 214 ESH protocols, 214 FQP, 215, 216 MAD, 214

239 MAPD, 214 reference vs. test device, 213 standard, defined, 213 sequential, 209 simultaneous, 210 stability test, 207 standards noninvasive BP monitors, 206 static test, 206 statistical analysis BP ranges, 211 gender distribution, 211 invasive testing, sample size, 212 measurement frequency, 211 noninvasive testing, sample size, 211, 212 sample size, 211 special-subject population, 212 time duration, measurement procedure, 208, 209 Valsalva, 84 Valsalva experiments, 96–98 Valsalva maneuver, 171, 183 Vascular unloading/method of Penaz, 53 Videoplethysmography (VPG) method, 69 Vierordt’s instrument, 20 ViSi Mobile, 56 Volume clamp method, 11, 53–55, 62 W Wave propagation, 3 Wearable, 43, 50, 55, 58 Wearable devices, 5 Wearable objects, 62 Wearable/unobtrusive system, 159 Wesseling model, 167, 168 World Health Organization (WHO), 193 Wrist and finger sensors, 27 Wrist-worn devices, 4, 57 Y Young’s modulus, 82 for zero arterial pressure, 65 Z Zero transmural pressure, 46