Intel® Technology Journal

In te l ®

tec hnology jo ur n a l | E N AB L I NG h ea l Intel® Technology Journal

SEPTEMBER 2009

Enabling Healthcare in the Home t h care i n t h e ho me

More information, including current and past issues of Intel Technology Journal, can be found at: http://developer.intel.com/technology/itj/index.htm vol 13 | issue 03 SEPTEMBER 2009

ISBN 978-1-934053-23-2

9 781934 053232 7735858 21143

$49.95 US

ITJ9-3_Cover_BFC_39spn_090409.indd 1 9/4/09 8:28:45 AM About the Cover

Enabling Healthcare in the Home is the theme of the Intel Technology Journal, Volume 13, Issue 3.

The physician in the foreground (1) is remotely performing a house call. The doctor is able to see the patient (2) and to obtain diagnostic information, such as blood pressure and pulse vitals. The patient is able to remain at home, which is helpful and efficient for those with ambulatory or transportation issues. Next door (3) is an elderly person who can safely live independently because the home is out- fitted with sensors (the yellow spots) that monitor motion. Family caregivers located elsewhere can be aware that this individual is performing routine tasks. Upstairs is (4) someone sleeping. Sensors measure nighttime activity, which can be an indicator of health risk. Also upstairs (5) is a child or an elderly person for whom reading is difficult. The child or elder can snap-shot a page and listen to the words. Thus, technology can enable healthcare in the home for all ages.

ITJ9-3_Cover_BFC_39spn_090409.indd 2 9/4/09 8:28:46 AM Intel® Technology Journal | Volume 13, Issue 3, 2009

Intel Technology Journal

Publisher Managing Editor Content Architects Richard Bowles David King Doug Busch Eric Dishman

Program Manager Technical Editor Technical Illustrators Stuart Douglas Marian Lacey Richard Eberly Margaret Anderson

Technical and Strategic Reviewers Steve Agritelley Alan Boucher Doug Bugia Doug Busch Eric Dishman Farzin Guilak Kristina Kermanshasche George Korinsky Brad Needham Kevin Rhodes

Intel® Technology Journal | 1 Intel® Technology Journal | Volume 13, Issue 3, 2009

Intel Technology Journal

No part of this publication may be reproduced, stored in a retrieval system or transmitted in any form or by any means, electronic, mechanical, photocopying, recording, scanning or otherwise, except as permitted under Sections 107 or 108 of the 1976 United States Copyright Act, without either the prior written permission of the Publisher, or authorization through payment of the appropriate per-copy fee to the Copyright Clearance Center, 222 Rosewood Drive, Danvers, MA 01923, (978) 750-8400, fax (978) 750-4744. Requests to the Publisher for permission should be addressed to the Publisher, Intel Press, Intel Corporation, 2111 NE 25th Avenue, JF3-330, Hillsboro, OR 97124-5961. E-mail: [email protected].

This publication is designed to provide accurate and authoritative information in regard to the subject matter covered. It is sold with the understanding that the publisher is not engaged in professional services. If professional advice or other expert assistance is required, the services of a competent professional person should be sought.

Intel Corporation may have patents or pending patent applications, trademarks, copyrights, or other intellectual property rights that relate to the presented subject matter. The furnishing of documents and other materials and information does not provide any license, express or implied, by estoppel or otherwise, to any such patents, trademarks, copyrights, or other intellectual property rights.

Intel may make changes to specifications, product descriptions, and plans at any time, without notice.

Third-party vendors, devices, and/or software are listed by Intel as a convenience to Intel’s general customer base, but Intel does not make any representations or warranties whatsoever regarding quality, reliability, functionality, or compatibility of these devices. This list and/or these devices may be subject to change without notice.

Fictitious names of companies, products, people, characters, and/or data mentioned herein are not intended to represent any real individual, company, product, or event.

Intel products are not intended for use in medical, life saving, life sustaining, critical control or safety systems, or in nuclear facility applications.

Intel, the Intel logo, Celeron, Intel Centrino, Intel Core Duo, Intel NetBurst, Intel Xeon, Itanium, Pentium, Pentium D, MMX, and VTune are trademarks or registered trademarks of Intel Corporation or its subsidiaries in the United States and other countries.

†Other names and brands may be claimed as the property of others.

This book is printed on acid-free paper.

Publisher: Richard Bowles Managing Editor: David King

Library of Congress Cataloging in Publication Data:

Printed in the United States

10 9 8 7 6 5 4 3 2 1

First printing September 2009

2 | Foreword Intel® Technology Journal | Volume 13, Issue 3, 2009

INTEL® TECHNOLOGY JOURNAL Improving Healthcare Through Technology and Innovation

Articles

Foreword �� 4

Powering Healthcare Visions: Taking Advantage of Complexity, Connectivity, and Consumerism �� 6

From People to Prototypes and Products: Ethnographic Liquidity and the Intel Global Aging Experience Study �� 20

Adapting Technology for Personal Healthcare �� 40

Healthcare IT Standards and the Standards Development Process: Lessons Learned from Health Level 7 �� 58

Healthcare Information Integration: Considerations for Remote Patient Monitoring �� 80

Personal Health Device Interoperability �� 104

A Common Personal Health Research Platform — SHIMMER™ and BioMOBIUS™ �� 122

Gathering the Evidence: Supporting Large-Scale Research Deployments �� 148

Assistive Technology for Reading �� 168

Developing Consumer and IT Platforms in a Regulated Medical Device Environment �� 188

Table of Contents | 3 Intel® Technology Journal | Volume 13, Issue 3, 2009

Foreword

Douglas F. Busch Vice President Chief Technology Officer Digital Health Group Intel Corporation Fundamental medical science has made enormous strides over the past several Eric Dishman decades, and the length and quality of life in developed countries has improved Intel Fellow remarkably. However, trends in demographics, lifestyles, and treatment Digital Health Group practices are making our healthcare systems unsustainable. Improving access Intel Corporation to healthcare and controlling healthcare costs are two of the most contentious current issues in United States politics, and are a huge concern in most other countries also. Better use of technology in healthcare is an obvious strategy to help solve these problems by applying innovation to how and where we deliver care.

Science and technology have been applied aggressively to clinical diagnostics and treatment. Diagnostic imaging, advanced surgical techniques, and sophisticated pharmaceuticals have all had strong positive benefits. We have not “Better use of technology in healthcare yet seen the benefits of applying everyday technologies that we take for granted in other aspects of our lives, though. Much attention is now being given to is an obvious strategy to help solve applying information technology to record keeping and communication in clinical settings. The recent American Recovery and Reinvestment Act (part these problems by applying innovation of the 2009 economic stimulus package) focuses strongly on deployment and to how and where we deliver care.” meaningful use of Electronic Medical Records.

This is not enough. To really impact the availability and cost of healthcare, and improve the quality of our citizen’s lives, we need to shift some of the focus of healthcare from the clinical setting to the home. Just as much of banking has moved from marble lobbies filled with tellers to a self-service model enabled by technology, we can use readily-available technologies to move some aspects of healthcare from clinical settings to the home. We can also use technology to allow our aging population to stay in their homes longer, living independently. Both of these approaches have the potential to improve quality of life and reduce the cost of providing the healthcare and living assistance our population needs. We need to really understand how to use technology effectively for these purposes, and we need to adapt common technologies for successful use in home settings. We need to make home health technology simple, unobtrusive, and low cost, while protecting the security, privacy, and safety of our most “We need to make home health vulnerable family members. technology simple, unobtrusive, and low cost, while protecting the security, privacy, and safety of our most vulnerable family members.”

4 | Foreword Intel® Technology Journal | Volume 13, Issue 3, 2009

In this issue of the Intel Technology Journal, we hope to provide a broad “We hope to provide a broad perspective on the use of technology for delivering healthcare at home. Dr. Mike McGee, a noted physician, speaker and author, describes a vision for perspective on the use of technology for lifelong healthcare delivered at home. Intel researchers describe their work and delivering healthcare at home.” some of the conclusions reached over several years gathering ethnographic data and conducting pilot studies of home-based sensor networks. The leaders of two major heath standards organizations describe the approaches being taken to create interoperability of data and devices for healthcare. Lessons learned about integrating healthcare data systems are shared by a team developing health data middleware. Intel’s experience entering the medical device product development industry is summarized by the leaders of our product development groups. The design team responsible for building an assistive device for reading disabilities shares their approach and learned lessons.

Intel is both a long-standing member of the aging research community and a “We hope that we can encourage the new participant in the medical device community. We hope that by sharing our insights, and those of our collaborators, we can encourage the development of development of powerful, easy-to-use powerful, easy-to-use technology for home healthcare and independent living. technology for home healthcare and We hope that we can also stimulate investment in — and innovation for — home-based care industries through broad-based R&D collaboration, standards independent living.” development, and political advocacy to ignite this important marketplace. Our future, and that of our children, depends on dramatically improving access to healthcare and reducing cost, which we believe is achievable if we can “go home again” by using technology and innovation to reimagine how we deal with the epidemics of chronic disease and injury that we all face from the enormous success of our global age wave. “Our future, and that of our children, depends on dramatically improving access to healthcare and reducing cost.”

Foreword | 5 Intel® Technology Journal | Volume 13, Issue 3, 2009

Powering Healthcare Visions: Taking Advantage of Complexity, Connectivity, and Consumerism

Contributors

Mike Magee, MD Opinion Over the past ten years, the World Health Organization and other groups have been actively engaged in trying to answer the question “What is health?” A large part of this thought process has involved defining what health is not. It is not the healthcare system. It is not the reactive elimination of disease. It is not a simple commodity to be weighed against all other commodities in society. It is different from these things, and it is more than these things. “A large part of this thought process Health is universal and common to the people of the world, independent of geography, race, income, gender, and culture. Health is an active state of well- has involved defining what health is being that encompasses mind, body, and spirit. It is the capacity to reach one’s not.” full human potential, and, on a larger scale, the key that unlocks a nation’s full productivity and potential for development.

The Modern Health World: Shaped By Intersecting Megatrends To lead and innovate in today’s healthcare world requires a modern vision that is based on a highly strategic understanding of the major intersecting and mutually reinforcing elements shaping health worldwide. Among these are aging demographics, the growth in home-based caregivers, the dual burden of disease, the Information Revolution, and consumer empowerment and activism.

“By 2030, there will be over a million For most people involved in healthcare, aging equals numbers (how many people are over 65, how many are over 85, and so on). But aging is not United States citizens over 100.” about the aging population only. It is more complex than that. Fifty percent of current sixty-year olds in the United States have a parent alive, and by 2030, there will be over a million United States citizens over 100. Before our eyes, the American family is growing from three generations to four and five generations. In 25 percent of American homes, a non-professional, family caregiver works without pay and without support to manage frail parents and grandparents on the one hand, and immature children and grandchildren on the other. (I call these “informal” caregivers, as opposed to “formal” caregivers, those who are trained and paid for what they do.) Almost all of these caregivers are third-generation women, ages 40 to 70. When we look at the whole question of providing healthcare, we must address these family dynamics efficiently and effectively, and our solutions must bring support and relief to beleaguered home-based informal caregivers.

6 | Powering Healthcare Visions: Taking Advantage of Complexity, Connectivity, and Consumerism Intel® Technology Journal | Volume 13, Issue 3, 2009

Coincident with an aging society is the Internet, an incredible tool. When “Doctors, nurses, and hospitals seem combined with ever-expanding computing power, remarkably innovative processor technologies, broadband, wireless, and converging powerful to agree that ‘the best patient is an applications pushing massive amounts of information at lightning speed, the educated patient.’ ” Information Revolution gives us a glimpse of what is possible in a modern world in terms of health and healthcare. The true significance of the Internet, however, in this context, is that it ignores geography, and in doing so, breaks all the rules governing communication: who gets what information from whom, and for what purposes.

Regina Herzlinger at Harvard Business School first suggested a consumer- driven, healthcare model nearly 30 years ago. She and others at the time saw a disconnect between those who paid for the care and those who consumed it. It was clear to all that consumers of healthcare were often at a disadvantage because they lacked adequate information to support decisions. Three decades later, with the support of new information technologies, we see a very different reality. Doctors, nurses, and hospitals seem to agree that “the best patient is an educated patient.” They acknowledge that paternalism must give way to partnership. It must be teams over individuals. There must be mutual decision making. Most people would agree that these are all good things. However, our access to health information is leading us into health activism, led mostly by our informal family caregivers who labor as both providers and consumers of healthcare in nearly 25 percent of all American homes: they have no formal support or even acknowledgment from healthcare providers. For them, lack of “Our access to health information is healthcare information is not what’s literally killing them. Lack of a support system is killing them. leading us into health activism.”

Finally, an aging population, the Information Revolution, and the consumer health movement intersect and reverberate within the context of a global interconnected world community. High-speed travel, overnight delivery, cultural integration, the rapid spread of inter-species microbes, and the mass marketing of products, good and bad, have created a wide range of health challenges. The developing nations struggle with warfare, migration, famine, child mortality, and infectious diseases, while the developed world is overwhelmed with largely preventable chronic diseases. As the developing world exports the next influenza virus, the developed world returns the favor by marketing tobacco and the American diet to developing nations. I believe that together we share a dual burden of disease and a range of environmental threats to our health from global warming to scarcity of clean, safe water.

Rather than fighting these accelerating trends that are connecting our world, opening up access to health information, but also making our world “As the developing world exports the increasingly complex, we must embrace them. Our vision must be sufficiently next influenza virus, the developed powered to take advantage of complexity, connectivity, and consumerism. world returns the favor by marketing tobacco and the American diet to developing nations.”

Powering Healthcare Visions: Taking Advantage of Complexity, Connectivity, and Consumerism | 7 Intel® Technology Journal | Volume 13, Issue 3, 2009

The Power is Shifting “I present here five visions Clearly, healthcare reform is in the air. Whether such reform can positively affect our health, however, remains to be seen. For true reform to take place, sufficiently powered to guide a true powerful visions must be joined by cross-sector innovation to serve, humanize, transformation in our health.” and realize future human potential. I present here five visions sufficiently powered to guide a true transformation in our health and to encourage cross- sector collaboration among traditional healthcare organizations and technology leaders, such as Intel, and home health caregivers with existing positions in American homes.

Vision 1: Home-Centered Healthcare Nearly 100 percent of the assets we currently include in our definition of the healthcare system have little to offer us, in their current form, that would help in the build-out of a truly preventive healthcare system. I am including assets such as the bricks and mortar of our hospitals and our patient offices; our human resources as embodied in our training, roles, responsibilities, and payment incentives; our educational curricula; and our continuously reengineered processes targeted at in-patient safety and efficiency. These assets “We have been unable to create are original, or second or third iterations, of a century-old interventional care or manage a more inclusive and system that stubbornly survives, largely in its original form, because we have been unable to create or manage a more inclusive and anticipatory healthcare anticipatory healthcare system.” system.

Prevention of illness is grounded in education and behavioral modification. It begins before birth and extends beyond death. Let me explain: the parents’ nutritional status as well as their use of alcohol, drugs, and cigarettes affects fertility, as well as conception and in-utero development. On the other end of one’s life, friends and family predictably suffer grief and anxiety at the death of an individual, and these friends and family can often benefit from formal bereavement support services.

To be successful, therefore, a preventive healthcare system must take advantage of multi-generational relationships to provide multiple, repetitive inputs in real-time that allow micro-adjustments in a person’s daily life. Such a system demands intimately informed, highly motivated, and deeply committed individuals willing to gently prod those under their charge toward health and wellness. I believe that the goals of a preventive healthcare system require guiding hands and a pervasive presence across several sectors: family and community linkages, and the ability to efficiently lay out lifecycle plans and execute lifespan management on the one hand, and ensure adherence to palliative treatment plans for patients with chronic disease on the other. “To be successful, therefore, a Realizing this vision of a truly preventive healthcare system will require an integrated approach to care delivery, supported by cross-sector collaboration. preventive healthcare system must take advantage of multi-generational relationships.”

8 | Powering Healthcare Visions: Taking Advantage of Complexity, Connectivity, and Consumerism Intel® Technology Journal | Volume 13, Issue 3, 2009

Beginning with New Technology In the build-out of this new healthcare system, there is only one location that is both geographically identifiable and politically viable as a candidate: the home. Most Americans have a very favorable view of the home and its value to quality of life. As a result, politicians frequently embrace the imagery of “home” and are likely to be willing to invest (through legislation and other means) in the creation of healthier homes. While the home may be where the heart is, it is most certainly not currently where the health is, and in this way it is the optimum site for constructive change to take place in our health and healthcare.

I first began to think about the “moldability” of the home when I saw a “If we can leverage all kinds of remarkable exhibit at the World Fair in New York in 1964. As a 16-year-old, I remember sitting in the revolving theatre and witnessing the marvels of technology, I believe that we can technology. The audience was transported to what life was like in 1904 and efficiently re-center our healthcare then brought forward through the decades to 1964, and then 1984, 20 years into the future, showcasing how technologies and products helped to improve system in the home.” our lives and would continue to improve our lives in the future. The exhibit convinced me that toasters and refrigerators had truly improved our lives. Can we leverage technology today to do the same for our health? America views homelessness as a social failure, and we have begun to view “healthlessness” as a social failure also. It doesn’t have to be that way. If we can leverage all kinds of technology, from diagnostic and imaging to entertainment and financial, to equitably re-outfit and at least partially improve the health of people in their homes, I believe that we can efficiently re-center our healthcare system in the home.

Moving into the Home The primary health information loop in our current healthcare system extends from a hospital to a physician’s office and back. The home is an after-thought. If you wish to obtain healthcare, you must leave your home and enter the hospital-office loop. What if technology allowed us to change the trajectory of that loop so that it goes from home to care team and back to home? In this new approach, informal caregivers would become fully enfranchised “In this new approach, informal members of physician-led and nurse-directed care teams. These family caregivers would not only be linked virtually to their multi-generational caregivers would become fully families and to their care teams, but also to other informal family caregivers, enfranchised members of physician-led thereby effectively addressing the profound sense of isolation that comes with the informal caregiver role. A wide range of secondary loops would evolve from and nurse-directed care teams.” generalist to specialist, from clinician’s office to hospital, and from care team to insurer or pharmacy. However, the primary loop, where data would originate and from which privacy access would be granted, would be home-centered.

Powering Healthcare Visions: Taking Advantage of Complexity, Connectivity, and Consumerism | 9 Intel® Technology Journal | Volume 13, Issue 3, 2009

“The healthy home would also have The data flowing out of the home of the future would be rich, varied, real- time, and widely available. It would include vital signs and diagnostic and ubiquitous, low-cost sensors that imaging results sent wirelessly to care teams. The healthy home would also could track motions, actions, and have ubiquitous, low-cost sensors that could track motions, actions, and interactions. The data produced by these sensors would be interpreted by interactions.” artificial intelligence software and measured against predicted healthy living plans. The results would be fed in a continuous stream to the care team. The feedback loop would consist of a human team partner communicating through a friendly interface of one’s choice, such as a wristwatch, phone, radio, TV, or computer: this interface would act as a guide and companion. For example, the interface might remind the person to bathe if he or she had forgotten to do so; to increase fluid; to alter diet or to exercise; to take medication or vary dosage that day; or even to call his or her daughter as promised.

The Ideal Scenario Beyond home healthcare for our aging population, this model would make it “For example, the interface might possible for more healthcare to take place in the home for all the generations remind the person to take medication within the home. Ten outcomes can be skillfully integrated into my vision of a calm, well-organized, and healthy home: or vary dosage that day.” 1. A home health manager, previously the informal family caregiver, is designated for each extended family. 2. Nearly all Americans have health insurance, and a universal, integrated medical information highway is constructed primarily around the patient, and integrated with caregivers, rather than the other way around. 3. The majority of prevention, behavioral modification, monitoring, and treatment of chronic diseases takes place at home. 4. Physician-led, nurse-directed, virtual health networks of home health managers provide a community-based, 24-hour, seven days a week, clinical, educational, and emotional support team. 5. Healthcare insurance premiums for families are lower due to expert performance of the home health manager, as reflected in evidence-based, outcome measures of family members. 6. Basic diagnostics, including blood work, imaging, vital signs, and therapeutics are performed by the home health manager and transmitted electronically to the physician-led, nurse-directed educational network which provides feedback, coaching, and treatment options as necessary. 7. Sophisticated behavioral modification tools, age adjusted for each generation, are present and utilized in the home. These are funded in part by diagnostic and therapeutic companies that have benefited from expansion of insurance coverage and health markets, as early diagnosis and prevention takes hold.

10 | Powering Healthcare Visions: Taking Advantage of Complexity, Connectivity, and Consumerism Intel® Technology Journal | Volume 13, Issue 3, 2009

8. Physicians have more time to respond to health issues that truly require an office visit, as more and more health concerns are increasingly managed in the home. Monetary compensation for physicians increases in acknowledgment of their roles in managing clinical and educational teams and multi-generational complexity. Nursing school enrollment goes up, as the critical role as educational director of home health manager networks becomes a major magnet for the profession. 9. Family nutrition is carefully planned and executed; activity levels of all five generations rise; weight goes down; cognitive functioning goes up; mental and physical well-being are also up. 10. Hospital size continues to be reduced to appropriately meet the needs of the community. Scientific advances allow earlier diagnosis and more effective treatment, making the need for hospitalization increasingly rare. Hospitals are more specialized and safer: that is, the concentration of specialized services in Centers with high volume and expertise ensures safer outcomes from select procedures.

Vision 2: Reconnecting the Family — Unlocking Human Capital When virtual health networks first appeared, clinical visionaries at the time “The true purpose of innovative imagined an age of “telemedicine.” Dreamers thought the most pressing application for these networks would be to connect knowledge and skills in hardware and software is to integrate developed nations with pressing needs in the developing world. In reality, knowledge and skills in our own we now understand that the true purpose of innovative hardware and software, combined with pervasive broadband and wireless networking, is not homes.” connecting the developed with the developing world, but rather to integrate knowledge and skills in our own homes.

If we can reconnect the multigenerational American family, we can plan efficiently for health and reorganize community resources, thereby addressing both accessibility and affordability of health resources. If the lessons learned in one generation can benefit other generations downstream, and if we can efficiently utilize a multi-purposed health workforce, we will be able to afford wellness, prevention, and scientific and technologic progress in our healthcare and our health.

Segmentation as a Strategy No Longer Works The most pressing health demographic in the United States and worldwide “We will be able to afford wellness, is aging and its associated burden of chronic disease. This is not news. We have seen it coming and tried to plan for it. For decades our training schools, prevention, and scientific and hospitals, regulators, insurance industry, housing industry, legal system, long- technologic progress in our healthcare term care providers, and elder advocacy associations have focused on our aging populations and their unique needs and vulnerabilities. and our health.”

Powering Healthcare Visions: Taking Advantage of Complexity, Connectivity, and Consumerism | 11 Intel® Technology Journal | Volume 13, Issue 3, 2009

“We must now use technology and Rather than segment the American family into old and young, we must now use technology and social networking to aggressively reconnect the generations social networking to aggressively within the modern family. By doing this, I believe that we can manage the reconnect the generations within the complexity, mobility, and health data of the modern family, and we can distribute the responsibility for taking care of our health, as we move toward modern family.” the provision of universal healthcare.

Five Strategic Problems with Segmentation First off, generational segmentation within a family, especially as the number of people over 65, 85, and 100 continues to grow, becomes increasingly problematic. For one thing, it pits one generation against another in a “civil war of health financing,” that, absent coordinated efficiencies, creates a downward cycle with all roads leading to health rationing.

A second problem with generational segmentation is that it traps family learnings and health traditions in multiple intergenerational divides. For example, the 58-year-old, third-generation mother caring for her fourth- generation 82-year-old mother with a fragile fracture from osteoporosis, “Generational segmentation traps should ideally not only be an experienced and sympathetic caregiver, but also family learnings and health traditions a wise source of health information. While she is focusing on her mother’s independence, dignity, and pain control, she should also be asking “How can in multiple intergenerational divides.” I avoid becoming vulnerable to the same disease as my mother?” Her research and responses to questions of her doctors and other caregivers should tell her that silent osteoporosis exists in 52 percent of third-generation women by age 50, which is to say, she herself is vulnerable and should have a bone scan.

Her research should also make her aware of her own 27-year-old daughter’s vulnerability, since 98 percent of a woman’s skeleton is formed by age 20. Further, it should bring her 5-year-old granddaughter’s activities into sharp relief, since diet, exercise, and health-related behaviors in these early years determine whether this first-generation family member’s skeleton will be in good shape 15 years from now. Thus, in an enlightened, connected multi- generational family committed to each other’s health and well-being, health learnings should continually and seamlessly flow down the inter-generational ladder, with the goal of constantly shrinking the entire family’s future disease burden.

A third problem inherent in a segmented family care system is that informal family caregivers are trapped in a cavern of social isolation. Many ignore their own health and become ill during this period. Some die before those for whom they are caring. They are disconnected from the formal care team and “A third problem is that informal from other informal family caregivers who struggle in isolation themselves. Finally, they are often disconnected from their siblings and from children and family caregivers are trapped in a grandchildren, as the chaos, depression, and complexity of what they are doing cavern of social isolation.” overwhelms them.

12 | Powering Healthcare Visions: Taking Advantage of Complexity, Connectivity, and Consumerism Intel® Technology Journal | Volume 13, Issue 3, 2009

A fourth problem that affects the health and healthcare of a generationally “Segmentation works against the segmented family is that such segmentation works against the adoption of electronic health records and electronic health planning. These electronic adoption of electronic health records records and plans are required to move us from a system of intervention to one and electronic health planning.” of prevention, and they are the tools that will propel us ahead of the disease curve, by customizing and personalizing individual health planning.

A fifth and final argument for family connectivity is that it allows us to better define the human resource needs for a modern healthcare team; emphasizes individual roles and responsibilities; better assigns and utilizes these human resources with an eye toward multi-use for cross-generational assessments and services; and it enables a more informed application of unique community resources to fill in the gaps where they exist.

Vision 3: Lifespan Planning Record — An Organizing Planning Platform To manage the complexity of our multi-generational families and to make optimum use of our increasingly connected world requires an organizing platform with sufficient power and flexibility to adjust in real-time, while customizing and personalizing strategic health planning. Over the last decade, the discussion about healthcare records has focused on converting paper-based patient records of hospitals and doctors into electronic medical records (EMR). “The concept of a ‘Personal Health The goal was to improve accuracy and efficiency, a worthy goal. However, Record’ (PHR) began to emerge.” as this conversion process began, it became increasingly clear that what constituted a “Record” was not only what was in hospitals’ and doctors’ files. Rather, much of the real data on health resided with the patient; the EMR was only part of the overall health picture. Thus, the concept of a “Personal Health Record” (PHR) began to emerge, and this PHR may well someday subsume the focus on an EMR.

An Integrated Long-Term Plan I believe that in a truly preventive system, “health” is not a collection of late- stage, reactive interventions. Rather, I believe that health should be defined as a life fully lived — hopeful, productive, fulfilling, rewarding, and manageable. The determinants of such a life begin before birth, embedded in the healthful behaviors of one’s future parents, and are extended beyond death to one’s survivors.

“I believe that health should be defined as a life fully lived — hopeful, productive, fulfilling, rewarding, and manageable.”

Powering Healthcare Visions: Taking Advantage of Complexity, Connectivity, and Consumerism | 13 Intel® Technology Journal | Volume 13, Issue 3, 2009

Death

Birth Life Plan (100 years)

Figure 1 Source: Intel Corporation, 2009 Considering this broader view of health, the right concept for our health record system should be a Lifespan Planning Record (LPR), a virtual online organizing scaffold for health information and health planning.

The LPR for a single individual born today would extend out at least 100 years. It would include all of the baseline medical information needed by patients, and much more. It would consider economic, social, educational, and spiritual goals and milestones as well as medical and scientific objectives.

“The LPR for a single individual born Born today, the newborn child’s LPR would already contain a great deal of data. Some reasonable compilation of the health records of parents, today would extend out at least 100 grandparents, and siblings would be represented. Future diagnostic and years.” preventive therapeutic measures, based on familial information, would be flagged on the timeline. Print, video, and graphic information from other accessible intelligence databases would be seamlessly interwoven for easy use by the people caring for each other and this new global citizen.

As time passes, this “living” record would grow and adjust to assist in informed decision making, preventive behavior, and full and complete human development.

Obviously, many issues will need to be sorted out in order for such a vision to be realized, not the least of which are confidentiality, patient privacy, and control over records. However, the bottom line is that as quickly as the EMR is being subsumed by the PHR, the PHR is now being subsumed by the need for an LPR. The LPR is the best way to move us toward a preventive healthcare system.

“Future diagnostic and preventive therapeutic measures, based on familial information, would be flagged on the timeline.”

14 | Powering Healthcare Visions: Taking Advantage of Complexity, Connectivity, and Consumerism Intel® Technology Journal | Volume 13, Issue 3, 2009

Vision 4: Collapsing Databases — Accessing and Integrating Knowledge If the home is the center of our future healthcare universe, the connected “In order for healthcare consumers family an integral member of the health delivery team, and the LPR its organizing framework, where will the health knowledge come from and how to have full trust in their physicians’ will it be used to the best advantage? recommendations, they need to have Three enormous health databases are currently in the process of becoming full disclosure regarding conflict of available electronically and therefore, easily accessible. The first of these is the Clinical Research Database (CRD). Access to the latest information from interest.” research discoveries is critically important as we shift more responsibility for decision making to consumers’ shoulders. Health product companies appreciate that anything short of full and immediate disclosure of all study results could increase liability into the future and run the risk of collapsing a successful product (Merck’s Vioxx drug is a case in point) Moreover, in order for healthcare consumers to have full trust in their physicians’ recommendations, they need to have full disclosure regarding conflict of interest: if a researcher or physician receives payment for doing drug or product research and giving talks in support of that product, device, or drug, then consumers need to know this. As a result, therefore, of conflict of interest concerns and legitimate health consumer desires for early access to discovery information, major research databases are moving toward open transparency.

For better or worse, the public will soon have ready access to the vast “American people, as part of good majority of positive and negative results at the time of completion of medical research efficacy and effectiveness studies of new products and health devices citizenship, will soon be able to studies. These results will be electronic and therefore capable of being widely voluntarily contribute their de- distributed. In addition, the American people, as part of good citizenship, will soon be able to voluntarily contribute their de-identified health information identified health information to to nationally coordinated Clinical Effectiveness Research, effectively becoming data sources and thus co-contributors in the ongoing effort to improve health nationally coordinated Clinical in America. Effectiveness Research.” The second database is the Continuing Medical Education (CME) database which will also soon be widely accessible. In fact, nearly 20 percent of the U.S. CME database is already electronic and has been demonstrated to be effective. It is likely that within ten years, the vast majority of the CME database will be accessible and will be applied in real-time rather than in episodic segments. Hand-held devices are increasingly standard medical equipment in caring encounters, providing immediate database support to the patient-physician relationship during the evaluative and joint decision-making process. This allows experts to quite confidently predict that in a preventive healthcare system, where information is overwhelmingly the dominant healthcare “CME will be seamlessly integrated product, CME will be seamlessly integrated into the whole doctor-patient encounter and become an integral part of the diagnosis and outcome. into the whole doctor-patient encounter and become an integral part of the diagnosis and outcome.”

Powering Healthcare Visions: Taking Advantage of Complexity, Connectivity, and Consumerism | 15 Intel® Technology Journal | Volume 13, Issue 3, 2009

“CME and CCE would in many ways Which brings us to the third database, one that I envision will be called Continuing Consumer Education (CCE). As the consumer movement become one and the same.” continues to evolve from educational empowerment to active engagement and inclusion in the healthcare team, patients and their families will demand access to the same hand-held hardware and information software that the other care team members are using. This will help avoid any confusion that might arise from multi-tracked information and accelerate the need for simple and well- designed educational and organizational products.

Bridging the Translation Gaps As CRD, CME, and CCE become widely accessible, we are left with two translation gaps between new medical discoveries and those discoveries being translated into practice. The first gap is between CRD and CME. For example, if a study reveals that it is safer and more effective for the mother and the child if the mother is given an epidural anesthesia at 2 cm rather than at 5 cm dilation, and that doing so not only does not increase C-section rates but ensures safer, more comfortable labor and better Apgar scores for the baby, under our past system, this knowledge transfer to practice would take many years. However, with integrated CRD and CME databases, practice behavior changes would become coincident with a new discovery.

“Do people not loan these data to the If CRD and CME were to collapse upon each other, CME and CCE would in many ways become one and the same. Over the past decade patients and their people in whom they have the greatest caregivers have been gaining expertise and experience with new information trust and confidence?” technologies and applications at roughly the same rate. Yet the movements to organize an individual’s health data electronically was originally conceived as a provider controlled function. The EMR was viewed as a hospital and office- based platform. The EMR concept has more recently broadened into the PHR, a patient-centric organizing framework. But are separate approaches defensible?

Are these records not, after all, one and the same? Do not all clinical data originate with the people? Do people not loan these data to the people in whom they have the greatest trust and confidence: their physicians, nurses, and other caregivers? Moreover, if our records are one and the same, should we not also use the same informational resources to support our joint decision making? Wouldn’t this method be the best way to help us stay on the same page and avoid miscalculations, misinformation, or mistakes?

The elimination of the translation gap between CME and CCE has enormous economic implications also. Studies indicate that patients adhere to the plans they establish with their physicians only 25 percent to 50 percent of the time. Lack of adherence means costly re-treatment, and often hospitalization, when “Lack of adherence means conditions predictably worsen. In addition, many of the safety issues that have been the focus of attention in the healthcare community in recent years costly re‑treatment, and often are based on miscommunication between healthcare providers and between patients and their caregivers. hospitalization, when conditions predictably worsen.”

16 | Powering Healthcare Visions: Taking Advantage of Complexity, Connectivity, and Consumerism Intel® Technology Journal | Volume 13, Issue 3, 2009

Finally, the emergence of Clinical Effectiveness Research and the culture of continuous process improvement are dependent upon evidence-based protocols and processes shared and managed by patients and their clinicians. Together, these trends carry with them the promise to transform our health delivery system and to improve quality and efficiency simultaneously. Common shared databases, used effectively in real-time and customized to patient need, are critical to improving healthcare.

Vision 5: Techmanity — Technology as a Humanizing Force Back in 1983, Dr. John A. Benson, Jr., then President of the Board of Internal “Common shared databases are Medicine, voiced these words when questioned about technology’s impact on the patient-physician relationship. “There is a groundswell in American critical to improving healthcare.” medicine, this desire to encourage more ethical and humanistic concerns in physicians. After the technological progress that medicine made in the 60’s and 70’s, this is a swing of the pendulum back to the fact that we are doctors, and that we can do a lot better than we are doing now.” He accurately described the mood of clinicians then, and for most of the 20th century, toward technology; that is, a complex love-hate relationship. While they have rejoiced and cheered on progress, they have struggled to accept and master change in a manner that would avoid driving a wedge between them and their patients.

Medical Informatics Meets Consumers It is fair to say that, as the health consumer movement has matured over the past 30 years, and physicians have moved away from paternalism to partnerships and team-based approaches to care, that outright resistance and abject fear of technology has subsided, and clinicians have progressed to and beyond grudging acceptance. People and their caregivers have developed computer skills together, pursued broadband and wireless connectivity together, and discovered the value of personalized and customized computer search engines together.

Alongside this evolution, the specialty of Medical Informatics has risen to “People and their caregivers have legitimacy within the medical hierarchy, and its leaders have reinforced the need to take advantage of technology and informatics in support of humanistic developed computer skills together.” care. One such voice is that of Warner V. Slack, who heads the Center for Clinical Computing at Harvard Medical School. His first published paper on Medical Computing appeared in the New England Journal of Medicine in 1966. His book, Cybermedicine: How Computing Empowers Doctors and Patients for Better Health Care, is considered a classic, and argues, as Health Informatics expert Kevin Kawamoto, from the University of Washington has said, “Computers can be mutually beneficial for both the patient and the healthcare provider.”

Powering Healthcare Visions: Taking Advantage of Complexity, Connectivity, and Consumerism | 17 Intel® Technology Journal | Volume 13, Issue 3, 2009

“In embracing technology in medicine, If we have managed to move as caregivers from resistance to acceptance of technology in healthcare, I would suggest we have not moved far enough. As we must view it as both assistive and I stated in a reference paper as chair of the technology sub-committee for the transformational.” National Commission for Quality Long Term Care: “In embracing technology in medicine, we must view it as both assistive and transformational.”

Powering Up The Vision The revolutionary strength of modern information and scientific technologies is that they ignore geography. In so doing they allow us to reorient and connect beyond the limits of a range of barriers whether they be physical, social, financial, or political. The danger is not in over-reaching but in under- reaching. Our vision must be sufficiently forward looking and expansive to challenge technology innovators. Where are the “killer applications” that would allow lifespan planning to move us ahead of the disease curve? How can we target technologic advances in healthcare to first reach our citizens most at risk? How do we, in powering the health technology revolution, broaden our “Our vision must be sufficiently social contract to include universal health insurance? How do we unite the forward looking and expansive to technology, entertainment, and financial sectors (previously locked out of the healthcare space) with the traditional healthcare power players, and incentivize challenge technology innovators.” them to work together to create a truly preventive and holistic health delivery system that is equitable, just, efficient, and uniformly reliable? How can each citizen play a role in ongoing research and innovation, and help define lifelong learning and behavioral modification as part of good citizenship? What can corporate America do to advance the health information infrastructure, and in “doing good,” do well financially, thereby serving Main Street as it serves Wall Street?

Moving Beyond Acceptance Health Information leaders of the 21st century need to be more revolutionary. Were they to perform at their full capacity, our healthcare system would be transformed. Healthcare would be oriented around relationship-based care, cementing the people to the people caring for the people. If such a transformation were to take place, we would see improvement on ten different fronts simultaneously: access, efficiency, team-care coordination, multi-generational family linkages, inclusion of informal family caregivers in healthcare teams, targeted interventions for vulnerable populations, informed mutual decision making, lifespan health planning, evidenced-based personalized care, and a palpable presence of physicians, nurses, and care team members in the home. “Health Information leaders of the 21st century need to be more revolutionary.”

18 | Powering Healthcare Visions: Taking Advantage of Complexity, Connectivity, and Consumerism Intel® Technology Journal | Volume 13, Issue 3, 2009

Further Reading Kawamoto K. “Computer Technology in Health Care Settings.” The Journal of Education, Community, and Values. May-June, 2003.

Magee M. Home-Centered Health Care: The Populist Transformation of the American Health Care System. Spencer Books. NY, NY. 2007. Available at www.spencerbooks.com.

Magee M. “Connecting Healthy Homes to a Preventative Healthcare System: Leveraging Technology For All It Is Worth.” Harvard Health Policy Review. Fall, 2007, pp. 44-52. Available at http://www.hcs.harvard.edu.

Magee M. “Health Records of The Future.” Health Politics. November 18, 2006.

Nelson, B. “Can Doctors Learn Warmth?” New York Times. September 13, 1983. Available at query.nytimes.com.

Slack, Warren V. Bibliography. Harvard Medical School. Beth Israel Deaconess Medical Center: Center for Clinical Computing. February 13, 2008.

Author Biography Mike Magee, MD, is Editor of HealthCommentary.org. He is a Senior Fellow in Health Policy at the Center For Aging Services Technologies and a Commissioner of the National Commission for Quality Long Term Care. He is a widely recognized public speaker and author of ten books including Positive Leadership, Health Politics: Power, Populism and Health, and Home Centered Health Care: The Populist Transformation of the American Health Care System. He can be reached at www.mikemagee.org.

Copyright Copyright © 2009 Intel Corporation. All rights reserved. Intel, the Intel logo, and Intel Atom are trademarks of Intel Corporation in the U.S. and other countries. *Other names and brands may be claimed as the property of others..

Powering Healthcare Visions: Taking Advantage of Complexity, Connectivity, and Consumerism | 19 Intel® Technology Journal | Volume 13, Issue 3, 2009

From People to Prototypes and Products: Ethnographic Liquidity and the Intel Global Aging Experience Study

Contributors

Tim Plowman Abstract Intel Corporation This article documents how a large-scale, multi-site, ethnographic research project into aging populations, the Global Aging Experience Study, led to David Prendergast the development of concepts, product prototypes, and products for the Intel Corporation independent living market. Successfully leveraging the output of ethnographic research within large organizations and product groups is often fraught with Simon Roberts challenges. Ethnographic research produced within an industry context can Intel Corporation be difficult for an organization to thoroughly capitalize on. However, careful research design and sound knowledge transfer activities can produce highly successful outcomes that can be thoroughly absorbed into an organization, and the data can lend itself to re-analysis. Our research was conducted by the Product Research and Innovation Team in the Intel Digital Health Group, and the work was done in Europe and East Asia, eight countries in all. Using a mixed methodology, our research examined health and healthcare systems in order to chart the macro landscape of care provision and delivery. However, the core of our study was ethnographic research with older people, and their Index Words formal (clinical) and informal (family and friends) caregivers in their own homes and communities. Data from this study were organized and analyzed to Ethnography produce a variety of tools that provide insight into the market for consumption Aging by teams within the Digital Health Group. As the results of the research Product Development were driven into the Digital Health Group and other groups within Intel, it Ethnographic Liquidity became clear that the Global Aging Experience Study possessed what we term Healthcare ethnographic liquidity, meaning that the data, tools, and insights developed in the study have layers of utility, a long shelf life, and lend themselves to repeated and consistent use within and beyond the Digital Health Group.

20 | From People to Prototypes and Products: Ethnographic Liquidity and the Intel Global Aging Experience Study Intel® Technology Journal | Volume 13, Issue 3, 2009

Introduction In 2006, researchers within the product research and innovation team within “We explore how this research has the Digital Health Group launched the Global Aging Experience Study. This is a multi-year, multi-site, ethnographic, research project designed to develop affected how we imagine the new a global, comparative understanding of the practices and meanings associated independent living technologies of the with growing older. To date, we have conducted intensive, qualitative research into older people, living in eighty-five households in eight countries. These future.” data, supplemented by additional interviews with dozens of informal caregivers and healthcare professionals, resulted in a detailed database comprising field and analytic notes, thousands of photographs, and five hundred hours of video. Long-range research design, innovative reporting, and knowledge transfer activities by the team have resulted in the data and analysis providing ongoing value. In this article, we explore how this research has affected how we imagine the new independent living technologies of the future.

Demography: the Age Wave and its Implications The world’s population is growing older. According to the latest United Nations biennial population forecasts, 11 percent of the world’s current population of 6.9 billion are over 60 years old. By 2050, this will be 22 percent of a total population of 9 billion people. However, the developed world will be older, with 33 percent of its population over the age of sixty [1]. There are two key demographic drivers of this change: longer life expectancy and lower fertility rates. “The developed world will be older, with 33 percent of its population over People are now living longer lives and, in general, are healthier and more active. During the course of the twentieth century, average life expectancy the age of sixty.” within the developed world rose from 50 to 78 years [2]. Populations have benefited from better basic healthcare and from the decline in infectious diseases. However, fertility rates have continued to fall in the developed world. The rate is now 1.6, which is below the 2.1 rate, regarded by demographers as the requirement for a population to replace itself. This low fertility rate will mean that global population growth will level out during the middle of the century. However, it also means that the ratio of people who work versus those who do not work, the dependency ratio, will continue to fall. For example, in Britain in the early 1900s when old-age pensions were introduced, there were 22 people of working age for every retired person. In 2024, there will be less than three [3]. Such a ratio has fiscal implications: governments will get less revenue to support the health and social care needs of their aging populations, and they will be less able to meet the pension entitlements and expectations of “Governments will get less revenue retired people. Political and policy discourse around the inevitable bankruptcy to support the health and social care of the social security system in the United States is perhaps the most familiar manifestation of the impending problem to U.S. residents. needs of their aging populations.”

From People to Prototypes and Products: Ethnographic Liquidity and the Intel Global Aging Experience Study | 21 Intel® Technology Journal | Volume 13, Issue 3, 2009

The ways in which different countries’ populations will age in light of the Percentage of population over 65 years old, demographic shifts just described are likely to differ widely, both from a 2006 Percentage of population over 65 years old, sociocultural and a material perspective. For example, in the United States, the 2050 Projection shift will be amplified by the aging and retirement of the so-called baby boomer Percentage of population over 80 years old, 2003 and 2004 generation; i.e., the 80 million people born between 1945 and 1965. In Italy, Percentage of population over 80 years old, Spain, and Germany, the continued decline of fertility rates is resulting in 2050 Projection smaller families, which itself results in these countries’ populations growing 12.5% older, faster. But whatever the structural constraints (i.e., social, demographic, USA 20.6% cultural, or economic) on the aging experience, the anticipated challenges 3.35% 7.3% presented by aging, both to the individual and societies, will be profound. See Figure 1 for the percentage of populations in the developed world who are over 19.7% Italy 35.5% 65 and 80, respectively, and also projections of aging as a percentage for those 4.87% populations. 15.2% Looking at the 27 member states of the European Union alone, the rate at 11.6% Ireland 25.9% which Europe’s population is aging relative to the number of younger people 2.65% available to provide support for elders is predicted to diverge. While the falling 7.6% dependency ratio is also due to falling fertility rates, the net outcome is that a 17.6% shrinking workforce threatens the fiscal base of numerous healthcare systems. Sweden 24.7% 5.32% As the broad demographic picture changes globally, the health landscape 9.7% is being transformed. The decline of deaths due to infectious disease is 19.4% accompanied by the rise in the number of people with chronic diseases, Germany 28.4% 4.24% conditions increasingly associated with affluent and sedentary lifestyles. Among 12.2% these chronic diseases are, for example, Type 2 diabetes, chronic obstructive

17.7% pulmonary disease (COPD), and congestive heart failure (CHF). These diseases Spain 34.1% are expensive to manage and represent a significant strain on the healthcare 4.22% systems of the developed world. Annual U.S. costs of COPD were $37.2 12.3% billion in 2004, which included $20.9 billion in direct healthcare costs, $7.4 16.4% billion in indirect morbidity costs, and $8.9 billion in indirect mortality costs France 27.1% 4.28% (from the American Lung Association). However, many of the developing 10.9% nations are experiencing significant increases in the incidences of such diseases

15.8% also. It is also important to note that individuals with these conditions are UK 23.2% not the only impacted parties: caregivers, public health systems, payers, and 4.32% employers are also negatively affected by the rise in the incidences of these 8.8% diseases. 0% 10% 20% 30% 40%

Figure 1: Percentage of Population Over 65, and Over 80, and Projections Source: Intel Corporation, 2009

22 | From People to Prototypes and Products: Ethnographic Liquidity and the Intel Global Aging Experience Study Intel® Technology Journal | Volume 13, Issue 3, 2009

Health and Welfare Systems “Of critical interest to our research While differences exist in the ways that countries will age, there are also team was the ways such systems have significant variations in the ways that countries are organized to respond to the challenges and opportunities that an aging population present. Of critical evolved.” interest to our research team was the ways such systems have evolved — how they are funded and structured — and how people comprehend, experience, and behave in relationship to these systems. A base assumption of the project was that there is a dynamic relationship between health and care systems and the ways people think about, and respond to, health challenges and the process of aging in general.

Spain, Sweden, and Italy appear to receive comparatively high dividends for their levels of investment in healthcare; i.e., the overall measures of national health relative to the percentage of GDP spent suggests that those euros are well spent. This suggests that cultural, environmental, and other factors are at least as important as formal healthcare systems in shaping health outcomes. However, it is worth remembering also that a high proportion of healthcare expenditure is directed towards treating ill-health, particularly for those over the age of 75 [4].

For example, in Europe there is considerable diversity in the ways that welfare “The interplay between people, states are organized and how health and social care for older people are practices, policy and politics, and delivered. A range of factors underpin this variety: the historical development of welfare states; local, regional, and national ideas about the role of the state economics has created a complex versus the individual, family or third-sector (voluntary) in providing care; and the fiscal and economic frameworks in which development has occurred within and mutable health and healthcare these countries. The interplay between people, practices, policy and politics, landscape.” and economics has created a complex and mutable health and healthcare landscape.

Analysts have created a series of models that seek to describe and categorize the different welfare systems of Europe. These models provide a good guide to understanding the fundamental principles or structure of how care is organized and financed. Notably, Esping-Anderson’s [5] work identified three distinct social welfare regimes. These are the social democratic model (Nordic countries), conservative or state corporatist model (Germany or the Netherlands), and the liberal welfare or Anglo-Saxon model (United Kingdom). Ferrera [6] subsequently added a fourth category — the Southern European model including Italy, Greece, Spain, and Portugal. In the United States, the system can best be characterized as a liberal, capitalist model that “endorses equality of opportunity but not of outcomes” and is arguably outperformed by the state corporatist model in terms of social and economic objectives [7]. “These models provide a good guide to understanding the fundamental principles or structure of how care is organized and financed.”

From People to Prototypes and Products: Ethnographic Liquidity and the Intel Global Aging Experience Study | 23 Intel® Technology Journal | Volume 13, Issue 3, 2009

“In short, these models render human These models represent high-level views of systems and so fail to capture the regional and local variations in delivery that often makes care provision deviate and economic activity in brush from these ideal, typical models. In short, these models render human and strokes too broad to support nuanced economic activity in brush strokes too broad to support nuanced insights into phenomena such as the relationship between policy and the experience of aging insights.” in a given socio-economic structure. Over recent years and in anticipation of dramatic transitions within the demographic landscapes of Europe, there have been significant changes in policy planning and rhetoric surrounding how welfare and care systems operate. European long-term care systems demonstrate that different attitudes and expectations exist about the relative roles of family, state, and volunteers.

To answer the questions of shifting demographics and varying approaches to healthcare, many countries are moving away from hegemonic, expensive, institutional, and medicalized responses towards a more community- and home-centered approach to care. The propensity of bio-medicine to extend its domain over a range of real and perceived maladies, despite any empirical basis for doing so, has been well documented within the social science literature. “This shift towards a more Scheper-Hughes and Lock [8] argue that “the funnelling of diffuse but real complaints into the idiom of sickness has led to the problem of medicalization community- and home-centered and to the overproduction of illness in contemporary advanced industrial approach to care is being forged, with societies.” In addition to the gradual bankrupting of healthcare systems, owing to shifting workforce ratios, medicalization has also contributed to healthcare varying degrees of success, through systems being overburdened and dysfunctional, as well as to inadequate attempts to create the optimal mix institutional and social responses to the challenge of aging. This shift towards a more community- and home-centered approach to care is being forged, of funding and provision of care by with varying degrees of success, through attempts to create the optimal mix citizens, family, state, and private and of funding and provision of care by citizens, family, state, and private and voluntary sectors of society. This multi-faceted approach frequently entails a voluntary sectors of society.” repositioning of the state as a financier, rather than as a direct provider of care to individuals, through the amplified use of market mechanisms, the growth of the private and commercial care industries, and increased regulation and funding of the not-for-profit and voluntary sector [9]. Additionally, a central component of the rethinking that is underway in the way that care for older people is organized is a commitment to home-based models of care in which technology plays an enabling role.

24 | From People to Prototypes and Products: Ethnographic Liquidity and the Intel Global Aging Experience Study Intel® Technology Journal | Volume 13, Issue 3, 2009

The Intel Global Aging Experience Study The need for innovation today in order to prepare for tomorrow’s demographic “The work was designed to support the change was the primary context in which the Global Aging Experience Study was conceived and conducted. The work was designed to support the Digital Health Group’s commitment Digital Health Group’s commitment to affecting a paradigm shift within to affecting a paradigm shift within the independent living market. Our research built on a foundation of earlier work conducted in the United States, and we sought to extend that the independent living market.” work into Europe and East Asia thereby providing concrete local detail, while simultaneously developing a more thorough global perspective on the experience of aging. In addition to deepening our understanding of the myriad social and cultural differences in people’s subjective experiences of aging and health, our other research objectives were to challenge prevailing assumptions about what it means to grow old and to identify strategic opportunities for appropriate technologies and services for older people. Grounded in a thorough review of the gerontological literature, and utilizing ethnographic techniques such as open-ended interviews, observations, and multi-day visits at multiple households in each country, our team gathered thousands of images, stories, and insights about what it means to grow old from the perspective of older people themselves. This was the first step in a long process of understanding the needs over time for real people. Local researchers and healthcare professionals helped us recruit older people from diverse socioeconomic backgrounds and with a wide range of health conditions. Efforts were also made to ensure a reasonable distribution of ages was represented, ranging from 60 to 100 years old. Researchers would typically spend two days per household. The first day was usually spent getting to know the occupants and exploring their life and health histories, while the second visit involved an exploration of the home and the meaning and use of material objects within it. The ethnographic interview ended with a mapping exercise and frequently, it ended with trips around the neighborhood with the senior to meet their friends, families, and community members. Our style of questioning was purposively fluid and relaxed, providing encouragement and time for participants to think through and formulate their responses, often revisiting sensitive or painful memories. Over the course of the ethnographic encounter the following domains, among others, were gradually probed and explored:

•• Older people’s relatives, acquaintances, and care networks. •• Life review and biography. “Our team gathered thousands of •• Experiences and expectations of aging. images, stories, and insights about •• Health and housing histories. what it means to grow old from the •• Older people’s relationships with their built, material, and social environments. perspective of older people themselves.”

From People to Prototypes and Products: Ethnographic Liquidity and the Intel Global Aging Experience Study | 25 Intel® Technology Journal | Volume 13, Issue 3, 2009

•• Perceived and actual availability of formal and informal support networks following critical health events or along chronic disease pathways. •• Health information sources used by older people and their carers in different parts of the world to make informed decisions about both their treatment and their future. •• Everyday life, technology, and home. We provide a vignette to demonstrate the wealth of material and insights that can be gained from just one of these case studies.

Widowhood and Aging in an Italian Village “Anna brought up her own children Anna (85) shares a house with her unmarried oldest son Dario in a village about an hour from Milan. She has lived in this area all her life and soberly to work hard and be financially recalls the poverty of her youth. “I remember a widow came to my mother independent.” and told her that her children didn’t want to sleep because they were hungry. And my mother gave her some flour...” Anna developed an abhorrence of debt after seeing her father, a seed tradesman, in difficulties, and she brought up her own children to work hard and be financially independent. Life wasn’t easy. To make a profit from the small café bar she and her husband ran demanded their attention eighteen hours a day. When her husband suddenly died from acute kidney failure in 1973, she managed to close their business for eight days and then the family had to go on working. Times were hard, but they were able to survive, as her three children were already working. Now well into her eighties, Anna reflects how she feels differently inside in comparison to how she felt in her youth, “I am different in the way life is different now. In the past I had to sacrifice a lot. I managed to get a house to pass onto my children. I worked hard for it. My husband liked to have a motorcycle but work was secondary for him. It was always first with me.”

As she ages, Anna feels she has a safety net in her children and especially her oldest son who has taken on the role of family caregiver. Italian law stipulates that an elderly parent in need has the legal right to demand either accommodation or financial help from children with resources. Many would not press this right however. Anna’s daughter now lives in another part of Italy and so can only visit for holidays, but they talk a great deal on the telephone.

Anna believes older Italians, though still reliant on the family, live better today “As things stand today, young Italians than in the past due to the basic state pension — a welfare benefit the country may have trouble supporting for future generations. As things stand today, entering the workforce will have to young Italians entering the workforce will have to pay contributions equivalent pay contributions equivalent to 127 to 127 percent cumulatively of their salaries over the next fifteen years to guarantee the same level of benefit support received by the current older percent cumulatively of their salaries generation. Anna is able to subsidise her income a little with a small pension over the next fifteen years to guarantee from the days when her husband worked in a factory, but aside from this, like many women and small business owners, she regrets they were not able to the same level of benefit.” contribute to personal pensions.

26 | From People to Prototypes and Products: Ethnographic Liquidity and the Intel Global Aging Experience Study Intel® Technology Journal | Volume 13, Issue 3, 2009

Anna feels her health has seriously started to decline over the last year with “She felt it demeaning to use a walker recurrent problems with her knees causing her the greatest concern. In 2002, her arthritis was causing her so much pain she opted for a knee prosthesis in public and preferred a single operation against her doctor’s advice. She was in hospital for three weeks and walking stick.” then had to spend six weeks at home in bed being cared for by her son and sister. Her treatment included an hour of physiotherapy per day and a passive exercise machine to prevent her muscles from atrophying. Her rehabilitation was paid for by the healthcare system for two months and she feels this was not long enough. She neglected her exercises, and now problems have begun to emerge with her other knee due to long periods of over compensation. This is partly because she felt it demeaning to use a walker in public and preferred a single walking stick.

Anna suggests the chronic pain in her legs is helped by massage and creams. “Anna feels alienated from her doctor Unfortunately she is unable to get massages on the public health service and private practitioners charge €40 an hour, a sum she feels guilty about spending. so she avoids visiting her local clinic.” Anna feels alienated from her doctor so she avoids visiting her local clinic. Her shoulder has been troubling her recently and when she asked for more of the cream that works on her knee to fix the problem, he refused on grounds of expense. Nowadays, Anna just phones in when she needs her prescription and her son picks it up. She finds it very irritating that she has to do this every six to twelve days, as she is only allowed one box of twelve blood pressure pills at a time and would like more pills in each box to save her the trip.

Decreasing mobility is having a significant effect on her quality of life. “I feel old because I am dependent on other people. The problems with my legs make me feel old. I don’t have any plans for the future now. I don’t want to go out dancing but I would love to visit a specialist to be able to move around better.” For the last eight weeks, Anna has not left her apartment. Instead she spends her days watching the world go by from her balcony and trying to do the little she can in terms of housework: her family decided to pay for a home help worker to visit four hours a week to clean. Her son has taken over the shopping duties. Anna is not able to conceive of using an electric buggy to get around as she no longer believes she is safe going out by herself. Having given up on travel, she now regrets not seeing more of the world during her life. The main variation in routine occurs when her daughter or sister comes to help look after her once a year when Dario takes a respite break from his care-giving duties and visits a health spa. “I would consider a nursing home during these times, but only for a month, for a change, never for good.” “For the last eight weeks, Anna has

Anna uses sticks to get around and has a stick located in every strategic position not left her apartment. Instead she in the household. She sometimes tests and pushes herself by attempting to walk spends her days watching the world go between resting points. The main obstacle for her in the house is the slippery marble staircase from her apartment down to the ground floor. by from her balcony.”

From People to Prototypes and Products: Ethnographic Liquidity and the Intel Global Aging Experience Study | 27 Intel® Technology Journal | Volume 13, Issue 3, 2009

“As Anna’s mobility has declined and Anna has heard of alarm pendants but does not have one. She carries a mobile phone with her in case of emergencies but never takes it into the bathroom her social circle has narrowed, her with her, one of the most dangerous rooms in the house. Last year, Anna had phone has become an increasingly a fall in her bedroom and could not get up. Her son was not able to hear her cries over the living room TV and her mobile phone battery was flat. She important lifeline.” managed to crawl to her charger but didn’t realize that she could use the phone while it was plugged in. It was two hours before she was discovered.

As Anna’s mobility has declined and her social circle has narrowed, her phone has become an increasingly important lifeline. Her son works long hours, and she receives few visitors: she, therefore, calls her sister as often as she can afford it. She can still cook a little and loves to sew. She used to make clothes for her family, but she has stopped as nobody wears what she makes. Reading is another hobby, as is listening to the radio. The real highlight of her month is a visit from her grandchild. This occurs only rarely, however, as she dislikes her second son’s wife and they do not feel she is capable of taking care of the child on her own. “I don’t think I would want to spend another twenty years living like this, though I would love to see my grandchild married. Although I am alone a lot, I am not lonely. I can always watch the street from my balcony.”

Developing Insights and Tools for Knowledge Transfer “The real highlight of her month is a As the selected field notes of the interview with Anna demonstrate, visit from her grandchild.” ethnographic fieldwork requires a balance between having an appropriate focus, but not being so focused on one topic that you lose sight of what is important to the participants themselves. Ethnography, in this respect, is not like requirements gathering, usage model development, or design research, where the goal is to determine the boundary conditions for a particular product. Instead, we are trying to understand what it is that people want and need through a holistic exploration of their subjective experience, and the associated meanings and behaviors that interpellate [10] with wider socio-economic structures. The requirements come later.

When ethnographers return to the office, they bring with them images, audio recordings, video recordings, and hundreds of pages of field and analytic notes to both review and log for later analysis. As depicted in Figure 2, transcribed voice recordings and field notes are then entered into qualitative data analysis software to be coded into themed categories that allow for a horizontal view across the data set to be developed. “We are trying to understand what it is that people want and need through a holistic exploration of their subjective experience.”

28 | From People to Prototypes and Products: Ethnographic Liquidity and the Intel Global Aging Experience Study Intel® Technology Journal | Volume 13, Issue 3, 2009

Figure 2: Qualitative Software for Coding Raw and Analytic Field Notes Source: Intel Corporation, 2009

As important as systematic coding of interviews are, group analytic working “This kind of analysis requires very sessions, or harvest sessions, are equally important. Harvest sessions are where data from the field, the photos, and other collateral are shared, in an attempt to active participation and constant inductively identify patterns in the data, as well as to note the important critical engagement.” dimensions of variability within and between countries. These tasks are accomplished by working through the data and actively listening for recurring insights, themes, and other aspects of the data. This kind of analysis requires very active participation and constant critical engagement. Post-it notes are used to capture ideas as they emerge. After hours, sometimes days, of sharing stories from the field, the team begins to organize what might be hundreds of post-it notes into clusters, and robust patterns begin to appear. Identification of the patterns in the data leads to grounded theory insights about what is most important to people. For instance, one persistent idea that came up in story after story in our Global Aging project was simply this: people don’t want to be treated as if they are no longer competent or valued members of “This kind of analysis requires very society. They are human beings, with aspirations, goals, and a sense of identity active participation and constant and self-worth. And yet, so many healthcare devices and services treat older people as if they were children, or worse. critical engagement.”

From People to Prototypes and Products: Ethnographic Liquidity and the Intel Global Aging Experience Study | 29 Intel® Technology Journal | Volume 13, Issue 3, 2009

Feeling safe Supporting cognition “Since I fell I’m a lot more “You have to challenge afraid of falling.” your mind and brain.”

Environments Supporting of choice THE MIND-BODYphysical CONTINUUM activities “Despite my lack of mobility “I want to move easily, I can turn on garden lights be independent.” with this remote control.”

Meaningful & Bringing useful life healthcare home “It would be great to maybe “People think that doctors have an exhibition of my can cure everything— work sometime.” that’s not ture.”

Enabling Help social interaction getting care “I don’t like to be alone.” “We know the pharmacist very well, it’s easy to get my medicine.”

Figure 3: Initial and Final Versions of an Opportunity Map Source: Intel Corporation, 2009 As more and more of these needs are noted, the team begins to cluster them into what we call the opportunity map. The map, featured in Figure 3, helps us organize many of the possible ideas about needs and technology ideas that come up during sharing of field stories, or ideas about services or other interventions. We repeatedly refer to this map when thinking about products already on the market, or products in development within the Digital Health Group. By doing this we can identify gaps or places that need more attention.

Beyond the identification of needs and opportunities, when thinking through the ethnographic data, the study team identifies insights that are neither ideas for new products nor areas of explicit need; rather, they are insights into the required characteristics of any device or solution we might develop. We capture these as design principles (see Figure 4).

Figure 4: Design Principles Derived During the Analytic Process Source: Intel Corporation, 2009

30 | From People to Prototypes and Products: Ethnographic Liquidity and the Intel Global Aging Experience Study Intel® Technology Journal | Volume 13, Issue 3, 2009

With all these materials and insights in hand, we finally have a basis for actually thinking about products, services, or other interventions. We follow up harvest sessions with design brainstorms. During these work sessions (see Figure 5), we use the opportunity maps to identify which slice of the map we want to address. Keeping in mind the values and design principles from our data, we begin to explore ways of addressing particular needs. Here is where mapping other products, technologies, or research onto the map ahead of time really helps inform our thinking.

Key Aging Experience Themes A range of important themes emerged from the analysis of the Global Aging Experience data set. Seven of these are outlined next.

People want to focus on what they CAN do, not what they CAN’T do Figure 5: GAE Brainstorming Session As one woman told us, succinctly and unambiguously, “You are sick when you Source: Intel Corporation, 2009 are lying in bed.” That is why so few people self-identify as either ill or old. Many people chose not to use canes or assistive devices in the home. This is not just because these devices are socially stigmatizing in appearance, but because these devices reinforce a personal identity as someone who is sick. Many people sought out challenges as ways of keeping themselves sharp: in fact it seemed that it was people’s energy level, their will to pursue such challenges, that “Many people chose not to use canes or most correlated with self-identification as ill or old. Still, we cannot avoid the assistive devices in the home.” fact that many people will and do need assistance. The key is to provide this assistance in ways that people recognize “as helping me do what I want to do,” not as a constant reminder that I am no longer capable.

We identified a sort of denial that is a healthy part of the natural aging process that makes some interventions very difficult. We call this adaptive optimism. Often, people will regard most interventions as not necessary until it is too late, unless we can find a way to introduce these interventions into their lives earlier. The clear implication of this is that we need to be thinking about technologies that adapt and age with a person, changing the mix of offerings as the person ages. “Often, people will regard most interventions as not necessary until it is too late.”

From People to Prototypes and Products: Ethnographic Liquidity and the Intel Global Aging Experience Study | 31 Intel® Technology Journal | Volume 13, Issue 3, 2009

“These factors can seriously impact Healthy aging and independent living mean far more than health There are many factors that enable a person to remain at home and relatively a person’s ability to live the life they independent. These include a wide variety of capabilities, not just health-related desire in later years.” ones. The ability to take care of one’s own home maintenance or gardening, the ability to get groceries, prepare meals, and clean up afterwards, the ability to get around the neighborhood or town, are among some of the most important ones. All of these factors can seriously impact a person’s ability to live the life they desire in later years. We need to be thinking about all of these factors when we design interventions. A few of the key areas for enabling services are these:

•• Mobility and transportation in the community are crucial. Community- based transportation initiatives are catching fire both in the United States and in Europe. Different attitudes towards private transportation and different availability of public transportation will mean very different systems in different places. •• Services for the home are becoming increasingly important. This is particularly true for widowed husbands or wives: the loss of a spouse can mean a serious adjustment of lifestyle, at a time when the ability to adapt is not what it once was. Cooking, cleaning, home maintenance, and other forms of routine care will become increasingly needed. An opportunity exists to provide an infrastructure to enable a broader, less centralized marketplace for trusted providers. •• The need for finding trustworthy providers is going to increase. Technology can play a major role in helping communities identify and enable trusted providers. This may be particularly important considering that some forms of care that one would expect to come from family members will come from strangers. Health is not an objective quality; it is defined collaboratively and culturally Health is defined through interactions and negotiations among various people, including informal caregivers, family members, hired in-home caregivers, and medical caregivers. We saw many differences within and between these networks in our assessments of an elder’s health.

•• Assessments of health can vary from person to person. In Germany, for instance, an elderly husband insisted that his wife (who had suffered a mild stroke) needed much more help with driving, shopping, and taking her medications than the woman herself felt she needed. Both she and her daughter felt that the husband was being overly protective. In multiple cases, children had put their parents in nursing homes even though those parents did not feel they needed to be there. And yet, the parents acquiesced, mostly because they “did not want to be a burden.”

32 | From People to Prototypes and Products: Ethnographic Liquidity and the Intel Global Aging Experience Study Intel® Technology Journal | Volume 13, Issue 3, 2009

•• Notions of health vary from place to place in both major and minor ways. The cultural, social, and political systems in which we are embedded shape our attitudes and behaviors with regards to health. In England, people behave in particular ways as a result of the presence of the National Health System’s guarantee of life-long, free healthcare. In Mediterranean countries, people do not recognize eating as a health-related practice. They simply eat what people outside the region recognize to be very healthy foods. People mark the progression of aging by watershed events People do not experience aging as a gradual loss of capability, but rather as a series of threshold events. It is most obvious in the case of falls, where a combination of physical or cognitive decline might go unacknowledged for a long time before the precipitating event, namely, the fall, makes it official. While the obvious solution might seem to be monitoring and intervening earlier, we suggest it is not that simple: people who are in this state of adaptive optimism or denial might not willingly submit themselves to this kind of intervention.

•• Some watershed events initiate decline. Non-health-related events can have a serious health impact. For example, the loss of a loved one or a child, resulting in grief and depression, may have measurable, negative health “Meaning is made through social consequences. interaction and that does not cease in •• Sometimes external events leave deficits that expose health decline. The loss of a spouse may entail the loss of a person who has for years helped older age.” one cope with other deficits. A wife who has prepared her husband’s meals, done his laundry, and helped with other daily life needs over a period of forty or fifty years will, upon death or disability, leave a huge gap deficit that the husband or other informal caregivers will need to overcome. Healthy aging is inextricably linked to social participation People of all ages aspire to have a sense of belonging, a legitimate role in the life of their family and community. We are meaning makers all through life: meaning is made through social interaction and that does not cease in older age. At the same time, people become more acutely aware that ongoing engagement with the community or their family might actually increase their sense of being a burden, not a contributor. Thus, enabling social participation means lowering the costs associated with social engagement in at least three ways:

•• Feeling useful or productive, or at least having something to offer and the sense of identity that comes with that. This could be either in the form of caring for grandkids or being able to tell one’s story, to put one’s life’s history in context. •• Enabling emotionally satisfying contact with the people the elderly care about, people they share relations or a history with. At the same time, note that there are costs associated with such social contact, and those costs seem to rise with age (for example, as you lose your ability to drive, or when cognitive impairments affect your ability to remember names on the telephone).

From People to Prototypes and Products: Ethnographic Liquidity and the Intel Global Aging Experience Study | 33 Intel® Technology Journal | Volume 13, Issue 3, 2009

•• Enabling recognition and lightweight engagement with a broader community. As one of our participants told us, “Sometimes you just want to look out your window and see what’s happening in your neighborhood.” In many European communities people placed a lot of value on being known in a place, of interacting with people with whom they share some history or interests. This could come from such activities as going regularly to the local market, or participating in a club. The lived-in space is crucial to the experience of aging People’s choice of objects to inhabit their homes, as well as less malleable aspects of architecture, location, and other aspects of the physical environment were important in how people experienced aging and health. One interesting theme to emerge from our work was a deeper discussion of why nursing homes or institutional residences were generally regarded as bad, while staying at home is generally regarded as good. As we teased the attributes apart we realized that these elements characterize differences across many kinds of environments, including movement from one private residence to another. These elements “One interesting theme to emerge from suggest interesting possibilities for technology: our work was a deeper discussion of •• Houses reflect cultural norms around physical privacy. There is a reason why bathrooms are hidden in European and American homes. There are why nursing homes or institutional cultural notions of privacy and intimacy associated with care and nurturing of the body, particularly hygiene. We need to be cognizant of how different residences were generally regarded as cultural values around bodily intimacy may shape acceptance or resistance bad, while staying at home is generally to technologies. regarded as good.” •• Houses are the sites of spiritual and personal expression. Expressions of spirituality and the emplacement of memories (for example, in shrines to lost loved-ones) abounded in the houses we visited. •• Enacting of micro-routines. Simple daily activities, from checking the barometer, to drinking a glass of juice at bedtime (and using the empty glass as a medication reminder in the morning) to gardening, were crucial for people for multiple reasons simultaneously: as a way for people to fill little slots of time, as a way of grounding behaviors in some structure that aids memory, to provide some lightweight ways of finding meaning and having something to talk about, get a little bit of physical exertion, keep one’s mind sharp, and generally have some sense of small purpose to focus on each day. •• The home is not just the four walls. It is the physical situation of the house in the neighborhood, proximity to other services and opportunities for social engagement. One woman we visited in a nursing facility in Spain was quite happy there, partly because the home was in her old neighborhood, near the same market, the same services, and the same social network that she had always had. Healthcare networks are large and increasingly complex just about everywhere People in virtually every country we visited struggled to get the best value out of their healthcare systems. People struggled with healthcare bureaucracy and sought to bring alternative approaches to care into alignment:

34 | From People to Prototypes and Products: Ethnographic Liquidity and the Intel Global Aging Experience Study Intel® Technology Journal | Volume 13, Issue 3, 2009

•• Help people figure out their options. In places where coverage is universal, people still seek out alternatives. In the United Kingdom, despite universal coverage, people find it necessary to pursue alternative (private) healthcare systems to speed up access. In some cases, people had to leave their country to gain access to treatments that were unavailable because of resources or regulations. •• Help people help each other. Several people we visited had been forced to learn a lot about the healthcare system, because of a chronic condition or the need to care for someone who is ill. More than one noted that it was unfortunate that, when their situation changed (a patient died or got better) they had no place to share this hard-won knowledge. Simple tools might make a collaborative user-based system to help people share that knowledge and help each other. The identification of key themes such as these provides broad insight into the issues, problems, and desires of participants to help guide the development of effective, culturally sensitive, and age-appropriate technologies. We now outline some examples of how the global aging experience research program has stimulated more in-depth studies and has contributed to the cycle of product incubation and development, constantly ongoing within the Intel Digital Health Group. “Over the past decade, a somewhat different style of ethnographic research Ethnographic Liquidity has been developed in the context of Since the beginning of the twentieth century, the vast majority of ethnographic research has been conceived and consumed within an academic context. The industry and commerce.” structure of ethnographies as academic texts, and the tools and techniques used to produce them, are well fitted to an audience of scholars that, to a greater or lesser degree, sees ethnographic research as cumulative and contributing to anthropological and other related fields of inquiry. Over the past decade, a somewhat different style of ethnographic research has been developed in the context of industry and commerce. This work differs in scope and duration. It does not bear the same implied scholarly and cumulative relationship to academic ethnographic research.

Within the context of industry, ethnographic research has different demands placed on it and different gauges are used to assess the relative success or failure of an ethnographic project. A key demand placed on ethnographic research in this new context is that of immediate relevance. Equally, multiple stakeholders expect the research to be actionable. The term actionable refers “A key demand placed on ethnographic to the requirement that the research must be problem-directed and result in an analysis that produces results that are easily consumed, understood, and research in this new context is that of acted upon by other stakeholders in an enterprise (for example, research that immediate relevance.” produces engineering requirements leading to product specifications).

From People to Prototypes and Products: Ethnographic Liquidity and the Intel Global Aging Experience Study | 35 Intel® Technology Journal | Volume 13, Issue 3, 2009

For research to have an impact in any organization it needs to circulate so that the findings can be understood, conclusions discussed, and the implications fleshed out by relevant stakeholders. It could be argued that research that does not travel through an organization can only ever have limited impact, since it is on this journey that audiences can create for themselves a meaning for the research. However, the impact of ethnographic research in the context of large organizations, and beyond, can often be impaired by issues of circulation. Creating outputs that travel well is all important to developing what we call ethnographic liquidity, that is, the ability of research to be converted into something of value by its audiences.

Therefore, another way of thinking about the actionability of ethnographic research is through the lens of liquidity. If research activity and its output is “If research activity and its output is liquid, it is readily exchangeable. The research creates debate, is able to inform existing activities, and creates the basis for new endeavors. Liquid research will liquid, it is readily exchangeable.” perform different functions for different members of an organization, because it is multivalent and contains different layers of utility. For example, engineers and designers might focus more on design principles as a means of guiding their work. Marketing may take more note of needs and compelling ways of telling their business story.

In short, the best ethnographic research in an industrial context is research that has continued resonance and meaning. In essence, there are two competing demands placed on ethnographic liquidity. First, the research must be conceived and conducted to meet current organizational objectives. Second, the research must have prolonged and ongoing relevance for the organization.

When we commenced the design of the Global Aging Experience Study, the team was new to a recently created business division. The organization to some degree was still, to borrow the concepts of Tuckman [11], storming and forming. As a result, the research was conducted as much to shape the activities of a business division as it was to meet a pre-defined set of research needs. An explicit goal was to create a piece of liquid research that would endure and have long-term value.

The Global Aging Experience Study was designed accordingly. A foundation of the research was an extensive review of gerontological literature and its dominant theories of aging. This review lead to the creation of domains of experience that were used to create the interview guides and the coding “The research was conducted as much system later applied to our data. While this grounded theory research approach to shape the activities of a business is commonplace in the academy, it is less so in industry. (Researchers and management in industry generally perceive this approach as too academic and division as it was to meet a pre- too resource intensive. The members of the Digital Health Group, however, viewed this approach as a wise, long-term investment). The domains we created defined set of research needs.” remain relevant to the aging experience and have enabled other researchers to access and mine the original data we collected for other projects within the Digital Health Group, a practice that will continue for the foreseeable future.

36 | From People to Prototypes and Products: Ethnographic Liquidity and the Intel Global Aging Experience Study Intel® Technology Journal | Volume 13, Issue 3, 2009

Another aspect of the ethnographic liquidity of our research is the multiple “Booklets are used to draw together artifacts we have created to disseminate the research. These include video, PowerPoint* presentations, academic papers, printed internal and external large projects presenting ethnographic reports, press releases, conference talks, posters, and an opportunity map. Key profiles, country overviews, and to this strategy of dissemination has been the production of materials in an analog format. illumination of the analysis of the

We have focused on the production of high-quality, printed material in the material.” form of booklets. Booklets are used to draw together large projects presenting ethnographic profiles, country overviews, and illumination of the analysis of the material. The booklets aim to provide content that will become on-going resources that diverse readers find useful (for example, by including relevant statistics and demographic and market data). Our experience has been that focusing on materializing research in this way has allowed us to increase the liquidity of our work. Perhaps counter-intuitively, moving ethnographic material out of digital formats into paper formats can make it easier to circulate. Further, its existence in an organization becomes more durable. Materials in this format have been crucial to supporting the persistence of our research in the geographically dispersed, digital workplace. Furthermore, we have found that producing versions of our documents [12, 13] for external audiences further enhances the liquidity and durability of the original research.

Because of the ethnographic liquidity characterizing our work, the organization has made repeated use of it. The following represents a sample of activities to which the Global Aging Experience Study has contributed, either in part or in whole:

•• Deeply informed the initial conceptualization and subsequent development of two major product incubation efforts, one of which has developed into a key program within the Digital Health Group. •• Defined a further research agenda examining social health and social participation in more detail within the context of the European Union and the United States. •• Resulted in the submission of 60 or more invention disclosures for the Digital Health Group. •• Informed major research efforts in the Technology Research for Independent Living (TRIL) research partnership between Intel and the Irish government within the areas of social connection, cognitive decline, and falls. One specific example is how older people negotiate movement through their homes and develop pathways in the household. This insight informed the development of a sensor strategy. •• Developed an operational blueprint for ethnographic research of this scale and complexity and identified best practices for knowledge transfer. •• Launched deeper research efforts in the areas of sleep and sociality, transportation and mobility, social care, active retirement groups, and the Figure 6: Front Covers of Two External-facing reinvention of retirement. Documents Source: Intel Corporation, 2009

From People to Prototypes and Products: Ethnographic Liquidity and the Intel Global Aging Experience Study | 37 Intel® Technology Journal | Volume 13, Issue 3, 2009

Conclusion In this article, we have discussed the Global Aging Experience Study in the context of ethnographic liquidity and knowledge transfer. The project afforded our team an opportunity to develop tools and artifacts that have led to the continued relevance of our research to the Digital Health Group. One can argue that conducting fieldwork and analyzing data represents the comparatively easy parts of any research project. The real challenge when conducting ethnographic research in the context of industry, comes in the practice of sustained knowledge transfer and in the ability to drive the insights gained and communicated through research into a program for development.

References [1] Beck, Barbara. “A Slow Burning fuse: A Special Report on Aging Populations.” The Economist, June 27th, 2009, page 3.

Ibid.[2]

[3] “Aging population: Facts behind the fiction.” At news.bbc.co.uk

[4] World Health Organization. At www.who.int

[5] Esping-Anderson, Gøsta. 1990. The Three Worlds of Welfare Capitalism. Princeton University Press.

[6] Ferrera, Maurizio, ed. 2005. Welfare State Reform in Southern Europe: Fighting Poverty and Social Exclusion in Greece, Italy, Spain and Portugal. Routledge: New York.

[7] Goodin, Robert E., Headey, Bruce, and Muffels, Ruud. 1999. The Real Worlds of Welfare Capitalism. Cambridge University Press: Cambridge.

[8] Scheper-Hughes, Nancy and Lock, Margaret. “The Mindful Body: A Prolegomenon to Future Work in Medical Anthropology.” Medical Anthropology Quarterly, Volume 12, Issue 3, 1987.

[9] Timonen, Virpi, Doyle, Martha and Prendergast, David. No Place Like Home: Domiciliary Care Services for Older People in Ireland. Liffey Press: Dublin, Ireland, 2006.

[10] Althusser, Louis. 1969. For Marx. Translated by Ben Brewster. Verso: London.

[11] Tuckman, Bruce. 1965. “Developmental sequence in small groups.” Psychological Bulletin 63 (6): 384-99.

[12] Roberts, Simon and Prendergast, David. 2009. “Community Supports for Aging: Care Pathways for Older People in Europe.” Digital Health Group, Intel Corporation.

[13] Plowman, Tim, Roberts, Simon and Prendergast, David. 2007. “The Global Aging Experience Project: Ethnographic Research.” Digital Health Group, Intel Corporation.

38 | From People to Prototypes and Products: Ethnographic Liquidity and the Intel Global Aging Experience Study Intel® Technology Journal | Volume 13, Issue 3, 2009

Author Biographies Tim Plowman, PhD, is a Senior Research Scientist and a Medical Anthropologist for Product Development in the Digital Health Group at Intel. He has been leading innovation from a social science perspective in Silicon Valley for over a decade. His current work focuses on aging and possible instrumental and non-instrumental applications of information and communication technologies in support of successful aging. He holds multiple patents and has published widely on social science, technology, and design. His e-mail is tim.plowman at intel.com.

David Prendergast, PhD, is a Social Anthropologist in the Digital Health Group at Intel and a Principal Investigator of the Social Connection Strand of the Technology Research for Independent Living (TRIL) Centre. His research focuses on later-life course transitions, and he has authored a number of books and articles on aging and health. David utilises ethnographic research to co-design and iteratively develop culturally appropriate, independent-living technologies for older people. His e-mail is david.k.prendergast at intel.com.

Simon Roberts, PhD, leads the European social science and design team for Product Research and Incubation in Europe and is the Intel Principal Investigator on the ethnographic research strand of the TRIL Centre. He is an Applied Anthropologist with over ten years of experience translating research into product, service, and business opportunities. Simon has published widely on ethnography, the application of anthropology, technology, design, and the aging experience. His e-mail is simon.w.roberts at intel.com.

From People to Prototypes and Products: Ethnographic Liquidity and the Intel Global Aging Experience Study | 39 Intel® Technology Journal | Volume 13, Issue 3, 2009

Adapting Technology for Personal Healthcare

Contributors

Terrance (Terry) Dishongh Abstract Intel Corporation The increase in the aging population in the United States, coupled with healthcare costs that consume nearly 20 percent of US Gross Domestic William (Bill) DeLeeuw Product, are straining the ability of the government to deliver quality care. Intel Corporation One way to ease this strain on our national budget would be to use technology to enable the aging population to age at home. Technology could be used to Mark Francis support the independence of the elder through monitoring the activities of Intel Corporation daily living (ADL), tracking cognitive health, proactively managing chronic health conditions, alerting to falls, and providing a system to promote social engagement. This alerting, monitoring, and tracking can be done via home- based technologies that provide information to formal and informal caregivers. Effectively, with the use of technology, the care model shifts from a large centralized, provider-centric one to a home- and community-based model of distributed care.

Index Words Introduction In this article, we review a variety of tried and tested technical solutions to Sensors address the care and needs of the elderly in the United States. Our solutions are Surveys broadly characterized into three types: surveying, sensing, and social care. Social Care Personal healthcare •• The surveying solution involves periodic interactive dialogues with the Long-term Care senior to assess functional, physical, and mental health while also providing a communication platform for health education. Queries may be as simple as a repeated list of standard questions or as complex as a system of personalized, condition-specific questions with branching logic. •• The sensing solution is achieved by using active or passive sensing technology to collect information about the person’s behaviors, activities, and functional status. •• The social care solution involves encouraging seniors and providers to actively participate in social communities, nurture relationships, and to break down social isolation as individuals age-in-place alone. We show how these three technical solutions were used by seniors and providers to enable personal healthcare, enhance care, and improve quality of life.

40 | Adapting Technology for Personal Healthcare Intel® Technology Journal | Volume 13, Issue 3, 2009

A Stressed Healthcare System In the past, the majority of people in the United States did not live beyond 50 “The existence of hundreds of millions years of age. In 1900, for example, life expectancy was 47 years of age. In 2010, life expectancy in the United States will exceed 80 years of age [1]. As a result, of individuals aged 65 and older is a there are now more than 39 million Americans over the age of 65. Globally, the 21st-century phenomenon.” current population of individuals aged 65 years and older exceeds 500 million. By 2030, it will exceed 1.5 billion. The existence of hundreds of millions of individuals aged 65 and older is a 21st-century phenomenon [2].

Historically, when the need for long-term care occurred, this need was met by family members. It was only in the last century that long-term care emerged as an industry. The 1940s were marked by a surge in the construction of traditional nursing homes and the institutionalization of care. The 1960s saw the establishment of Medicare (the government-run, healthcare system for the elderly), Medicaid (a government-funded, healthcare system for certain low-income individuals and families), and the establishment of the Older Americans Act that launched a series of services designed to support aging-in- place and home-based care. Over the past twenty years, we have witnessed the shift toward home- and community-based care, as seen in the rapid growth of the senior housing industry and the creation of numerous, innovative models of aging-in-place services.

Today, long-term care is prevalent and the costs are staggering: “These factors create a tremendous •• Over 40 percent of US households are involved in providing elder care [3]. opportunity for technology to play •• The average out-of-pocket cost to family caregivers exceeds $5,500 annually [3]. a role in enabling personal care, •• The cost to employers due to lost productivity and absenteeism exceeds enhancing healthcare, improving $1,142 per year per employee [4]. quality of care, and relieving some of •• The average annual cost of nursing homes exceeds $72,000 [5]. •• The average annual cost of assisted living services exceeds $30,000 [6]. the stress on the healthcare system.” As baby boomers (a term used to describe those born during the post WWII baby boom) move into their 70s, the demand for long-term care services will rise dramatically. If nothing is done, this additional demand could break the long-term care system. However, these factors create a tremendous opportunity for technology to play a role in enabling personal care, enhancing healthcare, improving quality of care, and relieving some of the stress on the healthcare system.

Adapting Technology for Personal Healthcare | 41 Intel® Technology Journal | Volume 13, Issue 3, 2009

Users of a Technology Enabled Personal Care System: Seniors, Caregivers, and Family

Independent Assisted Skilled Nursing Living Services Living Services Services Demographic Active Boomer adults in the workforce; sandwich generation; Demographic new healthcare Beginning of functional consumers decline; living w/chronic Average Length disease – Demographic 2+ ADL issues; of Service: 20 years “uh oh moments” Alzheimer’s; Faller Demographic Services Average Length Seriously ill, end • Cognitive fitness of Service: 4–6 years Average Length of life; need for • Informational portals of Service: 2–3 years constant supervision Services Independence • Healthy aging and • Cognitive fitness Services Average Length maintaining youth • Medication • Support of 2+ ADLs of Service: • Discounts management • Cognitive Health 0.5–2 years • Lite IALD and • Active Falls ADL support • Wandering Services • Chronic Disease • Post-acute • Active medical care Management lite: episode care • 24/7 monitoring Healthy Heart • Functional foods/meals • Active ADL support • Meals • CDM • Meds management • PERS/Safety • PERS/Safety • Chronic Care

55+ 75+ 80+ 85+ Average Age

Figure 1: Transitioning through the Aging Services Continuum Source: Intel Corporation, 2009

“The senior, caregivers, and family As individuals transition though the second half of life, they traditionally proceed through stages of increasing functional, physical, and cognitive decline form a triad to address emerging and over time, as shown in Figure 1. For some individuals, this is a slow, gradual evolving needs.” process. Other individuals maintain healthy and active lifestyles until the very end of life. In either case, throughout this process, the senior, caregivers, and family form a triad to address emerging and evolving needs.

At-home technology can complement the work of caregivers such as physicians, nurses, wellness coaches, care coordinators, family members, neighbors, and friends. Technology provides new and innovative ways to monitor the wellbeing of older people, increase the levels of communication between all concerned, and enable response to rapidly changing conditions. The new awareness that long-term care includes clinical and non-clinical caregivers as well as local and distant family and friends is playing a large role in transforming long-term care services.

Home-care systems allow people to monitor themselves with devices that give proactive warnings of illness, thereby enabling patients to turn to their “For caregivers, home-care systems caregivers earlier, when intervention can be the most effective. For caregivers, mean more efficient and effective home-care systems mean more efficient and effective healthcare, driven by seniors who take greater responsibility for their own health. William healthcare, driven by seniors who take Herman, director of the Division of Physical Sciences in the Food and Drug greater responsibility for their own Administration’s Center for Devices and Radiological Health (CDRH), calls home-care systems “the fastest growing segment of the medical device health.” industry” [7].

42 | Adapting Technology for Personal Healthcare Intel® Technology Journal | Volume 13, Issue 3, 2009

The Home Environment In developing technologies to support aging-in-place and to enhance long-term care, the following key principles should be kept in mind:

•• The focus is healthcare; often a new technology is not what is needed; “Two of the biggest challenges to the instead, the application of an existing technology might be appropriate. •• At the center of this healthcare issue are the senior and the caregiver, not utilization of technology in a home are the technology. Without cooperation from the users of the system, the usability and installation.” system may never be useful. •• There is often more than one way to achieve a clinical or care objective: the first technology solution that appears may not be the best. •• The simpler the technology, the better. •• Technology solutions for healthcare are mission-critical; reliability is of paramount importance. •• The technology has to work in the home, not just in the lab. •• Usability by the senior and the caregiver is paramount. Two of the biggest challenges to the utilization of technology in a home are usability and installation. The usability issue is critically important in that compliance often hinges on usability. For what we call our survey and social care solutions to work, issues regarding user interfaces, choice of display colors and font sizes, method of interaction, device type, device size, and portability have a significant impact. For a solution to reside in an individual’s home, we must understand where a system will be used and the available space in the home for that system.

The installation issue is particularly challenging. Often a new technology requires multiple devices, batteries, cables, cable management, and wireless data connections. Integrating all of these components and installing them is often more challenging than anticipated.

Moreover, variability in home construction material, room size, room layout, “Drops, strikes, water damage, and and environmental factors (lightning and air flow) can impact the technology deployed and the reliability of the results. For our sensor solutions, ongoing tampering, among other things, are maintenance requirements, such as changing batteries, repositioning sensors, or updating the type of sensor, may also create obstacles to adoption. very likely to occur in a home and

Ironically, while technology may feel threatening to those in the home, the must be anticipated when technology home environment is actually a fairly inhospitable place for technology. is being designed.” Drops, strikes, water damage, and tampering, among other things, are very likely to occur in a home and must be anticipated when technology is being designed. Children, in many cases more technology savvy than their parents or grandparents, are an active threat and may drive new security requirements that require biometric authentication and tamper-detection technologies.

Adapting Technology for Personal Healthcare | 43 Intel® Technology Journal | Volume 13, Issue 3, 2009

“A technology will benefit from a Finally, another goal of in-home monitoring technology should be that it be hidden in plain sight. Being readily available and convenient are critical design that enables the parts of the elements for fostering regular use, but many people with chronic disease have technology to function, potentially all no desire to share that part of their life with others. Therefore, a technology will benefit from a design that enables the parts of the technology to function, over a home, but not draw unwanted potentially all over a home, but not draw unwanted attention to the resident’s attention to the resident’s health.” health. The use of Personal Emergency Response System (PERS) devices is a classic example of a market reaching a premature plateau because of the stigma attached to wearing a PERS pendant. PERS market penetration has flattened to a 16 percent adoption rate due to this usage issue and 40 percent of PERS users do not wear their pendants.

We now explore in more detail the three technology solutions that we evaluated and tested: surveying, sensing, and social care.

Surveying “Wellness surveys are one way to It is now known and accepted that proactively managing health can lead to healthier choices, improved accountability, increased quality of life, and approach behavior assessment, enhanced health. In an era where chronic disease is the leading cause of hospitalization and the major driver of healthcare expenditures, engaging encourage change, and achieve better individuals to modify behavior and make better lifestyle choices can have a outcomes.” profound impact on health, wellness, utilization of services, and system costs. Wellness surveys are one way to approach behavior assessment, encourage change, and achieve better outcomes.

Wellness Coach and Surveying The increased awareness of the importance of proactively engaging individuals in managing their own health has led to a proliferation of health coaching services. From the Mayo Clinic Embodied Health Services to numerous mobile health applications, these services run the gamut from clinical to consumer- focused programs.

Through the efforts in the Product Research and Incubation Group, we have conducted numerous research initiatives around the concepts of wellness coaching and surveying. We also employed non-clinical wellness coaches to use technology to improve the cognitive and social health of seniors, as well as to use surveys to explore how seniors could stay at home and age with as “Our surveying research involved much dignity, independence and functional, cognitive, and physical health as periodic interactive testing or queries possible. of the elder in their homes to estimate Our surveying research involved periodic interactive testing or queries of the elder in their homes to estimate physical, functional, and mental health. physical, functional, and mental This interactive testing involved using either dedicated testing hardware or measurable extensions of existing household items or activities. health.”

44 | Adapting Technology for Personal Healthcare Intel® Technology Journal | Volume 13, Issue 3, 2009

Our surveying methodology varied. In some cases, we used queries ranging “Sensors strategically placed on the from as simple as a repeated list of questions. In other tests, we used queries as complex as a system of branching logic. The ability to have flexibility in human body create a wireless body the surveying methodology allows for the healthcare or eldercare provider area network (BAN).” who monitors survey answers to customize future queries or test sequences to specific patient situations.

Sensing Any at-home healthcare solution must detect and respond to the activities and characteristics of the older person. A network of sensors (either worn, carried, or environmental) is a good platform for detecting and responding to health-relevant parameters, such as movement, breathing, pulse, heart rate, or social activity. Sensors strategically placed on the human body create a wireless body area network (BAN) that can monitor various vital signs while providing real-time feedback to the user and medical personnel. Further, wireless sensors can be deployed in a patient’s home environment to provide real-time and extended monitoring of activity and wellbeing.

When coupled with communications technologies, such as mobile phones and “For example, changes in gait might the Internet, the wireless sensor network (WSN) can keep family, caregivers, and doctors informed, while also establishing trends and detecting variations indicate the onset of Parkinson’s disease in health. Appropriate technology design can minimize intrusion and protect in advance of visible symptoms.” privacy, maximize user friendliness, and encourage long-term adherence to medical regimens.

In some cases the wireless sensor network requires no direct patient interaction (this is ideal where the patient may be suffering from some level of cognitive decline). Because WSNs are deployed where the patient spends most time, they can deliver long-term data sets to assist in diagnostics and patient response to interventions. The data collected by a WSN can be stored and integrated into a comprehensive health record of each patient that may allow identification of subtle changes in a person’s health. For example, changes in gait might indicate the onset of Parkinson’s disease in advance of visible symptoms [8]. Other example applications include medicine reminders, social connectivity, and emergency communications.

BANs can detect small changes in vital signals — such as heart rate and blood- “Measurements can be made oxygen levels — that are not obvious in a one-off visit to a doctor. Further, unobtrusively over an extended period BANs address the issues associated with so-called “white coat syndrome,” skewed physiological and cognitive test results that are related to the stress in a supportive home environment and anxiety associated with a clinical visit. Measurements can be made unobtrusively over an extended period in a supportive home environment that that more accurately reflects the true more accurately reflects the true values for a given parameter. values for a given parameter.”

Adapting Technology for Personal Healthcare | 45 Intel® Technology Journal | Volume 13, Issue 3, 2009

“WSNs can provide notice of As well as offering excellent long-term care benefits, the always-on nature of WSNs means that they can be used to detect and respond to health crises in a significant shifts in key physiological timely manner. In particular, WSNs can provide notice of significant shifts in parameters and may save lives, or key physiological parameters and may save lives, or initiate preventative care in order to forestall a health crisis. initiate preventative care in order to The potential value of WSNs for healthcare has not yet been fully explored. A forestall a health crisis.” significant research effort will be required to mine the vast data sets collected by WSNs and to find patterns that reveal information of interest to the healthcare professional. Breakthroughs in this area may support semi-automated analysis, diagnosis, and treatment processes in the future. Doctors will be assisted by an electronic counterpart, and patients’ health may benefit as a result of faster diagnosis and treatment of diseases. Other quality-of-life issues, such as privacy, dignity, and convenience, are supported and enhanced by the ability to unobtrusively provide services in the patient’s own home [9].

Overview of our Methodology In this section we present an overview of our general methodology for a WSN solution. Specifically, here are criteria for developing and deploying a WSN in healthcare [10]. •• Understand the problem. A clinician may already have analyzed the problem, and have come up with a technology outline for the engineer to develop. However, further analysis may pay dividends. At-home healthcare means that the appropriate environment to formulate any technology solution is not the clinic or the laboratory, but the home. Ethnographic observation of user behavior, where an understanding of how people live and why they live as they do, their constraints and their priorities day to day, can be very beneficial and has informed many of the solutions in which we are involved. •• Understand the end user. Who will use the solution in the longer term? What are their constraints or priorities? How do they feel about the type of solution envisaged? Usage modeling, where multi-disciplinary teams use personas to explore the user experience from several perspectives, can be useful here. •• Understand what data must be collected. In order to test a clinical hypothesis or to achieve a care objective, some information about the activity or health state of the patient must be collected. It is important that a clear understanding of what data will be collected by the sensor network is shared by the clinician and the engineer, from the start. The information that the clinician needs is important, not the data that the sensor has the technological capability to collect. Focus on what is needed, rather than on what the technology allows. “Breakthroughs in this area may •• Understand the environment. Many WSN deployments do not achieve their support semi-automated analysis, objectives because they fail to take into account the differences between the laboratory and clinical environment, and the home environment. Building diagnosis, and treatment processes in methods and materials vary from location to location; these may impact the future.” the technology. Other equipment in the home may also interfere with the sensor technology.

46 | Adapting Technology for Personal Healthcare Intel® Technology Journal | Volume 13, Issue 3, 2009

•• Consider sensor location. Are sensors to be worn, to be held, to be embedded in walls, floors, beds, or furniture? Keep in mind that the sensor network must be as unobtrusive as possible — solutions that require the patient to change his or her day-to-day behavior or which affect the comfort, privacy, or dignity of the elder person are unlikely to be successful on a long-term basis. Almost all at-home solutions aim to be long-term. •• Select sensors and actuators. Take into account the data, the environment, and the placement. What impacts do these have on power consumption, size and weight, form factor, radio communications, antenna type and computational capability? Where at all possible, use off-the shelf components, and avoid experimental or prototype technology. Remember this is a healthcare problem, not a technology research project. Aim for a low-cost, light-touch solution. •• Specify and build the aggregator. The aggregator receives data from all the sensors in the network, and may transmit them to an analysis engine or data visualization system. Alternatively, the aggregator may itself process the data and trigger responses by actuators or by communications with caregivers or doctors. In general, actuators take the form of local PCs or mobile devices, such as PDAs and smart phones. We do not cover aggregator design in any detail in this article, because one typically uses mass-market devices such as phones and PCs. •• Identify and deploy analysis and visualization capability. Ensure that the technology is in place to convert the data from the WSN to clinical information, and to allow the clinician to access this information in an appropriate manner. This may be out of the scope of a purely sensor- network project, but it is important that in such a case the precise nature of the interface between the network and the back-end data analysis and clinician system is clearly defined. The technology used for visualization is not explored in this article; analysis and visualization techniques and technologies are quite standard across many application domains. •• Build the solution in the laboratory and verify that all elements work as planned. Discuss the solution with clinicians and end users. •• Prototype and pilot test the solution in a controlled, friendly environment. The solution should be used in the home of a study participant who has agreed to act as a tester. In these friendly trials, the importance of proper safety testing and precautions cannot be understated. Before deployment to the home even for prototyping purposes, any Federal Communication Commission licenses or local radio licenses, UL or CE requirements, human subjects approvals, Institutional Review Boards (IRBs, signed agreements, etc.) must be obtained. •• Deploy the solution in the real world. Continue to engage with end-users to detect and correct any issues that may lead to failure to use the solution as planned.

Adapting Technology for Personal Healthcare | 47 Intel® Technology Journal | Volume 13, Issue 3, 2009

“Medication compliance is a vexing Overall, we have learned that deploying an effective sensor solution is a complex endeavor. As such, proactive planning, clear objectives, and issue at the core of many challenges in forethought of the end-stage solution are critical to a successful design and healthcare.” deployment of a WSN solution. Context Aware Medication Compliance (CAMP) Trial Medication compliance is a vexing issue at the core of many challenges in healthcare. In a recent report in the New England Journal of Medicine, Osterber and Blaschke [11] report that typical adherence to a medication regime ranged from 43 percent to 78 percent, depending on the age of the population researched and the type of medication taken. In other words, 50 percent of patients are typically failing to comply with their medication regimen.

Our team engaged in a context aware medication compliance (CAMP) trial to develop a sensor network system that is contextually aware. We wanted to test adherence to a medication regime over several months, in the hopes of improving the rate of adherence over time.

“Our system was designed to track Typical products on the market for medication adherence fall into two categories. The first is the product that acts like a timer or alarm clock. These the location of users, monitor the devices issue alarms when the time set for people to be reminded to take sleep activity of users, monitor the their medication is reached. Although these devices are highly accurate in measuring time and delivering an alarm, they often were not co-located with adherence to medication regimens, the medication at the time of alarm. Moreover, these alarms prompted so often that they became a nuisance and were often overlooked. Other devices are store and drive prompts to various devices and dispense medications. These are common in many households and take throughout the home.” the form of a system with cells in multiples of seven for the days of the week. Medication is stored in these devices, and the device is often located near the activity center associated with the time of day the medication should be taken. These systems fail when the user’s routine is disturbed.

Our system was designed to track the location of users, monitor the sleep activity of users, monitor the adherence to medication regimens, and drive prompts to various devices throughout the home, based on the appropriate time, location, and activity that would ensure the best adherence.

Planning Sensor Deployment A key issue for planning is the layout of the deployment environment (the subject’s home) and the identification of ideal sensor locations. Ideal locations reflect the range of the radios being used (Zigbee*, Bluetooth*), the size of the rooms, and the sensitivity of motion sensors. Figure 2 shows an example floor plan with sensor locations. The sensor locations must be verified and validated after the installation, to ensure that data are being captured and communicated as planned.

48 | Adapting Technology for Personal Healthcare Intel® Technology Journal | Volume 13, Issue 3, 2009

Base station

Livingroom

Bed sensor

Imed Pillbox Bedroom Closet

Kitchen Watch charger

DSFridge Bathroom

Closet

Door sensor Motion sensor Ceiling motion sensor Activity beacon Phonejack Cable connection

Figure 2: Floor Plan and Sensors Placement Source: Intel Corporation, 2009

The choice of radio technologies can have a profound impact on the success of “The choice of radio technologies can communications within a project. Similar devices from different suppliers may perform in dramatically different ways. have a profound impact on the success

Moreover, we learned through deployed trials in US homes throughout the of communications within a project.” Southwestern United States and the Pacific Northwest that construction methods and materials profoundly affect wireless technology. Homes with significant wood and lumber construction create a very different environment than do homes constructed of stucco, steel frames, and other materials.

Surprising at the time, although obvious now, we learned that variation in the materials and methods of home construction had a significant impact on sensor performance. As such, understanding the materials used in home construction has a direct impact on both the type of sensors selected and the manner in which a WSN will be designed and deployed.

Participant Tests In the course of the CAMP project, we deployed a pillbox with sensor- “Each time a door in the pillbox equipped doors. Each time a door in the pillbox was opened, a signal was sent was opened, a signal was sent to the to the aggregator, indicating that a pill had been taken. However, one subject left the door open, as a reminder to himself to take the second pill; this was not aggregator, indicating that a pill had as envisaged by the team, and it led to unexpected inferences by the system. been taken.”

Adapting Technology for Personal Healthcare | 49 Intel® Technology Journal | Volume 13, Issue 3, 2009

Fluorescent Lamps and Infrared Signals “Frequent at-home monitoring of Common fluorescent lighting fixtures driven by electronic ballasts emit low-frequency infrared signals modulated in the kilohertz range. This signal disease state tends to be far superior interferes to a very significant degree with infrared communications between to infrequent but highly detailed in- devices and aggregators, or among devices. Interference from fluorescent lighting needs to be taken into account when choosing the technologies to clinic assessments.” deploy, and also when positioning devices in the home. It should be borne in mind that some applications of fluorescent lights mean that they are used only occasionally, such as for lighting on kitchen counters.

Parkinson’s Disease At Home Test One option where technology in the home may improve upon the modern hospital and clinic framework is in the area of interactive monitoring of chronic or degenerative diseases. An ecosystem of technologies has grown around a few individual diseases, such as diabetes, and frequent at-home monitoring of disease state tends to be far superior to infrequent but highly detailed in-clinic assessments.

“Designers and engineers at Intel In 2004, we were approached by a not-for-profit Parkinson’s Disease research support foundation interested in technologies to better monitor the symptoms created an At Home Test (AHT) for of Parkinson’s disease. This interest was fostered by a series of meetings between Parkinson’s disease.” the foundation and prominent neurologists from seven different institutions. Each of the neurologists present had recent experiences with various techniques to quantitatively measure the symptoms of Parkinson’s disease. Together, the neurologists expressed interest in a single apparatus, designed around multiple measurement techniques, and capable of daily or weekly evaluations of the symptoms of Parkinson’s disease outside of a clinic or lab [12].

Historically, Parkinson’s disease testing has consisted of twice or three times yearly visits to a neurologist, where somewhat subjective estimates were made of the patient’s condition by using the Unified Parkinson’s Disease Rating Scale (UPDRS). While the UPDRS has been a gold standard for years, the neurology team and the foundation felt its typical application lacked frequency and did not produce truly quantitative measurements.

Intel, the Parkinson’s Disease foundation, and the team of neurologists formed a set of requirements and a study protocol to (1) establish evidence that in-home interactive sensing of Parkinson’s disease is viable and would be used by patients on a regular basis, (2) assess the effectiveness of the various quantitative measurement techniques brought forward by the team of neurologists, and (3) establish if those measurement techniques could show variation in measured symptoms from week to week. Based on this set of requirements and the new protocol, designers and engineers at Intel created an At Home Test (AHT) for Parkinson’s disease.

Figure 3: The Parkinson’s Disease At Home Test Apparatus Source: Intel Corporation, 2009

50 | Adapting Technology for Personal Healthcare Intel® Technology Journal | Volume 13, Issue 3, 2009

The AHT was designed to measure the following: “The study consisted of 52 subjects, •• Large limb motion all in the early stages of Parkinson’s •• Fine motor skills disease, who each took an AHT •• Fatigue apparatus home for six months of •• Ability to maintain rhythm •• Tremor frequency and duration once-weekly testing.” •• Reaction time •• Free speech (voice inflection) •• Continuous vocal tone Metrics designed around gait measurement or other ambulatory situations were rejected from final design, due to the potential for possible subject injury when subjects were in an unsupervised environment.

The study consisted of 52 subjects, all in the early stages of Parkinson’s disease, who each took an AHT apparatus home for six months of once-weekly testing. Prior to the six months of weekly tests, the subjects each took a test once every day for a week to ensure they were well practiced with the movements and to ensure that future study results would not reflect a learning curve. After the practice sessions, the subjects were evaluated by their clinician to see if they could actually follow the study directions. All but one subject was fully enrolled after practice, with the one missing subject disqualified due to other health reasons. In general, subjects proved capable of using the AHT apparatus.

Due to the relatively late onset of Parkinson’s disease symptoms (average age at diagnosis is 59) for most subjects, certain design considerations were taken into account:

•• Failing eyesight. •• Reduced range of hearing. •• Discomfort with complex technology. •• Presence of grandchildren who might tamper with or damage the equipment. Furthermore, because a movement disorder was known to afflict the study subjects, we employed large buttons in a widely spaced configuration for all navigation and selection tasks. We did not use any mouse or other pointing device. Instead, subjects used just up, down, and select interaction buttons.

The likelihood of participants having failing eyesight was mitigated by integrating a super bright and relatively large liquid crystal display (LCD) into the Parkinson’s disease AHT apparatus. The LCD had large fonts and provided a lot of empty space within screens. Furthermore, we provided all instructions “We employed large buttons in a and cues in both text and through a voice reading of the text. Reduced hearing widely spaced configuration for all ability was accommodated by using a large speaker and selecting an instruction narrator with a lower-toned voice. navigation and selection tasks.”

Adapting Technology for Personal Healthcare | 51 Intel® Technology Journal | Volume 13, Issue 3, 2009

“We developed significant data Finally, we made the design decision not to use a standard computer, where booting and peripheral connection could interfere with regular use of the security to ensure the authenticity of AHT apparatus. Instead, we prototyped a series of small netbook-like devices the information could be established with integrated tests. The form-factor we selected in the end was rugged, small enough for regular use on a counter or kitchen table, easily transported, and and that the health information nondescript so as to encourage subjects to leave it out and plugged in. within the test data could be protected The initial intent was to have the Parkinson’s disease AHT apparatus connect to from unauthorized viewing.” a home network. After extensive review of the availability of high-speed Internet, however, together with concerns about integrating the AHT apparatus with existing home networks, we made the design decision to simply store test data for later delivery to a central storage facility either physically or through a computer Internet connection. We later decided to include a very basic and inexpensive laptop computer equipped with a mobile broadband card to facilitate more regular uploads of study data.

We anticipated that the AHT apparatus might incur system failures in the field, so we also made the design decision to store all application software and study data on an external storage device. This decision allowed us to replace a defective system in the field without having to modify it for use by a specific At-Home Testing Program study participant or extract test data from a damaged unit. New systems could simply be sent to replace defective or damaged ones, while the damaged 3:49 PM Take Tests systems could simply be discarded. We decided on a USB storage device as the Friday, May 30 medium and we developed significant data security to ensure the authenticity • Instructions of the information could be established and that the health information within the test data could be protected from unauthorized viewing.

Next test session Move Up Because the systems were going to be deployed over much of the United States, Saturday 5/31 Select 4:00pm – 8:00pm Move Down we designed the software system to be easily upgraded. Upgrades were provided either through issuing new USB storage devices or by updating the software on the storage devices during the data upload process. We also designed packaging Reaction Test to protect the apparatus during air or ground freight.

The final AHT apparatus design was very well received [12] and even won a “BEEP” major design award in 2006. All but one study participant found the device to be easy to work with.

Although careful consideration and a great deal of organizational effort went into building the AHT apparatus, we learned many lessons the hard way. As an initial foray into the healthcare field, these lessons were as valuable, if not more so, than anything the actual study might tell us.

First and foremost, we learned the value of very careful industrial design and its impact on end-user acceptance. The system was beautiful and people wanted to use it. Something less refined would have begged to be put in a drawer.

Figure 4: Parkinson’s Disease At Home Test Screen Shot Source: Intel Corporation, 2009

52 | Adapting Technology for Personal Healthcare Intel® Technology Journal | Volume 13, Issue 3, 2009

On the study side of things, we found that training was critical, especially for “What seemed excessive at the time, the clinicians who would oversee the study subjects. They were the first line of support for subjects and, in many cases, solved problems without our input flying all of the clinicians to our offices (until the end of the study). What seemed excessive at the time, flying all of the at Intel for a day of intense training, clinicians to our offices at Intel for a day of intense training, turned out to be money and time very well spent. turned out to be money and time very

Finally, we found that the sheer volume of data collected by the AHT well spent.” apparatus to be intimidating, even to seasoned statisticians. There were so many dimensions for analysis that forming a coherent plan to correlate measurements with UPDRS scores was difficult.

In the end, we found that individuals with Parkinson’s disease would likely use this type of interactive technology on a regular basis in their homes if it was widely available. We also found that some quantitative measures showed a lot of promise in predicting disease state [12], but the initial study was not sized to prove any such assertion. The data, arguably one of the largest collections of Parkinson’s disease measurements, have proven useful for additional investigations into the progression of this disease.

We now move on to a discussion of the third aspect of our technology healthcare solutions, what we call social care.

Social Care Studies have shown that social isolation can lead to depression and loneliness “There have been many other articles and can accelerate the onset and manifestation of illnesses of aging. In May 2009, the New York Times ran a front-page story on the increased use of social and reports discussing the link between media by senior citizens — and the direct and indirect health benefits of such social engagement. There have been many other articles and reports discussing an individual’s health and the size the link between an individual’s health and the size and activity of their social and activity of their social network.” network [13].

In April 2009, The California Healthcare foundation released a report titled: The Wisdom of Patients: Health Care Meets Online Social Networking [14]. This report illustrated the strong growth of web 2.0 offerings focused on providing information and community to seniors and aging baby boomers. The report indicated that the active participation in social networks would be the hallmarks of a consumer-centric approach to healthcare. The access to health information, quality data, and provider rankings, coupled with the establishment of forums, sites, and support groups to rapidly exchange information, are fundamental shifts in how healthcare is provided.

Adapting Technology for Personal Healthcare | 53 Intel® Technology Journal | Volume 13, Issue 3, 2009

Social Health Trials “We hypothesized that we could One of Intel’s earliest research projects focused on examining social networks and social health. We hypothesized that we could detect and track the social detect and track the social activity of activity of a patient with Alzheimer’s disease, while also providing feedback a patient with Alzheimer’s disease, on that activity to the patient and a family caregiver. The desired result was to keep the patient as socially engaged as possible. To that end, we queried while also providing feedback on that each participant to define a social network of approximately ten friends and activity to the patient and a family family that were to be tracked. We built a sensor network in the home of each participant that was designed to detect and track physical visits and phone calls caregiver.” by anyone in the predefined social network.

We also built two interfaces that the patient and family caregiver could interact with: one was a caller ID application that rendered a picture and provided contextual notes regarding the last conversation between the patient and friend or family member, as well as a social networking galaxy model. The galaxy model provided a picture of the patient in the center of the graphical universe with pictures of the social network participants orbiting around the patient. As more or less social activity was detected by the sensor network, the orbit of each participant changed and became closer or further away from the patient, respectively [15]. This provided a simple model for the patient to use to reflect on their level of social activity with any given participant in their network. For example, if a patient’s daughter hadn’t called in a few weeks, the patient could see in the galaxy the orbit becoming more distant. Conversely, the orbits of those with whom the patient had more frequent contact would become closer.

“For example, if a patient’s daughter Participants found the interfaces compelling, keeping them more mindful of changes in their social activity than they otherwise would have been [15]. hadn’t called in a few weeks, the New models of interaction, particularly those focused on what one might call interactive journaling of patterns and daily activities are promising solutions for patient could see in the galaxy the keeping home-bound seniors and patients engaged and connected in their daily orbit becoming more distant.” lives.

54 | Adapting Technology for Personal Healthcare Intel® Technology Journal | Volume 13, Issue 3, 2009

Conclusions In this article, we examine the impact of the transformation occurring across many parts of the world with regard to healthcare. Increased longevity and the aging of the baby boomer population are creating a sea change in the demographic composition of populations across the globe. Correspondingly, societies are shifting from youth-orientated cultures to age-aware cultures. Healthcare systems are shifting from treating the acute illnesses of youth to managing the chronic illnesses of aging.

All of this is occurring at a time when healthcare is consuming more than 17 “Increased longevity and the aging percent of the United States’ GDP, the quality of care is deteriorating, and the healthcare delivery system is more challenged than ever. Meanwhile, more than of the baby boomer population 40 million people in the United States lack healthcare insurance. Millions of are creating a sea change in the others lack access — or cannot afford access due to the cost of care. demographic composition of One solution to ease some of these strains on the healthcare system may be to use technology to enable the aging population to age at home. Technology populations across the globe.” can support independence of the elderly through monitoring activities of daily living (ADL), tracking cognitive health, proactively managing chronic health conditions, and providing a system to promote social engagement. All of these services can be provided at home with the help of informal caregivers. Moreover, informal home caregivers help to alleviate the shortage of skilled nurses in the workforce today. Effectively, this technology shifts the care model from a large, centralized, provider-centric care model to a home- and community-based model of distributed care.

We have shown how three technical solutions were used in trials, by adapting technology to enhance the quality of life for patients, while at the same time, providing a platform for proactive health and wellness management for providers. The adoption of these solutions can play a major role in changing the playing field, improving quality of care, and improving health outcomes. “The adoption of these solutions can play a major role in changing the playing field, improving quality of care, and improving health outcomes.”

Adapting Technology for Personal Healthcare | 55 Intel® Technology Journal | Volume 13, Issue 3, 2009

References [1] J.E. Cohen. “Human Population: The Next Half Century.”Science , 14 Nov. 2003, pp. 1172-1175.

[2] J.E. Cohen. “Human Population: The Next Half Century.”Science , 14 Nov. 2003, pp. 1172-1175.

[3] “Family Caregivers — What They Spend, What They Sacrifice. The Personal Financial Toll of Caring for a Loved One.” Evercare and the National Alliance for Caregiving, November 2007, page 9.

[4] “The MetLife Caregiving Cost Study: Productivity Losses to U.S. Business.” MetLife Mature Market Institute and National Alliance for Caregiving, July 2006.

[5] “Genworth 2009 Cost of Care Survey: Home Care Providers, Adult Day Health Care Facilities, Assisted Living Facilities and Nursing Homes.” April 2009. Genworth Financial, page 8.

[6] “Genworth 2009 Cost of Care Survey: Home Care Providers, Adult Day Health Care Facilities, Assisted Living Facilities and Nursing Homes.” April 2009. Genworth Financial, Page 18.

[7] C. Lewis, “Emerging Trends in Medical Device Technology: Home Is Where the Heart Monitor Is.” http://www.fda.gov/

[8] B. Schlender. “Intel’s Andy Grove: The Next Battles in Tech.” Fortune, 12 May 2003, pp. 80-81.

[9] J. Takamura and B. Williams. “Informal Caregiving: Compassion in Action.” An informal report, US Department of Health and Human Services, 1997.

[10] Dishongh, T. and McGrath, M. Wireless Sensor Networks for Health Care Application. Artech Publishing, New Jersey, 2009.

[11] Osterberg, Lars, and Blaschke, Terrence. “Adherence to Medication.” New England Journal of Medicine, Volume 353:487-497, August 4, 2005.

[12] Deleeuw, Bill. “Testing objective measures of motor impairment in early Parkinson’s disease: Feasibility study of an at-home testing device.” Movement Disorders, Volume 24(4): 551-6, March 15, 2009.

[13] “A Reason to Keep On Going.” New York Times, June 12 2009, Health Section, page D5. By Stephanie Clifford, Reporter.

[14] “The Wisdom of Patients: Health Care Meets Online Social Media.” April 2008. The California Healthcare Foundation. Health Report, by Jane Sarasohn-Kahn, THINK-Health.

[15] Morris, M. “Social Networks as Health Feedback Displays.” IEEE Internet Computing, vol. 9, no. 5, pp. 29-37, Sept./Oct. 2005.

56 | Adapting Technology for Personal Healthcare Intel® Technology Journal | Volume 13, Issue 3, 2009

Acknowledgements We acknowledge with thanks the hard work of our co-workers: Kevin Rhodes, Brad Needham, Margie Morris, Jay Lundell, Eric Dishman, Tim Plowman, Steve Agritelley, and Adam Jordan.

Author Biographies Terrance (Terry) J. Dishongh, Ph.D., is currently a Senior Principal Engineer in the Intel Digital Health Group. In his twelve years at Intel he has been awarded 48 patents with an additional 44 pending and has published books on wireless sensors and electronic packaging. Dr. Dishongh has held faculty positions at the University of Maryland, College Park and the State University of New York at Buffalo. His e-mail is terry.dishongh at intel.com.

William (Bill) DeLeeuw is an 18-year veteran of Intel and currently is an “Innovation Architect” in the Intel Digital Health Group. He is responsible for creating new health concepts consistent with Intel’s digital health strategy, prototyping those concepts, and working with study coordinators to establish feasibility. Bill has worked on PC FAX systems, applications sharing software, consumer video conferencing, distance learning, consumer electronics, experimental wireless systems, and user-centered design concepts. He received a Bachelors of Science degree in Electrical Engineering from the University of Texas in 1991. His e-mail is bill.deleeuw at intel.com.

Mark Francis, M.P.P., is a Product Marketing Manager in the Intel Digital Health Group. Prior to joining Intel, Mr. Francis held senior management or Board positions with Harrison Clinical Research, Health Hero Network, Age Wave, and the City of Boston. Mark conducted his academic work at the Harvard Business School, Harvard Kennedy School and the University of Pittsburgh. His e-mail is mark.r.francis at intel.com.

Adapting Technology for Personal Healthcare | 57 Intel® Technology Journal | Volume 13, Issue 3, 2009

Healthcare IT Standards and the Standards Development Process: Lessons Learned from Health Level 7

Contributors

Charles Jaffe Abstract Health Level 7 Standardization of information in healthcare is critical to the ability of a diverse community of caregivers to reliably exchange complex data without ambiguity. W. Edward Hammond Perhaps more importantly, every member of the healthcare team must be able Duke University to re-use the data within increasingly diverse applications and environments. In order to improve quality and to constrain the escalating costs of delivering John Quinn care and preventative medicine, the strategies for developing these standards Accenture have attempted to keep pace with the developments in information technology, Robert H. Dolin human biology, healthcare policy, social science, and economics. For more than Health Level 7 two decades, Health Level 7 has been at the forefront of those processes for an international community. New technologies, and new applications of existing ones, foster standards development against a complex backdrop of social and political demands.

Healthcare IT Standards: An International Perspective Index Words The history of standards has made the creation of standards difficult. The global community of today did not exist when the healthcare IT community Healthcare Information Standards recognized the necessity for health data standards to reduce the costs of Health Level 7 interfacing systems and to permit the exchange of data. The health-related community began to be interested in standards during the early 1980s, and several organizations were created during the late 1980s and early 1990s. For clinical systems, the needs were primarily within a hospital and included the domains of patient admission, transfer and discharge; lab test ordering and result reporting; prescriptions and dispensing; materials management; images; reimbursement; and other similar domains. New standards-developing organizations were created, mainly focusing on only one component of the many recognized needs.

Standards developing organizations (SDOs) that were creating standards in communications, banking, manufacturing, and other non-health areas were quite mature by this time. In the United States, ASTM American Society for Testing and Materials created E31 for health data standards, focused initially “Health Level Seven (HL7) was on standards for the reporting of laboratory test data. Health Level Seven (HL7) was organized in 1987 to create standards to support the development organized in 1987 to create standards of best-of-breed hospital information systems. to support the development of best-of- breed hospital information systems.”

58 | Healthcare IT Standards and the Standards Development Process: Lessons Learned from Health Level 7 Intel® Technology Journal | Volume 13, Issue 3, 2009

In Europe, the Comité Européen de Normalisation (CEN) formed Technical “The International Standards Committee 251 in the early 1990s to create health data standards. There was little competition between CEN and HL7, simply because there was no Organization (ISO) was founded apparent common market. In fact, the methodology that was used to create just after the end of World War II later HL7 standards was influenced by CEN. In 1995, Germany became the first international affiliate of HL7, followed shortly thereafter by the in 1947 and is an international, Netherlands. From then on, the competition between HL7 and CEN began. standard-setting body composed of The relationship between members of both organizations was cordial, but both HL7 and CEN were vying to have their standards used in Europe. representatives from these various All countries have a national standards body: the American National Standards national standards’ organizations.” Institute (ANSI) in the United States; the British Standards Institution (BSI) in the United Kingdom, the German Institute for Standardization (DIN), the French national organization for standardization, known as Association Française de Normalisation (AFNOR), and so on. The International Standards Organization (ISO) was founded just after the end of World War II in 1947 and is an international, standard-setting body composed of representatives from these various national standards’ organizations. ISO Technical Committee 215 was formed in 1998 through efforts of the U.S. and the U.K. to create standards in Health Informatics. Since most countries, by law, require the use of an ISO standard if one exists, ISO became a key player in international standards. In 1991, recognizing that there were not sufficient expert resources available for ISO and CEN to conduct their standardization activities independently, ISO and CEN signed the Vienna Agreement to work together to produce standards. The impact of this agreement, including the joint “After over five years of work, a single development of standards and the sharing of standards, was felt independently standard had still not been defined.” within each organization.

In 2002, HL7, following the lead of IEEE, signed an agreement with ISO, through ANSI, that permitted HL7’s work and standards to be brought into ISO upon approval of ISO TC 215. Although both of these agreements resulted in a move toward the consolidation of work, standards continued to be duplicated. In the spirit of harmonization, efforts were put into mapping one similar standard to another across organizations where multiple standards exist. An example of one such activity was the effort to create a single standard for defining data types, by combining similar work from ISO, CEN, and HL7. After over five years of work, a single standard had still not been defined. Similarly, approaches to define a single global Reference Information Model (RIM) failed, because of differences between the CEN standard EN 13606 and the HL7 RIM. This latter issue was further compounded when the HL7 RIM was approved as ISO/HL7 21731:2006-Health Informatics-HL7 version 3-Reference Information Model.

Healthcare IT Standards and the Standards Development Process: Lessons Learned from Health Level 7 | 59 Intel® Technology Journal | Volume 13, Issue 3, 2009

“Over the next few years, HL7 Over the next few years, HL7 submitted several HL7 standards to the ISO to become joint HL7/ISO standards. These included HL7 version 2.5 submitted several HL7 standards to Messaging Standards; HL7 version 3.0 Clinical Document Architecture, R2; the ISO to become joint HL7/ISO Common Terminology Server, R1; HL7 EHR–Functional Model; and Clinical Genomics-Pedigree. In late 2005, HL7 submitted four regulatory standards to standards.” ISO: Structured Product Labeling, Release 1; Individual Case Safety Report; Stability Study; and Annotated Electrocardiogram.

None of these standards were accepted by TC 215 (most nations abstained). In addition, the International Committee on Harmonization (ICH) was very concerned, because they themselves had standards in some of these areas. In 2006, at the International Conference on Harmonization of Technical Requirements for Registration of Pharmaceuticals for Human Use ICH meeting in Yokohama, leaders from ISO TC 215, CEN TC 251, and HL7 made presentations to the group about their respective organizations. As a result, ICH became Class D Liaison with ISO and developed relationships with both CEN and HL7. “There was a real interest among the There was a real interest among the leaders of ISO, CEN, and HL7 to work together, largely driven by the limited resources available to produce standards, leaders of ISO, CEN, and HL7 to as well as by the confusion in the marketplace because of multiple standards. work together.” At the Global Health Information Technology Standards Summit in Geneva in 2006, presentations by Dr. Yun Sik Kwak, Chair, ISO/TC 215; Kees Molenaar, Chair, CEN TC 251; and Dr. Ed Hammond, Chair-elect, HL7 suggested the three SDOs might be able to work jointly to produce a single global standard for a single business purpose. After additional discussions, a charter was written and was subsequently ratified by all three SDOs. The charter establishes a Joint Initiative Council (JIC) that includes the chairs of the three participating organizations plus two additional representatives from each SDO.

Figure 1 shows the global healthcare landscape. The JIC serves as a collaborative forum for the international standards community. Within the US, the Federal Health Architecture coordinates 55 Federal agencies. Vocabulary standards are depicted in the oval boxes. “The charter establishes a Joint Initiative Council (JIC) that includes the chairs of the three participating organizations plus two additional representatives from each SDO.”

60 | Healthcare IT Standards and the Standards Development Process: Lessons Learned from Health Level 7 Intel® Technology Journal | Volume 13, Issue 3, 2009

NHS

Infoway Joint Initiative Council (JIC)

NeHTA CEN HL7 ISO CDISC IHTSDO IEEE

SCO

NCPDP ASTM ASC X12 SNOMED HITSP ANSI CCHIT LOINC NQF FHA HIMSS/RSNA MedDRA ICH ONC NIST IHE CPT AMA HHS

NIH FDA VHA ICD WHO

NLM CDC DOD DICOM NCI CMS Home Sec

caBIG US Realm

Figure 1: Global Healthcare Standards Landscape Source: Health Level 7

The word “joint” initiative was chosen specifically to distinguish this activity “Harmonization was generally from harmonization, since harmonization was generally considered to be efforts to map one standard to another competing standard. Any joint project considered to be efforts to map one must be agreed to by this group. A Joint Working Group (JWG) was also created, managed by ISO, which includes members from the participating standard to another competing SDOs. The purpose of the JWG is to work through the details of any joint standard.” initiative project, identify potential new projects, monitor the work on joint initiative projects, and support, in general, the work of the JIC. Each JIC project would be hosted by one of the participating SDOs, with that SDO providing a project chair or lead for the project. Each of the other SDOs would provide a co-chair to ensure the resulting work met the needs of each participating SDO.

The work on any project would be done jointly by volunteer members from each participating SDO, working as a cohesive unit. The three SDOs would be considered equal in the work process. The resulting work product would be balloted simultaneously by each participating SDO, and the comments from each SDO would be aggregated and processed by the joint project team. The resulting standard would be the joint property of each participating SDO and would carry a shared copyright and the logo of each SDO.

Healthcare IT Standards and the Standards Development Process: Lessons Learned from Health Level 7 | 61 Intel® Technology Journal | Volume 13, Issue 3, 2009

“ISO 21090: Healthcare Data Types, The first, now successfully concluded, JIC project was a standard for data types. This standard, ISO 21090: Healthcare Data Types, was approved in 2009, after was approved in 2009.” many years of unsuccessful work to bring disparate groups together. Other projects, listed next, have been accepted as JIC projects:

•• Individual Case Safety Report •• Glossary Project •• 13606/HL7 version 3 Implementation Guide •• Identification of Medicinal Products •• Clinical Trial Registration and Result •• BRIDG Data Model There are a number of other items being proposed as JIC projects. Once the scope and definition of a project is defined, and the participants begin to establish a level of trust in one another and can work together, the projects do move ahead.

“The JIC has developed a policy and The JIC has developed a policy and procedures for defining projects, for governance, and for bringing other groups into the JIC. Since the creation procedures for defining projects, for of the JIC by the three SDOs, two other SDOs have joined the group: the Clinical Data Interchange Standards Consortium (CDISC) became a member governance, and for bringing other in 2008, and the International Health Terminology Standards Development groups into the JIC.” Organization (IHTSDO) became a member in 2009. Other SDOs are in the process of joining the JIC.

There are barriers that still must be overcome for the JIC process to work. The balloting scenario differs among the three SDOs. The length of the balloting period differs as does the publication process and form. Communication among the participating SDOs and their membership is very challenging. Understanding the process has been difficult for all those involved, and managing the process has also been a challenge.

The success of the JIC offers a lot of promise, however. In the United States, an organization similar, at least in purpose to the JIC, has been formed: the SDO Charter Organization (SCO). This organization promises to consolidate efforts within the United States to produce a single standard for a single purpose. Those involved are strongly motivated to work together for a number of reasons. They can share expertise and resources, such as tools and repositories; and working together makes more efficient use of limited funding. Moreover, most SDOs recognize that the global market is expanding, and that there is an urgent need to accelerate the development of standards to avoid both marketplace and regulatory ambiguity. The more confusion, the less likely “Understanding the process has been players will be to use standards in the marketplace. difficult for all those involved, and managing the process has also been a challenge.”

62 | Healthcare IT Standards and the Standards Development Process: Lessons Learned from Health Level 7 Intel® Technology Journal | Volume 13, Issue 3, 2009

The Role of Architecture in Developing Healthcare Interoperability Standards In this section, we look at the role of architecture and an architectural “While any healthcare organization framework in developing healthcare interoperability standards, and we present some example environments that use them. can use HL7’s interoperability

HL7 is a set of standards that supports interoperability of software applications standards, these standards are, for that are designed to support clinical and administrative processes in a the most part, designed for use by healthcare organization. While any healthcare organization can use HL7’s interoperability standards, these standards are, for the most part, designed for healthcare provider organizations.” use by healthcare provider organizations.

Simply put, the architecture of an interoperability standard is a high-level view of the components that make up the standard and the relationships between those components. An architecture framework, as discussed here, is a reference context that we use to view each HL7 standard product and its components. These useful analysis tools are new to HL7.

Over its 22 years, HL7 has both created new and adopted some existing standards. Because of the separate and distinct origins of these standards, a different architecture could have been written for the internal components and their inter-relationships for HL7’s major standards products: these products include the HL7 version 2.x; HL7 version 3.0; HL7’s EHR standards; CCOW; and Arden Syntax. To begin this remedial effort of creating a common housing “To hold the architecture definitions to hold the architecture definitions of all HL7’s major products, HL7 created of all HL7’s major products, HL7 an architectural framework called HL7’s Services Aware Enterprise Architecture Framework (SAEAF). created an architectural framework

We shall now provide a high-level introduction to HL7’s SAEAF. We view called HL7’s Services Aware Enterprise SAEAF as it applies to the HL7 version 3.0 standards, including the major Architecture Framework.” components that they share (for example, the HL7 RIM), and the somewhat different purposes that they serve. Our customers, including those working in the UK, Canada, and now the United States, all begin their efforts with provider organizations that use the different architectures of the HL7 version 2.x and the HL7 version 3.x products, that going forward must co-exist and interoperate to achieve country-wide interoperability goals.

HL7 and Architecture: a Brief Introduction In 1987 the HL7 organization was created with the purpose of creating interoperability standards for the exchange of electronic information between IT systems within and between healthcare providers. Not surprisingly, the technology, workflows, and systems architectures created by HL7 in 1987 were the ones in use by the information technology (IT) industry in 1987. The standard created at that time (HL7 version 2.1) achieved widespread use, “No formal architecture was ever starting in 1991 and has evolved through a series of major and minor updates planned for any element of the HL7 to HL7 2.6 today. No formal architecture was ever planned for any element of the HL7 version 2. version 2.”

Healthcare IT Standards and the Standards Development Process: Lessons Learned from Health Level 7 | 63 Intel® Technology Journal | Volume 13, Issue 3, 2009

About five years later, HL7 volunteers expressed an interest in and began working on a RIM for healthcare. There were no formal expectations at the time for the scope of the effort, the amount of time it would take, or even the expected difficulty that we would encounter. This experience was quite eye-opening, because it forced HL7 developers to document the vastness and complexity of healthcare. Nevertheless, no formal architecture was defined for any HL7 standard that uses the RIM, although the RIM itself did assume a formal internal architecture that addresses its governance, definitions, and the tools related to testing conformance.

“This experience was quite eye- The HL7 RIM was never intended to stand on its own — although as a reference model of healthcare information, it certainly does. Once the RIM was opening, because it forced HL7 recognized as possible, the organization also turned to the effort of developing a developers to document the vastness new standard for data interchange that would have these features: and complexity of healthcare.” 1. Model-driven (i.e., all data items that can be exchanged or used to generate messages, documents, or services have a place and are taken from the RIM where cardinality and relationship(s) to other data are clearly defined. 2. It must be formally connected to a methodology that itself could be instantiated into software tools, giving us the following potential: a. Using software tools to manage, validate, assemble, and publish the version 3 Standard’s products, including the eXtensible Markup Language (XML) schemas that would be needed to hold the metadata of the information being exchanged. b. Automating the generation of interchange specifications and creating an environment where two or more individuals working from the same set of requirements could generate identical specifications that could be measured against a similar, software-generated set of conformities. “Messages support connections between These features evolved into the goals of HL7 version 3, which was first published in 2004. Our goals for HL7 version 3 have also evolved into IT systems that are involved in three distinct delivery mechanisms that can be used for a version 3-based supporting a workflow.” interchange. These delivery mechanisms are discussed next. An interchange can be a message similar to an HL7 version 2 message. version 3 messages today are produced in XML syntax. Messages support connections between IT systems that are involved in supporting a workflow. HL7 version 2 describes a workflow through a pre-defined and somewhat vague descriptor called a “trigger event.” For example, within an institution, a trigger event could be a physician’s lab order placed at the patient’s bedside to the hospital pathology laboratory. This workflow may be clearly understood by everyone involved in every possible action step required, from the composition of the order to the reporting of the signed final test results to the assigned individual. However, the same workflow within an ambulatory setting can be far more difficult to describe, because the actors in the workflow do not have an integrated understanding of each other’s role.

64 | Healthcare IT Standards and the Standards Development Process: Lessons Learned from Health Level 7 Intel® Technology Journal | Volume 13, Issue 3, 2009

Most importantly, it is not ideal that we lack a formal behavioral model “A formal dynamic model could be (i.e., a dynamic or workflow model), but we compensate for this by analyzing the existing workflow and then programming our interchange steps to modified and re-used when slightly accommodate it. Secondly, we can only create different interchange patterns as new or different workflows are we encounter them. In both cases, a formal dynamic model could document the supported workflow; assign a unique identifier to that workflow that could encountered.” be used by all parties; and serve as a base platform to be modified and re-used when slightly new or different workflows are encountered.

Messages have an expected lifetime bound by the instant in time that they are created and used. A message has no further use once the content of a message is consumed by the target system and optionally used to change the state of its database. Hypothetically, a message might be used as a possible means of assisting in a database recovery action, something that, hopefully, is never needed.

The next delivery mechanism is Clinical Documents, based on the HL7 “Clinical Documents contain Clinical Document Architecture (CDA) standard, which is a part of HL7 version 3, and which shares and uses the same underlying artifacts (for structured clinical data that also example, HL7 RIM, R-MIMS, Data Types, etc.). Clinical Document conform to document requirements templates are implementation guides based on the CDA. Clinical Documents contain structured clinical data that also conform to document requirements of being signed (electronically), of being signed (electronically), immutable, and so forth. Documents may immutable, and so forth.” indirectly be the product of and/or support a workflow, but they stand on their own and typically have an indefinite life.

The third delivery mechanism is a service that is based on the SAEAF. This mechanism (as messages and documents) will be defined by the relevant HL7 Work Group responsible for the interchange’s content (for example, Structured Documents for CDA). In HL7, services are currently based on the principles of a Services Oriented Architecture (SOA) that is, itself, based on the antecedent concepts developed in the Referent Model of Open Distributed Processing (RM-ODP). “This mechanism (as messages and documents) will be defined by the relevant HL7 Work Group responsible for the interchange’s content (for example, Structured Documents for CDA).”

Healthcare IT Standards and the Standards Development Process: Lessons Learned from Health Level 7 | 65 Intel® Technology Journal | Volume 13, Issue 3, 2009

HL7’s Services-Aware Enterprise Architecture Framework SAEAF was created by the HL7 Architecture Board (ArB) as a response to a request from the CTO for an HL7 architecture that is driven by HL7’s commitment to the following three principles:

1. HL7 produces specifications to enable Computable Semantic Interoperability (CSI) between users of systems implementing those specifications. 2. Instances of CSI between two or more HL7-based systems may cross department, enterprise, and/or national boundaries. 3. An HL7 Enterprise Architecture Specification (EAS) is required, if HL7 is to produce durable specifications that enable CSI in an effective, efficient, and scalable manner. The ArB recognized early on that three critical components were missing from HL7 and, therefore, they needed to develop SAEAF. These are the missing components:

1. A Behavioral Framework to express interaction semantics. 2. A layered Enterprise Conformance/Compliance Framework (ECCF) to support service integration and run-time assessment of CSI. 3. A Governance Framework to oversee the development and implementation of service (and other HL7 Interoperability Paradigm) specifications. Computable Semantic Interoperability (CSI) In order to understand the aforementioned, it is necessary to understand a little of what HL7 sees as the requirements to achieve CSI. These necessary, but not inclusive, requirements include the following:

1. Common static models (for example, the HL7 RIM) across all domains of interest including: a. An information model versus a data model. b. The semantics of common structures. c. Models based on robust data-type specifications. 2. A mutually understood behavioral (or dynamic) model that enables sufficient (as defined by the problem space) understanding of the “use context” of the creation of the data by the producer and its intended use by the consumer. 3. Methodology for binding to concept-based ontologies that support these constraints: a. Domain-specific semantics. b. Country, regional, or use-domain selection of appropriate ontologies. c. Rigorous versioning release cycle management to ensure that individual terminologies are consistently interpreted by both the producer and consumer of the data.

66 | Healthcare IT Standards and the Standards Development Process: Lessons Learned from Health Level 7 Intel® Technology Journal | Volume 13, Issue 3, 2009

4. A formally-defined process for specifying structures that contain both data and defined actions to be exchanged between machines, i.e., a data exchange (such as a message or a service) or an electronic document or services specification. Looking at the first three requirements of CSI just outlined, it is useful to understand them as the dimensions of variability for any data element exchanged between two HIT systems. As Figure 2 illustrates, three properties must be defined to exchange data elements: the data, terminology, and process or behavior. Taken together, these three dimensions provide a context for the information to be shared. We need to know data type (e.g., integer, text, image), terminology (i.e., the specific ontology being used), and process or behavior (e.g., a diagnostic process).

These three requirements, in addition to the fourth requirement of actually moving the data, are necessary for CSI. Figure 2 depicts this.

Process/Behavior

(D,P,T)

Data T

Terminology

Figure 2: Information Context for Data Elements Source: Health Level 7

Summary and Next Steps HL7 has begun the effort of defining an architectural framework called SAEAF that is both accommodating of services-based interoperability in an SOA and capable of providing a framework for holding, comparing, and analyzing the similarities and differences among HL7’s products. Going forward, SAEAF is also the basis for EAS that will be used in the future to apply HL7 to user SOA environments.

In our next section, we move on to describe the HL7 CDA, and we examine its role in healthcare in the United States today.

Healthcare IT Standards and the Standards Development Process: Lessons Learned from Health Level 7 | 67 Intel® Technology Journal | Volume 13, Issue 3, 2009

“HL7 is also developing standards The Templated CDA Strategy: Scalable and Incremental Interoperability for the representation of clinical Many people know of HL7 as an organization that creates healthcare messaging documents (such as discharge standards. HL7 is also developing standards for the representation of clinical summaries and progress notes).” documents (such as discharge summaries and progress notes). These document standards make up the HL7 clinical document architecture (CDA). The HL7 CDA, Release 1, became an ANSI-approved HL7 standard in November 2000. CDA Release 2 became an ANSI-approved HL7 standard in May 2005, and it is now in widespread use across the globe.

What follows is an introduction and overview of the CDA specification, along with a description of how and why CDA has emerged as a cornerstone component of the United States healthcare interoperability strategy.

The need for a clinical document standard stems from the desire to unlock the considerable clinical content currently stored in free-text clinical notes, and to enable comparison of that content in documents created on widely “Approximately 1.2 billion clinical different information systems. Approximately 1.2 billion clinical documents are produced in the United States each year. Dictated and transcribed documents documents are produced in the make up around 60 percent of all clinical notes. These documents contain the United States each year. Dictated majority of physician-attested information and are used as the primary source of information for reimbursement and proof of service. and transcribed documents make The challenge, addressed by CDA and the templated CDA strategy, is to up around 60 percent of all clinical continue to meet the needs of front-line clinicians today, who are heavily notes.” dependent on largely narrative documents, while at the same time providing a migration pathway for greater and greater discrete data. In other words, the goal of CDA is not simply to provide yet another format for the exchange of clinical documents. If that were all we wanted, we could use PDF, MS Word*, or any other file format. Instead, our goal is to get at the clinical content within those documents and make it computable, accessible to a computer, so it can be used for such things as decision support and quality reporting. We want to do this in a way that fits in to real-world clinical workflows so that we can tackle the problem incrementally, without the need for massive process redesign.

From a technical perspective, the HL7 CDA is a document markup standard that specifies the structure and semantics of a clinical document. A CDA document is a defined and complete information object that can exist outside “Our goal is to get at the clinical of a message, and it can include text, images, sounds, and other multimedia content. Just as you can create a document in MS Word, in PDF, etc., you can content within those documents and create a clinical document in CDA format. CDA documents are encoded in make it computable, accessible to a XML and derive their machine processable meaning from the HL7 RIM. computer, so it can be used for such things as decision support and quality reporting.”

68 | Healthcare IT Standards and the Standards Development Process: Lessons Learned from Health Level 7 Intel® Technology Journal | Volume 13, Issue 3, 2009

CDA is based on a principle of incremental interoperability, whereby an “All CDA documents must use the implementer can begin with a simple CDA, and then add structured data elements over time. CDA R2 consists of a single CDA XML schema, and same narrative markup so that you the architecture arises from the ability to apply one or more templates that can receive a CDA document from serve to constrain the richness and flexibility of CDA. Professional society recommendations, national clinical practice guidelines, and standardized data anyone in the world.” sets can be expressed as CDA templates.

From a practical perspective, incremental interoperability means that one can easily create a minimally-conformant CDA document — not really much different than creating an HTML document. CDA documents have a header that identifies and classifies the document and provides information on authentication, the encounter, the patient, etc. There are a handful of required fields and a number of optional fields. The body of the CDA can be purely narrative, by using markup very similar to XHTML. All CDA documents must use the same narrative markup so that you can receive a CDA document from anyone in the world, and, following a defined algorithm, render the document such that the receiving clinician correctly views content that was attested to by the originating clinician.

In addition to its narrative markup, CDA provides XML markup for formally “CDA provides XML markup for representing the clinical statements within the narrative. A complete encoding of all clinical utterances can be hard if not impossible, and there is no model formally representing the clinical of healthcare that can fully and formally represent everything a clinician might say. While the HL7 RIM is a richly expressive model that can represent statements within the narrative.” much of clinical narrative, the templated CDA strategy shields developers from the need to learn all the RIM nuances. Developers only need to understand those templates that have been recommended by the Healthcare Information Technology Standards Panel (HITSP) or put on the Certification Commission for Healthcare IT (CCHIT) roadmap. Developers simply map their internal data stores to the prioritized templates.

Assume that next year, HITSP defines the way in which we should communicate medical conditions and allergies. Assume that CCHIT takes these HITSP patterns and creates corresponding certification requirements. For those using CDA, there is no need to change to a new XML schema, and there is no need to change the approach to communicating narrative notes. CDA templates for medical conditions and allergies are created. An application maps their internal data stores to these templates, and it includes corresponding structured markup for medical conditions and allergies into CDA, whilst making no change to the narrative. On the recipient side, there will continue to be those who only render the document, whereas there may also be those “For those using CDA, there is no who can parse out the formally encoded medical conditions and allergies for use in decision support, disease management, personalized medicine, and need to change to a new XML schema, many other critical healthcare delivery requirements. Figure 3 shows the CDA and there is no need to change the constrained by the data elements of the Continuity of Care Record (CCR, in red) named the Continuity of Care Document (CCD). In this example, the approach to communicating narrative CDA template includes additional data elements not found in the CCR (such notes.” as Chief Complaint and Discharge Diagnosis).

Healthcare IT Standards and the Standards Development Process: Lessons Learned from Health Level 7 | 69 Intel® Technology Journal | Volume 13, Issue 3, 2009

Clinical Document Architecture Next year, HITSP and CCHIT will prioritize new templates; the following year more templates will be produced, and so it goes — scalability arises from the Continuity of Care Document fact that one XML schema does it all; incrementalism arises from the fact that Allergies only well-described changes need to be introduced over time, on a defined roadmap. Problems CDA is attractive for these reasons: Meds •• Scope. There is a single XML schema for all CDA documents. Personal Information •• Implementation experience. CDA has been a normative standard since 2000, and it has been balloted through HL7’s consensus process. CDA is also Insurance widely implemented. … •• Gentle on-ramp to information exchange. CDA is straight-forward to YYYY implement, and it provides a mechanism for incremental semantic Narative and tabular interoperability. One can begin with a simple CDA narrative document XXX Narative implementation and add discrete data elements over time, based on national priorities. TTTT Narative and tabular •• Improved patient care. CDA provides a mechanism for inserting evidence- based medicine directly into the process of care (via templates), thereby ZZZZ Narative enabling the application of evidence-based medicine.

Mode of transport •• Lower costs. CDA’s top-down strategy lets you implement CDA once, and reuse it many times for new scenarios. Discharge Diagnosis These attributes of CDA have lead to its adoption as a core healthcare information technology component in many countries, including the United Chief Complaint States, in particular by HITSP. Interoperability on a use-case by use-case basis can lead to a disjointed set of standards that do not support re-use or internal consistency. What is needed is an overarching approach to interoperability, one that is both widely encompassing and scalable, as new use cases are developed. Figure 3: CDA constrained by Additional Data Elements A strategy being exploited by HITSP, as well as the Integrating the Healthcare Source: Health Level 7 Enterprise (IHE) and other international HIT initiatives, is to base a growing number of Interoperability Specifications (ISs) on the CDA, or, more precisely, on templated CDA. From its inception, CDA has supported the ability to represent professional society recommendations, national clinical practice guidelines, and standardized data sets as “templates” or constraints on the generic CDA XML. Perhaps the best known example of a templated CDA specification is the ASTM/HL7 Continuity of Care Document (CCD) specification (see Figure 4), where the standardized data set defined by the ASTM Continuity of Care Record (CCR) is used to further constrain CDA, specifically for summary documents. “What is needed is an overarching approach to interoperability, one that is both widely encompassing and scalable, as new use cases are developed.”

70 | Healthcare IT Standards and the Standards Development Process: Lessons Learned from Health Level 7 Intel® Technology Journal | Volume 13, Issue 3, 2009

Subsequent to its adoption of CCD as the basis for HITSP/C32 “Summary Clinical Document Architecture Documents using HL7 CCD,” HITSP recognized that a top-down strategy, Continuity of Care Document whereby one learns CDA once and then re-uses it in other ISs, leads to greater Allergies cross use-case consistency. HITSP has since created a library of CDA templates A CCD (HITSP/C83 “CDA Modules Component”) that are used within a growing based document number of CDA-based specifications (for example, HITSP/C28 “Emergency Problems

Care Summary,” HITSP/C32 “Summary Documents Using HL7 CCD,” Meds HITSP/C105 “Patient Level Quality Data Document Using HL7 Quality Reporting Document Architecture (QRDA),” HITSP/C48 “Encounter Personal Information Document constructs,” HITSP/C84 “Consult and History & Physical Note

Document,” and HITSP/C78 “Immunization Document”). Insurance … Additional templated CDA specifications being developed within IHE and

HL7 include the QRDA, the Public Health Case Report Document, Operative YYYY Reports, Personal Health Monitoring Reports, and Minimum Data Set, Narative and tabular version 3. What is true across the spectrum of specifications is that all the XXX Narative specifications conform to the underlying CDA XML, and templates are re-used to the extent possible. CDA, coupled with a library of re-usable templates, TTTT forms the basis for a growing number of HITSP ISs, and CDA represents a Narative and tabular national interoperability strategy that is both widely encompassing and scalable ZZZZ Narative as new use-cases are developed.

Mode of transport Considerable discussion is taking place now in the United States around the notion of meaningful use, and about ensuring that our interoperability Discharge Diagnosis strategy supports meaningful use. At the heart of meaningful use is simply a requirement for data re-use as in re-using clinical trial data in the construction Chief Complaint of decision support rules, re-using clinician-captured data for quality reporting, public health reporting, etc. Imagine, for instance, that there were separate models and XML schemas for immunization data, medication administration, pharmacy dispensing, lab, and clinical summaries, and that the onus was on Figure 4: Templated CDA Specification the implementer to reconcile the differences in the data in order to support Source: Health Level 7 data re-use. CDA and templated CDA address the concept of meaningful re- use, by maximizing data re-use.

The value of the RIM, and the rationale for its use as the underlying formalism for encoding clinical statements in a CDA document include the following:

•• Consensus. A consensus process is used in the development of the RIM that encompasses many years, many vendors, many countries, and many use- cases. •• Expressivity. This allows for the formal representation of many types of clinical statements. •• Data re-use. All version 3 specifications are derived from a common model. “CDA and templated CDA address •• Concrete. While not perfect, RIM is here today for us to use. the concept of meaningful re-use, by •• Governance and maintenance. There is a defined consensus process for revisions. maximizing data re-use.”

Healthcare IT Standards and the Standards Development Process: Lessons Learned from Health Level 7 | 71 Intel® Technology Journal | Volume 13, Issue 3, 2009

“It is anticipated that several ballot CDA Release 1 became an ANSI approved HL7 standard in November 2000. CDA Release 2 became an ANSI-approved HL7 standard in May 2005. cycles will be required to work through Balloting on CDA Release 3 is slated to begin towards the end of 2009. Given the list of new requirements, and that HL7 has three ballot cycles per year, and given the widespread adoption of CDA Release 2, it is anticipated that several ballot cycles will be required that CDA Release 3 will become to work through the list of new requirements, and that CDA Release 3 will an ANSI-approved HL7 standard become an ANSI-approved HL7 standard sometime in 2011. CDA Release 3 requirements are cataloged as formal proposals and as suggested enhancements. sometime in 2011.” The main feature enhancement expected for CDA Release 3 is the inclusion of much more of the HL7 RIM, thereby enabling a wider range of clinician narrative to be expressed.

Standards and Standards Development Organizations: A New Collaboration Standards development is a political process. As described previously, the stakeholders are diverse but not every interest is equally represented. In many countries, the dominant voices in this process are the government agencies that “Standards development is a determine the business requirements and fund the development initiatives. The other end of the spectrum is represented by the vendor community political process.” that has always made a substantive contribution to the definition of the standards’ artifacts. In the past, the caregiver community has been significantly disenfranchised.

That is not to say that credentialed professionals have not been leaders of standards development organizations and government agencies. Even when clinicians lead these organizations, however, decision-making processes are often blunted by layers of administrative bureaucracy and regulatory overhead. Within the healthcare IT community, the chief medical information officer has often provided critical guidance and commanded substantive influence.

There is a growing body of critical data and published studies that describe the successful deployment of large-scale IT solutions within multi-disciplinary, healthcare systems. In fact, the application of eHealth processes has been shown to be transformative in various large, healthcare provider organizations. Unfortunately, recent exposés in the public media have recounted instances of dire unintended consequences of new or updated healthcare IT solutions. More often than not, the problem with these installations has not been technical. The apparent proximal cause has been the significant failure by system designers and software developers to understand the workflow and daily business requirements of the clinical end-users. “The application of eHealth processes has been shown to be transformative in various large, healthcare provider organizations.”

72 | Healthcare IT Standards and the Standards Development Process: Lessons Learned from Health Level 7 Intel® Technology Journal | Volume 13, Issue 3, 2009

At the technical end of the implementation spectrum lies the need to ensure “In simple terms, a physical therapist seamless, unambiguous interchange of data. The inability to achieve this is rarely the failure of the technical standards (specifications) themselves. Often may not apply the same meaning to there is a lack of formal agreement on vocabulary, the definition of individual a term as the orthopedic surgeon who data elements, and the application of the required terminology within the patient care continuum. This is exacerbated when requests are made or first used it.” requirements defined across clinical boundaries. In simple terms, a physical therapist may not apply the same meaning to a term as the orthopedic surgeon who first used it. In the in-patient setting, this is repeated daily in a failure to achieve unambiguous communication between physicians, nurses, pharmacists, and laboratory staff.

Attempts to rectify these ambiguities have often failed to overcome parochial and economic hurdles. In a practical sense, there is an ongoing tension among primary caregivers and the specialist community. The gatekeeper model of the managed healthcare system failed to achieve either fiscal or clinical outcome metrics because of the dissension that was exacerbated when practice variability was confronted with ambiguous, often counter-productive payment schema. In those managed care systems with closed practice systems and a single information system (e.g., Kaiser Permanente), the results are uniformly significantly more successful.

Outside of the United States, stringent government regulation has often led “Outside of the United States, to improved outcomes following implementation of enhanced healthcare stringent government regulation information systems. Perhaps this is more related to a single payer system and a more homogeneous patient population than to the technology itself. At times, has often led to improved outcomes regulatory oversight has been an enormous obstacle to data interchange. For example, the global regulated research community (chiefly pharmaceutical following implementation of enhanced and biotech industry) has embraced a structured vocabulary called the healthcare information systems.” Medical Dictionary for Regulatory Activities (MedDRA). This terminology is principally required for data encoding in the reporting of adverse events in clinical trials and in post-approval pharmacovigilance. The system is largely incompatible with SNOMED CT (Systematized Nomenclature of Medicine — Clinical Terms), which is in widespread use for patient care worldwide. Both clinical research and patient care suffer because of the artificial barriers to effective information exchange.

These problems transcend the functional requirements for interoperability. For our purposes, it is best to rely on the definition established by IEEE. Interoperability is the ability of two or more systems or components to “Interoperability is the ability of exchange information. Semantic interoperability is the ability to use the two or more systems or components information that has been exchanged. Faxing an EKG tracing between two professional offices provides a high degree of interoperability, if both clinicians to exchange information. Semantic agree on the parameters to use for interpretation. Like most data that are not interoperability is the ability to use the encoded, the EKG cannot be re-used in any meaningful way. Parenthetically, HL7 has developed a standard for encoding an annotated EKG. information that has been exchanged.”

Healthcare IT Standards and the Standards Development Process: Lessons Learned from Health Level 7 | 73 Intel® Technology Journal | Volume 13, Issue 3, 2009

“These practice guidelines are Reducing the ambiguity of the data is increasingly important as healthcare managers and researchers attempt to measure quality. For our purposes, the predicated on more data being derived enhanced clinical quality and patient-care outcomes are realized when practice from clinical research and have been guidelines are adhered to by individual caregivers and system-wide care requirements. Usually these guidelines are established by central authorities referred to as evidence-based (such as the US National Institutes of Health) and by professional societies medicine.” (for example, the American College of Cardiology). Globally, quality estimates (outcomes) are provided by the World Health Organization.

In the United States, these guidelines have been historically written by the Agency for Healthcare Research and Quality (AHRQ), an agency of the Department of Health and Human Services. More recently, non-profit organizations, such as the National Quality Forum, have been created to take a more proactive role in writing these guidelines. These groups establish priorities for quality evaluation as well as establishing the parameters for measuring success. These practice guidelines are predicated on more data being derived from clinical research and have been referred to as evidence-based medicine. These guidelines are critical to improving patient outcomes and reducing costs. “The development of treatment Unfortunately, the clinical guideline for one disease may be contraindicated algorithms that weigh the relative when complying with the guideline for a co-morbid disease. For example, the use of a relatively common anti-inflammatory drug for arthritis, such severity (importance) of two co‑morbid as ibuprofen, may be contraindicated in the management of hypertension. These two ailments are often concurrent in the Medicare-aged population, but diseases is the foundation for ongoing practice management algorithms would indicate a violation of quality patient research.” care. The development of treatment algorithms that weigh the relative severity (importance) of two co-morbid diseases is the foundation for ongoing research.

Other instances of conflicts of standard practice guidelines are less easily resolved. These conflicts may represent the differences in the interpretation of medical evidence between two medical societies or the same specialty in different regions or countries. It is easy to imagine that the recommendations offered by the American Academy of Orthopedic Surgeons would not coincide with those of the American Chiropractic Association. HL7 is working closely with the National Quality Forum and the Agency for Healthcare Research and Quality to standardize these metrics.

Delivering evidence-based medicine at the point of care (inpatient, ambulatory, “Decision support systems are highly emergent, or chronic home care) is enabled by a technology that is broadly classified as decision support. Decision support systems are highly complex, complex, predicated on the integration predicated on the integration of a vast amount of patient care and research of a vast amount of patient care data, and reliant upon the harmonization of technical standards. Within HL7, the Decision Support Working Group has advanced the technical and research data, and reliant upon functionality of these systems. Practical implementation is more complex. the harmonization of technical Individual system vendors have implemented decision support by using a vast array of technologies, alerts, graphical interface representations, and workflow standards.” modification. At the most simple level, this technology may be implemented

74 | Healthcare IT Standards and the Standards Development Process: Lessons Learned from Health Level 7 Intel® Technology Journal | Volume 13, Issue 3, 2009 in the form of a clinical alert, for example, when two drugs with a potential for “HL7 is working closely with several adverse interaction are prescribed concurrently. Another example of decision support might be a recommendation for an annual PAP smear at the time decision support initiatives.” of an outpatient visit. The most sophisticated systems guide complex disease management and require data from lab testing and medication administration, as in guidance for insulin dosing. HL7 is working closely with several decision support initiatives, such as the Clinical Decision Support Consortium.

The most innovative program does not directly involve the development of standards. It is an initiative in social engineering. Nearly four years ago, HL7 undertook a program for including specific clinical input into the standards development process. This Work Group recognized the contributions of a wide range of members of the caregiver community, including physicians, nurses, pharmacists, and even medical librarians. While important progress was made towards the identification of critical development pathways, the Work Group sorely lacked a focus on patient care, public health, and program management. In 2008, HL7 recognized the immediate need to align with the caregiver communities. With funding from the AHRQ, an experiment was first undertaken. The experiment went by the name of Bridging the Chasm.

At first blush, the chasm to be bridged was between the healthcare community “In 2008, HL7 recognized the and the organization of healthcare IT professionals. It was most evident to the immediate need to align with the clinicians that everyone was speaking a different language, filled with techno- speak, jargon, and acronyms, with which they were embarrassingly unfamiliar. caregiver communities.” Moreover, concepts, such as knowledge representation, so commonplace in IT, were lost on the clinicians. To re-purpose an old adage, “the clinicians were mad and they weren’t going to take it anymore.” Of course, there was another, much older, and often more acrimonious divide to bridge. That was the divisions erected over hundreds of years between specialists and primary care physicians, between physicians and nurses (you can substitute almost any other caregiver function), and between many of the ancillary but critical roles of pharmacist, dietician, physical therapist, pathologist, and many others.

In April 2009, a conference was convened in Washington to begin the bridge building. Funded by AHRQ and led by HL7, over 100 professional societies met with one objective in mind: begin to define the terminology, workflow processes, business requirements, and the like that everyone had in common. Of course, standards development was to be a collateral outcome, but only after the difficult task of bridge building was underway. This was not about fence mending, since it was hoped that the fences would be torn down eventually. For two days there was no jargon. Not an acronym was uttered. The results were unprecedented, because this heterogeneous body moved forward with unanimity of purpose. “Standards development was to be a collateral outcome, but only after the difficult task of bridge building was underway.”

Healthcare IT Standards and the Standards Development Process: Lessons Learned from Health Level 7 | 75 Intel® Technology Journal | Volume 13, Issue 3, 2009

“In the office setting, in the home, With the creation of a new professional society, the Clinical Information Interchange Collaborative, development of common terminology seems and in the hospital, the patients will possible. Integration of workflow into eHealth systems seems achievable. be the beneficiaries.” Articulating a business case for the harmonization of practice parameters and clinical guidelines has begun. In the office setting, in the home, and in the hospital, the patients will be the beneficiaries.

Standards development for healthcare will always be about politics. The problems to be solved have and will always have highly technical solutions. Interoperability is not a goal, but rather the means to improving quality and reducing costs. Creating innovative solutions to complicated issues of unified vocabulary and seamless data integration can be realized in an environment of collaboration.

Further Reading Blobel, Bernd. “Making hospital IT interoperable.” HITE 2008:1:2: Summer 2008:12-15. Available at www.hospitaliteurope.com

Clinical Data Interchange Standards Consortium (CDISC) This website contains standards and policy documents cited in this article. http://cdisc.org/ Hammond, William Edward; Jaffe, Charles; Kush, Rebecca Daniels, “Healthcare Standards Development: The Value of Nurturing Collaboration” Journal of AHIMA, July 2009 80/7, pages 44-40.

Health Level Seven, Inc. www.hl7.org On the website can be found more detailed supporting material cited in this article: CDA Tutorial, 2008. HL7 version 2.5 and HL7 2.6 Standards. Dolin RH, Alschuler L, Boyer S, Beebe C, Behlen FM, Biron PV, Shabo A. “HL7 Clinical Document Architecture, Release 2.” J Am Med Inform Assoc. 2006;13:30–39. Available at www.jamia.org.

HITSP/C32: “Summary Documents Using HL7 Continuity of Care Document (CCD); Component.” July 8, 2009. Version 2.5.

HITSP/C37: Lab Report Document Component. July 8, 2009. Version 2.3.

HITSP/C75: Healthcare Associated Infection (HAI) Report Component. July 8, 2009. Version 1.1.

HITSP/C83: CDA Content Modules Component. July 8, 2009. Version 1.1.

HITSP/C105: Patient Level Quality Data Document Using HL7 Quality Reporting Document Architecture (QRDA). June 30, 2009. Version 0.0.1.

76 | Healthcare IT Standards and the Standards Development Process: Lessons Learned from Health Level 7 Intel® Technology Journal | Volume 13, Issue 3, 2009

European Committee for Standardization (CEN) Technical Committee 251 (Healthcare) This site contains standards and policy information cites in this article. http://www.cen.eu/cenorm/sectors/sectors/isss/index.asp ISO Technical Committee 215 (Healthcare Informatics) This workspace contains documents and standards cited in this article: http://isotc.iso.org/livelink/livelink?func=ll&objId=529137&objAction=browse

Acronyms Agency for Healthcare Research & Quality (AHRQ) American National Standards Institute (ANSI) Association Française de Normalisation (AFNOR) British Standards Institution (BSI) Certification Commission for Healthcare IT (CCHIT) Clinical Data Interchange Standards Consortium (CDISC) Clinical Document Architecture (CDA) Comité Européen de Normalisation (CEN) Computable Semantic Interoperability (CSI) Continuity of Care Document (CCD) Continuity of Care Record (CCR) Enterprise Architecture Specification (EAS) Enterprise Conformance/Compliance Framework (ECCF) Extensible Markup Language (XML) German Institute for Standardization (DIN) Health Level Seven (HL7) Healthcare Information Technology Standards Panel (HITSP) Integrating the Healthcare Enterprise (IHE) International Health Terminology Standards Development Organization (IHTSDO) International Standards Organization (ISO) Interoperability Specifications (ISs) Joint Initiative Council (JIC) Quality Reporting Document Architecture (QRDA) Reference Information Model (RIM) Referent Model of Open Distributed Processing (RM-ODP) Services Aware Enterprise Architecture Framework (SAEAF) Standards Developing Organization (SDO) SDO Charter Organization (SCO) Services Oriented Architecture (SOA)

Healthcare IT Standards and the Standards Development Process: Lessons Learned from Health Level 7 | 77 Intel® Technology Journal | Volume 13, Issue 3, 2009

Author Biographies Charles Jaffe, MD, PhD, FACP. Since 2007, Dr. Jaffe has served as the Chief Executive Officer of HL7. He began his career at Intel in 2005 as the Senior Global Strategist for the Digital Health Group. He has also served as Vice President of Life Sciences at SAIC and as the Global Director of Medical Informatics at AstraZeneca Pharmaceuticals. He completed his medical training at Johns Hopkins and Duke Universities, and he was a post-doctoral fellow at the National Institutes of Health and at Georgetown University. Formerly, he was President of InforMed, a consulting firm for research informatics. Over his career, he has been the principal investigator for more than 200 clinical trials, and he has served in various leadership roles in the American Medical Informatics Association. He has been a board member on leading organizations for information technology standards, and he served as the Chair of a national institutional review board. Currently, he holds an appointment in the Department of Medicine at the University of California at San Diego. He has been the contributing editor for several journals and he has published on a range of subjects, including clinical management, informatics deployment, and healthcare policy.

W. Ed Hammond, PhD, is Professor-emeritus, Department of Community and Family Medicine and Professor-emeritus, Department of Biomedical Engineering, Duke University and Adjunct Professor in the Fuqua School of Business at Duke University. He has served as President of the American Medical Informatics Association (AMIA), President of the American College of Medical Informatics, and as Chair of the Computer-based Patient Record Institute. He is currently serving his third term as the Chair of HL7. He was Chair of the Data Standards Work Group of the Connecting for Health Public- Private Consortium. Dr. Hammond was a member of the IOM Committee on Patient Safety Data Standards. He was awarded the Paul Ellwood Lifetime Achievement Award in 2003 and the ACMI Morris F. Collen Award of Excellence in November 2003.

John Quinn is the Chief Technology Officer of HL7 and has served in this full-time position since September 2007. HL7 is one of several ANSI -accredited SDOs operating in the healthcare arena.

Quinn is also a Senior Executive, Chief Technology Officer and Thought Leader in Accenture’s Health Life Science’s Provider Practice focusing on healthcare information system’s technologies, architectures, and data and messaging standards; and he serves in his role as HL7’s CTO in part through a significant contribution by Accenture. He has over 32 years experience in Healthcare IT and 35 years experience in computer industries, respectively. He has participated on the HL7 Board of Directors since its inception 21 years ago and served as the Chair of its Technical Steering Committee from 1989–2007.

78 | Healthcare IT Standards and the Standards Development Process: Lessons Learned from Health Level 7 Intel® Technology Journal | Volume 13, Issue 3, 2009

Robert H. Dolin, MD, MS, serves as the Chair-elect on the HL7 Board of Directors and has been involved with the organization since 1996. Dr. Dolin is an internationally renowned and innovative physician expert in healthcare information technology standards development with more than 20 years of clinical experience. After receiving his medical degree, he served as chief resident at UCLA Department of Medicine in 1989, where he developed Harbor-UCLA first outpatient electronic medical record system. Dr. Dolin is a Fellow in the American College of Physicians, and he was elected into Fellowship in the American College of Medical Informatics in recognition of his work on standards development. Dr. Dolin currently owns Semantically Yours, a healthcare standards consulting firm. Previously, he spent 18 years as a Hospitalist at Kaiser Permanente, Department of Internal Medicine. At Kaiser, he was the physician lead for the enterprise terminology services team, responsible for deploying SNOMED in their national electronic health record. Dr. Dolin also co-chairs the HL7 Structured Documents Work Group, and is co-editor of the HL7 CDA and the CCD specifications. He has also served on the SNOMED CT Content Committee, and currently co-chairs the US HITSP Foundations Committee. Dr. Dolin’s work has also been published in both technical and clinical peer reviewed journals.

Healthcare IT Standards and the Standards Development Process: Lessons Learned from Health Level 7 | 79 Intel® Technology Journal | Volume 13, Issue 3, 2009

Healthcare Information Integration: Considerations for Remote Patient Monitoring

Contributors

Kristina M. Kermanshahche Abstract Intel Corporation Modern healthcare is no longer practicable without data integration. Without standards, integration costs soar and threaten the effectiveness of healthcare delivery. With standards, information becomes accessible in computable form, driving new levels of clinical research, patient-empowered healthcare, and innovative business models. In this article we examine a few of the latest developments in healthcare informatics standards with respect to remote patient monitoring, review recent learnings in deploying a remote patient Index Words monitoring solution, and identify key considerations and areas for further development. Remote Patient Monitoring (RPM) Semantic Interoperability Service Oriented Architecture (SOA) Introduction Healthcare Informatics Simple interventions can save lives and reduce the cost of healthcare. Home Chronic Disease Management (CDM) care for patients with co-morbidities can mean the difference between life and HL7 Clinical Document Architecture (CDA) death. For example, with ongoing monitoring of vital signs, patient condition, and medication levels, sudden changes outside of the patient’s established thresholds can be detected, which can mean the difference between extended hospitalization and rapid decline in health, versus maintaining a quality of life in the patient’s home, surrounded by family and friends.

Similarly, maintaining tight control over blood glucose levels during gestational diabetes can mean the difference between premature birth with a whole host of medical complications for both mother and child, versus a stable healthy maternity and normal delivery.

RPM can be significant to both examples just cited, and the effectiveness of RPM depends directly on the availability of standardized information from a variety of healthcare data sources, including patient health summary, prescriptions, lab results, daily vital signs collection, and functional assessments. “Home care for patients with co‑morbidities can mean the difference between life and death.”

80 | Healthcare Information Integration: Considerations for Remote Patient Monitoring Intel® Technology Journal | Volume 13, Issue 3, 2009

We first explore the standardized representation of RPM information by using Founded in 1987, Health Level Seven (HL7) the Health Level 7 (HL7) v3 CDA Release 2 [4] and the Personal Health is a not-for-profit, ANSI-accredited standards Monitoring Report (PHMR) Draft Standard for Trial Use (DSTU) [10]. Then, development organization dedicated to we consider recent developments with the American Health Information providing a comprehensive framework and Community (AHIC) Use Cases and Healthcare Information Technology related standards for the exchange, integration, Standards Panel (HITSP) Implementation Guides for Remote Patient sharing, and retrieval of electronic health information that supports clinical practice and Monitoring [6, 7], Consultation and Transfer of Care [5], Long-Term Care the management, delivery, and evaluation of Assessments [13], and the resulting HL7 Implementation Guide for CDA 2: health services. CDA Framework for Questionnaire Assessments DSTU [8]. We review several SOA principles and considerations critical to healthcare integration. Next, we discuss several technical and workflow considerations when designing and deploying a RPM solution. Along the way we identify multiple areas suitable for further development.

Remote Patient Monitoring Use Case

In this section we discuss a typical RPM use case, in order to set the context for AHIC is a federal advisory body, chartered in applying healthcare informatics standards and to explain how subtle nuances 2005 to make recommendations to the Secretary in clinical workflow and system interactions can have a profound impact, both of the U.S. Department of Health and Human on the design of the end-to-end information integration as well as on future Services on how to accelerate the development enhancements to informatics standards. and adoption of health information technology. The sequence diagram depicted in Figure 1 was adapted from the HITSP AHIC specifies prioritized healthcare use RMON Business Use Case [6], representing a superset of actors, steps, and cases, for which the HITSP then develops functionality across a number of finer-grained but related use cases. We interoperability specifications, leveraging, currently work with clinicians in a variety of regions worldwide but primarily harmonizing, and further constraining existing based in the United States and the United Kingdom. Clinical delivery and internationally-recognized standards. HITSP reimbursement models are very different between these two regions. Some identifies gaps in coverage and forwards new healthcare delivery models rely upon visiting nurses, community outreach, development areas to Standards Development and clinical call centers to perform primary patient engagement, monitoring Organizations (SDOs) for consideration patients remotely and within the home, and only referring the most acute cases in future revisions. While both AHIC and for physician or hospital intervention. HITSP are entirely U.S.-centric in focus, similar organizations exist in each major region Other delivery models rely upon physician practices to monitor patients worldwide and their common goal is to achieve directly, at times delegating these tasks to trained staff that operate as part of standards harmonization. The process of the practice under their clinical supervision. These two models exist in both international standards harmonization is covered regions regardless of patient acuity, frequency of monitoring, or reimbursement separately in this issue by Jaffe, et al. models. (In the healthcare context, patient acuity refers to the type and severity of illness, with acutely ill patients requiring emergency care.) It is important that use cases and the supporting informatics standards comprehend these types of variation in order to deliver an effective RPM solution. “We currently work with clinicians in a variety of regions worldwide but primarily based in the United States and the United Kingdom.”

Healthcare Information Integration: Considerations for Remote Patient Monitoring | 81 Intel® Technology Journal | Volume 13, Issue 3, 2009

Remote Monitoring Electronic Health Management System Record (EHR) System

Patient

Event: Collection of Patient Session Data (vital signs, assessments) according to prescribed Treatment Plan

Remote Monitoring Values Compared Against Reference Range

Alerts and Notifications Generated for Values Outside of Reference Range

Remote Monitoring Information Transmitted to Clinician’s EHR

Clinician Reviews Remote Monitoring Information from Within the EHR

Clinician Modifies Updated Treatment Plan Transmitted Treatment Plan as to Remote Monitoring System Necessary Updated Treatment Plan Conveyed to Patient

Figure 1: RPM Sequence Diagram Source: Intel Corporation, 2009. Adapted from HITSP RMON Business Sequence Diagrams [6]

Figure 1 represents three primary actors of concern – the Patient, the Remote Monitoring Management System (RMMS) that operates the RPM solution, and an external Electronic Health Record (EHR) system, which might represent anything from a clinical electronic medical record system to a purpose-built system developed specifically for chronic disease management (CDM) or home health monitoring. The primary monitoring of a patient’s wellbeing by a clinician can occur either within the RMMS itself or by extension, from within the EHR system. The sequence diagram is triggered by the event of patient data collection, including vital signs and assessments defined by a treatment plan, which specifies not only what data to collect but also the frequency and schedule.

Typical patients are asked to collect vital signs anywhere from one to four times per day (an average of eight vital sign readings per day), ranging anywhere from three to seven days per week. Patients work with a variety of peripheral “Patients work with a variety of device types including blood pressure cuff, glucose meters, weight scales, and pulse oximeters. Vital sign measurements are combined with assessment peripheral device types including blood questions that are geared to assess patient adherence to the treatment plan, functional status, and coping mechanisms with regard to their specific set of co- pressure cuff, glucose meters, weight morbidities, such as congestive heart failure, diabetes, and chronic obstructive scales, and pulse oximeters.” pulmonary disease (COPD).

82 | Healthcare Information Integration: Considerations for Remote Patient Monitoring Intel® Technology Journal | Volume 13, Issue 3, 2009

Once the RMMS aggregates the vital sign and assessment data, it compares “Optionally, alerts and notifications the values against a pre-established reference range or threshold to identify those readings that are out of normal range. Optionally, alerts and notifications may be generated to the EHR system may be generated to the EHR system and members of the clinical team, and members of the clinical team, notifying them of the abnormal readings. Next, both normal and out-of-range information is transmitted to the EHR system, either in raw or summary notifying them of the abnormal format. The clinician reviews the information, makes clinical notes, and as readings.” needed, generates referrals and modifies the treatment plan. The updated treatment plan is transmitted to the RMMS and subsequently communicated to the patient and the patient-local devices.

The significance of the use case and sequence diagram just described are as follows:

•• There is a high degree of variability in where patient data are maintained. The clinician and clinical support team review and annotate patient information in multiple systems of record. The RMMS is generally deployed in addition to multiple legacy systems used to manage patient healthcare, and typically, little to no integration exists. •• Some healthcare use cases and informatics standards assume that a clinician event is responsible for triggering the transfer of information from one clinical team to the next; however, this use case underscores several examples where data transmission is triggered instead by system events, and it is important for informatics standards to comprehend both scenarios. •• The type, frequency, and schedule of the information collected is also highly variable, which must be comprehended by informatics and system deployment designs alike. The more standardized the information, the more it will drive machine computability, advanced analytics, and improved healthcare at lower costs. However, informatics standards must always balance the goal of perfect information requiring every data element, with a pragmatic approach to incremental adoption, relying instead on implementations to deliver best effort results in populating standard data elements with carefully designed constraints. •• A consistent, appropriately encoded specification of the treatment plan is just as important as the data collection itself. Initial focus is rightfully on capturing and transmitting RPM data, yet to truly measure patient adherence and prognosis over time, we also need a consistent method to specify the treatment plan in a semantically meaningful way.

Healthcare Information Integration: Considerations for Remote Patient Monitoring | 83 Intel® Technology Journal | Volume 13, Issue 3, 2009

Cloud Services: Entity Identity Service Controlled Terminology Service Remote Patient Monitoring Solution Boundary 6 Patient Consent Service Rx, Lab Results 7 Remote Monitoring SOA Integration Management System Services

9 1 8 4 Data Data Data 12 11 Data 10 Personal Health Synchronization Monitoring Device Electronic Health Clinician Record (EHR) System Patient 2 3 5

Care Manager

Figure 2: RPM Interaction Diagram Source: Intel Corporation, 2009

Figure 2 reveals several additional deployment considerations with respect to RPM. Step 1 represents the transmission of patient session data to the RMMS. The RMMS aggregates trend information, compares vital signs against reference ranges, and generates alerts and notifications. It has its own data repository, consisting of historical patient treatment plans, reference ranges, clinical notes, patient monitoring data, and customer-specific configuration information. The RMMS is necessarily designed to be an online transaction processing (OLTP) system, since its primary goal is to drive RPM in a high- performance, transaction-oriented manner.

Optionally, Step 2 and 3 depict a care manager who reviews the patient’s status from within the RMMS. The care manager may modify the treatment plan directly, or annotate the record and refer to a clinician for review and follow- up. The care manager may also play a key role in assessing the viability of the data collected prior to escalation. For example, in the case of a very low weight reading generating an alert, the care manager might confirm that the patient’s grandson had in fact triggered the scale.

Step 4 represents some form of data synchronization between the RMMS and the Integration Services environment. The primary goal of the Integration “The RMMS is necessarily designed to Services environment is rapid query, retrieval, translation, transformation, and be an OLTP system, since its primary guaranteed delivery (push/pull) of healthcare information between a number of participating entities (Step 5). Integration Services, therefore, are commonly a goal is to drive RPM in a high- mix of online analytic processing (OLAP) combined with logical mechanisms performance, transaction-oriented of extract, transform, and load (ETL). Trend analysis and summarization may also be performed at this stage. The Integration Services environment is best manner.” developed according to SOA principles, largely due to the complexity and

84 | Healthcare Information Integration: Considerations for Remote Patient Monitoring Intel® Technology Journal | Volume 13, Issue 3, 2009 number of end points, the variability of legacy and standard interfaces, and “Note that many providers will the number of different workflows and external services required. We discuss the rationale and benefits of applying SOA to RPM in more detail later. The require manual review of the data for mechanism of data synchronization depends upon patient acuity, data latency, clinical accuracy prior to committing and a number of other considerations, also discussed later in this article. it to the EHR.” As a part of the transformation that occurs in Step 5, multiple calls may be made to services across the Internet (Steps 6 and 7) in order to complete the healthcare dataset, including identity match; terminology encoding; record locate; patient consent; or data enrichment, incorporating the latest lab results, prescription history, or drug-to-drug interaction checks.

Finally, the RPM information is transmitted to the EHR (Step 8), where it is reviewed and annotated by the clinician (Step 9). Note that many providers The HL7 v3 RIM is an abstract healthcare will require manual review of the data for clinical accuracy prior to committing informatics model, generated by using Uniform it to the EHR. This review requirement poses significant workload and process Modeling Language (UML) and consisting of implications to the future viability of RPM, discussed in more depth later. a set of classes, attributes, and relationships. The HL7 CDA and other v3 specifications are The clinician enters modifications to the treatment plan and generates requests derived from the RIM and encoded by using for consultation (Step 10). Changes to the treatment plan are returned to the XML. The RIM is capable of representing Acts, RMMS (Step 11) and transmitted to the patient-local devices (Step 12). The Entities, Relationships, Roles, and Participants, inbound data flows undergo similar decomposition and transformation as and it is refined by using standard code sets and outbound flows (Steps 5-7), remapping identity and terminology to local code controlled medical vocabulary. sets.

System designers should anticipate the need to support a large number of end points with a high complexity and variability of transforms (e.g., HL7 v2.x, HL7 v3, proprietary XML, proprietary delimited files) and translations (e.g., SNOMED CT, LOINC, RxNORM, ICD9/10) required across an array of different transport protocols (e.g., SOAP over HTTPS, SFTP, MLLP, and proprietary sockets). A powerful approach is the implementation of the HL7 v3 CDA [4] in the Integration Services environment (Step 5) as a normalized view of RPM information, leveraging the full richness of the HL7 v3 Reference Information Model (RIM). The CDA becomes in effect a Rosetta Stone, making the subsequent translation to legacy, proprietary, and standards- based systems predictable, reliable, and semantically correct. The incremental adoption model inherent in CDA ensures the relative ease with which we can populate optional segments with new information, such as Plan of Care, while the SDOs work through the optimal encoding scheme. It also ensures Continua Health Alliance is a non-profit, open straightforward compliance with new CDA document types, such as the industry coalition of more than 200 healthcare Continua Health Alliance Personal Health Monitoring (PHM) Report [3], as a and technology companies collaborating to minor transform from the baseline CDA. improve the quality of personal healthcare. Continua is dedicated to establishing a system of interoperable personal health solutions, with the knowledge that extending those solutions into the home fosters independence, empowers individuals, and provides the opportunity for personalized health and wellness.

Healthcare Information Integration: Considerations for Remote Patient Monitoring | 85 Intel® Technology Journal | Volume 13, Issue 3, 2009

Using HL7 to Encode Telehealth Data “The standard is both precise and The HL7 v3 CDA Release 2 [4] is the ideal vehicle for healthcare data integration. It is an informatics standard based on years of industry expertise, it adaptable, as demonstrated by the represents healthcare documents in their purest form, ranging from a progress Continua xHR Implementation report, to a discharge summary, to capturing an MRI study as a standard image set. The HL7 CDA can scale easily from early to advanced stages of Guide.” adoption: it can represent a container with basic summary information (such as basic patient demographics, ordering physician, and a facsimile of a lab result stored as an attachment), to a container with a longitudinal health record. The standard is both precise and adaptable, as demonstrated by the Continua xHR Implementation Guide [3], a variant of the CDA developed to define the Personal Health Monitoring Report (PHMR) [10].

The real ingenuity of the CDA is the overlay of a series of templates or constraints to the v3 RIM, to uniquely identify and encode sections of a clinical document in a standard and semantically meaningful way. The HL7 v3 Continuity of Care Document (CCD) [9] was one of the first broadly implemented set of constraints applied to the HL7 CDA Release 2 [4], and it was selected by HITSP as the basis of their implementation guides. There are now a whole complement of implementation guides based on either the umbrella CDA or on the further constrained CCD, all of which aim to deliver both human-readable and machine-consumable clinical information in a systematized fashion.

Assigned Author The HL7 CDA supports the concept of the Assigned Author being either a human or a device, as depicted in Code Listing 1.

“The real ingenuity of the CDA is the overlay of a series of templates or constraints to the v3 RIM, to uniquely identify and encode sections of a … clinical document in a standard and semantically meaningful way.” …

Code Listing 1: Machine-Computable XML: Assigned Author Source: Intel Corporation, 2009

86 | Healthcare Information Integration: Considerations for Remote Patient Monitoring Intel® Technology Journal | Volume 13, Issue 3, 2009

In some cases, a summary report of RPM is generated along with clinical notes, as part of a transfer of care or a request for consultation, and the use of a human Assigned Author makes perfect sense. In other cases, information is automatically collected by RPM devices, compiled into an HL7 CDA, and forwarded to other healthcare systems for subsequent clinical analysis. In this latter case, there is no human author who can be assigned, as the information has yet to be reviewed by any clinical personnel. Similarly, while some regions may require legal authentication of a clinical document, other regions may delegate legal authentication to a device or system, and still others may opt to release a clinical document prior to legal authentication. The HL7 PHMR does an excellent job of both considering and supporting each of these variations in usage models [10].

Medical Equipment The HL7 PHMR [10] (based upon the work of the Continua Health Alliance “The HL7 PHMR does an excellent [3]) does a thorough job of specifying required peripheral manufacturer information. Table 1 and Code Listing 2 show the Medical Equipment section job of both considering and supporting from the HL7 PHMR [10], depicting both the XML-rendered table and a each of these variations in usage subset of the machine-computable section for a single device, respectively. models.” Medical Equipment System System System System ID Production Regulated Type Model Manufacturer Spec Blood Pulse Nonin 1F-3E- Unspecified: Regulated Pressure Master 46-78- Serial Monitor 2000 9A-BC- Number: DE-F1 584216 Part Number: 69854 Hardware Revision: 2.1 Software Revision: 1.1 Protocol Revision: 1.0 Prod Spec GMDN:

Table 1: XML-rendered Table: Medical Equipment Source: Health Level Seven, 2009 [10]

Healthcare Information Integration: Considerations for Remote Patient Monitoring | 87 Intel® Technology Journal | Volume 13, Issue 3, 2009

“Systems can transform the Medical Equipment terminology to other standard and … the standard is encoded in the first … …

Code Listing 2: Machine-computable XML: Medical Equipment Source: Health Level Seven, 2009 [10]

“Semantic interoperability is the Note the significance of the use of standard terminology in Code Listing 2, which leverages both SNOMED CT and MDC code sets (refer to “code/code highest form of data integration.” system…translation code/code system”). This semantic encoding is what makes the information machine-computable and enables interoperability. Systems can transform the terminology to other standard and proprietary formats, precisely because the standard is encoded in the first place, just like the Rosetta Stone. Semantic interoperability is essential to enable clinical research and to facilitate queries across a wide variety of data sources; however, semantic exchange is only possible if the terminology has been first normalized to one of a few dozen international healthcare terminology standards.

Levels of Interoperability Semantic interoperability is the highest form of data integration, such that receiving systems can readily and precisely consume information, encoded with standard terminologies and code sets, with no loss of meaning or context for abstract terms and concepts. This is the goal of the HL7 v3 RIM.

Syntactic interoperability, the next lower form of data integration, exchanges information by using agreed-upon syntax. A set of fields are specified along with their syntax, but nothing is specified as to the possible range of values or “Prior messaging standards tended to meaning. Prior messaging standards tended to stop short at syntax: they did not address the crucial last step of semantics, and therefore left an incomplete stop short at syntax.” data standard wide open to conflicting interpretation and incompatible

88 | Healthcare Information Integration: Considerations for Remote Patient Monitoring Intel® Technology Journal | Volume 13, Issue 3, 2009 implementations. For example, the earlier versions of HL7 v2.x messages focused almost exclusively on syntax, and hence the expression “every v2.x interface is a new v2.x interface.” Each implementation added proprietary interpretations and extensions over time.

Integration at the lower five levels of the OSI model focus on standardizing transport, protocol, and security, which is also the primary focus of such groups as W3C, Continua, IHE, and IEEE. Technology standards are optimal to address the exchange on the wire, while healthcare domain expertise is optimal to address the syntax and semantics of the exchange.

Vital Signs Table 2 shows a sample of the Vital Signs section of a PHMR, depicting the XML-rendered table, while Code Listing 3 depicts a subset of the machine- “It is common in fact for information computable section for a single measurement. This example highlights the dual in the machine-consumable section role of the CDA construct: to provide highly accessible information viewable by clinicians in a concise, easy-to-read format, along with a fully encoded, to be richer than that depicted in the machine-consumable version of the same information, capable of supporting any degree of advanced analytics and clinical workflow. It is common in fact XML-rendered table.” for information in the machine-consumable section to be richer than that depicted in the XML-rendered table. For example, the XML-rendered table might present only summary or trend information, whereas it is perfectly acceptable for the machine-consumable segment to include both raw and summary information.

Vital Signs Date Captured Peripheral Measurement Value Condition Reference Range

2008-01-07 13:55:14.000 Blood Pressure Systolic 137 mmHg Alert 89-130

2008-01-07 13:55:14.000 Blood Pressure Diastolic 85 mmHg Normal 59-90

2008-01-07 13:55:14.000 Blood Pressure Pulse 89 BpM Normal 59-100

Table 2: XML-rendered Table: Vital Signs Source: Intel Corporation, 2009

Healthcare Information Integration: Considerations for Remote Patient Monitoring | 89 Intel® Technology Journal | Volume 13, Issue 3, 2009

Vital Signs … … …

Code Listing 3: Machine-computable XML: Vital Signs Source: Intel Corporation, 2009

“Inclusion of this information Note also the use of Observation/interpretationCode and Observation/ referenceRange in Code Listing 3, to indicate whether a particular reading is crucial to be able to quickly is considered within or outside of the reference range for that measurement type. These data elements are optional, but are strongly encouraged as industry perform patient triage as well as best practice (i.e., “SHOULD”) by the HL7 PHMR [10]. Inclusion of this trending analysis and population information is crucial to be able to quickly perform patient triage as well as trending analysis and population management. management.”

90 | Healthcare Information Integration: Considerations for Remote Patient Monitoring Intel® Technology Journal | Volume 13, Issue 3, 2009

Functional Status Remote patient monitoring has a clear need for including questionnaire “There are widely recognized assessments as a part of Functional Status. Assessments can range from standard question-answer sets, recognized regionally as the authoritative instruments worldwide for managing protocol for a given disease condition, to proprietary question-answer sets, chronic disease conditions such as designed by particular institutions along with customized care plans. The standard assessments are routinely used by payers and providers alike to gauge diabetes, congestive heart failure, individual patient functional status and assess overall population trending. asthma, or depression.” The HITSP Long Term Care–Assessments AHIC Gap/Extension [13] lists several assessments applicable to the United States, including the Long-Term Care Minimum Data Set (MDS), the Continuity Assessment Record and Evaluation (CARE), the Outcome and Assessment Information Set (OASIS), and the Inpatient Rehabilitation Facility-Patient Assessment Instrument (IRF-PAI). There are widely recognized instruments worldwide for managing chronic disease conditions such as diabetes, congestive heart failure, asthma, or depression. Assessments can also be utilized as part of pre-qualification and patient recruitment for clinical trials.

AHIC and HITSP acknowledged the gap in the lack of specifications related to assessments, and worked with HL7 to develop the CDA Framework for Questionnaire Assessments DSTU [8]. Standardized information, templates, guidelines, schedules, and scoring mechanisms are all needed to fully comprehend assessments within the HL7 CDA. Currently, work is divided amongst several different teams, including HITSP Consultation and Transfer of Care; Quality; and Long Term Care Assessments: all are working in conjunction with the HL7 Patient Care Work Group, the HL7 Structured Documents Work Group, and a broad array of clinical and technology industry experts. Some assessment instruments have the concept of weighted points associated “An extension of this is the concept of a with particular question responses, which can then be used to triage patients who require immediate intervention. An extension of this is the concept of a Scored Care Plan Report, with higher Scored Care Plan Report, with higher patient acuity associated with higher patient acuity associated with higher points assigned to more significant health indicators. For example, Congestive Heart Failure patients might score a 0 if they are coping well on a particular points assigned to more significant day, versus a 5 or 6 if they suddenly put on additional weight or notice swelling health indicators.” in their legs. The ability to encode this information by using standard CDA templates, constraints, and appropriate terminology would be extremely powerful to drive both analytics and clinical workflow.

Healthcare Information Integration: Considerations for Remote Patient Monitoring | 91 Intel® Technology Journal | Volume 13, Issue 3, 2009

“The framework is very thorough and The great news is that both the CDA and specializations such as the HL7 PHMR permit the addition of optional sections such as Functional Status nuanced in its handling of diverse use to represent this type of information. However, for a system to accurately cases.” consume and mediate the information, it would still require extensions to the specification. The work to date from the CDA Framework for Questionnaire Assessments DSTU [8] is outstanding in this regard, in making such rapid progress in such a short period of time. The framework is very thorough and nuanced in its handling of diverse use cases, both enabling early adoption while parts of the specification are yet to be finalized, as well as demonstrated capability to instrument very complex assessments, as evidenced by the derivative work on the Minimum Data Set Version 3 (MDSV3) [8]. We urge the SDOs to continue to aggressively pursue work in this area, essential to all forms of RPM, long-term care, and other clinical settings.

In Table 3, we show an example of how we might render assessment responses in human-readable form.

Functional Status Session date/time: 2008-01-07 13:53:04.000 Assessment Answer Score How are you feeling today compared to yesterday? Worse 5 Have you taken all of your prescribed medication Yes 0 over the past 24 hours? Did you need to take any extra diuretic medication Yes 10 in the past 24 hours? Did you need to use any extra oxygen in the past Yes, more 5 24 hours? than normal Total Score: 20

Table 3: XML-rendered Table: Functional Status Source: Intel Corporation, 2009

High priority should be placed on standardizing the templates to encode patient assessments, including terminology, along with the concept of assessment scales or scores where relevant.

Plan of Care “A Plan of Care is used to capture Multiple SDOs and HITSP working groups are planning to address Plan ‘What did we ask the patient to do?’ of Care as it relates to Consultations and Transfers of Care [5]. This is another example of the power and flexibility of the CDA architecture. whereas the combination of daily While a traditional treatment plan might be best characterized by a set of assessments and vital sign collection is multidisciplinary protocols and a follow-up planned upon hospital discharge, in RPM it takes on a different character altogether. Plan of Care includes the answer to ‘What did the patient functional and other nursing assessments (not to be confused with physician assessments of patient status, which are located in the CDA “Assessment and actually do?’ ” Plan” section).

92 | Healthcare Information Integration: Considerations for Remote Patient Monitoring Intel® Technology Journal | Volume 13, Issue 3, 2009

A Plan of Care is used to capture “What did we ask the patient to do?” whereas “We would like to see the SDOs create the combination of daily assessments and vital sign collection is the answer to “What did the patient actually do?” Thus, the combination of Plan of Care, Plan of Care section constraints that Functional Status, and Vital Signs can be used to gauge patient adherence cover both definition and scheduling to the agreed-upon treatment plan, and can be used as a basis for long-term outcomes analysis with respect to RPM. We would like to see the SDOs create of standard assessments and vital signs Plan of Care section constraints that cover both definition and scheduling collection.” of standard assessments and vital signs collection. Such constraints would be machine-consumable by different systems and encoded, by the use of international terminology standards, to promote advanced analytics.

Service Oriented Architecture (SOA) Principles Centralized Remote patient monitoring truly represents a superset of healthcare integration use cases, sitting at the hub of a complex coordination of care — across clinical specialists, home care, lab, pharmacy, hospital, and assisted living facilities.

PYREX 400 550ml 300 Chronic disease management adds a further layer of complexity, given that Lab 200 most home-care and remote monitoring systems were developed as proprietary systems with only a passing acquaintance with healthcare and technology Clinic standards. Also, it is not uncommon for these CDM systems to be used RX alongside one or more of the major commercial EMR systems. Hospital SOA provides the essential grounding, agility, and extensibility to manage and reduce complexity when integrating healthcare systems, services, and Federated information. It is adaptable to a challenging and ever-changing business climate, and represents a proven return on investment [1, 2, 11, 12]. SOA

400 provides the necessary bridge between legacy and proprietary environments PYREX 550ml 300 Lab 200 and moves us towards standards-based deployments that can scale to handle many network end points and a rich diversity of healthcare data services. Clinic In this section we outline key considerations and SOA principles when establishing an RPM solution. RX Hospital Flexible, Scalable Architecture to Support Any Deployment Model With healthcare integration, it is important to leverage architecture that is Hybrid flexible across different deployment models and data-use agreements. Figure 3 depicts three common deployment models, wherein health information is maintained in a centralized, federated, or hybrid database model. The model

400 selected largely depends upon data-use agreements in the region — whether to PYREX 550ml 300 Lab 200 maintain data centrally to a given region, remotely (federated), or as a hybrid Clinic model of the two. The centralized model is optimum both in terms of performance and access to a consistent, normalized data set, suitable for both healthcare delivery and clinical research. However, the centralized model RX Hospital requires the political will by all participants, public and private alike, to agree to centralized data sharing and data-use agreements.

Figure 3: Data Origin Flexibility Source: Intel Corporation, 2009

Healthcare Information Integration: Considerations for Remote Patient Monitoring | 93 Intel® Technology Journal | Volume 13, Issue 3, 2009

“Due to concerns over performance Due to concerns over data sharing, ownership, and local control, the federated model is frequently chosen over the centralized or hybrid. When a discharge and availability, deployment models summary or lab report is required, a record locator service query is issued from frequently shift over time from the center to each of the federated or member organizations. If the federated service is available and the appropriate agreements are in place, one or more federated to hybrid.” relevant documents are returned that pertain to the patient in question. Due to concerns over performance and availability, deployment models frequently shift over time from federated to hybrid, maintaining a small set of demographic data and centralized pointers to records maintained at the perimeter.

The federated model can be quite effective for very large, distributed data sets, but the records must be normalized at the edge by the use of agreed-upon terminology standards. The political will to use such standards can be even more challenging to achieve than that required for the centralized model. An excellent example of the federated model is the caBIG (Cancer Biomedical Informatics GRID at cabig.nci.nih.gov), which links physicians, patients, and researchers to clinical, cancer, and genomics repositories distributed worldwide in a standard normalized fashion.

“A key success factor in deploying Remote patient monitoring requires the integration of health information from a variety of disparate sources. SOA is well suited to adapt and extend to this RPM is to leverage a flexible level of service, data origin, and terminology complexity. A key success factor in architecture which can scale from deploying RPM is to leverage a flexible architecture which can scale from small institutions, to regional health information exchanges, to national networks. small institutions, to regional health Finally, the deployment model and technology selected must readily scale to information exchanges, to national processing at the core or at the edge of the network. networks.” Service Extensibility, Virtualization of End Points Traditional peer-to-peer approaches to integration lead to the N2 problem as depicted in Figure 4, in that each and every application requiring integration causes a geometric expansion of up-front cost and ongoing maintenance. In healthcare integration, the N2 problem is made manifest by the inconsistent adoption of healthcare and technology standards by legacy and proprietary systems. When a new application joins the network, each and every adapter has to be modified, in addition to the new application.

100 90 80 70 60 50 40 n = 2n = 3 30 20

n = 6 Custom Interfaces 10 2 0 “In healthcare integration, the N 12345678910 HIT Systems problem is made manifest by the inconsistent adoption of healthcare 2 and technology standards by legacy Figure 4: N Problem in Healthcare Source: Intel Corporation, 2009 and proprietary systems.”

94 | Healthcare Information Integration: Considerations for Remote Patient Monitoring Intel® Technology Journal | Volume 13, Issue 3, 2009

Integration Brokers change the cost model from geometric to linear, but have their own share of challenges, with the risk of establishing heterogeneous “islands” of integration. Integration Brokers rely upon a hub-and-spoke architecture, creating a single point of failure by routing all messaging traffic through a central hub. Only with a standardized information model, service extensibility, and virtualization of end points, provided by an Enterprise Service Bus (ESB), can one completely address the N2 problem. The heart of the 2N problem lies in the simplistic framing of integration as a “The heart of the 2N problem lies in two-dimensional use case. When we frame the exchange of health information as a simple, bidirectional exchange between a total of two points, we obfuscate the simplistic framing of integration as the actual complexity involved. In reality, the RPM exchange requires multiple a two-dimensional use case.” data sources, or network end points, in order to complete the CDM view of the patient, including vital signs, assessment responses, functional status, lab results, prescriptions, diet, exercise, treatment plan, and clinical notes, to name just a few. The information needs to be addressable by using a standardized information model, and over time, each of the data services should be exposed by using a standard set of query and retrieval methods.

A service network architecture allows for building the “on-ramp” once, with no adapter maintenance required, as other applications join or leave the network. Service extensibility serves to virtualize the end points, abstracting the details of transport protocols and peer-to-peer connections. Services can be dynamically registered, discovered, and rerouted to scale as performance and reliability needs dictate.

Network Compute Model The HL7 v3 CDA Release 2 [4] constrains the v3 RIM and leverages the full “Services can be dynamically richness of its healthcare informatics model and standardized terminology, delivering computable, healthcare information as well as human-readable registered, discovered, and rerouted to clinical documents. By first composing all of the telehealth data to the HL7 v3 scale as performance and reliability CDA Release 2 (i.e., the network on-ramp in Figure 5), it becomes a repeatable exercise then to perform any secondary transforms to legacy or proprietary needs dictate.” messaging protocols and local terminology (i.e., the network off-ramp).

Healthcare Network Service using On-Ramp HL7 CDA Healthcare Service Network Canonical Transform

Canonical

Figure 5: Service Network Architecture Source: Intel Corporation, 2009

Healthcare Information Integration: Considerations for Remote Patient Monitoring | 95 Intel® Technology Journal | Volume 13, Issue 3, 2009

This integration pattern accelerates adoption of the latest healthcare informatics standards, while lowering the barriers of adoption for smaller organizations that need to proceed at their own pace.

A robust network informatics exchange model can be used to establish trust at the network level — a healthcare dial tone. Peer systems may validate, authenticate, and audit the encrypted XML payloads at any point in the network. Moreover, the network compute model enables the ability to route, compose, and decompose fine-grained business objects according to national and regional, legal and regulatory, privacy, data protection, and patient consent requirements.

Pluggable Services in the Cloud “Since the service location is SOA provides the ability to leverage services available within the data center or across the Internet. New services can be brought online and directly utilized by virtualized, and the service network participants, without requiring additional modifications to each end implementation is encapsulated, point. Since the service location is virtualized, and the service implementation is encapsulated, a service can be readily created or replaced without impacting a service can be readily created or existing service consumers. The OMG/HL7 Healthcare Services Specification http://hssp.wikispaces.com/ replaced without impacting existing Project (HSSP at ) is working to define standard Web service methods to access critical healthcare infrastructure services, service consumers.” including entity identity, controlled terminology, record locator, decision support, and clinical research filtered query services. Similarly, there is a need for advanced healthcare data services, such as drug interaction checks, adverse event reporting, clinical trial recruitment, and public health reporting. SOA design methodology allows for incremental implementation, at first utilizing simple data match routines and then, when the complexity of exchange dictates, readily switching to industry-strength identity and terminology services, all without changing the service interface.

Deployment Considerations “What is required, however, is deep Effective RPM deployment requires the application of industry best practices with respect to information technology, data center operations, enterprise healthcare domain expertise, a keen service delivery, and robust security and privacy measures. The technical sense of customer requirements, and an challenges in healthcare are not insurmountable; rather, they can be solved by using well-known solutions and design patterns. What is required, however, is understanding of the context in which deep healthcare domain expertise, a keen sense of customer requirements, and an understanding of the context in which the system will be used. In order to the system will be used.” fully realize the potential of RPM, we must arrive at the right combination of technology and process.

96 | Healthcare Information Integration: Considerations for Remote Patient Monitoring Intel® Technology Journal | Volume 13, Issue 3, 2009

Patient Acuity and Mode of Healthcare Delivery Patient acuity and the concomitant mode of healthcare delivery are arguably “The mode of healthcare delivery tends the most important determinants of RPM requirements. Patient acuity determines the level of monitoring and likelihood that intervention will be to be matched to patient acuity and required, the frequency of data collection, the criticality of reference ranges provider service level agreements.” and alerting mechanisms, and the relative intolerance for data latency. The mode of healthcare delivery — whether it be a wellness coach, a visiting nurse, community outreach, an assisted living facility, hospital discharge monitoring to avoid readmission, population management, or patient-empowered healthcare — tends to be matched to patient acuity and provider service level agreements. Contact with the patient, and the relative need to process the clinical findings, will therefore range from occasional, monthly, quarterly, daily, hourly, or perhaps even more frequent when a sudden decline in health is detected. The duration of RPM deployments could be measured in weeks in the case of hospital discharge monitoring, months in the case of high-risk pregnancies, or years when monitoring elderly patients with co-morbidities.

Data Latency and Modes of Transmission As discussed earlier, tolerance for data latency is largely determined by patient “Alerts and notifications can be acuity of the target population. Relatively healthy patients, coupled with wellness coaches, can tolerate high data latency with summary reports over generated via real-time, event-driven periods as much as a few months at a time. High-acuity patients require triggers, whereas batch operations more frequent monitoring, with data latency approaching near real-time. High data latency can readily be accommodated by scheduled, file-oriented, and monthly summary reports can and store-and-forward processes to perform data integration. In contrast, be scheduled to occur during off-hour low data latency requires end-to-end integration performed via Web services every few minutes, where the incremental transmission is closer to a single processing.” patient session, containing the latest raw measurements and assessment responses, rather than a complete trend analysis of the past period. Alerts and notifications can be generated via real-time, event-driven triggers, whereas batch operations and monthly summary reports can be scheduled to occur during off-hour processing.

Volume and Quality of Data RPM involves significantly more data than what is typically anticipated for an EHR, such as a clinical encounter. Patients may be instructed to take vital sign measurements multiple times per day in addition to responding to various assessment questions. One of the chief areas of ongoing investigation is the optimal level of summarization of the PHMR. Different clinicians will likely want a full range of options, from daily, monthly, or quarterly to a filtered “Different clinicians will likely want summary, depending upon patient acuity and the mode of healthcare delivery. a full range of options, from daily, monthly, or quarterly to a filtered summary, depending upon patient acuity and the mode of healthcare delivery.”

Healthcare Information Integration: Considerations for Remote Patient Monitoring | 97 Intel® Technology Journal | Volume 13, Issue 3, 2009

Once the report design is optimized, additional adaptations may be necessary to the recipient system in order to fully leverage the additional rich data types, process alerts, and triage patients based on clinical findings. In particular, it is unlikely that recipient systems are prepared to work with patient-specific reference ranges, threshold violations, assessment questionnaires, or weighted scores for industry standard protocols. Some systems are unprepared to process datetime stamps on individual measurements, because of the prior exclusive focus on clinical encounters in office settings. In other words, while a given office medical system might record the date of the office visit, it rarely records the time of an individual measurement. While the concept of an RPM “patient session” can be likened to an office visit, the recipient medical system is unprepared to process the sheer volume of RPM sessions and measurements.

“All points along the end-to-end data As clinical systems and processes evolve to process data from RPM, the need for more sophisticated methods of patient triage, alert, and notification will also be flow should be instrumented and required. For example, a large number of normal readings from a moderately- sized patient population will quickly outpace the most efficient of care-manager monitored for effective operation and organizations, if workflows require the manual acknowledgement of all patient safety.” readings, rather than triage based on out-of-range measurements. Conversely, if an organization becomes overly dependent upon the direct integration of RPM data without also developing adequate means for systems monitoring, an undetected outage or transmission failure might inadvertently create a false impression that patient readings are within normal limits. It is critical, therefore, to develop adequate systems monitoring and failsafe methods. For example, reports should be appropriately annotated with synchronized datetime stamps, indicating both the time of report generation and the time of the last data transmission. All points along the end-to-end data flow should be instrumented and monitored for effective operation and patient safety.

“We have identified the need to define Threshold violations, especially life-threatening ones, need to trigger specific workflows that are customizable by practice, by co-morbidities, by target tiers of thresholds to separately drive populations, and by individual patients. We have identified the need to define patient and clinician workflows and tiers of thresholds to separately drive patient and clinician workflows and modes of intervention, ranging from patient education, to clinician referral, modes of intervention.” to emergency hospital admission. There is also the need to capture both the trigger event and the clinical intervention as part of analytics. For example, a patient’s oxygen saturation falls dangerously low, which triggers an alert and results in some form of patient intervention — whether a phone call, an SMS text message, a video conference with a clinician, or a house call from a visiting nurse. The clinical intervention may result in a change in protocol, a lab order, a medication change, or hospitalization. Each of these events needs to be associated with a standard measure of outcomes in order to support analytics for evidence-based medicine and drive further improvements to healthcare.

98 | Healthcare Information Integration: Considerations for Remote Patient Monitoring Intel® Technology Journal | Volume 13, Issue 3, 2009

Clinician Workflow and Reimbursement Model Another important consideration in the integration of RPM information is “Annotating the data stream with the clinician workflow and the reimbursement model. Some jurisdictions require that a clinician manually review and accept each and every measurement prior source and method of reporting helps to importing the data into the institutional EHR. This follows standard clinical to account for these differences in practice of signing off when reviewing external lab results, yet again, the volume of data is fundamentally different when considering RPM. policies.”

Some institutions extend the system boundary of the EHR to encompass any automated data capture but draw the line at information that is patient-reported or otherwise manually entered, such as a PHR containing diet and exercise journal entries. Data that are grandfathered in as automatic data capture might not require the manual approval step, whereas patient-reported data may be reviewed but perhaps not incorporated into the institution’s legal record. Annotating the data stream with the source and method of reporting helps to account for these differences in policies.

Accommodations are required for both the level of summary and raw information in a given report. Streamlined mechanisms are required to process messages through the clinical inbox, along with careful consideration as to what level of clinical staff might be able to process what level of data on behalf of the doctor, in order to potentially offload this manual step.

Further, the clinical reimbursement model is frequently called into question “Some reimbursement models attempt with respect to RPM. Some reimbursement models attempt to equate RPM with an office visit, while others only reimburse when the patient establishes to equate RPM with an office visit, and maintains good tolerance of pre-established thresholds. Each of these while others only reimburse when the considerations will have an impact on the rate of adoption of RPM, especially when combined with additional processing overhead on the part of the patient establishes and maintains good physicians to periodically review the results. tolerance of pre-established thresholds.” Trans-border Data Flow Considerations Careful attention to detail will be required for any deployment in which integration is planned across borders of state/provinces, regional, or national boundaries. Privacy and data protection laws are rapidly changing worldwide, with significant penalties for mishandling of data and breach of privacy. Advanced workflow, transformation, and routing engines will be required to comply with local data protection regulations and policies. Special consideration is due when determining where to locate a primary or alternative “Privacy and data protection laws are data center hosting patient data, since a number of countries require that protected health information (PHI) not cross national boundaries. To manage rapidly changing worldwide, with and track patient consent and negotiate appropriate data use, business associate and data controller/supplier agreements are all essential, regardless of whether significant penalties for mishandling or not information crosses any recognized governmental boundaries. of data and breach of privacy.”

Healthcare Information Integration: Considerations for Remote Patient Monitoring | 99 Intel® Technology Journal | Volume 13, Issue 3, 2009

Identifying Systems of Record “While clinical data might be easily Identifying a single system of record (i.e., a single authoritative source for each and every data element) is essential to any successful integration project segregated between different systems and frequently overlooked in applications such as RPM. It is typical for even of record, it is highly likely that every small organizations to already have multiple systems in place for purposes of chronic disease management, population management, a primary EHR data system maintains its own copy of repository, a separate system for lab results, etc. The addition of telehealth data likely represents the addition of one or more systems to an already complex and demographic data.” disorderly mix.

A key area of concern is patient and clinician demographics. While clinical data might be easily segregated between different systems of record, it is highly likely that every system maintains its own copy of demographic data. Considerations must be given both from a systems and a workflow perspective to demographics synchronization, import, and ongoing maintenance. The older systems likely do not have a method to disable manual edits to demographics, yet one must ensure that clinicians are always working from an authoritative source of patient and clinician demographics and contact information.

“Provisions are made to disable In an advanced integration deployment, the demographics system of record updates the recipient systems with the latest information, including translation manual edits in the recipient systems, of identity to the recipient system, via an entity identity service (EIS). or at least ensure that processing Provisions are made to disable manual edits in the recipient systems, or at least ensure that processing detects, logs as an exception, and overrides any detects, logs as an exception, and unauthorized changes. overrides any unauthorized changes.” In mixed environments with both legacy and newer systems, a complex scheme of automated demographics integration along with a carefully designed business process is required. A central system can be configured to synchronize demographics to each of the recipient systems. When manual edits in each system cannot be disabled, they must be controlled via business process, training, and careful oversight to ensure that changes in demographics are only entered into the central system. Common identity mismatch errors tend to require a small staff to resolve and maintain on an ongoing basis.

As healthcare systems integration becomes more complex, encompassing multiple end points and service providers, each with their own independent systems of record, it becomes paramount to employ an industry-strength EIS to accurately address the identity match problem. Industry leaders in EIS leverage advanced stochastic algorithms for matching against multiple demographics attributes to disambiguate identity. The OMG/HL7 Healthcare Specifications “Common identity mismatch errors Services Project (HSSP) is working to define standard Web service interfaces for common capabilities like EIS, such that different commercial services may be tend to require a small staff to resolve deployed without requiring a change to the implemented interface. and maintain on an ongoing basis.”

100 | Healthcare Information Integration: Considerations for Remote Patient Monitoring Intel® Technology Journal | Volume 13, Issue 3, 2009

Customization Requirements There are a number of deployment considerations when establishing a remote “While anything is technically feasible, patient-monitoring solution. The AHIC use cases are foundational to defining routine healthcare interactions and processes, required data elements, and it is not practical for a business to terminology constraints to standardize the exchange. It is also important to develop a system to be all things to all distinguish the areas of local variation and future development, such as those identified in the RPM sequence and interaction diagrams (Figures 1 and 2, people.” respectively). It is important to control the level of customization required for any given solution to something that delivers real value to the customer, both in the short-term and in the reasonable future, yet is practical enough to represent low maintenance over time. While anything is technically feasible, it is not practical for a business to develop a system to be all things to all people. It is critical to establish up front some of the key business drivers, including patient acuity, target mode of healthcare delivery, and the relative tolerance for data latency. From this baseline, customization mechanisms can be established to allow for local variation, leveraging codeless configuration changes to metadata, rather than requiring a code recompile and system overhaul for each deployment.

Conclusion Remote patient monitoring represents a critical intersection of healthcare information integration, embodying the need for healthcare informatics standards, careful consideration of workflow and system deployment tradeoffs, and direct engagement with the patients and clinicians who work with the system. The use of HL7 CDA helps to accelerate adoption of healthcare standards by properly constraining the rich schema and vocabulary of the RIM, and lowering the barriers of entry through incremental evolution. The use of SOA design principles enables us to respond to changes in business drivers and adapt to the complexity of healthcare integration end points. Terminology standards drive computable information which in turn enables advanced analytics, clinical research, and transformational healthcare delivery. We are actively working with the SDOs in pursuit of future enhancements to the existing healthcare informatics standards.

Healthcare Information Integration: Considerations for Remote Patient Monitoring | 101 Intel® Technology Journal | Volume 13, Issue 3, 2009

References [1] Bass, Christy and Lee, J. Michael. 2002. “Building a Business Case for EAI.” eAI Journal, January 2002, pp. 18-20.

[2] Booz Allen Hamilton. 2005. “Canada Health Infoway. Pan-Canadian Electronic Health Record: Projected Costs and Benefits.” March 2005. At www.infoway-inforoute.ca.

[3] Continua Health Alliance. 2009. “Continua Design Guidelines.” V1.0, June 2009. At www.continuaalliance.org.

[4] Dolin R.H., Alschuler L, Boyer S, Beebe C, Behlen FM, Biron PV, Shabo A, (Editors). 2005. “HL7 Clinical Document Architecture, Release 2.0. ANSI-approved HL7 Standard.” May 2005. Ann Arbor, Michigan: Health Level Seven, Inc. At www.hl7.org.

[5] Healthcare Information Technology Standards Panel (HITSP). 2008. “Consultations and Transfers of Care Interoperability Specification.” HITSP/IS09. Released for Implementation. 20081218 V1.0. At www.hitsp.org.

[6] Healthcare Information Technology Standards Panel (HITSP). 2008. “Remote Monitoring Interoperability Specification.” HITSP/IS77. Released for Implementation. 20081218 V1.0. At www.hitsp.org.

[7] Healthcare Information Technology Standards Panel (HITSP). 2009. “Remote Monitoring Observation Document Component.” HITSP/ C74. Released for Implementation. 20090708 V1.1. At www.hitsp.org.

[8] Health Level Seven (HL7). 2009. “Implementation Guide for CDA Release 2: CDA Framework for Questionnaire Assessments (Universal Realm) Draft Standard for Trial Use (DSTU) Release 1.0.” April 2009. At www.hl7.org.

[9] Health Level Seven (HL7). 2007. “Implementation Guide: CDA Release 2 – Continuity of Care Document (CCD).” April 01, 2007. At www.hl7.org.

[10] Health Level Seven (HL7). 2009. “Implementation Guide for CDA Release 2.0 Personal Health Monitoring Report (PHMR) Draft Standard for Trial Use (DSTU) Release 1.” At www.hl7.org.

[11] OMG SOA Consortium, CIO Magazine. 2008. “OMG SOA Consortium and CIO Magazine Announce Winners of SOA Case Study Competition.” At www.soa-consortium.org.

102 | Healthcare Information Integration: Considerations for Remote Patient Monitoring Intel® Technology Journal | Volume 13, Issue 3, 2009

[12] Schindler, Esther. 2008. CIO Magazine. “Service-Oriented Architecture Pays Off for Synovus Financial.” September 30, 2008. At www.cio.com.

[13] U.S. Dept. of Health and Human Services (HHS), Office of the National Coordinator for Health Information Technology (ONC). 2008. “Long Term Care–Assessments. AHIC Extension/Gap.” December 31, 2008. At www.hitsp.org.

Acknowledgements Special thanks to Dr. Robert H. Dolin, co-chair of the HL7 Structured Documents Committee, who assisted us in developing our initial CDA models.

Author Biography Kristina M. Kermanshahche is Chief Architect for Intel’s Digital Health Group. Her research interests include SOA as applied to regional and national healthcare architectures, CDM from home health to clinical trials, and translational biomedical informatics. She was the Lead Architect for the Intel SOA Expressway and contributed to the Intel Mobile Clinical Assistant and the Intel Health Guide; she has a patent pending in healthcare semantic interoperability. Her twenty-five year career in software development spans nearly every industry, with an emphasis on complex distributed systems, multithreading, high availability, networking, and database design. She received her B.A. degree in 1988 from the University of Oregon in Political Science and International Studies, magna cum laude. Her e-mail is Kristina.M.Kermanshahche at intel.com

Copyright Copyright © 2009 Intel Corporation. All rights reserved. Intel, the Intel logo, and Intel Atom are trademarks of Intel Corporation in the U.S. and other countries. *Other names and brands may be claimed as the property of others..

Healthcare Information Integration: Considerations for Remote Patient Monitoring | 103 Intel® Technology Journal | Volume 13, Issue 3, 2009

Personal Health Device Interoperability

Contributors

Douglas P. Bogia Abstract Intel Corporation Healthcare costs continue to spiral upwards to the point of prompting national mandates for change in the whole healthcare system. Technology has Rick Cnossen advanced to a point where personal telehealth systems provide viable, cost- Continua Health Alliance effective solutions and represent a very real opportunity to help control costs. In order for deployment to become widespread, an ecosystem of standards- Chris Gough based interoperable components (starting with the consumer-facing device) is Intel Corporation essential. In this article, we discuss the challenges related to creating such an ecosystem and examine the efforts of the Continua Health Alliance to address this need.

Introduction Several factors have converged to provide new opportunities to address user needs in the healthcare domain:

Index Words •• The dramatic rise in costs associated with established healthcare services. •• A global increase in the number of older adults who are most likely to Continua Health Alliance require these services. Healthcare Technological advances, such as innovations in networking technologies and Interoperability a rapid increase in the number of Internet-connected devices, enable the Standardization development of solutions that address user needs in a cost-effective manner. Communication These technologies also allow people to remain safely in their own homes for a Cost Reduction longer period of time. Personal telehealth systems composed of an ecosystem of standards-based interoperable components are the building blocks of these solutions. In this context, we define personal as systems that are used in a home, and we define telehealth as the way in which communication technologies, such as a phone line or broadband Internet access, are used in healthcare services.

In this article, we describe in more detail the underlying need for interoperability between these components, the approach that was used by the Continua Health Alliance [1] to define and design such a system, and some of the key challenges that were encountered along the way.

“Personal telehealth systems composed of an ecosystem of standards-based interoperable components are the building blocks of these solutions.”

104 | Personal Health Device Interoperability Intel® Technology Journal | Volume 13, Issue 3, 2009

Need for Device Interoperability While there are many challenges associated with the successful design, implementation, and deployment of personal telehealth systems, one of the more obvious problems present in early telehealth solutions was a lack of device interoperability or well-defined standards to which technology companies could develop their products. In the next sections we discuss these issues in more detail from both the integrator and purchaser perspective as well from the product designer perspective.

Integrator and Purchaser Telehealth stakeholders, both system integrators (companies producing “The presence of many component telehealth solutions made up of components from a number of different vendors) and system purchasers (healthcare providers that purchase these vendors competing for business drives solutions and offer them to their patients or members) require a wide variety down costs, improves the quality of of system vendors and components to select from. These stakeholders produce or consume, respectively, a large number of telehealth systems and need to solutions, and encourages adding protect themselves from undesirable situations, such as a component vendor differentiating features to these going out of business or being unable to deliver the desired volume of system components in a given timeframe. Furthermore, the presence of many solutions.” component vendors competing for business drives down costs, improves the quality of solutions, and encourages adding differentiating features to these solutions.

However, the ability for integrators or purchasers to select from a wide variety of personal health devices offered by multiple vendors is only possible if the devices are compatible with one another. This compatibility is enabled by well- defined communication interfaces between the various system components.

Product Designer From the perspective of a product designer, device interoperability is also an “Without interoperable standards, a essential need. For instance, without interoperable standards, a system designed to communicate with a wide range of telehealth peripherals (for example, system designed to communicate with weight scales, blood pressure monitors, glucose meters) must implement proprietary data modeling and communication protocols over a variety of a wide range of telehealth peripherals transport technologies , such as serial, Bluetooth*, and infrared. must implement proprietary data Product designers need to create a device-specific software adapter such that modeling and communication their system can communicate with each of the peripherals. This leads to inefficiencies in terms of both software development and system validation. protocols over a variety of transport Likewise, designers of peripherals must design data and communication technologies.” protocols. Since these are specific to their device, the implementation of device driver(s) for target operating system(s), user applications with GUIs, and documentation is necessary.

Personal Health Device Interoperability | 105 Intel® Technology Journal | Volume 13, Issue 3, 2009

“The necessary, interoperable solution The necessary, interoperable solution minimizes cost, improves design and development efficiencies, and enables separation of concerns for each of the minimizes cost, improves design and parties involved. Consequently, device vendors can focus on devices, software development efficiencies, and enables vendors can focus on software development, and service providers can focus on service delivery, without requiring each party to have intimate knowledge separation of concerns for each of the of every component that makes up the overall solution. Such an interoperable parties involved.” solution can be achieved by defining communication interfaces in an unambiguous set of standards and guidelines.

Challenges to Standardization There are many challenges and barriers to standardization of device interfaces in the personal health domain. In this section we explore some of the more difficult ones.

Confluence of Domains “It is imperative to converge on a The scope of the personal health domain is quite large and ranges from regulated clinical disease management devices to unregulated consumer-grade single unified solution yet difficult to fitness devices. From an architectural and practical standpoint it is imperative define a single solution that is both to converge on a single unified solution yet difficult to define a single solution that is both efficient and optimum for all options in the spectrum. efficient and optimum for all options Regulated Environment Versus Consumer Environment in the spectrum.” Devices that are intended for a regulated environment (for example, US FDA Class 2 Glucose Meters for a medical professional to treat diabetes) and consumer grade devices (for example, a personal pedometer to track activity levels) are starkly different. These differences show up in the requirements (security levels, privacy restrictions, data reliability, and logging) as well as in the level of testing required (stress testing, off-nominal conditions, etc.). These differences make the decision-making process for interoperability standards difficult as there are always tradeoffs.

Diversity of Applications The current state of the personal health market is such that there are independent vertical application solutions (for example, health and fitness, disease management, and aging independently). The broad vision for an interoperable solution is to eliminate these divisions in order to allow for innovative solutions that span domains and create optimum value for the consumer. For example, many consumers of disease-management applications could benefit from an integrated application that accepts fitness data in order to evaluate the effectiveness of a particular exercise against a particular disease. The challenge is in creating a generic framing protocol that can be applied to “The challenge is in creating a generic all application areas in order to normalize collected data. This can be difficult as different domain applications have different approaches for some of the framing protocol that can be applied common elements. For example, the timestamp granularity for disease- to all application areas in order to management applications tends to be higher than for aging-independently applications. normalize collected data.”

106 | Personal Health Device Interoperability Intel® Technology Journal | Volume 13, Issue 3, 2009

The Nascent Personal Health Industry The personal health industry is relatively young and lacks mature standards- “It takes a notoriously long time for based solutions for interoperability. This necessitates developing new standards or profiles in many cases instead of using existing standards that have been used the standards to be created, accepted and validated over time. It takes a notoriously long time for the standards to by the industry, augmented with tools, be created, accepted by the industry, augmented with tools, and understood by engineers. Such a lengthy process runs counter to the desire to quickly define and understood by engineers.” solutions for industry build-out.

International Need With the ubiquity of the Internet, common communication advances, multi- national companies, and global markets there is an opportunity and a strong desire to create solutions that can be deployed internationally. International deployment can capitalize on a company’s initial development investment and increase overall revenue by expanding market opportunities. Within the international standards organization, these solutions require geographic differences among standardization processes to be harmonized. Harmonization complicates and slows down the standardization process. For example, US- “The United States approach tends centric solutions tend to use US Standard Development Organizations (SDOs) to be industry-driven, while the such as IEEE, ANSI, and W3C, while EU-centric solutions tend to use EU SDOs, such as CEN, ETSI, and CENLEC. There are some fundamental European Union approach tends to differences in approaches also. For example, the United States approach tends be more government-driven.” to be industry-driven, while the European Union approach tends to be more government-driven. These differing approaches impact the process and must be reconciled. Additionally, there are regional and demographic differences that must be overcome. For example, demographics data (address, gender) vary among countries, and a common structure must be defined to allow for all system instantiations.

Business Motivation Despite the best intentions, there are many existing standards that are incomplete or unused. The bottom line for many companies is that investing in the development and implementation of standards must benefit them to some degree economically. For new, nascent markets, such a return on investment can be difficult to project and demonstrate which, in turn, can lead to a dearth of volunteers or to slow uptake in product build-out.

The Continua Health Alliance Solution This section introduces the Continua Health Alliance and describes how it “For new, nascent markets, such a addressed the challenges and issues described in the preceding sections. return on investment can be difficult to project and demonstrate which, in turn, can lead to a dearth of volunteers or to slow uptake in product build-out.”

Personal Health Device Interoperability | 107 Intel® Technology Journal | Volume 13, Issue 3, 2009

Approach The Continua Health Alliance is an open industry coalition of over 200 healthcare and technology companies focused on interoperability in the personal healthcare domain. Founded in 2006, the Continua Health Alliance addresses the unique combination of challenges just discussed. The Continua Health Alliance leveraged examples from other solution domains such as home networking (for example, Wi-Fi Alliance and Digital Living Network Alliance) in order to help define its overall approach. The Continua Health Alliance adopted the following methodology: •• Select an existing, applicable set of industry standards. •• Extend these selected standards where required to meet user needs identified in use-cases and requirements. •• Eliminate ambiguity in interpretation of these standards through a collection of interoperability guidelines. In the following sections we describe the process, architecture, and final implementation that address the challenge of enabling an ecosystem of interoperable, standards-based personal-healthcare devices.

Process The Continua Health Alliance process begins with use-case collection and refinement. After agreeing upon the use-cases, requirements are elaborated upon and architecture is defined. Existing standards are then reviewed to determine if a standard already exists that meets most, if not all, of the requirements. If so, it is selected. If not, a new one is created. Next, the Continua Health Alliance creates interoperability guidelines to define profiles over the standards to serve as a basis for product certification. To assist companies that make products adhering to the guidelines, the Continua Health Alliance hosts interoperability plugfest events to ensure products from multiple vendors work together. The Continua Health Alliance also establishes a certification and testing program that ensures product compliance with the guidelines. It certifies that products pass the certification and testing program, and if they do, the products may use the Continua Health Alliance logo.

Use-Case Collection With more than 200 companies in the Continua Health Alliance, converging on a set of priority use-cases for each iteration can be a challenge.

To deal with this challenge, the Continua Health Alliance maintains a rigorous process that includes the following steps:

•• Crisply define candidate use-cases by using a template that includes title, date, version, description, scope, actors, preconditions, flow, and technical feasibility. •• Consolidate and generalize use-cases, segmenting work along market boundaries in order to encourage participation in areas of expertise and business interest. This eliminates redundancy and brings like-minded companies together to truly describe a consolidated industry need.

108 | Personal Health Device Interoperability Intel® Technology Journal | Volume 13, Issue 3, 2009

•• Review use-cases to carefully constrain scope (for example, consider product instantiations, resourcing, schedule) and appropriate focus (for example, ensuring component interfaces are defined but do not overly address product capabilities as the interfaces are opportunities for differentiation). •• Include a review cycle with domain experts and consumers to ensure use- cases represent real and useful solutions. Requirements Elaboration Once the use-cases are complete, they are decomposed into system requirements. This process entails analyzing the use-case to pull out actors, actions, and assumptions and converting these into corresponding requirements.

Standard Selection and Development Once the requirements are defined, members of the Continua Health Alliance “In cases where there is no standard, search for existing SDOs and standards that best satisfy the requirements and desired characteristics. In cases where there is no standard, the members must the members must select an SDO for select an SDO for hosting a new standard and apply resources to create the new standard. The following questions are considered to evaluate the selected hosting a new standard and apply standard: resources to create the new standard.” •• Does the standard adequately address the requirements from the selected use cases? •• Does the SDO create international standards or a path to generate them? •• Is the SDO open for participation from member companies? •• If a standard exists, how well is it harmonized with related domain standards? If standards do not exist, how open is the SDO to harmonizing with other standard organizations? •• How is the standard accessed and what are the control mechanisms for the standard from the SDO? •• What are the intellectual-property (IP) rights for standards from the SDO? •• What are the costs pertaining to licensing, implementation, and access? •• Does the SDO provide tool support for creating or testing implementations of their standards? •• What is the level of adoption and maturity of standards from the SDO? Guidelines Development Once the standards are available, Continua Health Alliance members compare the standard against the identified system requirements to identify remaining gaps. The Continua Health Alliance creates interoperability guidelines to address these gaps, expand on the standards, minimize optionality, and “The Continua Health Alliance creates facilitate tight interoperability. interoperability guidelines to address these gaps, expand on the standards, minimize optionality, and facilitate tight interoperability.”

Personal Health Device Interoperability | 109 Intel® Technology Journal | Volume 13, Issue 3, 2009

“The lab uses an automated test Test Plans and Tools After standards and the interoperability guidelines are completed, the Continua tool, a system simulator, and mature Health Alliance creates test assertions and procedures. The test procedures certified reference devices to verify verify that a product adheres to the appropriate standards and guidelines. All procedures can be run manually; however, as many as possible are also interoperability.” automated to reduce the level of product certification effort.

Certification Lab When a product is completed, it is tested in a certification lab. The device is run through various procedures to receive certification. The lab uses an automated test tool, a system simulator, and mature certified reference devices to verify interoperability. Once testing in the lab produces a satisfactory result, a certification can be issued and the product can use the Continua Health Alliance logo for marketing purposes.

Architecture “The architecture defines the subsystems One of the key activities that goes hand-in-hand with the collection of requirements discussed in the preceding Requirements Elaboration section that make up the overall system and is the definition of the system architecture. The architecture defines the subsystems that make up the overall system and describes how they interact. describes how they interact.” Just as important, it provides a useful model for organizing the requirements and offers guidance to engineers who look for applicable standards and technologies during the subsequent implementation phase.

Early on in the process of designing the Continua Health Alliance Device Architecture, there were two critical challenges that needed to be addressed:

•• Resolving differing viewpoints from a large number of member companies and participants. •• Describing the architecture in a manner that provided enough technical guidance to contributors working on the follow-on implementation, but could still be comprehended by non-technical stakeholders such as those in marketing, and also by usability experts. The first challenge was solved in large part by following a relatively simple process: a comprehensive set of standard architecture principles was selected, and referenced terms were defined (for example, extensibility, portability, time to market, and scalability). The team came to the realization that the system architecture could look quite different, depending on how these principles were prioritized, so each member company furnished a prioritized list that “A comprehensive set of standard effectively served as their ballot. The lists were compiled, and the resulting unified prioritized list of architecture principles served as a useful tool during architecture principles was selected, architecture sub-team meetings where various design alternatives and tradeoffs were discussed. and referenced terms were defined (for example, extensibility, portability, time to market, and scalability).”

110 | Personal Health Device Interoperability Intel® Technology Journal | Volume 13, Issue 3, 2009

The second challenge primarily arose from the fact that the team needed Application Hosting Device to create a logical view of the system architecture composed of logical (or Control Application functional) components, but it was much easier to describe concepts to a non- technical audience in terms of physical components (for example, a weight scale or a personal computer). The solution in this case was to create a hybrid view of the architecture that represents both logical and physical entities, as Service Virtualization Middleware shown in Figure 1. PAN Client Component In the following subsections we look more specifically at the architecture depicted in Figure 1.

Personal Area Network Devices PAN Interface A personal area network (PAN) device is a real-world physical device that a PAN Service person interacts with to obtain a measurement, for example. Examples of PAN Component devices include blood glucose meters, weight scales, and pedometers. Note that PAN Device the term PAN device is synonymous with the term agent introduced later in the Implementation section. Physical Device Logical Component

PAN Service Component A PAN device must contain a PAN service component. This is a logical Figure 1: Continua Device Architecture component that offers one or more services over the PAN interface. Source: Continua Health Alliance, 2009 Application Hosting Device An application hosting device is a physical device that communicates with one or more PAN devices to receive measurements, for example. Examples of application hosting devices include purpose-built, healthcare appliances, personal computers, and cell phones. Note that the term application hosting device is synonymous with the term manager introduced further on in the Implementation section.

PAN Client Component An application hosting device must contain a PAN client component. This is a logical component that consumes one or more services over the PAN interface.

Service Virtualization Middleware Service virtualization middleware is a logical component that normalizes “Examples of application hosting details associated with a number of PAN client components, such that control applications can consume PAN services from a wide variety of devices in a devices include purpose-built, consistent fashion. For example, if a system has two PAN devices, a Bluetooth healthcare appliances, personal blood-pressure monitor and a USB weight scale, control applications would receive measurements through the unified interface offered by the virtualization computers, and cell phones.” middleware without requiring detailed understanding of the specific transport technologies involved.

Control Applications Control applications consume the interface implemented by the service virtualization middleware to communicate with a wide variety of PAN service components.

Personal Health Device Interoperability | 111 Intel® Technology Journal | Volume 13, Issue 3, 2009

PAN Interface PAN 1…* 1 Application Device Hosting Device The PAN interface is the standards-based communication interface between PAN service components and PAN client components that enables device 11interoperability in the Continua Health Alliance ecosystem.

Figure 2 shows the multiplicity of relationships between the various 1…* 1…* components that make up the Continua device architecture. A PAN device PAN Service 11PAN Client Component Component contains one or more PAN service components. An application hosting device contains one or more PAN client components. There is a one-to-one Physical Device Logical Component relationship between PAN service components and PAN client components. There is a one-to-many relationship between an application hosting device and a PAN device. Figure 2: System Multiplicity Source: Intel Corporation, 2009 We now move on to describe how the PAN interface was realized during the implementation phase of the Continua Health Alliance development process.

Implementation There are multiple standards that are utilized when implementing the PAN solution. Figure 3 depicts the standards and their relationships.

Continua Design Guidelines

Device Specialization Layer

04 – Pulse 07 – Blood 08 – Thermometer 15 – Weighing Oximeter Pressure Scale

17 – Glucose 19 – Insulin 21 – Peak Flow +Phase III Meter Pump

41 – Cardiovascular Fitness & Activity 42 – Strength Fitness Equipment

71 – Independent Living Activity Hub 72 – Medication Monitor

Optimized Data Exchange Protocol Layer Higher Layer ISO/IEEE Std 11073–20601 Protocol Lower Transport Layer Layer Protocol Bluetooth* Health USB Personal ZigBee* Personal Device Profile Healthcare Device Healthcare Device

Figure 3: Standards for PAN Interface Source: Continua Health Alliance, 2009

112 | Personal Health Device Interoperability Intel® Technology Journal | Volume 13, Issue 3, 2009

The higher-layer protocols are standardized by the International Standards “The highest-layer protocols are Organization (ISO) and the Institute of Electrical and Electronics Engineers (ISO/IEEE). The higher-layer protocols consist of a common, optimized data specializations that tailor the usage of exchange protocol that provides a broad toolbox of interactions in support of the toolbox to specific device types.” any type of device. The highest-layer protocols are specializations that tailor the usage of the toolbox to specific device types (such as to a weighing scale) in order to maximize interoperability between the device sensing the health data and the one receiving the data. Each of the device specializations begins with ISO/IEEE Std 11073-104zz where zz is the two digit number in each of the boxes in Figure 3.

The lower transport layers are specified in other organizations such as Universal Serial Bus (USB) Implementers Forum, Bluetooth Special Interest Group (SIG), or ZigBee* Alliance. However, the IEEE Personal Health Device (PHD) work group has worked closely with the other organizations to harmonize the application levels with the transport levels.

As a backdrop to all the standards, the Continua Health Alliance produces “One key goal was to support a broad design guidelines that further clarify and if necessary, narrow optionality to increase interoperability. Again, there is a close working relationship between range of device types that fit within Continua Health Alliance membership and all the standard organizations and special interest groups. one or more of these categories: health and wellness, disease management, Principles of Implementation When initially creating the IEEE standards, the PHD work group surveyed and independent living.” existing devices to understand the needs in the market.

One key goal was to support a broad range of device types that fit within one or more of these categories: health and wellness, disease management, and independent living. To do this, it was felt that a common exchange protocol was necessary to allow a single manager (for example, health appliance, set- top box, personal computer, cell phone) to accept data from a broad range of device types such as glucose meters; fitness equipment; and home sensors, such as motion sensors, light sensors, and gas sensors. At the same time, since the sensing agents often run on small batteries, efficient transmission of the data is also paramount. IEEE Std 11073-20601-2008 attempts to balance the architecture between these two goals so that it is flexible enough and yet has optimized packet sizes.

In order to optimize data exchange, the IEEE Std 11073-20601 has three different event-reporting formats that range from very flexible (but bigger) to very compact (but less flexible). In the very compact forms, all of the information about what the bytes mean is extracted and sent once during configuration. This allows the event reports containing the health data to “At the same time, since the sensing report primarily just data and remove redundant bytes of meta-data. agents often run on small batteries, efficient transmission of the data is also paramount.”

Personal Health Device Interoperability | 113 Intel® Technology Journal | Volume 13, Issue 3, 2009

“When there was a choice between An overarching principle when creating the standards was that when there was a choice between placing a feature on the sensing agent or on the manager, the placing a feature on the sensing agent feature should be placed on the manager. This choice helps to keep the cost of or on the manager, the feature should building agents as low as possible. Therefore, in general, the agent is “in charge” and determines when it has data, when it wants to transmit those data, and be placed on the manager.” what functionality it selects to expose. The manager must adapt to the agent.

Due to the range of device types, the standards need to support three different styles of data delivery: episodic – data delivered based on an action, such as stepping on a scale; periodic – the device sends data at a known frequency, such as updating the blood oxygen saturation every second; and store and forward – the device collects data while not connected to the manager and then the data can be uploaded on a later connection, such as when a person walks with a pedometer for an exercise session. Thus, there is a range of tools in IEEE 11073- 20601 for supporting these modes of data transfer.

Our design goal from the outset was to keep the higher-layer protocol “Our design goal from the outset independent from the transport layer. There are several reasons for this was to keep the higher-layer protocol independence. First, we recognized that some use-cases were best supported with a wired transport, and others were better supported with a wireless independent from the transport layer.” transport. Second, transport technology continues to advance, so it is advantageous to adopt new transports as they become available. Third, by utilizing a common application layer, the transport details can be abstracted deep within a software stack, allowing most of the upper portions of software to be re-used, regardless of which transport is used.

Another big difference between the PHD standards and other healthcare standards is that PHD standards focus on personal health in the home and mobile environments. Such a focus allows for a reduction in complexity. For instance, when blood pressure is measured in the home, there is a limited set of ways it is measured (for example, it is measured non-invasively on the arm or wrist); whereas, in a clinical setting, there are additional ways it could be measured (for example, invasively across heart valves or by measuring umbilical pressure). By focusing on the needs of home health devices only, the standards process is less complex, and implementations are easier to build, test, and make interoperable. In addition, solutions can meet the price points required by the consumer market.

Common Optimized, Data-Exchange Protocol Within the IEEE Std 11073-20601 there are five major elements as shown in “By focusing on the needs of home Figure 4. We describe these elements in detail in the following sections. health devices only, the standards 1. Abstract Syntax Notation.1 (ASN.1). Language used in the abstract modeling process is less complex, and of data and interactions. implementations are easier to build, 2. Nomenclature. Provides binary codes for terminology. 3. Domain Information Model (DIM). Describes the device and physiological test, and make interoperable.” data.

114 | Personal Health Device Interoperability Intel® Technology Journal | Volume 13, Issue 3, 2009

4. Service Model. Defines interactions with the device and data. ISO/IEEE Std 11073–20601

5. Communication Model. Manages the connection state machine and ASN.1 Nomenclature communication characteristics. Domain Information Model Abstract Syntax Notation

Abstract Syntax Notation.1 (ASN.1) is an existing standard for describing Service Model concepts in an abstract notation that is independent of an underlying implementation language. Communication Model By using ASN.1 to describe the modeling of the Domain Information and Service Models, it is possible to generate code automatically for different computer languages as well as code that converts from internal data structures into different transmission representations (such as binary or XML). Figure 4: Components of IEEE Std 11073-20601 Source: IEEE, 2008 Nomenclature The nomenclature represents binary values that are used to describe, unambiguously, the concepts being modeled. For example, there are separate codes for pulse rate, heart rate, systolic blood pressure, and diastolic blood pressure. By defining specific codes for all the concepts being modeled, two devices are able to exchange information by using a common language. By using binary codes rather than strings, the nomenclature also aids in supporting localization efforts so the concepts can be represented to the operator in the local language. “By defining specific codes for all the The Domain Information Model concepts being modeled, two devices The Domain Information Model (DIM) contains objects that represent both device-level information (for example, battery status, current time, model and are able to exchange information by serial number), physiological information (for example, weight, blood pressure, using a common language.” number of steps), and environmental information (for example, gas or water left on, house too cold or too hot). The DIM uses object-oriented concepts in specifying the model. This eases the maintenance and modeling effort for standard creation. There is no requirement to implement the model by the use of object-oriented languages and, in fact, there are product implementations built by using Standard C.

The object classes that are available are generic in nature in order to support a wide range of usages. There are currently classes of objects to describe the device itself (for example, battery or wall-powered device), numeric objects (for example, physiological measurements such as temperature), real-time sample arrays (for example, usually for waveforms such as an ECG), and enumerations (for example, at this time the door was opened). There are also classes that support the storing and forwarding of measurements and scanners that assist in “The Domain Information Model compacting data streams. (DIM) contains objects that represent Each class defines the object type, the standardized attributes (variables) available in the class, and the actions (methods) that can be executed for both device-level information, objects of the class. The agent represents its data by creating one or more physiological information, and objects and then communicating the starting values of the attributes. As data changes, the agent sends event reports to the manager to inform it of the environmental information.” changes.

Personal Health Device Interoperability | 115 Intel® Technology Journal | Volume 13, Issue 3, 2009

“Attributes of an object are set by Attributes of an object are set by the agent to represent the current conditions (for example, there is an observed value attribute that defines the person’s the agent to represent the current weight and another that describes the unit-code for pounds or kilograms). conditions.” Attributes have qualifiers in the standard that describe whether the attribute is: a) mandatory, that is, it must be present, b) conditional, that is, if the condition is met, it is mandatory; otherwise, it is optional, or c) optional, that is, where the agent implementers determine whether they want the attribute in their product. In addition, additional attributes can be added to an object beyond those that are standardized. This allows vendors to model other items to differentiate their product, if desired.

Service Model The service model describes the services that are available for communicating data to and from the agent. There are three categories of services:

•• Event reporting. This is used by the agent to send updates to the manager about its configuration or any new measurements that are available. •• Object access. This is used by the manager to get, set, or invoke actions on the agent. •• Association. This is used by agent and manager to create an association with one another so they are synchronized on their operational and communication state. Communication Model The communication model is the abstraction between the upper application layers and the transport layers. The key items that are described in the communication model include the following:

•• Communication characteristics. While it would be nice to be agnostic to the underlying transport, there are certain requirements that the application- level protocol requires of transports. These are described to make it easy to assess whether a transport can work natively or if a shim is required to make the transport conformant. •• Connection state machine. This describes the states that a manager and agent move through during an association. The agent’s state diagram is shown in Figure 5. •• Legal interactions in each state. These describe the actions that are and are not allowed while in each state of the state chart. •• Conversion service. This describes the method required to convert from the ASN.1 modeling into a transmission format encoding. The standard supports three different styles of encoding, but it mandates that the most basic one shall always be present to ensure the agent and manager can interoperate at an encoding level.

116 | Personal Health Device Interoperability Intel® Technology Journal | Volume 13, Issue 3, 2009

Disconnected Transport connect Transport disconnect indication indication

Connected

Disassociating Associated

assocRelReq

+entry/TxAssocRelReq Operating

TxAssocAbort RxAssocAbort RxAssocRelRsp RxAssocRelReq/ TxAssocRelRsp

Unassociated TxAssocAbort

Configuring RxAssocAbort RxConfigEventReportRsp RxAssocRelReq/ (accepted-config) TxAssocRelRsp

Waiting Approval

assocReq RxAssocRsp RxAssocAbort (accepted) Or RxAssocRsp RxConfigEventReportRsp TxAssocAbort (rejected) (unsupported-config)

Associating TxConfigEventReportReq

RxAssocRsp Sending Config +entry/TxAssocReq (accepted-unknown-config)

Figure 5: Agent State Diagram Source: IEEE, 2008

Example Device Specialization (Glucose Meter) Figure 6 shows the glucose meter modeling. To be conformant, the only mandatory objects are the glucose reading and the object representing the device. However, many glucose meter models available today allow the user to enter additional contextual data that helps when interpreting a reading. For example, did the person exercise? Did the person eat a meal recently? Was medication taken near the reading? The optional objects of the glucose meter standard allow a vendor to model this type of context data. Finally, there is a persistent metric store that allows the device to store readings for a longer time, such as the time between doctor visits.

Personal Health Device Interoperability | 117 Intel® Technology Journal | Volume 13, Issue 3, 2009

PHD-Glucose meter object instances

Glucose: Numeric Glucose Meter: MDS Observations: PM-Store 11 1 0…1

1 ContextExercise: Numeric 0…1 0…1 ContextHealth: Enumeration

ContextDiet: Numeric 0…1 0…1 EventDeviceErr: Enumeration

ContextMedication: Numeric 0…1 0…1 ContextMeal: Enumeration

ContextHbA1C: Numeric 0…1 0…1 ContextSampleLocation: Enumeration

0…1 ContextTester: Enumeration

Figure 6: Glucose Meter Modeling Source: IEEE, 2009

Transport “This specification takes into The Continua architecture separates the transport from the protocol, data, and nomenclature. This abstraction allows for various transports to be leveraged consideration personal health wired as usage requires without impacting the upper-layer applications. As of this writing, there are three different transport technologies available and the requirements and defines a mechanism specification used for each one is explained next. to exchange data between devices over USB USB.” If a wired transport is desired, then the USB Personal Healthcare Device Class (PHDC) specification has been selected. This specification takes into consideration personal health wired requirements and defines a mechanism to exchange data between devices over USB. The USB Implementers Forum has an automated test tool that is in place to provide USB Personal Healthcare Device class transport certification of personal health devices.

Bluetooth If a wireless transport is desired for point-to-point personal area network “This specification takes into connectivity, then the Bluetooth Health Device Profile specification has been consideration personal health wireless selected. This specification takes into consideration personal health wireless PAN requirements and defines a mechanism to exchange data between devices PAN requirements and defines a over Bluetooth. The Bluetooth SIG has an automated test tool that is in place mechanism to exchange data between to support a Bluetooth transport certification of personal health devices. devices over Bluetooth.”

118 | Personal Health Device Interoperability Intel® Technology Journal | Volume 13, Issue 3, 2009

If a low-power version is necessary (for example, extreme sensitivity to battery “This specification takes into life), then the Bluetooth Low Energy (LE) specification has been selected. This technology is anticipated to be combined with classic Bluetooth on the consideration personal health wireless aggregation device in order to accept input from either HDP or LE. LAN requirements and defines a ZigBee mechanism to exchange data between If a low-power wireless transport is desired for local area network connectivity, then the ZigBee Health Profile specification has been selected. devices over ZigBee.” This specification takes into consideration personal health wireless LAN requirements and defines a mechanism to exchange data between devices over ZigBee. The ZigBee Alliance has a tool in place to support a ZigBee transport certification of personal health devices.

Lessons Learned To enable the personal health market during the development of the Continua Health Alliance over the past three years, several key lessons have been learned. We discuss several of the important ones.

•• Start out with market-viable use-cases and requirements. It is easy to let technology lead the way for a particular use-case, but it is imperative to ensure that there is broad market support (i.e., more than one company) and that the purchasers and domain experts are involved in the definition process such that what is eventually delivered is relevant. •• Carefully constrain scope and ensure appropriate focus. As the number of companies in an organization grows so does the appetite for new, complex capabilities, and it is easy to get distracted and lost. Since there are limited resources, it is important that the highest-priority capabilities are identified and that they become the main focus of the organization. •• In order to avoid confusion, it is important to agree on the philosophical positions of the organization ahead of time. These include choices, such as whether to profile by developing guidelines, or try and augment the standards; or whether to focus on simplicity (basic features) instead of complexity (rich features).

Conclusion As a result of the potential benefits of home-based healthcare, a growing “Excellent progress has been made number of companies have focused their efforts on developing telehealth solutions in the personal healthcare domain. While there are a number of toward the removal of one key challenges that must be addressed for this endeavor to be successful on a large obstacle: personal health device scale, excellent progress has been made toward the removal of one key obstacle: personal health device interoperability. interoperability.”

Personal Health Device Interoperability | 119 Intel® Technology Journal | Volume 13, Issue 3, 2009

The efforts of the Continua Health Alliance and standard development organizations, such as the USB IF, Bluetooth SIG, and IEEE, have culminated in a set of standards that specify the manner in which personal health devices communicate as well as in design guidelines that profile these standards, thereby removing ambiguity and ensuring true interoperability.

Armed with these design guidelines and standards, device vendors, software developers, service providers, integrators, and purchasers are now in the position to focus on their respective areas of expertise and collectively deliver on the vision of home healthcare delivery based on systems composed of interoperable personal health devices.

Reference [1] Continua Health Alliance. Available at www.continuaalliance.org.

Author Biographies Doug Bogia is a Standards Architect in Intel’s Digital Health Group. In 1995, Doug received his Ph.D. from the University of Illinois, Urbana-Champaign in Computer Supported Collaborative Work. Since joining Intel in 1995, Doug has held a number of positions ranging from implementing collaborative business and personal products, to implementing small business support services, to creating telecommunication products. During 2002, he assisted in creating the Advanced Telecom Computing Architecture standard enabling the telecommunications industry to begin interoperable implementations in 2003. In 2005, he shifted focus to the healthcare industry. Doug led the formation of the ISO/IEEE 11073 Personal Health Devices Work Group where he currently serves as the Work Group Chair. He is also an active participant in the Continua Health Alliance, Bluetooth Medical Devices Work Group, and USB Personal Health Device Profile. His e-mail is Douglas.P.Bogia at intel.com.

Rick Cnossen is the President and Chairman of the Board for the Continua Health Alliance. Continua is a large, international, cross-industry consortium, focused on the establishment of an ecosystem of interoperable, personal telehealth systems. He is also a Director of Personal Health Enabling for Intel’s Digital Health Group with a focus on medical device interoperability standards. Additionally, he serves on the HITSP Board of Directors and is a participating member of the ISO/IEEE 11073 Personal Health Data Work Group, the HITSP Consumer Perspectives Technical Committee, the Health Level Seven (HL7) Structured Document Technical Committee, the USB Personal Health Device WG, and the Bluetooth Medical Device Profile WG.

120 | Personal Health Device Interoperability Intel® Technology Journal | Volume 13, Issue 3, 2009

Mr. Cnossen holds an MSCS degree from the University of Southern California (USC). He worked at McDonnell Douglas (Boeing) for 15 years as a Software Engineering Manager on a variety of aerospace and defense projects and at Intel Corporation for 9 years as an Engineer Manager and Director. During this time, he has been involved in many industry technology and standards development efforts including the Advanced Telecommunications Architecture (ATCA) manageability specification, Tactical Defense data collection and analysis, Hierarchical Data Assimilation and Storage, and High Performance Computing. His e-mail is rick.a.cnossen at intel.com.

Chris Gough is an Enterprise Architect with the Product Development team in Intel’s Digital Health Group. Prior to his focus on product development, Chris was closely involved with the Continua Health Alliance Technical Work Group and had a key role in developing the Continua system architecture. Chris joined Intel in 1991 and, prior to his tenure with the Digital Health Group, has held various technical positions in the Digital Enterprise Group focusing on motherboard design applications and validation tools, emerging networking technologies, and systems architecture. Chris holds a bachelors degree in Computer Science and Engineering from the University of California at San Diego. His e-mail is chris.s.gough at intel.com.

Personal Health Device Interoperability | 121 Intel® Technology Journal | Volume 13, Issue 3, 2009

A Common Personal Health Research Platform — SHIMMER™ and BioMOBIUS™

Contributors

Michael J. McGrath, Ph.D. Abstract Intel Corporation As the cost of healthcare continues to rise, and the aging population demographic in the Western world continues to increase, the pressure on direct Terrance. J. Dishongh, Ph.D. clinical resources continues to grow at an alarming rate. The health of people Intel Corporation around the world is worsening and the global population is getting older. With this in mind the goal of SHIMMER™ (Sensing Health with Intelligence, Modularity, Mobility and Experimental Reusability) systems is to provide an Index Words extremely compact extensible platform for body-worn or ambient sensing applications based on connected or disconnected data strategies with proven Research Platform system building blocks. Additionally, to aid in the analysis of the data coming BioMOBIUS™ from SHIMMER, it is fully integrated with the software environment of the SHIMMER™ BioMOBIUS™ research platform. The combination of the BioMOBIUS and Sensing SHIMMER platforms give researchers an extensible and flexible systems that Extensible can be utilized to rapidly develop prototype research tools to acquire data in Flexible a reliable and timely manner. In this article, we outline how the combined BioMOBIUS and SHIMMER platforms can be used to develop sophisticated research tools.

Introduction Biomedical and clinical research and development projects have a growing need for technology solutions that are highly flexible, extensible, easy to use, and provide a comprehensive range of capabilities. The availability of such resources allows researchers to focus on the R instead of the D in R&D during the lifecycle of their research project. The technology development overhead that many biomedical researchers must address can detrimentally impact the pace and scope of the research projects. This problem has been compounded in recent years due to the growing interest in moving from laboratory to data “The technology development overhead collection and observation in home environments. A variety of applications have been reported in the literature including physiological monitoring [1-3], that many biomedical researchers physical rehabilitation [4], activities of daily living (ADL) monitoring [5, 6], must address can detrimentally impact falls detection [7, 8], cognitive function [9], and social engagement [10, 11]. These applications typically comprise hardware and software components the pace and scope of the research interconnected to form customized applications to address the needs of a projects.” specific research hypothesis.

122 | A Common Personal Health Research Platform — SHIMMER™ and BioMOBIUS™ Intel® Technology Journal | Volume 13, Issue 3, 2009

Many biomedical applications utilize wireless sensing capabilities in the form “Typically, multiple parameters are of body-worn sensing or non-contact sensing to acquire data of interest and to transmit them to an aggregation device. Fusion of both body-worn and of interest including physiological, ambient sensors has been reported to improve accuracy of inference based on kinematic, ambient, and activity or behavior identification [12, 13]. Typically, multiple parameters are of interest including physiological [14, 15], kinematic [16, 17], ambient [18, environmental measurements.” 19], and environmental measurements [20].

To address the need for multisensing capabilities while minimizing the complexity of the hardware and software components, platform-based approaches have emerged where the sensing capability of the sensors can be modified by changing the sensing element, normally in the form of a daughterboard that connects to a common baseboard that provides the computational and communications capabilities. Chen et al [2] report a sensor node platform for wireless biomedical sensing. They utilize a flexible expansion connector that supports additional daughterboards, including ECG with TinyOS* firmware. DuboisFerrière et al [21] describe the TinyNode*, which features two types of add-on boards. The Standard Extension Board (SEB) includes footprints for two optional sensors: a relative humidity and temperature sensor and a photodiode light sensor. The MamaBoard features a variety of external communications including, LAN, WLAN, and GPRS options with support for data storage via an SD card. The platform “Platform-based approaches have also features an XE1205* from Semtech with a reported four- to eight- fold communications range performance improvement over MICA2 and emerged where the sensing capability TelosB sensor nodes. Nokia has reported a wearable sensor platform called a Nokia Wrist–Attached Sensor Platform* (NWSP) that is based on a highly of the sensors can be modified by flexible Field Programmable Gate Array (FPGA) [22]. The platform features changing the sensing element.” an accelerometer, a gyroscope, and magnetometer sensing capabilities in a wristwatch-like form-factor.

For a body-worn sensor platform, size is a key consideration. IMECs Human++ research program has developed a highly miniaturized and autonomous sensor system for body sensor network applications [23]. The platform combines wireless ultra-low power communications, 3D integration, MEMS energy scavenging techniques, and low-power design techniques. Two-channel Electroencephalography (EEG), two-channel Electrooculography (EOG), and one-channel Electromyography (EMG) for sleep-monitoring applications and wireless ECG monitoring have been demonstrated [24].

From a software perspective, platforms for wireless sensor networks have been reported in the literature [14, 25, 26]. Walker et al describe a Java* and Java Agent based Development (JADE) platform; it is designed to provide sensor network integration and application development capabilities [27]. Other Java platforms include SunSPOT* from Sun Microsystems [28].

“For a body-worn sensor platform, size is a key consideration.”

A Common Personal Health Research Platform — SHIMMER™ and BioMOBIUS™ | 123 Intel® Technology Journal | Volume 13, Issue 3, 2009

“Both the C-type programming and to The Arduino* open-source electronics prototyping platform has become popular (www.arduino.cc) especially for under-graduate teaching purposes. A a lesser extent the SDK approaches are number of research applications have been reported, including e-Textiles*, that complex and time-consuming.” are based on the integration of accelerometers for motion capture and human motion capture in combination with tactile feedback in response to specific postures in real-time [29, 30].

A number of options exist for high-level application program development with associated benefits and difficulties. Typically there is a choice between traditional syntactically-oriented software languages such as C/C++, Java, Python*, etc., bespoke software development environments (SDKs); e.g., Eclipse*, and graphical development environments (GDE). Both the C-type programming and to a lesser extent the SDK approaches are complex and time-consuming. Firstly, they often result in applications that are harder to debug, integrate, and evolve. Secondly, they restrict researcher access to the development process. Graphical programming approaches can deliver a common interface to the sensor/network with which all users will be able to interact. These interfaces generally take the form of drag and drop development environments where functional blocks are selected from a palette of tools “The environment can cater for users and connected on a drawing area to describe data flow through a system or application. The environment can cater for users with widely varying levels with widely varying levels of technical of technical expertise by enabling design and development work to be carried out at several levels of abstraction. Using a GDE, a developer can build a expertise by enabling design and working system graphically by dragging and dropping blocks onto a diagram development work to be carried out at (resembling a data-flow diagram). Typically, each block features a number of input and output pins, with connections made between blocks by clicking and several levels of abstraction.” dragging from a pin on one block to a suitable pin on another block.

LabVIEW* from National Instruments, has been popular for many years, particularly for applications that required data acquisition via an analog-to- digital converter (ADC) or device actuation via a digital-to-analog converter (DAC). LabVIEW provides a GDE with a drag and drop capability for functional blocks wired together to form a functional application. It has been used for a variety of biomedical-type applications including interfacing pulse oximeters [31], ECG signal analysis [32], tri-axial accelerometers and galvanic skin response (GSR) [33], and gyroscopes for gait analysis [34]. Other graphical programming environments that have been developed for biomedical applications include Scicos: these can be utilized for signal processing, systems control, and for studying physical and biological systems [35]. The BioEra* environment provides a graphical block-based environment that can process dozens of channels simultaneously in real-time from various devices, including EEG and heart rate monitors [36]. BrainBay* is a bio- and neuro- “LabVIEW provides a GDE with feedback environment, again based on a block-type development environment a drag and drop capability for designed to work with the OpenEEG*-hardware platform [37]. Simulink* from MathWorks provides an interactive graphical environment that has been functional blocks wired together to applied to a variety of sensor simulation applications [38, 39]. form a functional application.”

124 | A Common Personal Health Research Platform — SHIMMER™ and BioMOBIUS™ Intel® Technology Journal | Volume 13, Issue 3, 2009

In order to address the issues associated with technology development such “TRIL grounds its research in as rapid prototyping, reuse, extensibility, distribution, etc., the TRIL Centre [40] has developed a modular, extensible, and reusable technology approach ethnography and clinical efficacy, with based on an open research platform concept called BioMOBIUS. The TRIL a common set of research tools based Centre is a collaboration between Intel Ireland, University College Dublin, Trinity College Dublin, and the National University of Ireland at Galway, with on BioMOBIUS.” support from IDA Ireland. TRIL grounds its research in ethnography and clinical efficacy, with a common set of research tools based on BioMOBIUS, and takes the research out of the lab into homes of older people. BioMOBIUS enables TRIL researchers to carry out their research in an effective and efficient manner and to have the ability to share their research solutions in a logistically effective manner. A modular, abstracted approach reduces the learning curve and prompts rapid application prototyping and reuse, thereby lowering development costs and reducing the barriers to technology adoption. The software components of BioMOBIUS are made freely available to the biomedical research community by the TRIL Centre. The BioMOBIUS research platform comprised of both low-level and “The BioMOBIUS research platform high-level software development environments is fully integrated with the SHIMMER sensor platform [41, 42]. BioMOBIUS also supports a number is fully integrated with the of other third-party sensors and off-the-shelf devices. This integration of SHIMMER™ sensor platform.” software and hardware provides a unique set of capabilities that enables developers, engineers, and researchers to rapidly develop research applications to investigate a variety of research hypothesis and to rapidly modify the applications based on the evolving needs of the researcher and end user. The key features of the platform to enable such capabilities are as follows:

•• Ease of hardware integration. •• Software component reuse. •• Highly extensible both from a hardware and software perspective. •• High quality user interfaces (UI). •• Supports data acquisition rates appropriate for kinematic and physiological data capture. •• Provides real-time data processing and data presentation. •• Data persistence to file and database. •• Stability and reliability. •• Supports rapid application prototyping, i.e., short development times in the order of one to two weeks. A typical BioMOBIUS application includes SHIMMER sensors, processing “A typical BioMOBIUS application functionality, and a UI. The sensors monitor biomedical indicators such as gait stability, and the processing functionality converts the sensor data into includes SHIMMER sensors, meaningful information. The UI allows clinicians to view the information and processing functionality, and a UI.” adjust the application settings or it allows the home-based participant to view feedback information on performance; e.g., attention level measurement via GSR.

A Common Personal Health Research Platform — SHIMMER™ and BioMOBIUS™ | 125 Intel® Technology Journal | Volume 13, Issue 3, 2009

BioMOBIUS Software Environment “Originally focused on visual arts Drag and drop environments ideally support the concept of rapid application prototyping through the development of applications, by selecting blocks from applications, EyesWeb provides fusion a palette of capabilities categorized by function, such as input/output, data of motion and music for emotion management, digital signal processing (DSP), etc. Sophisticated applications can be quickly assembled, including hardware interfacing and real-time data capture during artistic performance.” processing, by using the functional blocks that are connected in sequence to represent the real-time data flow.

EyesWeb*, developed by the University of Genoa, was selected as the drag and drop GDE [43, 44]. Originally focused on visual arts applications, EyesWeb provides fusion of motion and music for emotion capture during artistic performance [45]. It is also utilized to develop applications based on multimedia techniques for therapy and rehabilitation; e.g., Parkinson’s, Autism [46, 47]. The EyesWeb software platform was originally developed in EU th5 Multisensory Expressive Gesture Applications (MEGA, see www.megaproject. org) and 6th Framework Programme, the Tangible Acoustic Interfaces for Computer Human Interaction (TAI-CHI). The EyesWeb environment has been adapted and enhanced by TRIL for biomedical oriented applications as shown in Figure 1.

“The EyesWeb environment has been BioMOBIUS Applications Applications adapted and enhanced by TRIL for biomedical oriented applications.” BioMOBIUS User Interface Applications– GUI Designer Physiological, Kinematic

RCF Interface

Image Database DSP Services Services ThirdParty EyesWeb File Math Applications IDE Protocols Services Operations e.g. Databases Sensor Audio/Video Services Services BioMOBIUS High Level Application Development Env Application Development BioMOBIUS High Level EyesWeb Kernel

Windows API’s

BT X10 USB/ IEEE Hardware Stack Stack RS232 802.15.4 Driver Driver Stack Stack Firmware Interfaces Hardware Hardware Driver Driver Hardware Hardware

Figure 1: BioMOBIUS Architecture Source: The TRIL Centre (www.trilcentre.org)

126 | A Common Personal Health Research Platform — SHIMMER™ and BioMOBIUS™ Intel® Technology Journal | Volume 13, Issue 3, 2009

The EyesWeb environment comprises a GDE that supports drag and drop “The EyesWeb environment comprises application development in the form of patches. BioMOBIUS provides a Graphical User Interface (GUI) development tool that enables the user a GDE that supports drag and drop to design an application to control a patch and view its output. The GUI application development in the form Designer is built by using the JUCE* C++ class library [48]. This is a cross- platform, high-level, graphical application program interface (API) and of patches.” provides the potential to develop BioMOBIUS applications on non-Windows* platforms.

Use of the GUI development tool allows the user to be abstracted away from the underlying complexities of the connected patch. The resultant application is called a GUI application. The runtime environment within BioMOBIUS comprises two key components:

•• Kernel Runtime Server. The EyesWeb Kernel Runtime is a Windows component that supports the execution of EyesWeb patches outside of the EyesWeb GDE. It is used in conjunction with the BioMOBIUS Runtime to execute GUI applications and patches as if they are native Windows applications. •• BioMOBIUS GUI Runtime. This is a Windows console application that supports the execution of a BioMOBIUS application outside of the GUI development tool. Typically, a BioMOBIUS application is first developed and tested within the GUI development tool. Thereafter, the BioMOBIUS Runtime is used to run the application as a Windows executable in conjunction with the EyesWeb Kernel Runtime component, via a TCP/IP connection that supports execution of the patch and GUI on two physically separated networked systems, if required. Both patches and GUI applications are defined in XML and interpreted by their respective runtimes. A number of benefits are achieved by separating these two frameworks:

•• Runtime efficiency. Applications consume less system resources; for example, output and display rendering is automatically disabled in the runtime environment. •• Separation of duty. Patches and UI applications may be developed in parallel. •• Distribution. An application and its patch may execute on separate “Both patches and GUI applications platforms across a network. •• Extendibility. Other graphical development environments may be used to are defined in XML and interpreted develop feature-rich applications such as FLASH-based interfaces. by their respective runtimes.”

A Common Personal Health Research Platform — SHIMMER™ and BioMOBIUS™ | 127 Intel® Technology Journal | Volume 13, Issue 3, 2009

Output Blocks pin x Blocks, as shown in Figure 2, are the fundamental building units of the environment which are dragged and dropped onto a patch from a palette of t blocks grouped by functionality. A block is coded as a C++ class with inherited Input attributes and methods. The number of inputs and outputs for each block is pins determined by the developer during the development process. Inputs and outputs are referred to as pins. There are two categories of inputs:

Figure 2: BioMOBIUS block with input and output •• Main inputs (block receives its primary data for processing). pins •• Parameter inputs (configuration details determines block execution). Source: The TRIL Centre (www.trilcentre.org) Blocks are executed in two ways at runtime. Active blocks initiate their own execution by requesting the kernel to schedule its execution. Initiation of execution is triggered periodically by using kernel polling, i.e., a call-back + GUI GUI Objects mechanism. The second type of execution is used by reactive blocks. These + Kernel blocks are executed by the kernel when data arrives at its main input. The E Legacy kernel is alerted to the availability of these data by a system flag in the block + Steinberg Technologies setting. + Base + System New blocks can easily be developed for the GDE by using the EyesWeb + National Instruments software development kit (SDK) and integrated development environment + 3D (IDE) wizards. Typically, a user creates a new block to support a new hardware + BioMOBIUSBiosignals device or execute a specific DSP function. In addition, EyesWeb applies + BioMOBIUSBase data-type checking on block data pins, and the SDK supports the creation of + BioMOBIUSHardwareIO user-specific data types. The developer of a new block has the choice to either + BioMOBIUSMath distribute the source and/or binary to the development community.

Catalog View Blocks are organized in two ways with the GDE. Firstly, they can be aggregated into catalogs that are physical collections of blocks within a standard Windows + GUI GUI Objects environment. Secondly, they can be structured into libraries which are logical + Math + Audio collections organized according to functionality/use; e.g., mathematical + BioMOBIUS operations (Figure 3). + DataStructures + Filesystem Patches + FlowAndControlStructure In order to build a functioning application, blocks are dragged from libraries or + Geometric catalogs in the GDE environment and connected together to form a patch. A + ImageAndVideo patch comprises a series of blocks linked together in order of execution to form + Math a fully-functional application. Figure 4 shows an example of a patch designed + Operations to allow the user to manually trigger the acquisition of data from a SHIMMER + Peripherals sensor, process the received data, and to display the acceleration values sensed + String + TimeAndDate by the sensor to the user. A patch is executed via the Kernel Runtime Server. + TimeSeries The EyesWeb Kernel Runtime is a Windows component that supports the execution of EyesWeb patches both inside and outside the EyesWeb GDE. Library View

Figure 3: Library and catalog views within the BioMOBIUS GDE Source: The TRIL Centre (www.trilcentre.org)

128 | A Common Personal Health Research Platform — SHIMMER™ and BioMOBIUS™ Intel® Technology Journal | Volume 13, Issue 3, 2009

CONNECT

START

STOP x

DISCONNECT t x

t

Shimmer x

t

Figure 4: Sample patch using SHIMMER in the EyesWeb GDE Source: Intel Corporation, 2009

As the complexity of an application increases so too does the need to “BioMOBIUS features a variety of structure patch development into logical or functional units. Complex patch development can be managed through the use of the subpatch feature digital signal-processing capabilities.” within the EyesWeb environment. The subpatch supports the integration of a functional group of interconnected blocks into a single block-like subpatch structure.

Digital Signal Processing Capabilities for Biosignal Processing BioMOBIUS features a variety of digital signal-processing capabilities. For real-time video processing, an extensive range of processing options exists including, baricenter/centroid calculations, quantity of motion from a silhouette, background subtraction, image concatenation, and image splitting. A variety of more advanced techniques is also available including real-time, full-body motion tracking and blob extraction. Time domain digital filtering is served by Finite Impulse Response (FIR) filters. A variety of frequency domain digital filters are available including Fourier Transform (FT), Hilbert Transform, and Kalman. Support is included for matrix-handling operations. A “A variety of frequency domain digital variety of specialist filters, such as Cochleogram generation in both image and matrix format from an audio buffer, are available. filters are available.”

A Common Personal Health Research Platform — SHIMMER™ and BioMOBIUS™ | 129 Intel® Technology Journal | Volume 13, Issue 3, 2009

“The SHIMMER platform comprises SHIMMER Platform a baseboard that provides the The SHIMMER wireless sensor platform was developed by the Intel Digital Health Group over a number of years to support a variety of internal research sensors computational, data storage, projects. The SHIMMER platform comprises a baseboard that provides the communications, and daughterboard- sensors computational, data storage, communications, and daughterboard- connection capabilities. The core functionality of SHIMMER platform is connection capabilities.” extended via a range of daughterboards that provide kinematic, physiological, and ambient sensing capabilities. This range of contact and non-sensing capabilities can be reliably used both in clinical and in home-based scenarios.

SHIMMER Baseboard Design The core element of the baseboard is the Texas Instruments MSP430* MCU (Microcontroller Unit) [49], which has been widely used in wireless sensors [50-52]. The primary advantages of the MCU are its extremely low power during periods of inactivity and its proven history for medical-sensing applications. The 10Kb of RAM on the MSP430F1611 variant is the largest memory available within the processor family and offers improved buffering “The primary advantages of the capability for communication applications. MCU are its extremely low power The MSP430 has eight ADC channels for 12-bit A/D conversions. For the SHIMMER platform, the external ports are utilized for reading data from the during periods of inactivity and its XYZ accelerometer (three channels), the internal expansion connector (three proven history for medical-sensing channels), and the external expansion connector (two channels). To maintain the low-power usage capabilities of SHIMMER, the MSP430 ADC core applications.” is disabled when not in use and re-enabled when necessary. The baseboard has a Freescale Semiconductor 3-axis (XYZ) accelerometer (MMA7260Q*). The accelerometer is suitable for low-power systems: when active the current consumption is around 500μA, and it contains a sleep mode, with current usage of 3μA. It is robust, having high shock survivability, and is suitable for high-sensitivity applications (800mV/g for 1.5g setting).

A passive tilt/vibration sensor (SQ-SEN-200* – Signal Quest) has been added to later generations of the SHIMMER baseboard. It is sensitive to both tilt (static acceleration) and vibration (dynamic acceleration). The sensor is used to trigger power state transitions. For example, when the user takes an inactive sleeping sensor out of the charging dock, the firmware will automatically switch on the necessary components (accelerometer, radio, etc.) to begin capturing and transmitting motion data.

Communications “For the SHIMMER platform, the One of the key functions of the SHIMMER board is its ability to communicate as a wireless platform. SHIMMER has the dual functionality of having both external ports are utilised for reading 802.15.4 and Bluetooth® radio modules. data from the XYZ accelerometer, the 802.15.4 Radio internal expansion connector, and the For IEEE 802.15.4-compliant wireless communication, the SHIMMER platform uses a Chipcon CC2420* radio transceiver and a gigaAnt 2.4GHz external expansion connector.” Rufa* antenna.

130 | A Common Personal Health Research Platform — SHIMMER™ and BioMOBIUS™ Intel® Technology Journal | Volume 13, Issue 3, 2009

Bluetooth® Radio Module The SHIMMER platform uses the Roving Networks RN-41* Class 2 Bluetooth module to communicate via an integrated 2.4GHz antenna. This module contains a full Version 2 Bluetooth Protocol Stack and supports the Serial Port Profile that facilitates rapid application development. The Bluetooth module is connected to the MSP430 directly via the USART1 serial connection. It can also be controlled by ASCII strings over the Bluetooth RF link.

SHIMMER Daughterboards For further functionality the SHIMMER baseboard has a Hirose Electric DF12 series* connector that allows the user to connect daughterboards to the baseboard. Currently there are seven daughterboards available for SHIMMER as shown in Table 1.

Kinematic Sensing Physiological Sensing Ambient Sensing Gyroscope ECG PIR Motion Magnetometer GSR Temperature Light Table 1: BioMOBIUS SHIMMER-based sensors Source: Intel Corporation 2009

External Connections “The purpose of the PRIMMER The external expansion (HIROSETM ST Series*) allows the user to attach the daughterboard is to enable rapid board to the programming dock or multicharger. integration and prototyping of External Hardware Integration The SHIMMER platform provides two options for external hardware commercially available and existing integration, such as third-party sensors. The General Purpose Expansion physiological front-end and sensor Module (AnEx Board) is an analog expansion module that enables two analog signals to be plugged into the SHIMMER platform. These analog signals circuitry.” interface with the A0 and A7 of the MSP430.

PRIMMER™ Daughterboard Prototype Reconfigurable board for SHIMMER (PRIMMER) is a daughterboard that enables the breakout of SHIMMER’s internal connector. The purpose of the PRIMMER daughterboard is to enable rapid integration and prototyping of commercially available and existing physiological front-end and sensor circuitry. The PRIMMER daughterboard breaks out six ADC lines, power and ground prototype solder points, two general purpose input/ouput (GPIO) lines, and a Universal Asynchronous Receiver/Transmitter (UART). In addition to this, it provides a boosted power supply by way of a power- boosting circuit that takes in 3V from the SHIMMER baseboard and increases it to ±5V (Maxim MAX768EEE* charge pump). The board has been utilized to interface to an EMG pre-amplifier sensor (MA-411-000*, Motion Lab Systems) and to streamed EMG data at 500HZ.

A Common Personal Health Research Platform — SHIMMER™ and BioMOBIUS™ | 131 Intel® Technology Journal | Volume 13, Issue 3, 2009

GLIMMER™ “The GLIMMER display platform GLIMMER, the SHIMMER-based display, is a low-cost, wireless display platform that enables context-aware prompting in the home for ADL is an extension of the standard monitoring, medication adherence, journaling, social messaging, and coaching SHIMMER platform.” applications. The GLIMMER display platform is an extension of the standard SHIMMER platform. Prompts are displayed via a 128x64x4bit 2.7” organic light-emitting diode (OLED) display (OSRAM Pictiva™). The resolution is 128x64 with 4-bit grey for image display.

SHIMMER Peripherals A number of peripheral components have been developed for the SHIMMER platform, and among these are the following:

•• Programming dock. USB-bus-powered cradle using the FTDI FT232* UART. The dock also has a battery charger status, USB Power, and UART activity indicators. •• Multicharger. Cradle that can charge six SHIMMERs simultaneously. •• Advanced Dual UART programming board. For applications where wired “SHIMMER firmware is primarily serial communication or enhanced debug capabilities are desirable. •• Enclosures. The SHIMMER platform has two types of plastic enclosure. The developed by using TinyOS and standard housing holds the SHIMMER baseboard, lithium ion battery, and released into the public domain via kinematic or ambient sensor daughterboards if required. The physiological enclosure is utilized with the ECG and GSR daughterboards and provides SourceForge.” external connection points for electrodes.

SHIMMER Firmware Firmware provides the low-level capabilities within the SHIMMER platform to control the device functions. There are a number of lightweight operating systems that can be used to run on a sensor node, namely TinyOS [53], Contiki [54], Nano-RK [55], ERIKA [56], MANTIS [57], and SOS [58]. All SHIMMER firmware is primarily developed by using TinyOS and released into the public domain via SourceForge [59]. Firmware running on the sensor platform provides local processing of the sensed data, local storage of the data if required, and communications of these data to a higher-level application for advanced signal processing, display and data persistence.

TinyOS handles task scheduling, radio communication, time, I/O processing, etc. and has a very small footprint size. TinyOS is a lightweight, event-driven operating system designed for sensor network nodes that have very limited resources. “TinyOS is a lightweight, event- driven operating system designed for sensor network nodes that have very limited resources.”

132 | A Common Personal Health Research Platform — SHIMMER™ and BioMOBIUS™ Intel® Technology Journal | Volume 13, Issue 3, 2009

Real-time Data Streaming to BioMOBIUS SHIMMER data can be reliably streamed at rates up to 500HZ in real-time to “SHIMMER data can be reliably a BioMOBIUS application via the Bluetooth or 802.15.4 radios. To enhance the reliable communications in addition to standard Bluetooth link reliability, streamed at rates up to 500HZ a packet format that supports various types of sensor data and provides in real-time to a BioMOBIUS enhanced reliability to the data transmission through packet sequence numbers and a cyclic redundancy check, was developed. The SHIMMMER firmware application via the Bluetooth or (implemented in TinyOS version 1) uses the frame format shown in Figure 5 802.15.4 radios.” for BioMOBIUS application integration.

Packet Packet Packet Packet Framing Header Payload Framing

Raw Sensor Data Frame Time Data BOF Sensor CRC EOF ID Type Sequence Stamp Length Number Data

8 bits 8 bits 8 bits 8 bits 16 bits 8 bits 1-255 bytes 16 bits 8 bits

Figure 5: SHIMMER frame format Source: Intel Corporation, 2009

The content of packets is unrestricted binary data. This can lead to problems with control characters like BOF (Beginning of Frame, hex byte ‘0xC0’) or EOF (End of Frame, hex byte ‘0xC1’) appearing in the packet header or payload and being interpreted as control characters. To avoid this issue, a byte stuffing technique based on ISO’s asynchronous high-level data link control (HDLC) protocol is utilized.

Logging Data to MicroSD Card Many biomedical applications require subjects to have complete and unhindered mobility. To achieve this, capability support for Secure Digital (SD) card logging has been developed. The SHIMMER platform utilizes a MicroSD card slot that supports up to two GBytes of flash memory. For ambulatory ECG monitoring at 500 Hz from three channels, more than 80 “The SHIMMER platform utilizes a days of data can be stored in uncompressed format. There are a number of firmware programs available on SourceForge that facilitate recording of data to MicroSD card slot that supports up to the SHIMMER platform’s MicroSD card. two GBytes of flash memory.”

A Common Personal Health Research Platform — SHIMMER™ and BioMOBIUS™ | 133 Intel® Technology Journal | Volume 13, Issue 3, 2009

Power Management “The goal is to minimize the on-time A significant difference between most sensor nodes and standard computing platforms is the heavy focus on power management and consumption. A of all hardware subsystems on the large number of applications require battery-powered operation for extended sensor hardware platform.” periods of time [60]. Computational capabilities have far outstripped battery technologies. SHIMMER battery optimizations have been implemented to satisfy various project requirements. These optimizations implemented in firmware have been mostly, but not exclusively, focused on lowering the duty cycle of the 802.15.4 radio. The goal is to minimize the on-time of all hardware subsystems on the sensor hardware platform. For example, each gyroscope on the SHIMMER gyroscope daughterboard requires a supply current of 9.5mA. With two dual-axis gyroscopes on the board there is a substantial 19mA requirement that can be saved when the gyroscopes are off. Other auxiliary battery features have been implemented for integration purposes, such as a low battery indication sent by the SHIMMER platform to a BioMOBIUS software application that prevents corrupted data from the accelerometer being received. The accelerometer outputs are ratio metric which means the output offset and sensitivity will scale linearly with applied supply voltage. Therefore, useful data can only be generated if the supply voltage remains constant. The low battery indication is a SHIMMER firmware feature that halts sampling of data when the supply voltage falls below +3V.

Programming/Flashing SHIMMER An intuitive mechanism to support the download of firmware executable programs to SHIMMER sensors was highlighted as an important capability for biomedical researchers. The SHIMMER development kit includes an easy-to-use graphical application called the “SHIMMER Windows Bootstrap Loader” to allow users who do not have a TinyOS installation on their system to program SHIMMER sensors for their application needs.

SHIMMER Sensing Validation The operation of the SHIMMER baseboard and daughterboards has been tested through a number of validation processes to determine the accuracy of the boards’ function and their usefulness for biomedical-oriented research applications.

Electrocardiography (ECG) Validation “A number of tests were carried out A number of tests were carried out to validate the SHIMMER ECG daughterboard as a valid tool for acquiring ambulatory ECG. The tests to validate the SHIMMER ECG consisted of validating the ECG amplifier and ADC performance by using daughterboard as a valid tool for calibrated input signals and a resting ECG from a healthy subject in normal sinus rhythm. Simulated ECG signals as well as an ECG recording from a acquiring ambulatory ECG.” healthy non-resting subject were used to validate the performance SHIMMER ECG daughterboard for use in ambulatory monitoring. Figures 6 and 7 show a 1mV QRS amplitude (60 BPM) waveform generated by the Fluke MPS450

134 | A Common Personal Health Research Platform — SHIMMER™ and BioMOBIUS™ Intel® Technology Journal | Volume 13, Issue 3, 2009

Patient Simulator* captured by a SHIMMER ECG and a MAC 3500 ECG Analysis System* (GE Medical Systems). The plotted waveform (Figure 6) was recognizable by a clinician as Normal Sinus Rhythm. Figure 7 shows the MAC 3500 ECG Analysis ECG System waveforms for comparative signal quality purposes. Visual examination of the waveforms indicate that they compare well.

Lead I 1

0.5

0 mVolts -0.5

-1 3456789 Seconds Lead II 1

0.5

0 mVolts -0.5

-1 3456789 Seconds Lead III 1

0.5

0 mVolts -0.5

-1 3456789 Seconds

Figure 6: Simulated ECG signal of 1mV QRS amplitude captured by the SHIMMER ECG Source: The TRIL Centre (www.trilcentre.org)

Figure 7: Simulated ECG signal of 1mV QRS amplitude captured by a MAC 3500 ECG Analysis System Source: The TRIL Centre (www.trilcentre.org)

A Common Personal Health Research Platform — SHIMMER™ and BioMOBIUS™ | 135 Intel® Technology Journal | Volume 13, Issue 3, 2009

“A 5.9 minute ECG recording QRS Detection of Normal Sinus Rhythm for a Non-resting Healthy Subject containing 503 heart beats from a A 5.9 minute ECG recording containing 503 heart beats from a non-resting non-resting healthy subject during a healthy subject during a moderate walk was captured by the SHIMMER ECG and also captured by a Medilog Holter monitoring system. The R-R intervals moderate walk was captured by the and instantaneous heart rate (HR) identified by using the Medilog Holter ECG SHIMMER ECG and also captured monitor software were compared against the R-R intervals calculated by using previously reported QRS detection and R-R interval correction algorithms by a Medilog Holter monitoring [61, 62]. Each automatically detected QRS point on the SHIMMER ECG was manually verified to ensure correct detection by using the QRS detection system.” algorithm.

The mean R-R interval for the SHIMMER acquired ECG was 0.7049 seconds, while the mean R-R interval for the Medilog acquired ECG was also 0.7049 seconds. The percentage difference between the R-R intervals for each acquisition was calculated on a beat-beat basis. The mean percentage error between the R-R intervals calculated, by using each acquisition, was found to be 0.0192 percent which can be considered negligible. These results indicate that the SHIMMER ECG can be used to acquire ambulatory ECG from resting and non-resting human subjects for research application purposes.

Kinematic Validation In order to validate the SHIMMER platform for use in studies of human gait analysis, temporal gait parameters derived from a tri-axial gyroscope on the SHIMMER platform were compared against those acquired simultaneously, by using the codamotion* analysis system from Charnwood Dynamics Ltd., UK.

The gait of one normal healthy adult male (age 25) was measured simultaneously by using two SHIMMER sensors placed on each shank and the Cartesian Optoelectronic Dynamic Anthropometer (CODA) motion analysis system. Data were recorded whilst the subject performed multiple over-ground walking and running trials along a 15-meter walkway in a motion analysis laboratory. In all, ten walking trials at a self selected comfortable walking pace and four running trials at a self-selected jogging pace were completed. Heel strike and toe-off points were calculated from the medio-lateral angular velocity derived from the gyroscope signal by using an algorithm reported by Salarian et al. [63] as shown in Figure 8.

The heel strike and toe-off characteristic points derived from the SHIMMER and CODA systems were used to calculate the three temporal gait parameters “Data were recorded whilst the subject listed below: performed multiple over-ground •• Stride time •• Stance time walking and running trials along •• Swing time a 15-meter walkway in a motion analysis laboratory.”

136 | A Common Personal Health Research Platform — SHIMMER™ and BioMOBIUS™ Intel® Technology Journal | Volume 13, Issue 3, 2009

150 Angular Velocity 100 th 2 All Peaks 50 Mid-Swing

0 Angular Velocity Angular

-50 0510 15 20 Time [s]

150 Angular Velocity 100 th 4 th 5 50 Heel Strike Toe Off 0 Angular Velocity Angular

-50 0510 15 20 Time [s]

Figure 8: Shank angular velocity signal derived from the SHIMMER gyroscope Source: Intel Corporation, 2009

Results show an intraclass correlation coefficient (ICC(2,k)) [64] greater than “These results suggest that the 0.85 in stride, swing, and stance times for ten walking trials and four running trials. These results suggest that the SHIMMER platform is a versatile cost- SHIMMER platform is a versatile effective tool for use in temporal gait analysis. Full results from this study are cost-effective tool for use in temporal reported elsewhere [65]. Previous studies [66, 67] have validated the CODA system as a reliable platform for gait measurements, so these results are very gait analysis.” promising.

GSR Validation The SHIMMER galvanic skin response (GSR) sensor contains an internal resistor network that works as a potential divider and provides a voltage that can be converted by the SHIMMER’s ADC to a 12-bit value, used to measure external skin resistance. All skin resistance values were calculated in the SHIMMER platform firmware and transmitted to a BioMOBIUS patch for real-time display and persistence to file. The sensor performance was correlated with a commercial Nexus-10* system (Mind Media BV) utilizing a series of known resistors from 10KΩ to 2.2MΩ. SHIMMER GSR demonstrated an average mean percentage error of 2.3 percent versus the commercial Nexus-10 that had an average mean error of 4.1 percent.

A Common Personal Health Research Platform — SHIMMER™ and BioMOBIUS™ | 137 Intel® Technology Journal | Volume 13, Issue 3, 2009

BioMOBIUS Prototype Applications BioMOBIUS software has been utilized by the TRIL Centre for a number of prototype applications both in clinical and home environments. A typical BioMOBIUS-type application comprising the software and SHIMMER wireless sensors described in this article is shown in Figure 9. Application Development Graphical User Interface

Embedded Signal Processing

Sensor Platform

Low Power Radios Low Power Radios Data Analytics Streaming Data for signals broadcasted Over Radio from Kinematic Data sensors on arms and legs

Physiological Data ECG Data Storage and Kinematic Data

Figure 9: Typical BioMOBIUS application Source: Intel Corporation, 2009

We now look at three TRIL Centre projects that have utilized the BioMOBIUS research platform.

Gait Analysis Platform The BioMOBIUS platform has been used to develop a prototype gait analysis platform for use in a clinical environment and designed for early detection of gait and postural instability. The system is designed to unobtrusively capture gait parameters and physiological data. The platform comprises a number of integrated subsystems that provide a true six degrees of freedom analysis, linked to footfall data and visual gait data. Body-worn kinematic and sensor data are provided by SHIMMER sensors and via a Tactex pressure-sensitive floormat. Video data of the subject are provided via two Webcams. The hardware components are integrated into a BioMOBIUS high-level application that

138 | A Common Personal Health Research Platform — SHIMMER™ and BioMOBIUS™ Intel® Technology Journal | Volume 13, Issue 3, 2009 provides real-time visual feedback to clinicians. The system’s user interface “The system’s user interface allows allows clinicians to select and adjust which data are collected and how the data are processed. The software encapsulates data acquisition and signal processing clinicians to select and adjust which modules, and it allows customization of the sensors [68]. The system is being data are collected and how the data used in the TRIL Clinic in St James’ Hospital, Dublin, Ireland to evaluate scientifically the impact in non-normal gait changes and associated falls’ risks. are processed.”

The MuSensor Project The Tril Centre’s MuSensor project aimed to develop a cheap, unobtrusive, home-based wireless sensor network to monitor an elder’s gait velocity and doorway dwelling times over an eight-week period. The network was developed using BioMOBIUS software and SHIMMER’s low-power 802.15.4 communications module and Passive Infrared (PIR) motion sensor daughterboards. Each activation of a PIR sensor was time-stamped and temporarily stored in the SHIMMER’s RAM, and batches of PIR events were periodically uploaded via the 802.15.4 radio to a central data aggregator located in the home. Data were then transferred to a server via a 3G GSM network. The PIR sensor activation events were used to measure in-home velocity of fallers and non fallers to determine if non performing (i.e., measured outside a clinical environment) gait velocity of fallers and non fallers differs in the home.

Engineering Alertness This project is focused on the development of, and validation of, a prototype BioMOBIUS application that can detect and correct lapses in alertness. The study tests the viability of a self-alert training (SAT) protocol supported by a BioMOBIUS software application with SHIMMER electrodermal activity (EDA) sensors for measuring exosomatic EDA.

A small DC current was passed across two electrodes placed on the skin, and the change in the resistance of the skin to the current was recorded as a function of increased sweat gland activity (Figure 10). For DC measurements, the current is kept constant and skin resistance is measured as changes in voltage.

The protocol implemented in the application is designed to assist older people in increasing attention and alertness levels in their own homes.

Developing the Ecosystem BioMOBIUS software components can be downloaded for free from the BioMOBIUS website (www.biomobius.org) under license from the TRIL Centre. The website features a comprehensive set of documentation to support both application and block developers. A variety of tutorials and sample applications to assist new users are available. A user forum is also available where users can log questions for TRIL engineers. A number of international Figure 10. SHIMMER sensor measuring workshops have also been delivered to help with knowledge transfer in the electrodermal activity (EDA) research community. Source: The TRIL Centre (www.trilcentre.org)

A Common Personal Health Research Platform — SHIMMER™ and BioMOBIUS™ | 139 Intel® Technology Journal | Volume 13, Issue 3, 2009

“BioMOBIUS software remains under The SHIMMER wireless sensor platform was licensed by Intel to Realtime Technologies in 2007. SHIMMER development kits are now commercially active development.” available from SHIMMER Research [42].

Conclusions and Future Work BioMOBIUS software remains under active development. Version 2.0 of the platform was released in late summer 2009. It features improved stability, support for the X10* family of sensors, improved graphing capabilities in the GUI Developer, support for SQLite* and MySQL* databases, improvements to SHIMMER blocks, etc.

A number of new features are currently in development, including ODBC support, Continua-compliant device support (www.continuaalliance.org) and Web services data transports. New features and improvements will be available to download from the BioMOBIUS website. “Hardware components, such as the The BioMOBIUS research platform represents a set of highly integrated software and hardware components developed by the TRIL Centre for SHIMMER wireless sensors platform prototype biomedical research applications. The software components are can be purchased from third-party freely available for download to researchers. Hardware components, such as the SHIMMER wireless sensors platform can be purchased from third- vendors.” party vendors. The platform is open and extensible, allowing users to develop new blocks for the graphical development environment or to integrate new hardware components into the platform. The TRIL Centre has demonstrated the use of the BioMOBIUS platform in a number of research projects including the development of prototype gait analysis platform and EDA monitoring applications. We believe that the BioMOBIUS research platform will prove to be a useful and versatile tool in enabling the biomedical research community. The prototype applications we developed with the BioMOBIUS research platform and described in this article are for research purposes only and are not commercially available.

Acknowledgements We acknowledge the contributions of Barry Greene, Adrian Burns, Cliodhna Ní Scanaill, and the Digital Health Group Product Research and Incubation in the preparation of this article. We also recognize the TRIL Centre’s researchers and their research programs, some of which are reported in this article.

“We believe that the BioMOBIUS research platform will prove to be a useful and versatile tool in enabling the biomedical research community.”

140 | A Common Personal Health Research Platform — SHIMMER™ and BioMOBIUS™ Intel® Technology Journal | Volume 13, Issue 3, 2009

References [1] R. Hongliang, M. Q. H. Meng, and C. Xijun. “Physiological Information Acquisition through Wireless Biomedical Sensor Networks.” In IEEE International Conference on Information Acquisition, Hong Kong and Macau, China, 2005, pp. 483-488.

[2] X. Chen, M. Q. H. Meng, and H. Ren. “Design of Sensor Node Platform for Wireless Biomedical Sensor Networks.” In Proceedings 27th IEEE International Conference of Engineering in Medicine and Biology Society, Shanghai, China, 2005, pp. 4662-4665.

[3] G. S. Gupta, S. C. Mukhopadhyay, B. S. Devlin, and S. Demidenko. “Design of a Low-Cost Physiological Parameter Measurement and Monitoring Device.” In IEEE Conference on Instrumentation and Measurement Technology, Warsaw, Poland, 2007, pp. 1-6.

[4] A. Lymberis. “Smart Wearables for Remote Health Monitoring, from Prevention to Rehabilitation: Current R&D, Future Challenges.” In 4th International IEEE Engineering in Medicine and Biology Society Special Topic Conference on Information Technology Applications in Biomedicine, Birmingham, UK, 2003, pp. 272-275.

[5] J. Kaye, T. Hayes, T. Zitzelberger, J. Yeargers, M. Pavel, H. Jimison, N. Larimer, J. Payne-Murphy, E. Earl, K. Wild, L. Boise, D. Williams, J. Lundell, and E. Dishman. “Deploying Wide-Scale in-home assessment technology.” Technology and Aging: Selected papers from the 2007 International Conference on Technology and Aging, Toronto, Canada, 2007, pp. 19-26.

[6] E. M. Tapia, S. S. Intille, and K. Larson. “Activity Recognition in the Home Using Simple and Ubiquitous Sensors.” Pervasive Computing, vol. 3001, Berlin/Heidelberg: Springer, 2004, pp. 158-175.

[7] R. Jafari, W. Li, R. Bajcsy, S. Glaser, and S. Sastry. “Physical Activity Monitoring for Assisted Living at Home.” In 4th International Workshop on Wearable and Implantable Body Sensor Networks, Aachen, Germany, 2007, pp. 213-219.

[8] T. Degen, H. Jaeckel, M. Rufer, and S. Wyss. “SPEEDY:a Fall Detector in a Wrist Watch.” In Proceedings 7th IEEE International Symposium on Wearable Computers, White Plains, New York, 2003, pp. 184-187.

[9] J. Kaye. “Home-Based Technologies: A New Paradigm for Conducting Dementia Prevention Trials.” Alzheimers and Dementia, vol. 4, no. 1, pp. S60-S66, 2008.

[10] M. Morris, J. Lundell, and E. Dishman, “Catalyzing Social Interaction with Ubiquitous Computing: a Needs Assessment of Elders Coping with Cognitive Decline.” In Conference on Human Factors in Computing Systems, Vienna, Austria, 2004, pp. 1511-1514.

A Common Personal Health Research Platform — SHIMMER™ and BioMOBIUS™ | 141 Intel® Technology Journal | Volume 13, Issue 3, 2009

[11] P. Bagnall, G. Dewsbury, V. Onditi, and I. Sommerville. “Promoting Virtual Social Interaction with Older People.” In International Workshop on Cognitive Prostheses and Assisted Communication, Sydney, Australia, 2006.

[12] D. G. Mcilwraith, J. Pansiot, S. Thiemjarus, B. P. L. Lo, and G. Z. Yang. “Probabilistic Decision Level Fusion for Real-time Correlation of Ambient and Wearable Sensors.” In 5th International Summer School and Symposium on Medical Devices and Biosensors, Hong Kong, China, 2008, pp. 117-120.

[13] L. Atallah, M. ElHelw, J. Pansiot, D. Stoyanov, L. Wang, B. Lo, and G. Z. Yang. “Behaviour Profiling with Ambient and Wearable Sensing.” In 4th International Workshop on Wearable and Implantable Body Sensor Networks, Aachen, Germany, 2007, pp. 133-138.

[14] C. Otto, Aleksandar Milenkovic, Corey Sanders, and E. Jovanov. “System Architecture of a Wireless Body Area Sensor Network for Ubiquitous Health Monitoring.” Journal of Mobile Multimedia, vol. 1, no. 4, pp. 307-326, 2006.

[15] T. Westeyn, P. Presti, and T. Starner. “ActionGSR: A Combination Galvanic Skin Response-Accelerometer for Physiological Measurements in Active Environments.” In 10th IEEE International Symposium on Wearable Computers, Montreux, Switzerland, 2006, pp. 129-130.

[16] D. Roetenberg, P. J. Slycke, and P. H. Veltink. “Ambulatory Position and Orientation Tracking Fusing Magnetic and Inertial Sensing.” In IEEE Transactions on Biomedical Engineering, vol. 54, no. 5, pp. 883- 890, 2007.

[17] S. Bamberg, A. Y. Benbasat, D. M. Scarborough, D. E. Krebs, and J. A. Paradiso. “Gait Analysis using a Shoe-Integrated Wireless Sensor System.” In IEEE Transactions on Information Technology in Biomedicine, vol. 12, no. 4, pp. 413-423, 2008.

[18] S. J. Bellis, K. Delaney, B. O’Flynn, J. Barton, K. M. Razeeb, and C. O’Mathuna. “Development of Field Programmable Modular Wireless Sensor Network Nodes for Ambient Systems.” In Computer Communications, vol. 28, no. 13, pp. 1531-1544, 2005.

[19] L. Benini, E. Farella, and C. Guiducci. “Wireless Sensor Networks: Enabling Technology for Ambient Intelligence.” Microelectronics Journal, vol. 37, no. 12, pp. 1639-1649, 2006.

[20] W.-Y. Chung and S.-J. Oh. “Remote Monitoring System with Wireless Sensors module for Room Environment.” Sensors and Actuators B: Chemical, vol. 113, no. 1, pp. 64-70, 2006.

142 | A Common Personal Health Research Platform — SHIMMER™ and BioMOBIUS™ Intel® Technology Journal | Volume 13, Issue 3, 2009

[21] H. Dubois-Ferriere, R. Meier, L. Fabre, and P. Metrailler. “TinyNode: a Comprehensive Platform for Wireless Sensor Network Applications.” In Proceedings 5th International Conference on Information Processing in Sensor Networks, Nashville, Tennessee, 2006, pp. 358-365.

[22] T. Ahola, P. Korpinen, J. Rakkola, T. Ramo, J. Salminen, and J. Savolainen. “Wearable FPGA based Wireless Sensor Platform.” In Proceedings 29th IEEE International Conference of Engineering in Medicine and Biology Society, 2007, pp. 2288-2291.

[23] B. Gyselinckx, C. Van Hoof, J. Ryckaert, R. F. Yazicioglu, P. Fiorini, and V. Leonov. “Human++: Autonomous Wireless Sensors for Body Area Networks.” In Proceedings IEEE Conference on Custom Integrated Circuits, San Jose, California, 2005, pp. 13-19.

[24] J. Penders, B. Gyselinckx, R. Vullers, M. De Nil, V. Nimmala, J. van de Molengraft, F. Yazicioglu, T. Torfs, V. Leonov, P. Merken, and C. Van Hoof. “Human++: From Technology to Emerging Health Monitoring Concepts” In 5th International Summer School and Symposium on Medical Devices and Biosensors, Hong Kong, China, 2008, pp. 94-98.

[25] L. Girod, N. Ramanathan, J. Elson, T. Stathopoulos, M. Lukac, and D. Estrin. “Emstar: A Software Environment for Developing and Deploying Heterogeneous Sensor-Actuator Networks.” In ACM Transactions on Sensor Networks, vol. 3, no. 3, p. 13, 2007.

[26] S. Hengstler and H. Aghajan. “WiSNAP: A Wireless Image Sensor Network Application Platform.” In 2nd International Conference on Testbeds and Research Infrastructures for the Development of Networks and Communities, Barcelona, Spain, 2006, pp. 6-12.

[27] Z. Walker, M. Moh, and T.-S. Moh. “A Development Platform for Wireless Sensor Networks with Biomedical Applications.” In 4th IEEE Consumer Communications and Networking Conference, Las Vegas, Nevada, 2007, pp. 768-772.

[28] D. Simon, C. Cifuentes, D. Cleal, J. Daniels, and D. White. “Java™ on the Bare Metal of Wireless Sensor Devices:.” In Proceedings 2nd International Conference on Virtual Execution Environments, Ottawa, Ontario, 2006, pp. 78-88.

[29] R. Slyper and J. Hodgins, “Action Capture with Accelerometers.” In ACM SIGGRAPH/Eurographics Symposium on Computer Animation, Dublin, Ireland, 2008, pp. 193-200.

[30] D. Spelmezan, A. Schanowski, and J. Borchers. “Rapid Prototyping for Wearable Computing.” In International Semantic Web Conference, Karlsruhe, Germany, 2008.

A Common Personal Health Research Platform — SHIMMER™ and BioMOBIUS™ | 143 Intel® Technology Journal | Volume 13, Issue 3, 2009

[31] Y. Jianchu and S. Warren, “Design of a Plug-and-Play Pulse Oximeter.” In Proceedings Second Joint IEEE 24th Annual Conference on Engineering in Medicine and Biology and the Annual Fall Meeting of the Biomedical Engineering Society Conference, Houston, Texas, 2002, pp. 1752-1753.

[32] M. Lascu and D. Lascu. “LabVIEW Electrocardiogram Event and Beat Detection.” WSEAS Transactions on Computer Research, vol. 3, no. 1, pp. 9-18, 2008.

[33] T. Paing, J. Morroni, A. Dolgov, J. Shin, J. Brannan, R. Zane, and Z. Popovic. “Wirelessly-Powered Wireless Sensor Platform.” In European Conference on Wireless Technologies, Munich, Germany, 2007, pp. 241- 244.

[34] S. Scapellato, F. Cavallo, C. Martelloni, and A. M. Sabatini. “In-Use Calibration of Body-Mounted Gyroscopes for Applications in Gait Analysis.” Sensors and Actuators A: Physical, vol. 123-124, no. pp. 418- 422, 2005.

[35] S. Campbell, J.-P. Chancelier, and R. Nikoukhah. Modeling and Simulation in Scilab/Scicos, 1st ed. New York: Springer, 2006.

[36] PROATECH BioEra—Visual Designer for BioFeedback webpage. Available at http://www.bioera.net/

[37] C. Veigl and J. Wikerson. “BrainBay—an OpenSource Biosignal Project.” Available at brainbay.lo-res.org/

[38] S. A. Jawed, D. CattIn, M. Gottardi, N. Massari, R. Oboe, and A. Baschirotto. “A Low-Power Interface for the Readout and Motion- Control of a MEMS Capacitive Sensor.” In 10th IEEE International Workshop on Advanced Motion Control, Trento, Italy, 2008, pp. 122-125.

[39] J. Music, R. Kamnik, and M. Munih. “Model Based Inertial Sensing of Human Body Motion Kinematics in Sit-to-Stand movement.” Simulation Modelling Practice and Theory, vol. 16, no. 8, pp. 933-944, 2008.

[40] TRIL Centre, Technology Research for Independent Living webpage (2009). Available at www.trilcentre.org

[41] SHIMMER wireless sensor platform webpage (2009, January). Available at http://www.csi.ucd.ie

[42] Sensing Health with Intelligence, Modularity, Mobility and Experimental Reusability webpage (2009). Available at www.shimmer‑research.com

144 | A Common Personal Health Research Platform — SHIMMER™ and BioMOBIUS™ Intel® Technology Journal | Volume 13, Issue 3, 2009

[43] A. Camurri, P. Coletta, M. Demurtas, M. Peri, A. Ricci, R. Sagoleo, M. Simonetti, G. Varni, and G. Volpe, “A Platform for Real-Time Multimodal Processing.” In 4th Conference in Sounds and Music Computing, Lefkada, Greece, 2007, pp. 354-358.

[44] A. Camurri, P. Coletta, G. Varni, and S. Ghisio, “Developing Multimodal Interactive Systems with EyesWeb XMI.” In Conference on New Interfaces for Musical Expression, New York, 2007, pp. 305-308.

[45] A. Camurri, B. MazzarIno, M. Ricchetti, R. Timmers, and G. Volpe, “Multimodal Analysis of Expressive Gesture in Music and Dance Performances.” Gesture-based Communication in Human-Computer Interaction, vol. 2915, A. Camurri and G. Volpe, Eds. Berlin/ Heidelberg: Springer Verlag, 2004, pp. 357-358.

[46] A. Camurri, B. MazzarIno, G. Volpe, P. Morasso, F. Priano, and C. Re. “Application of Multimedia Techniques in the Physical Rehabilitation of Parkinson’s Patients.” Visualization and Computer Animation, vol. 14, no. 5, pp. 269-278, 2003.

[47] M. L. Rinman, A. Friberg, B. Benediksen, D. Cirotteau, S. Dahl, I. Kjellmo, B. MazzarIno, and A. Camurri, “Ghost in the Cave—An Interactive Collaborative Game Using Non-verbal Communication.” Gesture-based Communication in Human-Computer Interaction, vol. 2915, A. Camurri and G. Volpe, Eds. Berlin/Heidelberg: Springer Verlag, 2004, pp. 403-404.

[48] Raw Material Software—JUCE webpage (2005). Available at www.rawmaterialsoftware.com/juce/

[49] MSP430 16-bit Ultra-Low Power MCUs webpage (2009). Available at focus.ti.com/

[50] B. Lo, S. Thiemjarus, R. King, and G. Z. Yang. “Body Sensor Network—A Wireless Sensor Platform for Pervasive Healthcare Monitoring.” In 3rd International Conference on Pervasive Computing, Munich, Germany, 2005, pp. 77-80.

[51] J. Polastre, R. Szewczyk, and D. Culler. “Telos: Enabling Ultra- Low Power Wireless Research.” In 4th International Symposium on Information Processing in Sensor Networks, Los Angeles, CA, 2005, pp. 364-369.

[52] G. Werner-Allen, K. Lorincz, M. Ruiz, O. Marcillo, J. Johnson, J. Lees, and M. Welsh. “Deploying a Wireless Sensor Network on an Active Volcano.” IEEE Internet Computing, vol. 10, no. 2, pp. 18-25, 2006.

[53] TinyOS Community Forum webpage (2009 June). Available at http://www.tinyos.net

A Common Personal Health Research Platform — SHIMMER™ and BioMOBIUS™ | 145 Intel® Technology Journal | Volume 13, Issue 3, 2009

[54] A. Dunkels, B. Gronvall, and T. Voigt. “Contiki — A Lightweight and Flexible Operating System for Tiny Networked Sensors.” In 29th Annual IEEE International Conference on Local Computer Networks, Tampa, Florida, 2004, pp. 455-462.

[55] A. Eswaran, A. Rowe, and R. Rajkumar. “Nano-RK: An Energy- Aware Resource-Centric RTOS for Sensor Networks.” In 26th IEEE International Real-Time Systems Symposium, Miami, Florida, 2005, pp. 5-8.

[56] M. Cirinei, A. Mancina, D. Cantini, P. Gai, and L. Palopoli. “An Educational Open Source Real-Time Kernel for Small Embedded control systems.” Computer and Information Sciences - ISCIS 2004, vol. 3280, BerlIn/Heidelberg: Springer, 2004, pp. 866-875.

[57] S. Bhatti, J. Carlson, H. Dai, J. Deng, J. Rose, A. Sheth, B. Shucker, C. Gruenwald, A. Torgerson, and R. Han. “MANTIS OS: an embedded multithreaded operating system for wireless micro sensor platforms.” Mobile Network Applications, vol. 10, no. 4, pp. 563-579, 2005.

[58] V. Tsetsos, G. Alyfantis, T. Hasiotis, O. Sekkas, and S. Hadjiefthymiades. “Commercial Wireless Sensor Networks: Technical and Business Issues.” In 2nd Annual Conference on Wireless On-demand Network Systems and Services, St Moritz, Switzerland, 2005, pp. 166- 173.

[59] SourceForge webpage (2009, June). Available at http://sourceforge.net

[60] J. Hill, M. Horton, R. Kling, and L. Krishnamurthy. “The Platforms Enabling Wireless Sensor Networks.” Communications of the ACM, vol. 47, no. 6, pp. 41-46, 2004.

[61] D. Benitez, P. A. Gaydecki, A. Zaidi, and A. P. Fitzpatrick. “The Use of the Hilbert Transform In ECG Signal Analysis.” Computers in Biology and Medicine, vol. 31, no. 5, pp. 399-406, 2001.

[62] P. de Chazal, C. Heneghan, E. Sheridan, R. Reilly, P. Nolan, and M. O’Malley. “Automated Processing of the Single-Lead Electrocardiogram for the Detection of Obstructive Sleep Apnoea.” IEEE Transactions on Biomedical Engineering, vol. 50, no. 6, pp. 686-696, 2003.

[63] A. Salarian, H. Russmann, F. J. G. Vingerhoets, C. Dehollain, Y. Blanc, P. R. Burkhard, and K. Aminian. “Gait Assessment In Parkinson’s Disease: Toward an Ambulatory System for Long-Term Monitoring.” IEEE Transactions on Biomedical Engineering, vol. 51, no. 8, pp. 1434- 1443, 2004.

[64] P. E. Shrout and J. L. Fleiss. “Intraclass Correlations: Uses In Assessing Rater Reliability.” Psychological Bullet, vol. 86, no. 2, pp. 420-428, 1979.

146 | A Common Personal Health Research Platform — SHIMMER™ and BioMOBIUS™ Intel® Technology Journal | Volume 13, Issue 3, 2009

[65] K. J. O’Donovan, B. R. Greene, D. McGrath, R. O’Neill, A. Burns, and B. Caulfield. “SHIMMER: A New Tool for Temporal Gait Analysis.” In Proceedings 31st IEEE Engineering in Medicine Biology Conference, Minneapolis, MN, 2009.

[66] K. Monaghan, E. Delahunt, and B. Caulfield, “Increasing the Number of Gait Trial Recordings Maximises Intra-Rater Reliability of the CODA Motion Analysis System.” Gait & Posture, vol. 25, no. 2, pp. 303-315, 2007.

[67] V. Maynard, A. M. O. Bakheit, J. Oldham, and J. Freeman. “Intra- Rater and Inter-Rater Reliability of Gait Measurements with CODA MPX30 Motion Analysis System.” Gait & Posture, vol. 17, no. 1, pp. 59-67, 2003.

[68] K. O’Donovan, T. Dishongh, T. Foran, D. Leahy, and C. Ni Scanaill. “The Development of a Clinical Gait Analysis System.” In Conference on Ambulatory Monitoring of Physical Activity and Movement, Amsterdam, Netherlands, 2008, p. 155.

Author Biographies Michael McGrath is Senior Technologist in Intel’s Digital Health Group. He is also a co-Principal Investigator in the TRIL Centre focusing on development technologies to support independent living research. Michael McGrath joined Intel in 1999. He has held positions as an automation engineer working in Ireland Fab Operations and as a researcher in IT Research and Innovation. He moved to his current position as a technologist in the Product Research and Incubation group in 2006. His areas of interest include sensors, wireless communications, assisted living technologies, intelligent user interfaces, data fusion, and data management techniques. Michael received his B.Sc. in Analytical Science from Dublin City University in 1992 and a Ph.D. in sensors and instrumentation from Dublin City University in 1995. His e-mail is michael.j.mcgrath at intel.com.

Terrance (Terry) J. Dishongh, Ph.D., is currently a Senior Principal Engineer in Intel’s Digital Health Group. In his twelve years at Intel he has been awarded 45 patents with an additional 44 pending and has published books on wireless sensors and electronic packaging. Dr. Dishongh has held faculty positions at the University of Maryland, College Park and the State University of New York at Buffalo. His e-mail is terry.dishongh at intel.com.

A Common Personal Health Research Platform — SHIMMER™ and BioMOBIUS™ | 147 Intel® Technology Journal | Volume 13, Issue 3, 2009

Gathering the Evidence: Supporting Large-Scale Research Deployments

Contributors

Tamara L. Hayes Abstract OHSU & ORCATECH Independent living technologies have tremendous potential to facilitate the study of aging and chronic disease. By collecting medical and behavioral data Misha Pavel continuously as people live their daily lives, we are able to form a much clearer OHSU & ORCATECH picture of the changes in people’s health, and the efficacy of treatments over time, than is possible by using periodic clinical visits. However, deploying these Jeffrey A. Kaye technologies in large-scale clinical trials poses unique challenges. In this article OHSU & ORCATECH we review those areas in which independent living technologies can provide significantly improved research data, and the challenges that must be overcome to effectively gather those data. We propose approaches, both technical and procedural, which enable efficient management of large-scale research studies, and we discuss scalability issues and their possible solutions.

Introduction People aged 65 and older are the fastest growing segment of the United States Index Words population: it is predicted that this number will more than double by the year 2030. Forty-two percent of this age group reported some type of long-lasting Aging condition or disability [6]. Furthermore, over 20 percent of people 85 and Home Monitoring older have a limited capacity for independent living [5], requiring continuous Clinical Research monitoring and daily care. Almost 6 percent of private health expenditures Scalability and approximately 13 percent of all public healthcare expenditures in 2005 were spent on home healthcare and nursing-home care for older adults [4]. The convergence of growing numbers of seniors with attendant chronic illness, the rising costs and complexity of care, and the inability to effectively develop and apply increasing knowledge of how to treat or manage these health declines merge to create a well-recognized crisis of healthcare in America.

A major challenge to research into caring for the aged is the reliable assessment of behavior and clinical status across multiple related domains (for example, cognitive, social, physiological, environmental). Clinical research has not fundamentally changed its tools of conduct and assessment paradigm since the beginning of the computer age. Thus, clinical research is currently still rooted largely in traditional clinical practice methodologies, where the basic paradigm of assessment is a clinical interview or examination, relying on the recall of the volunteer, performed at a brief moment in time, at a location convenient “Over 20 percent of people 85 and for the researcher, not the volunteer. Even when technologies are deployed in older have a limited capacity for the home, they are often used in a brief or episodic way (for example, 24-hour heart monitoring or two weeks of actigraphy) and not fully integrated into the independent living.” clinical data stream and workflow.

148 | Gathering the Evidence: Supporting Large-Scale Research Deployments Intel® Technology Journal | Volume 13, Issue 3, 2009

There have been several research efforts that have developed solutions for monitoring and assessment of individuals in their homes. These approaches use unobtrusive wireless sensors placed in a subject’s home (such as motion sensors, door sensors, and environmental sensors), or technologies that the subject wears or uses (such as medication adherence monitors, telemedicine boxes, personal computers, and body-worn location and activity sensors). Studies that use these technologies have found that even simple technologies can be valuable in assessing behavior in the home [13], and there are a growing number of researchers developing and evaluating emerging technologies for in-home assessment and health intervention. These technologies have the potential to transform the clinical research landscape by enabling the simultaneous assessment of an individual’s activity, function, and physiologic Measured Function status in a particular location across a wide array of time scales (for example, seconds to seasons), and by creating the potential to detect subtle changes in that status, by comparing this rich multidimensional data stream with baseline data from prior observations.

The general approach to this kind of study is to monitor a subject (or patient) over a period of time to establish a baseline, after which deviation from that baseline can be used to detect clinically-relevant change. Without this baseline, Month 1 Month 6 Month 12 patients must instead be compared to a population norm–and what is normal for one individual may be a cause for alarm in someone else. Furthermore, by sampling infrequently, true trends may be masked, as a person has good Figure 1: Problems with Infrequent Monitoring days and bad days. Figure 1 illustrates how infrequent sampling can lead Source: Alzheimer’s and Dementia: Hayes TL, to substantial aliasing errors. In the figure, you will see that the top panel Abendroth F, Adami, A, Pavel, M, Zitzelberger TA, depicts test scores taken during a standard clinic visit, taken at six-month Kaye JA intervals, for two different patients. Then note that the bottom panel depicts how continuous assessment could reveal a very different picture. Thus, the continuous nature of the in-home sensor data collection allows many data points to be collected, providing better time resolution for looking at changes and for reducing the overall variance. This continuous collection also allows changes to be detected in a shorter period of time. The immediacy and frequency of the collection of these data form the basis for the transition to a true substantiation of personalized medicine and directly aligns with new healthcare delivery trends and mandates, including the wide-scale build-out of electronic medical or health records [19], the Medical Home [8], and comparative-effectiveness research (CER) [14].

The Role for In-Home Monitoring in Research In-home monitoring has wide applicability in routine care and intervention. In particular, in-home monitoring allows tracking of conditions that evolve slowly over time or that have poor demarcation of onset; and it allows the tracking of conditions that transition to new clinical states, such as depression, frailty, “In-home monitoring has wide and cognitive impairment. Suitable sensors placed in the home may also allow immediate detection of acute events that are rare or irregularly occurring, such applicability in routine care and as falls, naps, or transient neurological events. intervention.”

Gathering the Evidence: Supporting Large-Scale Research Deployments | 149 Intel® Technology Journal | Volume 13, Issue 3, 2009

Continuous monitoring would also greatly enhance clinical trials of both drug and behavioral therapies, by allowing the ongoing assessment of patient adherence to the trial or therapy regimen, and by providing those administering these trials the ability to assess side-effects and health changes throughout the day. Traditionally, both adherence and side-effects in clinical trials are measured through self-report, which at best corresponds modestly to actual medication taking and to true physiological changes [17, 20] (see Vignette 1 at the end of this article). The immediate identification of side-effects accurately aligned to the time of treatment would provide valuable insights into the efficacy of treatments. Data about patient behaviors would allow improved coaching of the patient to ensure better adherence to the treatment protocol. Furthermore, in studies of recovery (such as from a hip fracture or stroke), instead of assessing outcomes by using subjective measures, for example, arbitrary scale numbers at a fixed time point such as 3-6 months after the event, new measures could be generated from recovery curves that represent the rate of functional recovery (such as total activity or walking speed). One of the most interesting benefits of using the continuous home assessment platform is not only the “In many cases in aging research, the opportunity to examine the trends in these objective measures, but equally variance in a measure may be more importantly, to determine their variability over time. In many cases in aging research, the variance in a measure may be more informative than the absolute informative than the absolute value.” value [12, 16].

The ability to improve on self-report also has the potential to transform epidemiological studies. Currently, epidemiological survey and assessment methodologies tend to rely on sparsely spaced questionnaires that depend on recall of events. This approach makes it difficult to identify data or events with detailed temporal or spatial precision, and it limits their ecological validity. A way to improve this state is to bring the locus of assessment into the home or community by providing a means of recording continuously, in real-time, events as they occur and where they occur. If full sensor networks and on-line reporting capabilities were incorporated into thousands of households, there would be an opportunity to collect changing health outcomes at the point “The metrics of interest would become of occurrence. In this case the metrics of interest would become true rates of change and not simply binary “present or absent” estimates. Item-level data true rates of change and not simply would be much less reliant on recall (e.g., “In the last six months how often did you wake at night?”) and instead reflect actual events and more objective binary ‘present or absent’ estimates.” measures (e.g., the number of bathroom visits, hours in bed, and so on).

Achieving Large-Scale Studies Recent advances in ubiquitous computing and sensor network methodologies make large-scale, in-home research studies feasible. When coupled with other convergent technologies and methods, such as ecological momentary assessment and telemedicine, the potential to transform the practice of clinical research is staggering. However, in spite of the number of smart homes that are now being used to develop and evaluate in-home monitoring technologies, we are a long way from wide-scale deployment. Experience has shown that solutions that appear to work in a controlled laboratory environment, even

150 | Gathering the Evidence: Supporting Large-Scale Research Deployments Intel® Technology Journal | Volume 13, Issue 3, 2009 when that environment is intended to simulate a home living situation, never work as expected in the field. For this reason, we have developed a staged approach to designing solutions for large-scale, in-home monitoring.

Assess Attitudes and Beliefs

SURVEY

Evaluate in Point-of-Care Lab

Internet

Test in Community Living Lab

Internet

Deploy on a Large Scale

Internet Internet Internet

Internet

Figure 2: Scaling In-home Assessments from the Lab to Communities Source: OHSU

Figure 2 illustrates the stages we use from technology inception to wide-scale “We conduct focus groups and deployment. First, we conduct focus groups and interviews to inform the technology design, and in particular to understand attitudes towards the interviews to inform the technology technology and to identify key usability and acceptance issues. In the second design.” stage, technologies are tested in a smart-home environment (the Point of Care Laboratory) by using convenience samples of participants. Data collection can be done quickly and easily, under controlled experimental conditions, and therefore this approach is appropriate for initial evaluations of technologies. However, these tests will not typically allow the collection of outcome measures that are representative of particular patient populations. Therefore, part of this

Gathering the Evidence: Supporting Large-Scale Research Deployments | 151 Intel® Technology Journal | Volume 13, Issue 3, 2009

“Subjects brought into a smart home stage includes bringing patients or subjects from the target cohort into the smart home, to collect additional data about how the population of interest will not behave as they would in their will interact with the sensor systems. This is useful for improving our own home.” understanding of the limitations and capabilities of the sensor technologies. Clearly, subjects brought into a smart home will not behave as they would in their own home, and therefore data collected in such an environment has less ecological validity than those collected in the individual’s home. Hence, the third stage entails taking the technology into the field — that is, instrumenting a small number (10-20) of patients’ homes — to provide much better outcomes’ data. We use the ORCATECH Living Lab for this deployment. The Living Lab is a group of senior volunteers who have agreed to allow technologies to be evaluated in their homes on an ongoing basis. A core set of technologies is continually maintained in the homes, and clinical and neurological assessments are conducted on this cohort on a semi-annual basis. This deployment helps to reveal problems that may arise when the technology is used with different home construction materials, infrastructure support, living arrangements, or patient populations. Once a technology has been field-hardened in the Living Lab, we are ready to begin to scale it to larger studies. “The Living Lab is a group of senior In spite of the challenges of scaling emerging technologies, there is a clear need volunteers who have agreed to allow to do so. The high inter-individual variability in behavior and health outcomes technologies to be evaluated in their drives the need for large studies. For example, while small randomized clinical trials of cholinesterase inhibitors have shown modest improvements homes on an ongoing basis.” in cognitive function, trials with treatment and control groups of 200 – 300 people are typically necessary to have sufficient power to show improvements from baseline [18]. Population-based studies require a much larger sample size, since these studies typically control less for co-morbidities, and therefore the variance in those samples is much greater.

In the process of deploying various in-home monitoring and intervention technologies in more than 400 homes around the United States, we have learned a great deal about the challenges of scaling the technologies for research. These challenges relate in part to the complex, heterogeneous nature of the data that are collected, in part to the challenges of adding the use of technology into the already difficult problem of recruiting and retaining patient populations (who may be technophobic), and in part to the fact that the global infrastructure and sensors designed specifically for collecting behavioral health data do not yet exist. In the remainder of this article we review each of these challenges and propose possible solutions. “The high, inter-individual variability in behavior and health outcomes drives the need for large studies.”

152 | Gathering the Evidence: Supporting Large-Scale Research Deployments Intel® Technology Journal | Volume 13, Issue 3, 2009

The Nature of the Data Continuous, in-home monitoring studies generate a vast amount of data “These datasets may contain mixtures pertaining to the daily activity of subjects. These datasets may contain mixtures of complex data, ranging from rare and random events to continuous signals. of complex data, ranging from rare Uniquely, these data inherently describe multiple data domains simultaneously: and random events to continuous space or location, time, activity, physiological function, neuropsychological function, etc. These data provide more accurate and detailed information signals.” than can be captured by traditional methods of self-report, and they hold information about patterns of activity in the home that could be used to examine a multitude of hypotheses about behaviors, such as restlessness at night, nocturia, patterns of use of the kitchen (and refrigerator), outings from the home, and acute events such as falls or stroke. However, the very strengths of these data — such as the high sampling frequency and the many different types of information gathered — make them challenging to manage, study, monitor, and interpret.

There are three key types of data that are needed to support large-scale research studies. The first type of data that is of interest is the raw data from the sensors themselves. These data reflect physical responses of the sensors to the behaviors of the research participants. For example, the instantaneous signal strength from a body-worn radio frequency identification (RFID) sensor as an individual walks through the home, the time at which a motion sensor fires when somebody passes by, and the force measured by a load cell under the support of a bed as a person breathes, are all signals that must be captured for subsequent processing and inference. Some of these (signal strength, load “These data may also be tied cell force) are sampled at regular intervals and form a time series sampled at a relatively high frequency. Others (sensor firing time) are event-driven, and the to location, and therefore they amount of data collected therefore depends upon the activity being sensed. may require a spatio-temporal Furthermore, these data may also be tied to location, and therefore they may require a spatio-temporal representation. representation.”

The second type of data is metadata about the sensors, the state of the sensors, and participants themselves. If one is relying on signal strength to estimate location, for example, then the battery level of the sensor becomes important to determining the likely variance and accuracy of the estimate. Similarly, if a motion sensor appears to have stopped firing, it is important to know if there is a hardware or communications problem or if the person being monitored has simply not passed near the sensor.

The third type of data is derived data, which come from inference algorithms “The derivation of such measures that are used to extract information from the raw data. Often these derived and metrics from in-home sensors data actually arise from sophisticated statistical models that fuse data from multiple sensors to make inferences about behaviors and health outcomes. is an area of active research, and at Derived data could be as basic as an estimate of respiration or walking speed, this point, only a small number of or as complex as a determination of how well a person was performing a particular activity of daily living (ADL). The derivation of such measures and measures have been developed and metrics from in-home sensors is an area of active research, and at this point, only a small number of measures have been developed and validated. validated.”

Gathering the Evidence: Supporting Large-Scale Research Deployments | 153 Intel® Technology Journal | Volume 13, Issue 3, 2009

“As purposed sensors become more The applicability of each of these types of data typically relates to the maturity of the sensor technology. For example, a pulsed ultra-wideband location sensor common, what was once derived data could provide data in the form of the time-difference-of-arrival measures, or will become raw data.” could provide an estimate of the individual’s location, based on a sophisticated algorithm that integrates data from multiple nodes. If the sensor were well- tested and its behavior in home environments was well-understood, then the expense of storing the raw data might not be warranted. Instead, the location estimates would be sufficient. Thus, as purposed sensors become more common, what was once derived data will become raw data, which in turn will be used to develop more sophisticated and complex measures.

Managing Large-Scale Studies “Unlike standard clinical trials, Conducting a large-scale, in-home trial differs from conducting a small pilot study as well as from conducting standard clinical trials. First, in large studies, in‑home studies require central the recruitment of subjects is often done across multiple sites in different cities tracking and remote systems or states, which raises issues of staff training, forming relationships with local infrastructure providers at each location, dealing with regional differences in management capabilities.” attitudes towards the technologies, and cross-site coordination. Furthermore, unlike standard clinical trials, in-home studies require central tracking and remote systems management capabilities. Second, the innumerable possible variations on home configurations and environmental conditions means that in-home technology must be adaptable to many environments, which is not always straightforward. Third, short installation time and high equipment reliability become essential in a large study. Fourth, because in-home technologies can potentially produce a large amount of raw data, centralized data management and a means of visualizing and reviewing the data are essential as well.

“Short installation time and high In our studies, the data and the remote installations are managed by using the ORCATECH Management Console (OMC), a multi-tiered remote equipment reliability become essential monitoring application built from Open Source tools. This software tool was designed to support the activities of both clinical and technical research in a large study.” assistants. OMC provides a central view into the data and has the following capabilities:

•• View subject and home information (floor plans, sensor placements, demographic info, etc.). •• Track recruitment and subject status, including generating alerts for upcoming scheduled visits for assessments or equipment installation. •• Track contact with subjects, for assessing total cost of installations (for example, time and frequency of support phone calls and visits). •• Roll up the health of the installed systems (and raise alerts if sensors fail to fire, or data fails to transfer). •• Provide summary views of the data to enable data validation on an ongoing basis. Figure 3 shows an example in which computer use is displayed for the two residents in a home. In this graph, we also plot in-home activity, since we should not expect any computer use if nobody is home.

154 | Gathering the Evidence: Supporting Large-Scale Research Deployments Intel® Technology Journal | Volume 13, Issue 3, 2009

Alerts X10 MedTracker Events Phone Monitor Events Ekahau Computer Usage Data Outage Start Date: 2009-07-01 End Date: 2009-07-31 Submit Equipment Home Activity Graphs 21,000 20,000 12 19,000 11 Personal 18,000 17,000 10 Pets 16,000 9 15,000 Residents 14,000 8

13,000 Logins 12,000 7 11,000 6

X10 Events 10,000 9,000 5 8,000 4 7,000 6,000 3 5,000 4,000 2 3,000 1 2,000 2009-07-05 2009-07-12 2009-07-19 2009-07-26 Date X10 Events (738) (760)

Figure 3: Checking Data Integrity with the ORCATECH Management Console Source: OHSU

In the following sections we discuss the core issues this study management tool was designed to support — and those issues that still need to be addressed to make this tool ultimately scalable to studies that include thousands of subjects.

Cohort Management One of the greatest challenges for any large-scale study is the recruitment and retention of the participant cohort. The challenges are exacerbated when the participant population is asked to use or interact with unfamiliar technology. In aging studies, many participants have never used a computer or cell phone and are highly intimidated by the thought of learning to do so. In one of our longitudinal studies, we require the participants to use a personal computer on a weekly basis. Approximately 32 percent of the participants we recruited already owned their own computer. The majority of the remainder, 169 subjects, were provided with a computer and trained to use it. A significant effort was spent in developing a training program that included skills ranging “A significant effort was spent in from the use of the mouse and how to double-click, to sending and receiving e-mail (a definition of computer literacy). The training program comprised developing a training program that six hour-long lessons that were taught to small groups. Clearly, this is not an included skills ranging from the use approach that scales to tens of thousands of subjects. However, many computer classes for seniors are now available through libraries and senior centers, and of the mouse and how to double-click, through partnerships with such programs, it is possible that large-scale studies to sending and receiving e-mail (a requiring frequent computer use may be viable even in studies of older adults, such as our studies (where the mean age is over 80 years old). definition of computer literacy).”

Gathering the Evidence: Supporting Large-Scale Research Deployments | 155 Intel® Technology Journal | Volume 13, Issue 3, 2009

“For an individual with a tremor, Many of the technologies available commercially, such as cell phones and mice, are difficult to use for older adults. For an individual with a tremor, pressing pressing a single key on a cell phone or a single key on a cell phone or moving a mouse to a target on a screen may moving a mouse to a target on a screen be literally impossible to do with accuracy. For many elderly, the text on a cell phone is too small to see, even with glasses. Recent research has provided may be literally impossible to do with considerable insight into the usability needs of the aging population [2, 3, 15, accuracy.” 21], and large-scale studies must be prepared to adapt their technologies to accommodate the wide variety of needs in this population.

This difficulty with the technology translates into a retention problem for research studies: participants that become frustrated are more likely to drop out of the study. Therefore, in technology-based studies, even more than in traditional clinical studies, it is vital to maintain active personal relationships with the participants. In our own studies, this is done by maintaining continuity of contact as much as possible. Each participant has an assigned research assistant that they know well and with whom they are used to interacting. In addition, we maintain a help-line for the participants to call with any questions they may have about the technology. By knowing that they will receive a response about their problem within a day, and that somebody wants to help them through it, the participants are able to give voice to their frustrations, which is often all that is needed to defuse the situation. Of course, not all participants are frustrated by the technology.

“It is vital to maintain active personal In aging studies, transitions are part of life. Many participants will move to smaller homes, to assisted living, or they will move in with family over the relationships with the participants.” course of a study. Even more participants will become ill, have an exacerbation of a chronic disease, or undergo surgery. In a study of behaviors or health outcomes, these transitions will have a significant impact on the data. It is important to track these changes, and in some cases to collect additional data that will help in the interpretation of changes in the data that result from the life transition. For example, in our Intelligent Systems for Assessing Aging Changes (ISAAC) project, we are interested in changes in walking speed over time. In this project, 6.8 percent of our participants have moved in the last eight months. Often, their new homes are smaller, and therefore the participants do not need to move around as much to go from room to room. Alternatively (or in addition), the participants may have made the change because of changing health needs that themselves impact their daily activity. Therefore, in order to understand how the home move impacts our measurements of activity, it is important to study the time periods around these transitions carefully, and to understand what health changes may have happened at the same time. To help us understand these data, we ask our participants to complete weekly online questionnaires about any health changes, medication changes, or emergency room visits that may have occurred in the previous week. “In aging studies, transitions are part of life.”

156 | Gathering the Evidence: Supporting Large-Scale Research Deployments Intel® Technology Journal | Volume 13, Issue 3, 2009

One final area of cohort management is important to consider. Many “Many technology-based studies place technology-based studies place expensive equipment into the homes of the participants. Depending on the neighborhood, this technology can make expensive equipment into the homes of the participant a target for break-in and theft. Furthermore, in studies with the participants.” cognitively impaired subjects, quite a few of the subjects will forget what the technology is for or even that it is in their home. In these cases, participants may pack up the equipment or even give it away. While these situations are rare, their resolution takes time and money.

Technology Management There are three keys to successful technology management in a large-scale, in‑home study: a stream-lined and well-supported installation process, continuous remote tracking of system status, and the ability to do remote management and maintenance of installed technologies. Our approach in each of these areas has changed over the years, as we have partnered with study sites around the country and have learned many lessons. The installation of in-home technology often involves sensors and devices that “There are three keys to successful are unfamiliar to many study assistants. Even setting up a computer may be outside the experience of the typical research assistant (RA). Therefore, the technology management in a large- technology installation must be both orderly and as idiot-proof as possible. In scale, in-home study: a stream-lined addition, a live help line is essential to provide support for inexperienced RAs. For example, in one study in which medication tracking devices and computer and well-supported installation kiosks for administering monthly or quarterly neuropsychological evaluations were to be installed in 200 homes in 25 cities around the country, we created process, continuous remote tracking a laminated step-by-step installation instruction sheet that was packed in the of system status, and the ability to do box with each technology set. All connectors between the equipment were color coded, and the instructions included photographs of how those color- remote management and maintenance coded connectors should be connected. A set of pre-install steps were clearly of installed technologies.” specified, including forming a working relationship with the local broadband provider for billing and technical support prior to commencing installs. Study coordinators from all sites converged locally for a one-day training session in which we attempted to cover possible problems with the installation. In spite of these preparations, the help line assistant received 55 calls during the first 57 installations. While the need for support tapered off sharply by the time each site had installed two systems, about two-thirds of sites still called for additional assistance at least once in the first year. Clearly, an effective help line with well-trained staff is one of the most essential components of ensuring a successful multi-site, large-scale, study deployment.

The continuous monitoring of equipment status is another important component of ensuring good data collection in a large-scale study. In our studies we do this through the ORCATECH Management Console, which queries the status of numerous system variables in each home once per day. “The continuous monitoring of Figure 4 shows an example in which an alert screen provides visual indicators equipment status is another important of the status of the sensors. The system checks for problem indicators daily and displays them as a red bar. Technicians can then reclassify the problems when component of ensuring good data they have determined the cause. Colors reflect the problem status. Multiple days can be displayed to assist in tracking problem status over time. collection in a large-scale study.”

Gathering the Evidence: Supporting Large-Scale Research Deployments | 157 Intel® Technology Journal | Volume 13, Issue 3, 2009

Alerts Summary Select Colored Cells to Classify

Services Pass: 2009-08-04 (02:00:01) History: 16 Alert: Sensor PC Reload Print

Apartment Classify Home Id 1192 1194 1196 1198 1200 1203 1205 1207 1020 20 85 109P 94 204B 109 414S 112 127 317 150 212 160 301P Unread 161 64 Unread 413 Watch 304B Not an Issue 1558RR Create New Issue Oa Out of Home 229

E 303 237 259

B 106 261 312 335 334 336 315 348 1012 401

Figure 4: Screenshot of the ORCATECH Management Console Alert Screen Source: OHSU

A great deal of system problem triaging can be done programmatically. Communication status between the routers and the in-home computers, the status of services running on the sensor computers in the home, remaining device battery life, failure to receive sensor heartbeats, sensor data values out of expected range, and even failure of the participant to complete a weekly form, can all be flagged as possible issues and brought to the attention of the study staff. Serious problems, which may result in the loss of data, can then be identified quickly and addressed immediately. By tracking the time taken to triage and fix problems, we are able to assess the overall quality of the study support.

“Scientifically, it is important to not Finally, remote maintenance of installed technologies is an extremely important component of conducting in-home trials, both scientifically and financially. visit with the participants more than Scientifically, it is important to not visit with the participants more than necessary due to the study protocol.” necessary due to the study protocol, because that contact may itself influence the outcome of the study. For example, in one of our studies we are interested in changes in walking speed and motor activity over time. Each time we visit a participant’s home to repair activity sensors, they become aware of their own activity and may subtly alter their behavior for a short period of time as a result. Furthermore, frequent visits to the participants can be an irritant to them, which puts study retention at risk. Financially, it is expensive to travel to participants’ homes to make repairs. Clearly, some repairs require in-person visits (replacing batteries or broken equipment), but many things can be fixed remotely (restarting services, scheduling tasks, pushing new software versions, helping participants with computer programs, etc.). Wake-on-LAN capability has proven to be an extremely valuable feature of the in-home computer. In our

158 | Gathering the Evidence: Supporting Large-Scale Research Deployments Intel® Technology Journal | Volume 13, Issue 3, 2009 studies, we used a laptop to collect sensor data that did not have wake-on-LAN “Once the project reached steady state, capability in its wireless LAN card. As a result, in those homes in which we used 802.11g to connect to the router, we were unable to remotely boot many the number of fixes trended down.” of the sensor computers that went off-line.

Figures 5 and 6 show the repair history for one of our studies, the ISAAC project [9]. In this project participants have passive infrared sensors distributed throughout their home to detect activity and walking speed and a personal computer that we use to track inter-keystroke intervals (a surrogate for finger tapping speed) and their interaction with a series of games. This study has two technical RAs who are responsible for supporting the 248 participants and 190 homes in the study. In Figure 5, you can see the clear pattern of install/bug fix that we engaged in during ramp-up of the study. Once the project reached steady state, the number of fixes trended down, reflecting the fact that issues that spanned multiple homes had been identified and addressed. At this point 0.13 in the project, most fixes are maintenance issues, such as battery changes and In-home fixes Phone support issues with broadband. Remote fixes 0.11 200 Total number of 180 Installed homes 160 In-home fixes Phone support 0.09 140 Remote fixes 120

100 0.07 80

60 Number of installs/fixes 40 0.05 20 Average number of fixes per month home 0 May–07 Jul–07 Sep–07 Nov–07 Jan–08 Mar–08 May–08 Jul–08 Sep–08 Nov–08 Jan–09 Mar–09 May–09 Jul–09 0.03 Router Sensors Sensor PC User PC

Figure 5: Patterns of Installation and Fixes on the ISAAC Project Figure 6: Types of Fixes for Each of the Four Source: OHSU Major Technologies Installed in the ISAAC Project Source: OHSU Figure 6 shows the ratio of in-home visits to support calls with the participant, to remote fixes, for the main problem areas. Because participants in this study are required to use their computer weekly, we have put into place a help line (e-mail and phone) that they can call for support. This is reflected in the large number of remote fixes done on the PC, which often take the form of undeleting a subject’s file that they cannot seem to locate, restoring e-mail settings that they have changed by mistake, or deleting spy ware that was “Surprisingly, most computer problems inadvertently downloaded by their grandchildren. In spite of our best efforts to unrelated to participant use have provide FAQs, handouts, and in-person training on basic computer use issues, many relatively simple tasks remain a challenge to the individuals who have required on-site visits.” never used computers prior to our studies.

Surprisingly, most computer problems unrelated to participant use have

Gathering the Evidence: Supporting Large-Scale Research Deployments | 159 Intel® Technology Journal | Volume 13, Issue 3, 2009

required on-site visits. This is a major barrier to scalability at this time. In part, this reflects the limited capabilities of the in-home systems used in this study. We chose inexpensive laptop computers running Windows* XP* for our study, in part because of the need to keep costs under control, and in part because some of the commercial sensor software we were using runs only under Windows. However, these laptops turned out to have unreliable network cards as well as some obscure device driver issues. Furthermore, at the time the study started, Intel® vPro™ technology was not yet available. Had we been able to use such technology, we would have been able to remotely diagnose a number of sensor computer issues rather than visit the home. Overall, our considered opinion at this time is that a Linux*-based system would perform more reliably, as well as require significantly less memory.

Data Management The management of the data from in-home technology studies raises some unique issues. Of primary concern is the security of the data and the privacy of the participants. At all levels of data management (acquisition, transmission to a central site, ongoing storage, and retrieval for analysis), the data must be protected. Many organizations, such as the Veteran’s Administration, “The data must be appropriately have highly-regulated restrictions on where the data may be stored, when they need to be encrypted, and who may access them. However, even in anonymized to protect the subjects’ research organizations with fewer regulations, the data must be appropriately privacy, and they must be stored anonymized to protect the subjects’ privacy, and they must be stored securely. Data acquisition poses interesting challenges to maintaining this privacy, since securely.” often the data are not encrypted before being sent to a local receiving device.

As described in the section on the Nature of the Data, raw data, metadata, and derived data all have different properties. As a result, they also have different storage needs. Large-scale, longitudinal studies that follow hundreds or thousands of subjects over years need to consider options for data compression, data storage, and data retrieval. As derived measures are developed, are the raw data still needed? Is it appropriate to archive these data, or does it make more sense to create a hierarchical storage model in which older data are down- sampled or summarized over longer time periods? Do meta-data need to be kept beyond the period in which they are used to address technology issues? If they are discarded, what historical data should be kept to allow for evaluation of the study’s efficiency? As derived measures evolve over time, how should the versions be documented and stored? For large datasets, where processing the data to derive the measures may take hours or days, it may make the most sense to store all of the derived datasets. In other cases, where storage is at a “While none of these issues are premium, it may make more sense to store the algorithms used to derive the unsolvable, some choices may be measures. While none of these issues are unsolvable, some choices may be irreversible. As a result, it is important to have a clear data management policy irreversible.” in place at the start of any large, in-home study.

160 | Gathering the Evidence: Supporting Large-Scale Research Deployments Intel® Technology Journal | Volume 13, Issue 3, 2009

Scalability Issues There are many issues that need to be addressed if large-scale deployments of “The largest scalability issue is the in-home research studies are to be successful. One issue centers on the diversity of technology experience and acceptance in the aging population. A second lack of technology infrastructure, issue is the lack of available tools for interpreting the complex data that result specifically focused on supporting from such studies, and the need to enable data-sharing through visualization tools. However, the largest scalability issue is the lack of technology large-scale research studies.” infrastructure, specifically focused on supporting large-scale research studies. We discuss these issues in more detail next.

Managing Diverse Populations Differences in living environments, in attitudes towards technology, and even in the relative extent of in-home versus out-of-home activity across different groups may require modifications in the platform and approach for continuous in-home assessment. For example, a recent ORCATECH study [1] suggested that rural seniors may have several technological challenges. Of 128 seniors aged 80 or older that were surveyed, only 5 percent reported using a computer frequently, or sending or receiving e-mail daily. This cohort is a good example of the kind of research and environmental challenges that could be encountered during wider deployment in terms of subject training, and in terms of collecting and transferring of data. Another group that may “Another group that may provide provide unique challenges is low-socioeconomic status (SES) elderly. SES is a unique challenges is low-socioeconomic combined measure of an individual’s economic and social position relative to others. SES is often assessed based on a method proposed by Hollingshead, status (SES) elderly.” which considers level of education and occupation. Krousel-Wood [11] found that level of education adversely affected patients’ willingness to participate in a telemedicine study. In addition, due to economic constraints, this cohort is less likely to use a computer on a regular basis. Therefore, a particular challenge in this group could be recruiting and training on computer use. Finally, cognitively impaired elders are often less able to interact with technology.

The other interesting issue that arises when a study is scaled nationally is regional differences in home construction. For example, in our national studies using in-home technologies, we have found that the wiring in many of the older homes in New York City result in high data loss with DSL broadband. For commercial applications, such issues may mean loss of market opportunity and disappointment in some populations that the technologies do not work in their home. For a research study, such differences could result in loss of access to a key demographic or cohort. For example, participants with low socioeconomic status often live in older homes. This is a population notoriously underserved by the research community because they are less likely to volunteer for studies and must therefore be actively recruited. However, “We have found that the wiring in these individuals may benefit greatly from aging-in-place technologies, and therefore they are an important cohort for inclusion in research studies. many of the older homes in New York City result in high data loss with DSL broadband.”

Gathering the Evidence: Supporting Large-Scale Research Deployments | 161 Intel® Technology Journal | Volume 13, Issue 3, 2009

The Need for Visual Analytics “In the field of unobtrusive in‑home As mentioned previously, sensor-based, in-home studies generate large amounts of complex data. It is well accepted that sharing research data increases monitoring, there has been no collaboration, increases research activity, and reduces the cost of research [10]. However, in the field of unobtrusive in-home monitoring, there has been no coordinated effort to share data from coordinated effort to share data from past and ongoing studies. Even if the data past and ongoing studies.” were available, analysis of those data requires significant domain knowledge and an understanding of the sensors, their deployment, and the clinical and sensor datasets that have been collected during the study. Therefore, there is a clear need for extensible, modular tools that allow researchers to focus on the data, without being mired in the details of the sensor technologies.

Specifically, new data analytic tools must be created. Data analytics is the use of visualization and descriptive statistical techniques for initial data exploration. Visualization is the display of multiple dimensions of a dataset graphically, grouped by some feature or category of interest. Visualization is meant to take advantage of the human capacity to process visual information quickly and efficiently. In the case of complex data such as in-home monitoring data, visualization tools must allow exploration of the spatial and temporal properties of the sensor data, as well as fuse information from multiple sensor types to examine interactions between different behavioral measures. These tools will enable collaboration between researchers already focused on this emerging paradigm, and they will facilitate entry into this area of research by investigators who may lack the resources to undertake the necessary data collection, the access to key patient populations, or both, thereby increasing the opportunities for discovery and formation of new hypotheses.

The Need for Global Technology Solutions “More than 50 percent of technology One of the primary barriers to wide-scale deployment of in-home, technology- based research studies is the lack of universally available and supported Internet issues were a result of failed infrastructure. In the one multi-state study in which we have used in-home broadband.” technologies, more than 50 percent of technology issues were a result of failed broadband, ranging from a poor signal due to old wiring in the home, to difficulty on the part of the Internet service provider (ISP) in getting the broadband up and working in the home, and to the ISP discontinuing service inadvertently or because of billing errors on their part. There are no low-cost broadband options available universally across the country, and multi-site studies therefore require establishing a separate ISP relationship and billing agreement with each city, and sometimes each county, where the study is to be conducted. This is a significant administrative barrier to large-scale deployment for research.

Another significant barrier to wide-scale, in-home research studies is the lack of a platform supporting purposed sensors designed for gathering behavioral and “We have been forced to repurpose physiological data in the home. In our current studies, we have been forced to sensors developed for inventory repurpose sensors developed for inventory tracking and security applications for our activity monitoring. In those cases where we have developed a new sensor tracking and security applications.” (for example, with medication tracking [7]), it was done with a particular

162 | Gathering the Evidence: Supporting Large-Scale Research Deployments Intel® Technology Journal | Volume 13, Issue 3, 2009 project in mind and did not grow out of a principled architecture intended “What is needed is the integration to support the growing needs for different types of sensors. Ultimately, what is needed is the integration of sensors designed to optimize the collection of of sensors designed to optimize the specific behavioral and physiological data with a platform built on an open, collection of specific behavioral and standards-based architecture. It is our hope that such an architecture may ultimately be derived from the combined efforts of organizations such as physiological data with a platform Continua Alliance, industry leaders promoting interoperable standards for built on an open, standards-based personal health; Integrating the Healthcare Enterprise (IHE), an initiative sponsored by professional societies to provide a common framework for architecture.” multi-vendor system integration; and the International Organization for Standardization (ISO), which has specified numerous standards for health devices and is one of the sponsoring organizations of Health Level 7 (HL7).

The research community has an additional need: standards-based data sharing and storage models, such as those developed by the Clinical Data Interchange Standards Consortium (CDISC) that have been endorsed by the FDA for use with clinical drug trial data. With the advent of large-scale, in-home studies that use technology to aid in the assessment and detection of health and behavior changes, there will be increasing amounts of temporal and spatial sensor data, as well as health outcome data, to be gathered, stored, analyzed, and archived for efficient later retrieval. Sharing of these research data would allow investigators to develop new algorithms for deriving health status and “Data sharing is most effective when interpreting behavior from these data, without the considerable costs and resources that would be incurred if each investigator had to collect his or her a standard data format and set of own data. Data sharing is most effective when a standard data format and set of tools are available to facilitate data tools are available to facilitate data sharing. Unfortunately, current, continuous monitoring studies all use different methodologies, and thus far, none have sharing.” made their data widely available to the research community for further analysis of health outcomes. No tools currently exist to facilitate the sharing of continuous monitoring data. Thus, the development of a common architecture for sharing and visualizing sensor-based study data is another key requirement for true scalability.

Conclusions There is tremendous hope that independent living technologies may lead to a reduction in the cost of healthcare delivery in this country. These technologies have the potential not only to support aging-in-place, but to enhance our ability to assess and remediate health changes in the elderly. Although a number of mid-sized research studies using such technologies are underway, there are still significant barriers to their deployment in large-scale clinical trials. However, we are beginning to understand these barriers and to find solutions to overcoming them, which will lead some day soon to a significant evolution in the practice of clinical research.

Gathering the Evidence: Supporting Large-Scale Research Deployments | 163 Intel® Technology Journal | Volume 13, Issue 3, 2009

Vignette “I never miss a pill.” After taking care of herself for 23 years since her husband died, 85-year- old Ana is pretty sure she has things under control. Although she takes five medications each day, three in the morning and two at night, she maintains strict control of her medication regimen and is able to cheerfully confirm that she has taken her meds when she talks to her daughter in their nightly phone call. When she enrolled in our medication adherence study, we asked her to add a low-dose vitamin C to her regimen once in the morning and once at night when she took her regular meds. She put these into our Medtracker pillbox and agreed to take them twice a day. Ana scored very well on her cognitive test, achieving a score that was a little lower than some but still considered normal for her age.

After a couple of days, we noticed that she had stopped taking the vitamins and gave her a call to remind her to take them. She was adamant that she’d been taking them when she took her own meds, and so we went out to her house to replace the Medtracker, which clearly wasn’t working properly. However, when we got there, we found that the pillbox was still full of pills. Ana was sure she had taken them, and thought that somebody else must have put more pills into her box. She wanted to continue in the study so we reviewed how to fill up the pillbox each week. After five weeks in the study, Ana’s overall adherence was 38 percent. To this day she maintains that she “only missed a couple of pills.”

References [1] Calvert J., Kaye, J., Leahy, M., Hexem, K., and Carlson, N. 2009. “Technology use by rural and urban oldest old.” Journal of Medicine and Technology. In press.

[2] Chadwick-Dias A., McNulty, M., and Tullis, T. 2003. “Web usability and age: How design changes can improve performance.” In Proceedings of the 2003 Conference on Universal Usability, Vancouver, B.C., Canada, pp. 30-37.

[3] Chadwick-Dias A., Tedesco, D., and Tullis, T. 2004. “Older adults and web usability: Is web experience the same as web expertise?” In Aging by Design Workshop, Bentley, MA.

[4] Centers for Medicaid and Medicare Services. 2005. National Health Expenditures: web tables. Available at www.cms.hhs.gov.

[5] Erickson P., Wilson, R., and Shannon, I. 1995. Years of Healthy Life. US Department of Health and Human services, CDC, National Center for Health Statistics, Statistical Notes 7, Hyattsville, Maryland.

[6] Gist Y. J. and Hetzel, L. I. 2004. We the people: Aging in the United States (CENSR-19). Available at www.census.gov.

164 | Gathering the Evidence: Supporting Large-Scale Research Deployments Intel® Technology Journal | Volume 13, Issue 3, 2009

[7] Hayes T. L., Hunt, J. M., Adami, A., and Kaye, J. 2006. “An electronic pillbox for continuous monitoring of medication adherence.” In 28th Annual International Conference of the IEEE Engineering in Medicine and Biology Society, New York, NY.

[8] Iglehart J. K. 2008. “No place like home—testing a new model of care delivery.” New England Journal of Medicine, 359(12), 1200-2.

[9] Kaye J. A., Hayes, T. L., Zitzelberger, T. A., Yeargers, J., Pavel, M., Jimison, H. B., Larimer, N., Payne-Murphy, J., Earl, E., Wild, K., Boise, L., Williams, D., Lundell, J., and Dishman, E. 2008. “Deploying wide-scale in-home assessment technology.” In Technology and Aging, vol. 21. In Assistive Technology Research Series. A. Mihailidis, J. Boger, H. Kautz, and L. Normie, Eds.: IOS Press, pp. 19-26.

[10] Koslow S. H. 2005. “Discovery and integrative neuroscience.” In Clinical EEG in Neuroscience. 36(2), pp. 55-63.

[11] Krousel-Wood M. A., Re, R. N., Abdoh, A., Chambers, R., Altobello, C., Ginther, B., Bradford, D., and Kleit, A. 2001. “The effect of education on patients’ willingness to participate in a telemedicine study.” Journal of Telemedicine and Telecare, 7(5), pp. 281-7.

[12] Li S., Aggen, S. H., Nesselroade, J. R., and Baltes, P. B. 2001. “Short- term fluctuations in elderly people’s sensorimotor functioning predict text and spatial memory performance: The Macarthur Successful Aging Studies.” Gerontology, 47(2), pp. 100-16.

[13] Logan B., Healey, J., Philipose, M., Tapia, E. M., and Intille, S. 2007. “A long-term evaluation of sensing modalities for activity recognition.” Vol. 4717. In Lecture Notes in Computer Science, Innsbruck, Austria: Springer Verlag, pp. 483-500.

[14] Mitka M. 2009. “Studies comparing treatments ramp up.” Journal of the American Medical Association, 301(19), 1975.

[15] Murata A., Nakamura, H., and Okada, Y. 2002. “Comparison of efficiency in key entry among young, middle and elderly age groups— effects of aging and character size of a keyboard on work efficiency in an entry task.” In 2002 IEEE International Conference on Systems, Man and Cybernetics, Oct. 6-9 2002, Yasmine Hammamet, Tunisia, 2, pp. 96-101.

[16] Salthouse T. A., Nesselroade, J. R., and Berish, D. E. 2006. “Short-term variability in cognitive performance and the calibration of longitudinal change.” Journal of Gerontology Series B: Psychological Sciences and Social Sciences, 61(3), pp. 144-151.

Gathering the Evidence: Supporting Large-Scale Research Deployments | 165 Intel® Technology Journal | Volume 13, Issue 3, 2009

[17] Stone A. A., Pennebaker, J. W., Keefe, F. J., and Barsky, A. J. 2000. “Part VII: Self-Report of Physical Symptoms.” In The Science of Self-Report: Implications for Research and Practice. A. A. Stone, Ed. Mahwah, NJ: Lawrence Erlbaum Associates, pp. 297-362.

[18] Takeda A., Loveman, E., Clegg, A., Kirby, J., Picot, J., Payne, E., and Green, C. 2006. “A systematic review of the clinical effectiveness of Donepezil, Rivastigmine and Galantamine on cognition, quality of life and adverse events in Alzheimer’s disease.” International Journal of Geriatric Psychiatry, 21(1), pp. 17-28.

[19] Tang P. C. and Lee, T. H. 2009. “Your doctor’s office or the Internet? Two paths to personal health records.” New England Journal of Medicine, 360(13), pp. 1276-8.

[20] Tourangeau R. 2000. “Remembering What Happened: Memory Errors and Survey Reports.” In The Science of Self-report: Implications for Research and Practice. A. A. Stone, Ed. Mahwah, NJ: Lawrence Erlbaum Associates, pp. 29-48.

[21] Ziefle M. and Bay, S. 2005. “How older adults meet complexity: Aging effects on the usability of different mobile phones.” Behaviour and Information Technology, 24(5), pp. 375-389.

Acknowledgements The research described in this article was funded through grants from the National Institutes of Health (NIA grants P30AG024978 P30AG08017, and R01AG024059), by Intel Corporation, and by the Kinetics Foundation.

Author Biographies Tamara Hayes, PhD, is Assistant Professor and Associate Division Head of Biomedical Engineering at Oregon Health and Science University (OHSU). She received her MS degree in Electrical Engineering at the University of Toronto and her PhD in Behavioral Neuroscience from the University of Pittsburgh. She has worked in both industry and academia, on projects ranging from creating tools for remote management of distributed systems to delivering teledermatology care to rural Oregon. Dr. Hayes’ current research interests include the use of technology to deliver healthcare in the home, with the goal of changing the current paradigm of clinic-centered healthcare to a model that is less costly, more effective, and allows individuals to participate more fully in their own healthcare. This research entails the use of low-cost, unobtrusive sensors in the home for collecting behavioral data related to acute and chronic motor and cognitive changes, and meaningful analysis of these data to assist and inform the patient.

166 | Gathering the Evidence: Supporting Large-Scale Research Deployments Intel® Technology Journal | Volume 13, Issue 3, 2009

Misha Pavel, PhD, is Division Head and Professor of Biomedical Engineering at Oregon Health and Science University (OHSU). He is also Director of the Point of Care Laboratory, which focuses on unobtrusive monitoring and neurobehavioral assessment and modeling. He received his PhD in Experimental Psychology from New York University and his MS degree in Electrical Engineering from Stanford University. His research interests include analysis and modeling of complex behaviors of biological systems, including perceptual and cognitive information processing, pattern recognition, information fusion, and decision making in healthy and cognitively impaired individuals.

Jeffrey A. Kaye, M.D., is a Professor of Neurology and Biomedical Engineering and the Director of the Layton Aging and Alzheimer’s Disease Center, as well as Director of the Oregon Center for Aging and Technology (ORCATECH) at Oregon Health and Sciences University. Dr. Kaye received his medical degree from New York Medical College and trained in neurology at Boston University. His Alzheimer’s research is focused on understanding differing rates of progression and cognitive decline as compared to optimal cognitive health in the elderly. Through his work with ORCATECH, he seeks to facilitate successful aging and reduce the cost of healthcare through the use of in-home technologies for assessment and remediation.

Gathering the Evidence: Supporting Large-Scale Research Deployments | 167 Intel® Technology Journal | Volume 13, Issue 3, 2009

Assistive Technology for Reading

Contributors

Selena Chan Abstract Intel Corporation Over 82 million people in the developed world face challenges each day when attempting to read a variety of printed materials. The challenges may be due Ben Foss to blindness, partial vision loss, or reading-based learning disabilities, such as Intel Corporation dyslexia. Regardless of the cause, people of all ages suffer a huge loss in their independence and productivity when they must rely on others to read material David Poisner on their behalf or struggle to read at a rate so slow that they become frustrated Intel Corporation and discouraged. This article focuses on the design and technical challenges of addressing this user need and the more general complexities of designing an independent-living product for a user group that ranges from teens to seniors and from those who are conversant with technology to those who downright fear it.

The Intel® Reader is the output of years of research into end-users who experience difficulty reading text for a variety of reasons. The Intel Reader is a mobile device that combines a high-resolution camera with an Intel Atom™ Index Words processor to take pictures of text and read that text aloud to the user on the spot. Assistive Technology Designing the Intel Reader presented challenges in many areas: user interface, Independent Living mechanical and electrical, and usage flows and specification of the system-level Dyslexia requirements. In this article, we first review the diverse end-user population Visual Impairment for this technology, and we identify unique subpopulation requirements that Blindness were distilled into a set of common parameters. We explore many of the Specific Learning Disability design challenges, tradeoffs, and unexpected discoveries that led us to the final definition for the Intel Reader.

Introduction Printed text is a default form of communication in the developed world. Beginning in early elementary school, reading is a central component of education and is the foundation for building independent, analytical skills. As students grow and begin to enter the workforce, the ability to read memos, purchase orders, and industry-related articles is critical to promotions and successful performance reviews. As the working adult matures, the prevalence of vision impairment increases and poses a challenge to those who desire to live “For a large number of people, reading an independent lifestyle. And for a large number of people, reading printed text is a barrier. This group of people includes those with specific learning printed text is a barrier.” disabilities, those with partial vision loss, and those who are completely blind.

168 | Assistive Technology for Reading Intel® Technology Journal | Volume 13, Issue 3, 2009

“The Intel Reader is a mobile device that takes pictures of text, translates the image into digital text by using optical character recognition (OCR) technology, and reads the text aloud by using text-to-speech (TTS) software.”

Figure 1: The Intel Reader Source: Intel Corporation, 2009

The Intel Reader is a mobile device that takes pictures of text, translates the image into digital text by using optical character recognition (OCR) technology, and reads the text aloud by using text-to-speech (TTS) software. Figure 1 shows a photograph of the Intel Reader, after many rounds of prototype design iterations and user-focused group discussions. It is roughly 6.5” by 5.5” by 1.1”, about the size of a paperback book. The Intel Reader contains a high-resolution camera, designed to capture crisp images of text. To improve the accuracy of OCR, it is critical that the text images are of high resolution and as sharp as possible. When designing the user interface for the Reader, we took into account the findings from our ethnographic research and we incorporated many human factors into our design to provide print accessibility to the target audience: we included features such as image rotation and automatic correction to increase the accuracy of the conversion. The Intel Reader utilizes the Intel Atom processor and a Linux* operating system.

Disabilities and the Challenges of Reading Reading can be challenging for a variety of reasons ranging from learning disabilities to partial or complete blindness. “We took into account the findings from our ethnographic research and we incorporated many human factors into our design to provide print accessibility to the target audience.”

Assistive Technology for Reading | 169 Intel® Technology Journal | Volume 13, Issue 3, 2009

“Roughly 10 to 15 percent of any Reading-Based Learning Disability Reading can be a frustrating experience for people who have specific learning given population has specific learning disabilities. For example, dyslexia is a neurological disability, often characterized disabilities, with 80 percent exhibiting by difficulties with word recognition, word spelling, and word decoding. One participant in our research described reading as “having a bad cell phone reading-based learning disabilities connection with the paper — words drop out or come through garbled even consistent with dyslexia.” though I can see just fine.” Dyslexia, like most learning disabilities, is an inherited condition, and not the result of lack of effort, sensory impairment, or of inadequate education. Roughly 10 to 15 percent of any given population has specific learning disabilities, with 80 percent exhibiting reading-based learning disabilities consistent with dyslexia. In the United States, there are 55 million people with specific learning disabilities in the public schools, 8 million of whom are students under the age of 18 who have dyslexia. Although studies may vary, it is estimated that 30 – 45 million people are in this target population overall in the United States alone.

The majority of students first learn to read by phonemic awareness (how speech sounds make up words) and by connecting those sounds to letters of the alphabet (phonics). Next, they learn to blend the sounds into words, recognize words, and comprehend what they read. People with specific learning disabilities, however, have difficulty with phonemic awareness and phonics. Research has shown that dyslexia occurs because information is processed differently in the brain, and for this reason, reading becomes an uphill battle and remains slow and laborious. When students struggle with these early “Reading becomes an uphill battle and stages of reading, they are often made to feel inferior. Frustration and isolation resulting from this lack of independence can be intense, leading to high drop- remains slow and laborious.” out rates and associated social problems.

With intensive and early help, students with dyslexia are often able to learn the basic skills of reading and develop strategies that allow them to stay in the conventional classroom. Generally, however, they are not able to achieve a level of reading fluency commensurate with their level of intelligence and aptitude. While they may be able to get words off a page with great effort, they are generally at a great disadvantage, relative to the majority of their peers, especially when it comes to comprehension. However, the comprehension of spoken words is not affected by dyslexia, and as a result, audio processing of information is highly desirable for individuals with dyslexia.

Blindness and Visual Impairments For people with vision-loss, reading poses a real challenge. Anyone with non- correctable reduced vision is considered to be visually impaired. Cataracts, glaucoma, macular degeneration, corneal opacity, diabetic retinopathy, and trachoma are examples of pathologies that can cause vision acuity loss. Vision impairment can range from severe low vision (uncorrectable beyond 20/200 “Anyone with non-correctable reduced to 20/400) to near total blindness (less than 20/1000). At the extreme end, vision is considered to be visually individuals who completely lack light perception are considered totally blind. impaired.”

170 | Assistive Technology for Reading Intel® Technology Journal | Volume 13, Issue 3, 2009

Visual impairment is unequally distributed across age groups. More than 82 “Potentially blinding eye conditions, percent of all people who are blind are over 50 years of age, although they represent only 19 percent of the world’s population. There are an estimated such as age-related macular 1.4 million blind children below age 15 in the world. In the next 30 years, an degeneration (AMD), diabetic ever-increasing number of people will develop some form of visual impairment due to the aging of our population. Potentially blinding eye conditions, such as retinopathy, and glaucoma are age-related macular degeneration (AMD), diabetic retinopathy, and glaucoma increasing as the world’s population are increasing as the world’s population ages. ages.” In recent years, as technology has become more prevalent, a number of assistive tools have emerged for people with disabilities. There are text readers that help students build reading comprehension skills, a number of computer programs are available to help students with literacy, and audio books have emerged as an alternative format to help students read by listening. Audio books are typically available for common textbooks, selected novels, and popular magazines and newspapers. Additionally, for content that is in electronic format, such as websites and e-mails, text-to-speech software can be used to convert it to audio. However, there are documents that exist primarily in printed format, such as store receipts, coupons, business memos, personal mail, and specialty “The Intel Reader was developed publications, all of which require an alternative means of access. Alternative to address these needs, offering an access means reliance on a third party, be it a scanning service to get access to text, or a special educator to provide on-going support. The Intel Reader was evolution of the existing assistive developed to address these needs, offering an evolution of the existing assistive technologies to a more portable and intuitive usage model, as well as access to technologies to a more portable and text at a time and place of the user’s choosing. intuitive usage model.”

End-User Personas When we started developing the Intel Reader we talked with end users who allowed us to gain insight into their needs, and who gave us the opportunity to hear how they use current technology in their everyday lives. To obtain a deeper understanding for end-user needs, we began with usage research. As part of this phase, we interviewed targeted users, followed them through a day in their lives, and observed their daily challenges at home and at work, and asked for their insights on the challenges of daily living. Over the course of this project, we have interviewed and tested the product at various stages with over 200 individuals. “Over the course of this project, we have interviewed and tested the product at various stages with over 200 individuals.”

Assistive Technology for Reading | 171 Intel® Technology Journal | Volume 13, Issue 3, 2009

The process began by interviewing 20 end users from various target groups and developing a broad set of model personas to guide our design choices along the way. Initially, over 30 composite characters were generated, each representing key elements of the user population. The key characteristics, shown in Figure 2, articulate the key attributes of our user population and their user needs.

Key Characteristics Vision: Affects: VUI/GUI, tasks/usage blind VI LV sighted

Age: Affects: dexterity, tasks, aesthetics kid teen adult senior old

Settings: Affects: context, tasks/usage home street work school

Tech tolerance: Affects: controls/ui, complexity low medium high

Coping: Affects: tasks/usage, settings, adoption struggling coping conquering

Attitude: Affects: tasks/usage, adoption, aesthetics ashamed ambivalent proud

Channel: Affects: who we target; how they encounter this medicare insurance grants OOP

Figure 2: End-user Key Characteristics Source: Intel Corporation, 2009

Five potential target personas emerged from our research and findings, and two personas, described next, defined the boundary conditions for a successful product. Persona 1 “His vision is getting less and less clear James is 67 years old and lives with his wife in St. Louis, MO, in the United States. He is a diabetic with health complications and deteriorating vision. and he can no longer read unless he His hands are a bit unsteady and he is losing sensation in them. His vision is getting less and less clear and he can no longer read unless he uses very strong uses very strong magnification.” magnification; even then some common fonts are too small for him to read. James struggles daily with controlling his diabetes. He tends to go through cycles of being in control followed by periods where he doesn’t take care of himself. The guilt he feels about his health is contributing to depression.

James wants a device to read the newspaper so he can stay well-informed on current events. He likes to go out to dinner with his family, and he would like to pay for the meal without asking his children to read the total on the receipt, thereby regaining his head of the house role. In addition, James would like to use coupons, distributed in the mail or newspaper, when he goes shopping. If “If he had a reading device, he would he had a reading device, he would use it also to read a variety of print material, use it also to read a variety of print such as mail, so he could sort junk mail from bills, read greeting cards from his grandchildren, read street signs, and read prescription instructions from the material.” pharmacy.

172 | Assistive Technology for Reading Intel® Technology Journal | Volume 13, Issue 3, 2009

Persona 2 Ethan is 14 and lives in Swindon in the UK. He is a student, living with his “He is frustrated that he is in special grandmother, and he was recently identified as dyslexic. He was doing pretty well in school until the reading level increased when he turned 13. Now that education classes for a portion of the he has started high school, he is frustrated that he is in special education classes day.” for a portion of the day, separated from his friends.

Ethan values discretion and does not want everyone to know that he is dyslexic. Ethan wants a device that looks modern and similar to other gaming devices that his friends have so that he would not draw attention to himself when he uses the device at school. He has a lot of reading to do for school, so he needs a tool to help him read books — including text books, popular books, and magazines. In addition, he would like a device that would also allow him to listen to music. For both James and Ethan, having a portable device is important so that “Having a device that is light-weight, they have the freedom to take it with them wherever they go and use it in a variety of environments, such as in a restaurant or on the school bus. Having has a long battery life, and large a device that is light-weight, has a long battery life, and large storage capacity storage capacity are all basic features are all basic features that James and Ethan care about. In James’s case, he wants something sturdy and solid, and not too small so that it becomes hard that James and Ethan care about.” to operate. Ethan, on the other hand, wants something with a cool factor that looks high-tech and is small enough to fit into his backpack.

Our ethnographic research therefore highlighted for us the three key elements for a reading device: accuracy, convenience, and discretion. The device needs to accurately capture information; it needs to be easy to use anywhere, anytime; and it must allow users to maintain privacy in their everyday tasks. We, therefore, had our mission: design and develop an assistive technology product to meet the needs of both James and Ethan.

Development Stages of our Product As with the development of any product, the Intel Reader went through a number of design phases, with feedback loops influencing and often forcing a rewrite of earlier specifications. Outlined here are the stages we went through in the development of the Reader by Intel and our collaborators:

Fundamental R&D. Intel has been researching imaging technology for “Our ethnographic research therefore decades, based largely on the need to develop visual inspection tools for silicon microprocessor manufacturing. Image improvement, rotation alignment, the highlighted for us the three key integration of multiple images into higher resolution images, and fundamental elements for a reading device: research in optics have all been part of Intel’s broad research efforts. Much of this work aided in the design, testing, and debugging of the Intel Reader. accuracy, convenience, and discretion.”

Assistive Technology for Reading | 173 Intel® Technology Journal | Volume 13, Issue 3, 2009

Exploration. Intel engineers began experimenting with image and optical technology, by using lower-resolution cameras and existing OCR solutions, to test the feasibility of capturing images of text taken with hand-held devices and converting them into speech. Initially the team explored the use of cell phones, and then they moved on to off-the-shelf digital cameras. The team mocked up early prototypes and demonstrated the concept for approval. The inclusion of targeted users as members of the design team was critical, thereby allowing deeper insight into usage models.

Usage Research. The team expanded to include behavioral scientists who worked with end users in their homes and workplaces to learn about their daily lives. In this phase, as the diversity of the population was better understood, the core usages were updated and refined.

Human Factors. Following this phase, human factors engineers pulled together a number of potential designs that were modeled on cameras, books, gaming devices, and other hand-held units. Flows of software user interfaces, as well as audio cues and button placements, were proposed. These were then subjected to user testing, whereby additional end users were allowed to hold, touch, and provide input to plastic models and mock-ups. It was at this phase that the insight regarding a downward pointing camera intended for use with documents placed on a table (see Figure 4) rather than wall posters came to light, fundamentally changing the teams understanding of how a reading device needs to be designed. Similarly, the decision as to where to place the key pad and other buttons for easy reach by adults and teen hands came out of this phase of development.

Industrial and mechanical design. Following the human factors work, industrial and mechanical designers defined the shape, color, and materials, working through the stack-up of major subsystems and the placement of key features.

Electrical and thermal engineering. With emphasis on achieving a good end-user experience of holding and utilizing the device, the system designers crafted board layout and airflows to fit with the mechanical and industrial design.

Software architecture. Research into third-party applications, as well as internal experience drove the team to use Moblin (see moblin.org) as the Linux platform and begin working on an architecture that could include both new and existing modules.

Alpha testing. Following six months of development, the team tested the first integrated system, molded in plastic, using an older version Intel processor (as the Intel Atom platform was still in development at that time). These systems were tested with another set of users, refining each of the key design elements while changes were more feasible.

Manufacturing. The team engaged with the manufacturing group to develop the high-volume capacity and necessary test infrastructure.

Future development. After launch, we will continue to improve the device further and provide updates to the product.

174 | Assistive Technology for Reading Intel® Technology Journal | Volume 13, Issue 3, 2009

Technical Design Challenges Three fundamental steps must be followed in order for the user to hear the text that is printed on a piece of paper: the image must be captured, the image must be converted to text, and finally the text must be rendered as audible speech. In some cases, one or two of the above steps can be skipped. For example, if the user already has the text in electronic form, then the capture stage can be skipped. This is common for e-book files. Even the text-to-speech step can be skipped if the text has already been recorded by a human. These are commonly known as books on tape or audio books, with MP3 being the most popular audio format.

Camera Placement The first design and usability challenge centered on camera placement. Initial prototypes placed the camera on the back of the device (as shown in Figure 3) similar to the design of most general-purpose cameras.

Figure 3: End-user Key Characteristics Source: Intel Corporation, 2009

Assistive Technology for Reading | 175 Intel® Technology Journal | Volume 13, Issue 3, 2009

“We knew we needed to develop a When placed in the hands of end users, it was interesting to observe their intuitive response: they stood up in order to capture pages in a book placed on different solution to make the product a desk. Standing was the only way for them to see what was displayed on the ergonomically friendly for all end users device screen, as the camera was pointed to the text laying flat on the table. When asked how they would use the device when sitting down, we observed of various age groups.” a very awkward twist in their wrist, in order to position the camera above the text, and a strain in their neck to peer above the device to view the display. We knew we needed to develop a different solution to make the product ergonomically friendly for all end users of various age groups. Interaction designers and engineers worked together to implement a solution to the problem: design a vertically-mounted camera placed at the bottom of the device. With the camera in this location, users can wrap both hands around the device and view the display that is positioned at user eye-level when seated — with the camera aimed directly above the printed text (see Figure 4).

?

Figure 4: Camera Placement at the Bottom of the Device Source: Intel Corporation, 2009

176 | Assistive Technology for Reading Intel® Technology Journal | Volume 13, Issue 3, 2009

Unanticipated benefits followed from this design choice. For most users, placing their elbows on a table assured them that the unit was level to the surface, and such a posture provided stability while holding the units, important for accurate imaging. For seniors, this was especially important, enabling them to rest while holding the device. For blind users, the physical location of the document between their elbows and directly in front of them while seated allowed for more accurate framing of the image.

Form-Factor Considerations In the midst of the human factors and usage research, we evaluated a number “One significant challenge concerned of end-user requirements. One significant challenge concerned the size of the device. Teens wanted the thinnest, lightest device we could design, while low- the size of the device.” vision seniors needed buttons to be placed far enough apart to allow easy use. Blind users had limited use for the screen on the device, though it would turn out that most blind users wanted to have a visual display so they could share information with sighted people. Low-vision users wanted to have as large a screen as possible. Each of these needs pushed industrial design in various different directions.

Placement of Buttons There are many considerations in the placement of buttons, among which are these:

•• Placement of the buttons influences the user’s experience and learning curve. The buttons must be close enough together such that a user does not have to stretch uncomfortably; yet, they must be large enough and far enough apart to accommodate users with large fingers and limited dexterity, including users with limited range of motion due to arthritis and other rheumatoid conditions. •• The main function of the device, capturing images, pointed to making the image-capturing button the most prominent button on the device. The large button, at the cut corner on the right, also oriented the device for a blind person, pointing the camera downward. For the teen or the senior, pushing the button while aiming it with one or two hands was an important consideration. This placement also was found to reduce motion blur induced by the user’s finger movement during the image capture. •• The team tested two-handed and one-handed usage models, determining that having one hand free to hold content or keep a book page flat would be instrumental to success. Following from this, the main keypad on the device (see Figure 5) was designed to be within a thumb’s reach while holding the device with the right hand, as almost 90 percent of the population is right-handed.

Assistive Technology for Reading | 177 Intel® Technology Journal | Volume 13, Issue 3, 2009

•• Human factors research also pointed to the importance of being able to move up or down in menu structures easily. A four-way directional pad with variable function, depending on context, met this need. Consumer electronic devices typically included a fifth navigation button for Select or OK button, and therefore we located this button within the center of the four-way direction pad. •• For advanced usages, these five navigation buttons have multiple functions depending on context. While playing a document, the Enter button functions as a play/pause button. This functionality, not originally planned, was added after user testing showed that many users expected this behavior. Similarly, the right and left buttons have multiple functions depending on context and usage. Holding the right button down while paused in a reading document, will allow the users to jump one page. If the users are playing back text, the right and left button accelerate progressively or jump back to recently read text. •• For the Intel Reader, a sixth button, called the back button was added to allow users to jump from lower-level menus to higher-level menus. In subsequent software versions, buttons would take on additional functions. For example, the back button, when held down for an extended duration, would return the user all the way to the Home menu, the top level of the user interface, rather than to just go up one level. •• Grouping of the other buttons on the device by function improved discoverability. •• Low-vision-related buttons, of no use to a blind person, and moderately useful to a dyslexic user, were placed together on the lower left. These buttons allowed for font magnification and for toggling between image view and text view. •• Voice speed changes were important to power users, those who are very comfortable with using assistive technology: these users listen to content at speeds of over 400 words per minute (four times common speech) and slow down for complex content. Similarly, all users wanted to have Favorites and Help easily accessible, which made dedicated buttons critical. •• A tactile adjuster for volume allowed one handed use, with a rocker-type button on the right side. •• Design considerations also included responsiveness of buttons, durability, and tactile feedback. For example, the button on the lower left for decreasing font size on the screen is indented, while the one for increasing font size is out dented, signaling to the users zoom in and out. Similarly, each of the buttons is shaped uniquely by function, allowing users to distinguish one button from another. •• In some cases, buttons were eliminated. For example, initial designs included a voice recorder for audio notes.

178 | Assistive Technology for Reading Intel® Technology Journal | Volume 13, Issue 3, 2009

The resulting placement of the buttons is shown in Figure 5.

Hold

Power Play

Shoot

Back

OK

Volume

View Toggle Navigation Buttons Zoom Out Intel® Reader Options

? Zoom In

Favorites Voice Speed Guide

Figure 5: Placement of Intel Reader Buttons Source: Intel Corporation, 2009

Image Capture The first major step is to capture the image of the target, which in this case is printed text. As with any camera, the target must be in focus, properly illuminated, and the capture time must be fast enough to prevent motion blur. The Intel Reader’s camera subsystem was designed to be fully automatic, including auto-focus, auto-exposure, and auto-flash mechanisms. The major considerations in image capture are discussed next.

Illumination The image sensor has minimum illumination requirements. In general, a sensor “The flash strobe can be significantly that has higher sensitivity will be more expensive, and a sensor that can operate at very low-light conditions will not have sufficient dynamic range to function lower in power than what might be at very high illumination. For the usages we considered, ambient illumination found on a standard camera that will vary across a tremendous range, from as low as 100 lux in a typical residential setting, to over 10,000 lux outside on a sunny day. might be two to three meters from the

The standard industry solution has been to use an integrated Xenon* flash target.” strobe, with an exposure time of a few tens of microseconds. The Xenon strobe is effective, because it can produce a large amount of illumination over a short time, with a reasonably uniform light distribution. For the Intel Reader’s usage models, the target is generally 10 cm to 1 meter from the sensor, so the flash strobe can be significantly lower in power than what might be found on a standard camera that might be two to three meters from the target.

Assistive Technology for Reading | 179 Intel® Technology Journal | Volume 13, Issue 3, 2009

In selecting the strobe, we also had to consider the useful life of the product. Intel® Reader A consistently-used Intel Reader could easily work through fifty pages a day of text, requiring 17,000+ strobe flashes a year. Given a useful life of three years for the device, 50,000+ would be a likely usage scenario. This drove designers to consider long-life strobes to ensure product durability.

? Focus The lens in front of the sensors determines the field of view. This field of view can be considered as a pyramidal cone, as shown in Figure 6. Often this is thought of as a stack of planes, where the area of the planes increases with the distance from the sensor. However, not all possible positions within the field of view are in sharpest focus. For any given distance between the lens and the sensor, there will be a plane of ideal focus, as shown in Figure 6.

Since the distance between the Intel Reader and the target is adjusted by the user, the lens must be moved relative to the sensor in order to have the target in the sharpest focus. The auto-focus algorithm calculates the required lens position. This algorithm must also account for slight hand movements by the user, especially for our low-vision senior users.

While the region of sharpest focus is shaped as a plane, the corresponding Figure 6: Mockup of Field of View and Plane of planes immediately in front of and behind that plane are still in reasonably Focus sharp focus. The distance between the sharpest plane and the plane of just- Source: Intel Corporation, 2009 acceptable focus is known as the depth of field. For usage models associated with the Intel Reader, the target material will often not be perfectly flat. For example, with a thick book opened to a page near the beginning, the distance from the sensor to the left side of the left-facing page may be several centimeters greater than the right side of the right page. The lens must thus have enough depth of field to account for that difference.

Preventing Motion Blur “The distance between the sharpest As the user holds the device, some natural motion will be imparted to the camera. The complete image must be captured while the image is stationary plane and the plane of just-acceptable on the sensor; therefore, the exposure time must be set correctly relative to the focus is known as the depth of field.” potential motion of the camera. For some users, especially elderly users with tremors, the motion may be significant. If the exposure time is too long, the image will be blurred, resulting in poor accuracy when the optical character recognition is attempted.

To prevent this blurring, the exposure must be fast. Of course, as the exposure time is reduced, the amount of light hitting the sensor will also be reduced. This drives sensitivity requirements on the image sensor, as well as performance requirements on the illumination source.

“To prevent this blurring, the exposure must be fast.”

180 | Assistive Technology for Reading Intel® Technology Journal | Volume 13, Issue 3, 2009

Even with the relatively low weight of the device, when capturing a large number of images, such as an entire magazine, newspaper, or book, the user may become fatigued. In addition, the book might not naturally stay flat on its own. The user might need some significant dexterity to hold the Intel Reader in one hand while holding the target open with the other hand, and at the same time not obscure the text of the target with his or her hand.

To address the user requirement of avoiding fatigue, especially for our low- vision senior group, and to allow the bound printed material to be held open, the team developed an accessory called the Intel® Portable Capture Station, as shown in Figure 7. The Intel Portable Capture Station has a tray that is large enough to keep magazines or books open, with two pages exposed in the view angle of the Intel Reader’s camera. The Intel Reader is placed into a holder that determines the exact height above the tray and positions the camera’s field of view.

Design of the Intel Portable Capture Station included a number of considerations. User focus groups identified durability as a key requirement, since this device is likely to be used in a public school setting. Portability is another key requirement, since the user may choose to bring it to work, school, or home. Figure 7: The Intel Portable Capture Station Source: Intel Corporation, 2009 A key element of the design of the Intel Portable Capture Station and the Intel Reader included the ability to transition from mobile to fixed usages. When the Intel Reader is used in conjunction with the Intel Portable Capture Station, bulk capturing of text becomes easy. However, for single worksheets and news handouts, capturing the text by using the Intel Reader in hand-held mode is quicker and more convenient.

Initial design included the possibility of an illumination system in the Intel Portable Capture Station itself. The additional complexity and associated cost of the system, including the electrical subsystems and the cost of the illumination elements, made using the lighting system on the Intel Reader a preferable choice.

Other user design requirements included an indented tray on the Intel Portable Capture Station to give users a physical guide to the Intel Reader’s viewable area when docked in the Intel Portable Capture Station. This allowed both sighted and non-sighted users the ability to orient the target material and assume all text is within the camera’s field of view.

The Intel Portable Capture Station was designed to have a separate button to initiate image capture by using the Intel Reader. Because the capture button “A key element of the design of the is placed on the bottom of the tray, the user does not have to reach up high Intel Portable Capture Station and to activate the shoot button on the Intel Reader that is placed in the holder, about two feet above the text. Users with reduced mobility in their arms find the Intel Reader included the ability this feature valuable. In addition, seniors often have difficulty raising their arms to transition from mobile to fixed above their shoulders, and the decision to place the capture button on the tray base was based on end-user feedback. usages.”

Assistive Technology for Reading | 181 Intel® Technology Journal | Volume 13, Issue 3, 2009

Optical Character Recognition (OCR) “A key step involves finding the Once the image has been captured, the next major step is to convert the raw image to text. This process is known as Optical Character Recognition. natural boundaries: the spaces between OCR software packages are commercially available and one was selected for columns, paragraphs, words, and incorporation into the Intel Reader. characters.” The OCR algorithm must examine the pixels associated with the image and derive the corresponding text. A key step involves finding the natural boundaries: the spaces between columns, paragraphs, words, and characters. Individual characters may only be a few pixels apart.

Key challenges to a highly accurate OCR process include having a sufficient number of pixels per character, having clear focus on the characters, and minimizing interference such as glare, etc. The differences in individual characters, such as the lower-case letter i and the lower-case letter l may be very subtle; therefore, acquiring a very accurate image of each character is critically important.

When capturing multiple pages, such as chapters of a book, the software is “Acquiring a very accurate image of designed to perform background OCR processing and allow users to listen to the first few pages that have been processed, if they choose to. Typically, each character is critically important.” by the time a user captures three or four pages of a multi-page document, the first page is available to be read aloud. This was a design tradeoff between optimizing for time to first sentence versus optimizing for capturing multi-page documents; we choose the latter based on our usage models.

Text to Speech After capturing the image of a page of text, the final stage is reading it aloud to the user, often called text to speech. This requirement also drives a wide variety of design decisions. Naturally, as users start to listen to more and more content, their ability to comprehend that content will also improve. For example, a new user may start out being comfortable with a speed of 110 words per minute, roughly the rate of common human speech, but will later find that speed far too slow, and he or she may eventually find that speed frustrating. However, even users that are comfortable with 300 words per minute for simple or “The ability to scale to much higher familiar content may need to go much slower for complex or unfamiliar content, such as a technical journal. Thus, the ability to scale to much higher rates will be of great importance to the rates will be of great importance to the long-term appeal of any device that long-term appeal of any device that reads aloud to users. The Intel Reader operates at rates as slow as 70 words per minute and allows playback of up to 500 words per minute, with increments at reads aloud to users.” appropriate rates in between.

Rendering Existing Electronic Content In designing the Intel Reader, we focused on the ecosystem of other consumer electronic devices within which the device would function, as well as the environment of electronic files for reading materials, such as plain text files or DAISY format (described later). Connecting to a personal computer, playing

182 | Assistive Technology for Reading Intel® Technology Journal | Volume 13, Issue 3, 2009

MP3 audio content, or even generating MP3 to play on other devices were critical to the usage model. To enable this transfer, the designs included the ability to access files on the system and transfer them on or off, by using a USB connection to a PC or by using USB flash drives.

Imported text files can be played in the same manner as text that has been “Instead of using the text-to-speech derived by using the OCR capabilities. However, files in MP3 and other digital audio formats require an entirely different playback mechanism. Instead of engine, a digital audio decode engine using the text-to-speech engine, a digital audio decode engine is required. is required.” While MP3 is a popular format for audio files, it is not the default for accessible text. A consortium known as the Digital Accessible Information System (DAISY) has established standards for Digital Talking Books (see www.daisy.org). The DAISY format includes both digital audio files with human narration and corresponding text of that file. The Intel Reader was designed to play both text only and audio versions of these files.

Organizing Files Over time, the user may accumulate a very large number of files. A familiar usage model is a filing cabinet, where related files are stored in the same drawer or folder within a drawer. As a user captures files, the files must be identified. To simplify this identification process, we set a default where the title of a document is identified, based on text size and placement on the page. The first “To simplify this identification process, 25 characters of this title become the title of the book. In a case where there is no clear title, the first 25 characters identified become the default title. The we set a default where the title of a user may rename the file and place it in one of several categories. To distinguish document is identified, based on text between documents with similar content, the file created also records the date and time. size and placement on the page.”

Sorting and managing content can also be done through a computer-based architecture, by attaching the device to a computer via a USB. Using a standard Windows* or Mac* interface to do this allows users with preferred screen- reading software to navigate the device content through the tools they use on a daily basis. Our team made a specific decision not to develop a separate user interface for the computer, allowing users to navigate the Intel Reader through the tools they already use.

Display Issues For users with vision, even limited vision, an electronic display, such as an LCD, can provide significant information. Some uses are obvious, such as display of functional menus or a preview of the text to be captured. A less obvious but very important use is to display the text as it is being read aloud by the device. For users with limited vision, seeing the text along with hearing the words yields a significantly improved reading experience. The same holds true “For users with limited vision, seeing for users with specific learning disabilities: multiple forms of input increases the text along with hearing the words comprehension. Given that users have differing levels of sight, the font size for the displayed text must be adjustable, even up to the point where a single word yields a significantly improved reading covers the entire displayed area. experience.”

Assistive Technology for Reading | 183 Intel® Technology Journal | Volume 13, Issue 3, 2009

“For users with specific learning A number of factors drive the decision on the size of the LCD panel to include. First, the LCD panel must be large enough to be useful. Users with limited disabilities, stochastic reading, placing vision will require very large fonts. For some users, reading is easier when many one word at a time on the screen, can lines of text are on the screen; low-vision users may have enough sight to shift between words while reading, thereby gaining context as they go. For users be useful.” with specific learning disabilities, stochastic reading, placing one word at a time on the screen, can be useful as this allows them to avoid scanning across a line of text, a common challenge for readers with specific learning disabilities.

Overall, having a larger display incurs many burdens on the system: the larger the display, the higher the cost, weight, and power consumption. One also needs to consider the business implications. Settling on a size that conforms to industry norms, for example, those commonly used for gaming systems, allows the design to benefit from the economies of scale for these adjacent markets.

“The LCD panels must include The display will be viewed under a wide variety of lighting conditions, ranging from dark rooms to bright sunlight. Given this requirement, the LCD panels integrated backlighting, with must include integrated backlighting, with LED-based illumination being the LED‑based illumination being the most power efficient. In order to have sufficient intensity, the voltage must be raised beyond the 12 volts provided by the main power supply, and a pulse- most power efficient.” width modulation must be applied to allow for variation in the intensity.

A wide viewable angle is also important: it allows users to hold the device in a lap or observe it in the Intel Portable Capture Station while seated or standing 30+ degrees above or below the screen.

For our blind users, having a screen is not critical. Their typical usage is to turn the screen off to save battery power and predominantly use menu voicing to determine device status; however, a number of blind users indicated that they would prefer to have a viewable screen to share content with sighted people when necessary.

Framing printed text is a challenge without audio cues to indicate whether the text is within the field of view of the camera. When specific usage tips are provided, blind users have more success framing the document. For example, a tip could be advising users to hold the Intel Reader a distance from the paper equal to the largest dimension of the page being imaged: that is, for an 8X11 piece of paper, the user is advised to hold the device 11 inches above the document to ensure it is captured. Placing the Intel Reader in the center of the document to be imaged and raising it directly above the document also proved useful for blind and low-vision users in testing. When capturing documents at home or in the office, having the Intel Reader docked in the Intel Portable “Placing the Intel Reader in the center Capture Station removes the height variable, and the user is confident that of the document to be imaged and anything placed within the defined borders of the base will be captured by the Intel Reader’s camera. raising it directly above the document also proved useful for blind and low‑vision users.”

184 | Assistive Technology for Reading Intel® Technology Journal | Volume 13, Issue 3, 2009

Storage Requirements Storage requirements were defined based upon usage. For a multi-megapixel “For text capture, therefore, the color image, the raw images may be over 10 megabytes each. Text is a very compact format with 1 character generally only requiring 1 byte of memory. memory has to be large enough to That same 10 megabytes can store over 5,000 pages of text, and if the text is handle capture of a few dozen raw compressed, storage can typically be increased by 2 to 4 times more. For text capture, therefore, the memory has to be large enough to handle capture of a images.” few dozen raw images.

Some of the storage space must be allocated for audio files and text files. Audio data can be tightly compressed due to gaps between words, the low base frequency, and the relatively small dynamic range of the human voice. MP3- type compression can yield files as small as 500 kbytes per minute. By far, the least amount of storage is required for text, as one minute of text at 160 words per minute might require only 800 bytes.

Testing showed that power users were likely to move content on and off the device to an external drive of a computer. This pointed the team to consider cost and to specify the size of the internal drive. Compression of images was also a critical element, allowing the user to get good OCR accuracy from full resolution images and then store the content once read in a low-resolution JPEG format.

Creating a Complementary Environment To assist in moving content into and out of the Intel Reader, a USB interface “A USB interface allows direct allows direct connection to any modern computer. The Intel Reader appears to the computer as would any storage-class device, such as is done with many connection to any modern computer.” digital cameras or external disk drives. This allows direct re-use of existing tools for movement of data files. Users can thus download content from remote servers to their computer, and then move it to the Intel Reader. Conversely, text captured by using the Intel Reader can be archived on the computer or sent to other systems.

In some cases, the user may not have access to a computer. In examining alternative data transfer methods, the USB flash drive appeared to be a great choice. No cable is required, as the flash drive is plugged directly into the corresponding USB host port on the Intel Reader. In addition to data transfers, the USB host port also allows some useful USB peripherals to be connected, such as a keyboard. This proved important in user testing for people who use alternative input devices, such as pointer or breath-activated keyboards. The keyboards generally function according to standard USB interfacing, allowing people with limited mobility and special-needs computing devices to use the “In addition to data transfers, the Intel Reader. As an extension of this functionality, the Intel Portable Capture Station also allows for USB connectivity, enabling the user to drive the device USB host port also allows some useful while docked in the station. USB peripherals to be connected.”

Assistive Technology for Reading | 185 Intel® Technology Journal | Volume 13, Issue 3, 2009

Selecting the Computing Environment “The Intel Reader needs a highly From our architectural exploration, it became clear that the Intel Reader must be based on a highly flexible platform. Several different vendors have flexible computing environment.” developed OCR and text-to-speech algorithms for different languages, and additional languages are being added each year. In order to take advantage of this continuous improvement in the OCR and text-to-speech environments, the Intel Reader needs a highly flexible computing environment.

In addition to flexibility, the processor must provide sufficient performance to complete the OCR algorithm within a reasonable delay. However, the size and power consumption must be low enough to fit with the required hand-held form-factor.

The Intel Atom processor and its associated chipsets running a general-purpose Linux operating system provides this type of flexibility, performance, power consumption, and size. The Intel Atom processor is an X86 compatible processor with operating speeds above 1 GHz.

Conclusion When developing technology to help people with disabilities become more independent, developers face a dilemma: if they tailor the use too tightly, the number of potential users is reduced to the point that the device is not economically sustainable. On the other hand, if they keep the feature set too broad, it fails to address the needs of any one community. In short, narrow designs fail for lack of a market, and broad designs fail for lack of use. The Intel Reader is an attempt to navigate this design challenge. The core usage model, reading text from a page with a camera and converting it to speech, addresses a need shared by blind, low-vision, and dyslexic users. In this way, we are attempting to reframe the market from a subpopulation of people with one of these disabilities to a larger population, renamed as those with Print Disability.

We had to make choices, focusing on users with a basic set of technology literacy skills; for example, someone who might use a cell phone or communicate via e-mail. We also balanced the competing needs of seniors and teens and of the sighted and those with limited vision. “We are attempting to reframe the market from a subpopulation of people with one of these disabilities to a larger population, renamed as those with Print Disability.”

186 | Assistive Technology for Reading Intel® Technology Journal | Volume 13, Issue 3, 2009

The Intel Reader can also be seen as a step on the road to the future of reading “By 2020, we should expect to see a and writing. By 2020, we should expect to see a range of devices that will move content seamlessly between print, digital, and audio. Moreover, we should range of devices that will move content expect the cost of such technology to plummet. The key inflection point seamlessly between print, digital, and in the cost curve will be when this technology moves from a necessity for a disability population to a convenience for the general population. Crossing this audio.” line will drive economies of scale similar to other common communication and entertainment technologies. Where DVD players once cost over $1,000, they now cost under $50. Cell phones, once the toy of the very rich, are now included for free with a service contract. We expect that in the future, the evolution of the Intel Reader technology will follow a similar path. Indeed, we share a vision of this technology becoming a very low cost device, one that will serve the disabled population and the general population. It is in this moment of cross-over that we will see the true goal of the device come to fruition — the establishment of an equal playing field for all people who want access to knowledge and increased independence.

Author Biographies Selena Chan is a Principal Engineer and has been at Intel since 2001. She currently manages end-user trials and market trials for the Intel Reader. Prior work involved delivering Intel technology platforms to the medical research community and leading external technical collaborations. She holds a Ph.D and M.S. degree in Electrical and Computer Engineering and a B.S. degree in Chemical Engineering. Her e-mail is selena.chan at intel.com.

Ben Foss is Product Development Manager for the Intel Reader in the Digital Health Group, leading a team of industrial, mechanical, electrical, and software engineers in the development of new products to help people with disabilities be more independent. He has been at Intel since 2003. Previously, he worked in the National Economic Council in the Clinton White House and was a Marshall Scholar. He holds a JD/MBA degree from Stanford. Ben is dyslexic: he volunteers his time as a board member of the civil rights organization, Disability Rights Advocates, and Headstrong, a non-profit group focused on issues regarding learning disabilities. His e-mail is ben.foss at intel.com.

David Poisner is a Senior Principal Engineer and has been at Intel since 1986. He is currently the chief hardware architect in Intel’s Digital Health Group. Prior work involved architecting nine generations of Intel chipsets and various platform technology initiatives in the realm of power management, security, audio, USB, storage interfaces, and LANs. He holds a BSEE degree from the University of Kansas. His e-mail is david.i.poisner at intel.com. Copyright Copyright © 2009 Intel Corporation. All rights reserved. Intel, the Intel logo, and Intel Atom are trademarks of Intel Corporation in the U.S. and other countries. *Other names and brands may be claimed as the property of others..

Assistive Technology for Reading | 187 Intel® Technology Journal | Volume 13, Issue 3, 2009

Developing Consumer and IT Platforms in a Regulated Medical Device Environment

Contributors

Barbara Fullmer Abstract Intel Corporation The Intel Digital Health Group, formed in 2005, is executing a steep learning curve as we develop our capabilities as a medical device manufacturer. Applying Alan Boucher the development practices used in the semiconductor industry to complete Intel Corporation regulated medical devices revealed number of important process gaps that needed to be closed.

Some of these differences result from the very different customers served by the Index Words Digital Health group. Some were the result of the different usage models and potential risks in healthcare. Other differences result from the conventions, Consumer Device practices and requirements of standards organizations and regulatory bodies. Medical Device This led to a series of complex inside-out structural modifications to the Digital Regulatory Health Group’s Product Development Lifecycle, which we discuss in detail Design Controls in this article. We also examine the effects of these changes on our group and Product Development the challenges we faced during the development of Intel’s first FDA Class II Validation regulated medical device.

Introduction The Intel Digital Health Group was formed in 2005 with the goal of applying Intel’s technical and business strengths to meet important needs in the healthcare industry. This was largely a new domain for Intel. We had provided semiconductor components to healthcare customers for many years and we had previously experimented with offering information security services for healthcare. We also had active research programs in home health and biomedical/life sciences. But we had never before designed and delivered fully integrated products, directly to users, which were subject to medical regulatory processes.

The Digital Health Group is engaged in building products which enable individuals to receive the care and assistance they need at home. This focus is “The highest quality of life and grounded in a philosophy we refer to as “Shift Left.” The highest quality of life and the lowest cost of care occur when individuals are living independently the lowest cost of care occur when in their own home, as illustrated in Figure 1. We have focused on developing individuals are living independently products and services which allow individuals to receive care and support in their own home that might otherwise be delivered in a clinical setting or an in their own home.” assisted living facility.

188 | Developing Consumer and IT Platforms in a Regulated Medical Device Environment Intel® Technology Journal | Volume 13, Issue 3, 2009

Home Care 100% Healthy, Independent Living Community Clinic

Chronic Disease Doctor’s Management Office

Residential Care Assisted Living Specialty Acute Care Skilled Nursing Clinic

Quality of Life Facility Community Hospital

ICU

0% $1 $10 $100 $1,000 $10,000 Cost of Care/Day

Figure 1: Shift Left, Stay Left — the Continuum of Care Source: Digital Health Group, Intel Corporation

A large proportion of healthcare spending is dedicated to caring for older “The first regulated medical device the patients who have one or more chronic diseases, such as diabetes, congestive heart failure, or chronic obstructive pulmonary diseases (such as emphysema). Digital Health Group developed was The first regulated medical device the Digital Health Group developed was a system designed to help such patients a system designed to help such patients manage their chronic diseases. Our system was to enable self-management by the patient with the assistance of manage their chronic diseases.” clinicians working remotely, often as part of a dedicated disease management team.

This first product, the Intel Health Guide, needed to be designed for use by elders who are not familiar or comfortable with computers or other complicated technology, and for use by skilled nurses and physicians who provide care for these patients. The system allows the clinical care team to plan a series of health sessions for the patient, in which the patient checks their own vital signs, answers survey questions about their current condition, and may be presented with educational material (text, photos, audio and recorded video.) The patient is reminded proactively to complete their health sessions. The clinical care team can also use video conferencing to interact with the patient, in order to coach them and assess their condition. “Collecting data from these regulated In order to simplify the use of the system and ensure accuracy, data needed to be collected automatically from medical devices such as blood pressure cuffs, medical devices results in classification glucose meters and pulse oximeters. Collecting data from these regulated of the Intel Health Guide as a medical devices results in classification of the Intel Health Guide as a regulated medical device in the US, the UK, and most other countries. regulated medical device.”

Developing Consumer and IT Platforms in a Regulated Medical Device Environment | 189 Intel® Technology Journal | Volume 13, Issue 3, 2009

The Intel Health Guide system would include a purpose-built PC with a Solution Integration specially designed user interface (for use by the elder in their home), a server- • Database ETL • Custom export based data collection system with a management application used by the clinician care team, and a set of tools to manage the deployment and maintenance of the system. Intel would act as an Application Service Provider, Content and Applications operating the server infrastructure and back-end applications. Figure 2 • Disease Management and RPM • Administration and Remote Management illustrates the components of the complete solution. • Education Content, Care Plans The needs and characteristics of several different user types would have to be Network considered. Patients would often be elderly, unfamiliar with technology, and • Broadband over Ethernet living in homes with poor communications and electrical power systems. Care • TCP/IP (TLS), VPN • USB, Bluetooth™ managers (often nurses with varying computer skills) might be working in central offices or from their own homes. Care manager administrators would OS, Drivers and System Software be assigning care managers to patients, and performing other administrative

• Windows XPe duties. System Administrators would be responsible for technical maintenance • Device drivers • VC Engine, VPN of databases, network connections and other infrastructure. Figure 3 illustrates the relationships between system components and these user types. IHG Hardware Core All aspects of the system needed to be designed for the intended users, and the • Intel MB, CPU, Chipset, BIOS • LCD with touch panel entire system would need to meet the requirements for medical devices. • Plastics and HIDs e.g. Gloworm, Buttons We knew we had much to learn, but we believed the strong capabilities of Telehealth Peripherals Intel’s product divisions and manufacturing organization prepared us well for • WeightScale the regulated medical device industry. The development team had extensive • Blood Pressure Cuff, Pulse Ox • GlucoseMeter, Peak FlowMeter experience in designing PC’s, and in designing, building and operating high reliability distributed IT systems. We had experienced software developers Intel Developed Intel Validated who had built complex software systems for business, engineering and manufacturing. Our corporate culture placed a strong emphasis on discipline and quality. The complexity of Intel’s semiconductor products, the tight process Figure 2: The Intel Health Guide System, controls required in our manufacturing processes, the reliability and correctness Version 1.0 required of our products, and the high volumes of products we ship would Source: Digital Health Group, Intel Corporation surely give us the necessary capabilities for development of the relatively simple medical devices we planned. We found that we faced a steeper learning curve than we had anticipated.

The direct user of Intel products in our core semiconductor business is a design engineer or an original equipment manufacturer (OEM) who develops and builds computer systems from Intel’s silicon and subsystem building blocks. As such, these architects, engineers and manufacturing personnel work within the culture of the electronics industry. They have specific expectations of technical design criteria, product specifications, product errata, supporting documentation, etc. “We found that we faced a steeper learning curve than we had anticipated.”

190 | Developing Consumer and IT Platforms in a Regulated Medical Device Environment Intel® Technology Journal | Volume 13, Issue 3, 2009

Patient Home Lighthouse Application Service Provider Data Center

Bluetooth Intel® Health Guide

USB Patient oC

Lighthouse Provisioning Application Peripherals Internet Care Manager Office Application Service Provider Data Center

Care Manager ® User interface Intel Care Manager Care Manager Administrator User interface

Work Order User interface Care Manager Administrator System Administrator Intel® Health Care Management Suite

Figure 3: The Intel Health Care Management Suite v1.0 Source: Digital Health Group, Intel Corporation

The typical Digital Health Group user, however, is not a design engineer or manufacturing team. The users of a Digital Health Group product may be a healthcare provider organization, a clinician, a patient, or a family member. The other critical stakeholders in the Digital Health Group environment are the regulatory bodies that oversee medical device development and manufacturing. The product characteristics and the development and manufacturing processes for medical devices must conform to the requirements of these customers and stakeholders.

As this became clear, we learned that we needed a new set of core capabilities in the Digital Health Group.

Digital Health Platform Design and Development The Intel Health Guide system was initially conceived and developed under Intel’s corporate Product Life Cycle (PLC), which has been proven over many years of successful semiconductor design and development.

The corporate PLC extends across the business landscape at Intel – from early market segment analysis to conceptualization of product families, through “The corporate PLC extends across the selection, development, product delivery and finally, product discontinuance or End-of-Life (EOL). business landscape at Intel.”

Developing Consumer and IT Platforms in a Regulated Medical Device Environment | 191 Intel® Technology Journal | Volume 13, Issue 3, 2009

“Consistency and visibility improve The corporate PLC is designed (as described in training material) to “drive shared expectations and resources such as structured systems, specifications, Intel’s product development and documentation and procedures as well as data systems and tools. This roadmap stability.” consistency and visibility improve Intel’s product development and roadmap stability.”

The corporate PLC provides a pathway to accomplish several major goals; increasing development efficiencies, producing stable and synchronized platform roadmaps, improving time-to-market and enabling organizations with better decision making capabilities.

(P)PLC Design Beta Approvals/ PFA Complete H/W Deploy Milestones FA A0-tape out PPS (P)POPL2 PDC PD PSA PCA Alpha PRA/PLA OIA/OSA DIA IPA A1/A2 SRA CA (P)POPL1 (P)POPL3 PIE PRQ

Exploration Planning Development Production

Figure 4: The Intel Corporate Product Life Cycle Source: Intel Corporation, 2009

“The PLC funnel diagram depicts The PLC funnel diagram (see Figure 4) depicts the 4 phases that comprise the Intel corporate PLC. Slanted lines between the phases, such as between the 4 phases that comprise the Intel planning and development, represent an overlap of activities between cross- corporate PLC.” functional teams and in some cases, even organizations. Intel’s corporate PLC has evolved over the years to support not just semiconductor development, but all aspects of platform development from silicon design, to subsystem building block development, to software subsystems, and eventually, to complete client and enterprise platform design and delivery.

“These documents enable all Each phase carries with it a unique set of boundaries and expectations in the life cycle. For instance, in the initial Exploration phase of the PLC, teams functional areas in the organization focus on basic or applied research, which identifies and tests new product to perform the planning needed to concepts, architectures, features or design approaches. Uncertainty is gradually reduced over the course of the exploration process, leading to a decision to establish the program Plan of Record.” commit larger amounts of resources to plan a product or system.

During the Planning phase, the Product Development Team (PDT) incorporates business analysis, user modeling and architectural requirements to create a product definition. This takes the form of product requirements, design documents, and quality planning. These documents enable all functional areas in the organization to perform the planning needed to establish the program Plan of Record (POR).

192 | Developing Consumer and IT Platforms in a Regulated Medical Device Environment Intel® Technology Journal | Volume 13, Issue 3, 2009

The Program POR defines the PDT’s commitments regarding product features to be delivered at various releases, delivery dates for interim and final product releases, product quality levels, product and program cost and pre-release design wins. Ultimately a decision is made within the organization to proceed to the “This phase usually involves several Development phase of the program. Here the product is engineered, developed and evaluated against its initial requirements and other program pre-production release gates and criteria. This phase usually involves several pre-production release gates and formal product testing activities.” formal product testing activities. Testing is conducted both internally by Intel and in many cases, externally as our customers perform validation activities.

Once product development is complete and a release candidate is ready for the market, the design transfer activities begin under the PLC. Production line qualifications, integrated system testing and yield analysis and improvement all culminate in a series of phase gate handoffs known as Product Release Qualification (PRQ) and Ship Release Authorization (SRA). These activities are components of the final design transfer process, which moves the product or platform from development into the Production phase.

The corporate PLC includes all of the steps necessary for developing complex, “Bringing the FDA into Intel required high quality products. However, we learned quickly that the details of processes that had evolved for semiconductor and computer building block development us to change our historical application were not always well-suited for developing complete medical devices, and were not aligned well with regulatory requirements. Our most significant issues of the corporate PLC and rethink how occurred in the Planning and Development phases. It is here that Digital we operated within the Digital Health Health encountered its greatest challenges. Group.” Bringing the FDA into Intel required us to change our historical application of the corporate PLC and rethink how we operated within the Digital Health Group.

Regulated Product Development The design, development, manufacturing and support of any product classified as a medical device is closely regulated everywhere in the world. In the US, the Food and Drug Administration (FDA) is charged with this responsibility. The quality system regulations the FDA applies are documented in the Code of Federal Regulations (CFR), 21 CFR, Part 820 (21 CFR 820).

Outside the US, most medical device regulations are based on ISO 13485:2007, which is closely aligned with 21 CFR 820. In both of these documents, each medical device manufacturer is required to implement “The design, development, a Quality Management System (QMS) suitable for their business, which manufacturing and support of any complies with the requirements. Historically, FDA interpretation is much more strict than ISO for requirements worded much the same. product classified as a medical device is closely regulated everywhere in the world.”

Developing Consumer and IT Platforms in a Regulated Medical Device Environment | 193 Intel® Technology Journal | Volume 13, Issue 3, 2009

“The QSR requires medical device The FDA’s quality systems regulations (QSR) uses a total systems approach designed to satisfy the specific safety and performance needs of each manufacturers to consider quality manufacturer, product, and user market. It is intended to have a positive at the earliest stages of all areas of impact on the quality, safety, and efficacy of the device. The QSR requires medical device manufacturers to consider quality at the earliest stages of all product design and development.” areas of product design and development. The FDA expects management and employees of each medical device manufacturer to fully commit to the firm’s quality management system (QMS). This commitment is vital, if the quality system is to be effective.

There were three paradigm shifts that needed to occur when the Digital Health Group began development of regulated medical products.

The first paradigm shift was that we needed to deeply understand the regulations that were now applicable to us. A critical document, 21 CFR Part 820–Title 21 Food and Drugs, Chapter 1, Food and Drug Administration, CFR (Code of Federal Regulations), Department of Health and Human Services, Subchapter H Medical Devices, Part 820, Quality System Regulation [2], “Good “This document has 15 subparts, each Manufacturing Practice for Medical Devices” quickly became standard issue to all Digital Health Group employees. This document has 15 subparts, each with a different topic area, such as with a different topic area, such as Quality System Requirements, Design Quality System Requirements, Design Controls, and Labeling and Packaging Controls. In addition to the regulations themselves, there is a classification system for medical devices that, when Controls, and Labeling and Packaging layered on top of the regulations, determines the level of scrutiny and the process and objective evidence that is needed to manufacture a given product. Controls.” (Medical devices are classified as Class I, II, or III, based on the level of control required to ensure the safety and effectiveness of the device). While there are many requirements defined in the regulations, there is little prescriptive direction (e.g., “Do it this way.”). This lack of specific direction caused confusion and frustration, as employees felt the goal posts kept moving, while the game was still in play.

The second paradigm shift was the realization that the regulations apply, like it or not. Intel employees are known for being detail oriented and for challenging all authority, real or imagined. While the specific application of how we implement the regulations is subject to interpretation, the fact that we were fundamentally subject to the regulations was not up for debate or discussion.

The third paradigm shift was the notion of objective evidence: what proof do you have that you did what you said you were going to do? Intel employees are accustomed to pointing to an e-mail, or hallway conversation, or a meeting, as their reason for a given design or development decision. We had to learn “We had to learn that anything that is that anything that is not documented correctly and archived in our controlled document repository simply didn’t happen, according to the FDA. not documented correctly and archived in our controlled document repository simply didn’t happen.”

194 | Developing Consumer and IT Platforms in a Regulated Medical Device Environment Intel® Technology Journal | Volume 13, Issue 3, 2009

The Inside-Out Analysis Faced with the paradigm shifts listed earlier, we followed the Intel standard practice and dedicated some key individuals to own the establishment of our QMS, with help from outside consultants. The result was an avalanche of documentation, starting with a Quality Manual, which was supported by a plethora of Standard Operating Procedures and Standard Operating Instructions (SOPs and SOIs, respectively). We had to create this library of governance and process definitions at the same time that design and development work was being done, generally by different employees. The exhaustive detail in the SOPs and SOIs gave us a false sense of being in control of design and development. In practice, our maiden attempt at a QMS resulted in an impossibly complex system that was not sustainable, nor was it “In practice, our maiden attempt appropriate to support our business model. Our attempt at design controls was at a QMS resulted in an impossibly viewed as over thought, overpriced, and overwhelming by our engineers, who simply wanted to design a good product. complex system.”

As we became aware of the burden we had inappropriately placed on ourselves, we began to detangle the web, section by section, and to figure out what design controls really meant for the Digital Health Group. We started with the basics, reinforcing one of the key phrases in the CFR — “establish and implement.” We needed to establish how we were going to do our work; we needed to implement those processes that we defined, and we needed objective evidence to document that we did what we said we were going to do. There was a lot of education about the significance of writing down the plan for the work — the how for design and development. We had to figure out the answers to lots of questions: What did the plan need to say? Who was subject to the plan, and why? What was covered in the plan, and what was excluded? Where were the “What was the right level of proof excluded areas addressed, if anywhere? How do we ensure that each of the to confirm that the activities had individual plans made sense, when combined with the rest of the project? We needed more education on the importance of creating and approving the plan occurred?” before any work was started, so that the implementation of the approved plan was clean and clear. The initial impression within the group was that the plan creation was simply paperwork; in other words, someone else’s issue, while the engineers did the real work, and that the plan could be done later. It took a long time to dispel this notion. The third paradigm shift, objective evidence, was even harder than the “establish and implement” phase of our process. What was the right level of proof to confirm that the activities had occurred? What level of approval was necessary for the proof? How did we ensure that the evidence really was evidence, and not just opinion?

Developing Consumer and IT Platforms in a Regulated Medical Device Environment | 195 Intel® Technology Journal | Volume 13, Issue 3, 2009

Overview of Intel’s Standard Product Life Cycle Versus the Code of Federal Regulations “We set about decomposing the Initially, we set about our detangling in a piece-meal fashion, going to the areas of the most pain first. While this approach staunched the bleeding, it still did standard Intel PLC while at the same not address the fundamental problem: what did the Digital Health Group need time decomposing the CFR.” to do to satisfy the CFR, while designing and developing products inside of Intel? To get to the bottom of the issue, we set about decomposing the standard Intel PLC while at the same time decomposing the CFR. We had to see where we needed to make changes, and to see how we could leverage what we already knew and were doing, to satisfy the CFR. In short, we began bringing Intel to the FDA.

One of the first distinctions we recognized was that we were not obligated to satisfy strict design controls while still in the Exploration phase. This was amazingly freeing for our incubation teams — they could spend their energies thinking about new opportunities and getting detailed ecosystem and user understandings, without having to check they were following the regulations each step of the way.

“This was amazingly freeing for our Once we internalized the separation of exploration from design and development, we realized that there were other clear separations within the incubation teams — they could spend CFR which were not reflected in Intel’s standard PLC. We knew that we their energies thinking about new confused our regulatory inspectors when we spoke, because we used words that were familiar to the FDA and to ISO; but our words meant something opportunities and getting detailed completely different. We used the word planning for what the FDA considered design and we used the word design when the FDA meant development. ecosystem and user understandings.” Another new concept for our group was formal “traceability.” This is the process by which design inputs are matched to design outputs, and then verified (ensuring outputs match inputs) and validated (ensuring the final device satisfies the intended use) for completeness. Risks are identified and categorized for potential severity during the input gathering stage; they are traced to outputs, and the risk mitigations are verified. All of this linking — inputs, outputs, risks, verification, and validation needs to be documented, and exceptions need to be identified and dispositioned. Nothing is thrown out without justification, and you have to have evidence of that justification. Coming from an industry in which it is normal practice to delay features to satisfy a schedule imperative, this concept of accountability for all recorded inputs was a big shift for us. It meant we had to deliver wholly on the requirements we defined for ourselves: a new meaning for us for the word “Cutting back on product features was verification. Cutting back on product features was not permitted without not permitted without justification justification and evidence. This meant we had to validate that our product satisfied its intended use by the customer, through user test results documented and evidence.” with objective evidence.

196 | Developing Consumer and IT Platforms in a Regulated Medical Device Environment Intel® Technology Journal | Volume 13, Issue 3, 2009

Finally, and perhaps most importantly, we had to establish and document the “We had to establish and document objective evidence that we had accomplished all of these tasks prior to release of product to customers. Intel has always done excellent engineering work, but the objective evidence that we had we never had to prove each element of the process to an auditor or other third accomplished all of these tasks prior to party, as is needed for medical device manufacture. The evidence had to be reviewed, agreed to, and approved by senior staff, and anything that was not release of product to customers.” satisfied had to be identified and then dispositioned. The disposition had to be documented as well, and it had to be readily available to an inspector.

A Walk Through the Code of Federal Regulations As a medical device manufacturer who is making devices in the US but distributing them in Europe, we are subject to both ISO and the CFR. The two sets of regulations have many similarities, but often differ in their emphasis. For our purposes, we had to blend these two sets of regulations to ensure we were meeting all the imperatives we needed to. In many cases, the CFR takes the lead, so we began our analysis by walking through the section of the regulation that applies to design and development: 21 CFR Part 820.30: Design Controls for Medical Devices. There are 15 subsections that we needed “We had to blend these two sets of to internalize and then institutionalize. We needed to tell our employees regulations to ensure we were meeting the why, as well as the how. Deciphering the CFR was much of the why, along with some hard-learned lessons from our employees who had previous all the imperatives.” experience designing medical devices outside of Intel. We also learned that institutionalization of regulations was much more successful if we provided our engineering audience with templates and comprehensive examples of how to satisfy the imperative.

Each section of 21 CFR 820.30 was analyzed carefully.

§820.30 (a) General This is the section that defines who is subject to meeting the design control section of CFR, and who is not — the biggest why we had to satisfy. There are three classes of devices, as mentioned previously, as well as some special allocation of design controls in specific classes. The most significant learning for our group was that because we automate our device with software, we will always be subject to design controls. Once we were able to help people realize this, teams more or less quieted down on their protests against design controls. “Because we automate our device with software, we will always be subject to design controls.”

Developing Consumer and IT Platforms in a Regulated Medical Device Environment | 197 Intel® Technology Journal | Volume 13, Issue 3, 2009

§820.30 (b) Design and Development Planning “Each manufacturer shall establish and maintain plans that describe or reference the design and development activities and define responsibility for implementation.” This is where it began to hit home that we had to define our plans in advance, we had to assign ownership for the elements of the plan, “Once we established the Design and we had to maintain the plan so that it continued to drive design and and Development plan as part of development activities. We internalized this into our Design and Development Quality Plan, a template that ensured the key elements of the regulations the process, there was the continuing were consistently maintained across multiple programs. The development education for the engineers.” activities matrix is the real meat of the artifact, a table that lists each design and development activity, its owner, what the owner is responsible for, a list of applicable SOPs or SOIs for the activity, change acceptance criteria, and with whom the owner has to interact in order to ensure successful delivery of the element. Once we established the Design and Development plan as part of the process, there was the continuing education for the engineers: this was now what they were adhering to, not an e-mail or PDT conversation.

§820.30 (c) Design Input “Each manufacturer shall establish and maintain procedures to ensure that the design requirements relating to a device are appropriate and address the intended use of the device, including the needs of the user and patient. The procedures shall include a mechanism for addressing incomplete, ambiguous, or conflicting requirements.” “First and foremost, CFR’s focus is on This is where the real essence of “What is a requirement?” started to hit home. There are keywords in the CFR that mandate that we realign our traditional ‘needs of user and patient.’ ” Intel perspectives. First and foremost, CFR’s focus is on “needs of user and patient.” We therefore had to ask ourselves who our user was. We had several, ranging from the patient whose care is monitored by the device, to the installer who places the device in the home, to the IT administrator, who is responsible for the backend support of the device. These are very different from the standard user for Intel’s core business, the design engineer or systems OEM. Once we had the users defined, we could start to get specific on the requirements, what we came to speak of as the what. We had a version of a Product Requirements Document, and we had a Market Requirements Document, standard Intel PLC artifacts. Our issue quickly became that we had too many requirements, in too many different places, and with too many interpretations. We compounded this problem by the way we used the word requirement: an all-encompassing term for anything we wanted the device to do, versus the FDA’s much more narrow definition. In order to help our teams grasp the FDA’s meaning of requirement we had to first help them identify where requirements came from. “We had to first help them identify where requirements came from.”

198 | Developing Consumer and IT Platforms in a Regulated Medical Device Environment Intel® Technology Journal | Volume 13, Issue 3, 2009

MDRs User Market Clinical Complaints Studies Data Data

Defined User Needs “Requirements come from a variety of sources.”

Specification

Product

Figure 5: FDA Requirements for a Medical Device Source: Digital Health Group, Intel Corporation

FDA guidance is that requirements come from a variety of sources, such as “When the word should is used, it can user needs’ studies, marketplace needs, customer complaints, and clinical data. There may be more than one user for a product — for our first product, be interpreted in many ways and that we needed to consider not only the patient in the home environment, but can strangle a design team.” also the clinician and anyone else who came in contact with the product. Requirements can be derived from initial hazard analysis; they are functional expectations. Once the PDT understood where requirements came from, we had to simplify our requirements and put them into a single document, to address “incomplete, ambiguous, or conflicting requirements.” We needed to use strong, declarative statements, and we needed to define when a statement was a design intent (nice to have) or a requirement (device does not ship until we satisfy it). “The device SHOULD do something,” versus “The device SHALL do something.” When the word should is used, it can be interpreted in many ways and that can strangle a design team and paralyze Verification and Validation (V&V) later on. If a device should be capable of running 24 hours a day, but it is actually only capable of running 23 hours a day, does the device fail the requirement? Sorting through all the words of the combined MRD- PRD took a very long time, and so we engaged outside consultants to assist us with this task. We needed people with clear heads, who were familiar with the endgame of medical device manufacture, but who were not invested in the device, to ask the tough questions. We needed to frame our requirements with the idea that they should be answerable with a yes or a no. An example requirement: “The Care Management Suite functions shall be presented in English.” This can be answered with a yes or a no, versus other items that felt like requirements (but really were design intentions) such as “Menu structures “When we had completed the work on the PHS will be simplified.” We started out by just cleaning up our new PRD, tagging requirements, and making them testable. This experience was on the Intel Health Guide, we moved painful and time consuming, but was essential to our success. When we had completed the work on the Intel Health Guide, we moved onto creating a onto creating a template for future template for future programs, so that others could start off on a good footing. programs, so that others could start off Our template had clear examples, based on what we had learned, so that the future authors could understand the difference between a good requirement on a good footing.”

Developing Consumer and IT Platforms in a Regulated Medical Device Environment | 199 Intel® Technology Journal | Volume 13, Issue 3, 2009

“Now that we had clear requirements, and a poor one. We defined some mandates, to simplify life further on in development, and during the creation of the PRD: requirements to be how did our design satisfy them?” numbered sequentially, no deleted requirements without justification, no reuse of identification numbers, proper use of the word shall, no use of the word should, and most importantly, user requirements to live ONLY in our PRD.

§820.30 (d) Design Output “Each manufacturer shall establish and maintain procedures for defining and documenting design output in terms that allow an adequate evaluation of conformance to design input requirements.”

There were two new concepts here: documenting outputs and evaluating conformance of those outputs to the inputs. Now that we had clear requirements, how did our design satisfy them? This was where we redefined specification to help Digital Health Group employees use the word in the same way that the FDA did, so that the result of the specification could be answered in a measurement (versus the yes or no of a requirement). The “Our users for the device are patients final specification, for font size on the device, read like this: “The PHS computer related interfaces shall use a font visually equivalent in size to 14 who may have visual impairments, point Verdana.” This is an important specification; although at first it might limited mobility, and multiple disease not appear so. Our users for the device are patients who may have visual impairments, limited mobility, and multiple disease states. We could not states.” presume their vision to be 20/20, or that their dexterity was equivalent to an athletically fit person. We had to make it easy for our user (the patient) to read the information presented on the device. The first version of the specification was much less precise and not measurable: there was no specific mention of font size or type, nor any definition of where this would apply.

The second area, conformance of outputs to inputs, meant that we had to get skilled at translating a requirement into a specification. In the example just cited, the final version of the requirement read: “PHS user interface, PHS education content, and PHS training material shall be suitable for use by patients with corrected vision adequate to read a typical large print magazine, “We had to get skilled at translating a such as the Readers Digest large print edition.” In other words, this is the what. Now, the specification made sense: the how of satisfying the requirement was requirement into a specification.” 14 point Verdana font on the PHS interface.

§820.30 (e) Design Review “Each manufacturer shall establish and maintain procedures to ensure that formal documented reviews of the design results are planned and conducted at appropriate stages of the device’s design development.”

The CFR goes on to provide additional clarity on who needs to attend these reviews, and to stipulate that the results need to be archived in the Design History File (DHF). As an engineering company, Intel was familiar with performing design reviews during the course of product development, but the premise of a formal review, with specific attendees who could not be omitted, and which was documented and archived, was a new twist. E-mail meeting summaries, which our group was well accustomed to, were no longer sufficient.

200 | Developing Consumer and IT Platforms in a Regulated Medical Device Environment Intel® Technology Journal | Volume 13, Issue 3, 2009

These often had insufficient information on the attendees (what role were they “Design Input, Design Output, fulfilling), too much information on the dialogue, not enough information on any decisions that were made during the review, and not archived in our and Design Transfer reviews hold a controlled document repository. “If it isn’t documented, it didn’t happen” particular meaning for the FDA, and became a familiar refrain, particularly when the subject of design reviews was discussed. Our initial SOP was overly burdensome, people were not these happen only once in a product’s effectively trained on it, and we were haphazard in the archiving of the review and the materials covered in the review. When we did archive material, we development.” were inconsistent in tracking our actions required (ARs). Were the ARs really related to the form, fit, or function of the product? If so, then we had to have objective evidence that they were addressed and closed. If they were related to general process or activities (such as “Send Alan the e-mail on the new SW code.”), then it did not need to be archived in a design review. This distinction has taken some time to become clear in many people’s minds. In addition to the mechanics of design reviews, we had to clearly define types of review, and we had to start to use the review terms the way the FDA meant them. Design Input, Design Output, and Design Transfer reviews hold a particular meaning for the FDA, and these happen only once in a product’s development. The “Verification means to the FDA that employees in the Digital Health Group felt that many design reviews were design input reviews, so again, our training programs had to carefully clarify our outputs clearly match our inputs: the new meanings we were embracing, as a medical device manufacturer. did we build what we said we were Design Review is an area we continue to focus upon, balancing between simplification and detail, with the goal of realizing full business value. going to build?”

§820.30 (f) Design Verification “Each manufacturer shall establish and maintain procedures for verifying the device design.” Verification means to the FDA that our outputs clearly match our inputs: did we build what we said we were going to build? It was during our initial attempts at verification that we really started to understand the cost of a poorly written specification. If we had a specification that could not be measured, then how could we verify that the design met the specification? Therefore, we started to get our verification team involved in the refinement of the specifications themselves: we engaged them in the upfront work, so that they could better perform their roles later on.

We also realized we could perform verification more efficiently. We previously had the perception that verification was something we did only at the end of the project, and we did not realize that there was much valuable verification work happening upstream of the final verification. This work met the intent as well as the requirement for thorough verification, but we had not been taking credit for it by documenting the right objective evidence. For example, engineering verification tests, code walk-throughs, stack-up analyses and other engineering tasks all could play a role in substantiating verification. We did not “Engineering verification tests, code realize this at first, and so did some redundant tests. Recognizing that work walk-throughs, stack-up analyses and we were already doing would help us substantially in the end was helpful in getting engineers to embrace the QMS. This was a clear case where we were other engineering tasks all could play a already doing good work, but not giving ourselves proper credit (the objective evidence) for it. role in substantiating verification.”

Developing Consumer and IT Platforms in a Regulated Medical Device Environment | 201 Intel® Technology Journal | Volume 13, Issue 3, 2009

§820.30 (g) Design Validation “Each manufacturer shall establish and maintain procedures for validating the device design.” “Validation, to the FDA, means that Validation, as practiced in Intel’s core businesses, involves thousands of hours of laboratory testing with commercial software, and the use of engineering our final product satisfies the intended samples by validation partners. Validation, to the FDA, means that our use we defined.” final product satisfies the intended use we defined. Does our product “allow caregivers to remotely access vital sign measurements of patients at home”? Does it “collect and store information and display it accurately on the LCD screen”? We were so occupied with the design and development activities that we did not internalize that the only way to validate our design, in the FDA meaning of the word, was to test our device with real users or at least with people who fit the profile of our real users. How do we set up those tests? How do we recruit users? How do we document the results, and what do we do with all the feedback received? Sorting through the information derived from these tests, and determining which feedback meant the device did not meet the intended use, and which feedback was a potential future enhancement, took tremendous time that had never been planned into the project schedule.

§820.30 (h) Design Transfer “It is imperative that device design be “Each manufacturer shall establish and maintain procedures to ensure that the device design is correctly translated into production specifications.” transferred to the manufacturing floor Design transfer simply means that product designs must be converted accurately.” into production specifications that allow the product to be manufactured repeatably and reliably, and that was validated to meet product requirements. This requires a correct and complete product design, clear specification of required manufacturing process capabilities, appropriate procedures for moving the design from the development team to the production team, and well documented testing processes. It is imperative that device design be transferred to the manufacturing floor accurately, so that the device quality is not compromised. The detail necessary to accomplish this objective can vary considerably, depending on the type and class of device, intended use, and the knowledge transfer activities between the design and manufacturing teams. The Digital Health Group engineering team now had to learn how to effectively move the design into a High Volume Manufacturing environment. Further complicating the transition was our prototyping strategy, which involved transferring several revisions of the design to the manufacturing team in order to support field tests. Although the engineering team often felt we were constantly doing design transfer, we had to improve our processes to fully meet “The Digital Health Group FDA requirements. engineering team now had to learn how to effectively move the design into a High Volume Manufacturing environment.”

202 | Developing Consumer and IT Platforms in a Regulated Medical Device Environment Intel® Technology Journal | Volume 13, Issue 3, 2009

§820.30 (i) Design Changes “Each manufacturer shall establish and maintain procedures for the “We had to establish a formal change identification, documentation, validation, or where appropriate, verification, review, and approval of design changes before their implementation.” control process, so that we could ensure

We had to learn when to start tracking changes, as mandated by the CFR, that all the steps mandated were versus when changes were simply the natural evolution of the design and not satisfied.” subject to inspection. We had to establish a formal change control process, so that we could ensure that all the steps mandated were satisfied, and that we had the right objective evidence to support the decision to invoke the change. The change control process we defined had attributes that we were becoming quite familiar with, such as the definition of the right roles to be represented during the decision making, the definition of the evidence necessary to support the decision, and the appropriate archiving of the decision as well as the evidence. We had twice weekly change-control board meetings, and sometimes emergency meetings, to start to indoctrinate the team in this process. Our biggest problem, prior to the change control process improvement, was when to put a design element under formal design controls. Like many start-up 2 medical device manufacturers, we put everything under formal control, right Program Revision Control out of the gate. This added a significant amount of overhead and frankly, pain, to our design teams. It made the design and development work cumbersome 1 and difficult for the PDT. We streamlined the process, by giving PDT two forms of change control: Program Revision Control, leaving change control in Formal Revision Control the hands of PDT where it belongs, and Formal Revision Control, used for 0 those design elements that need full control early on during product design Design Development Transfer and development activities (see Figure 6). As the product evolves, so do the design elements that support the design and development activities. This variable and increasing approach has significantly reduced the amount of Figure 6: Change Control Process overhead associated with design changes. Source: Digital Health Group, Intel Corporation

§820.30 (j) Design History File (DHF) “Each manufacturer shall establish and maintain a DHF for each type of device.”

This mandate is the ultimate test of “if you didn’t document it, it didn’t happen.”

The intent of a DHF is to create a repository for all design and development work, one place where all design inputs, design output, tests, test output, etc. is gathered. No longer was this to be stored on someone’s computer — the information needed to be available to the team. The contents of a DHF provides objective evidence that we followed our design and development plan. “This mandate is the ultimate test of ‘if you didn’t document it, it didn’t happen.’ ”

Developing Consumer and IT Platforms in a Regulated Medical Device Environment | 203 Intel® Technology Journal | Volume 13, Issue 3, 2009

“Finally, we could go about the task Our early DHFs were haphazard. We complicated them by the fact that we had multiple document repositories, due to transitions within Intel for of ensuring that the records actually document storage, as well as the reality that each engineer had his or her own demonstrated that ‘the design was favorite repository or Web site. We had a controlled document repository that was difficult to use, and that only accepted physical signatures for approved developed in accordance with the documents, a true anomaly in an IT-based, e-mail-centric corporation. The first approved design plan and the task we faced with the DHF was corralling anything that could be identified as a DHF item. We cast a wide net, too wide as it later turned out. The second requirements of this part.’ ” task was to get the right level of physical approvals on the artifacts — not an easy task in a business group with employees in many separate physical locations. The third task was actually loading the items into our controlled repository and appropriately storing the physical copies. Finally, we could go about the task of ensuring that the records actually demonstrated that “the design was developed in accordance with the approved design plan and the requirements of this part.”

New Product Life Cycle, Expanded and Aligned with the Code of Federal Regulations Based on our new understanding of both the CFR and ISO, it became obvious that we needed our own unique PLC, separate from the rest of Intel, to ensure that we were following the CFR and the appropriate design controls (see Figure 7).

Production Release Authorization Production (PRA) Product Ship Candidate Release (PPC) Authorization (SRA) Product Overview Concept Product Product Proposal Approval Overview Overview Level 3 Product (CA) Proposal Proposal (POPL3) Discontinuance Level 1 Level 2 Regulatory Authorization (POPL1) (POPL2) End of Clearance (PDA) Life (EOL) Market Planning Market/Sell Product Manufacture/Distribute Explore Assess Feasibility Design and Develop Product

Research Product Support Product

Figure 7: The Digital Health Product Life Cycle Source: Digital Health Group, Intel Corporation

204 | Developing Consumer and IT Platforms in a Regulated Medical Device Environment Intel® Technology Journal | Volume 13, Issue 3, 2009

We spent several weeks decomposing the inputs and outputs for every major “We started with the Intel corporate activity in the Digital Health PLC, ensuring we had clear mapping to the entire Quality System Regulation, 21 CFR 820 and clearly mapping out PLC, because we knew that our dependencies and deliverables. Once we had a good handle on those aspects, engineers had been well trained on it.” we were ready to create our own PLC. We started with the Intel corporate PLC, because we knew that our engineers had been well trained on it and because we knew we would have to correlate back to it when we did the training on the Digital Health Group’s version. We renamed major blocks of activity to match FDA and ISO terminology.

We defined the phases as follows:

•• Research includes the study of user needs, ecosystems, and new technologies. •• Explore includes the activities that deliver the initial investigation of a product or platform opportunity. •• Assess Feasibility includes the activities that analyze the business, user, and technology feasibilities of a concept. •• Market Planning includes the activities that define the marketing approach and sales strategy for the product. •• Market/Sell Product includes the activities that execute the marketing and sales plan. •• Design and Develop Product includes the activities that lead to the creation of the product. •• Manufacture/Distribute Product includes the activities that create the final product for sale, and that transfer the product from the manufacturer to the customer. •• Support Product includes the activities that enable the successful use of the product, once it is received by the customer. We created separations between the blocks, so that the decisions, handoffs, and design control expectations were clear. Marketing a medical device, either in the United States or the European Union, is quite different than Intel’s standard marketing activity. We had new restrictions on what we could say, to whom we could say it, and at what stage in the product development life cycle we could say it. Therefore, we also created a separation between marketing and design and development, as well as a separation between the market planning and marketing and selling activities. “We created separations between the blocks, so that the decisions, handoffs, and design control expectations were clear.”

Developing Consumer and IT Platforms in a Regulated Medical Device Environment | 205 Intel® Technology Journal | Volume 13, Issue 3, 2009

From Feasibility to Design and Development “Our group needed the business Now that we had a clear distinction between the research, exploration, and design/development spaces, we needed to decide how to capture the early work analyses contained in those business simply, in a manner that supported management decision-making, but also artifacts to make solid resource supported the brainstorming and innovation we needed. We learned that there are business artifacts that are not required by (or reviewable by) the FDA or allocation decisions.” ISO, but we knew our group needed the business analyses contained in those business artifacts to make solid resource allocation decisions. Intel’s approach to these decisions focuses on the three vectors of Technology, User, and Business, so we used those vectors as the basis of our Concept Requirements Document, the template and artifact for capturing the early thinking. The template has five major sections: overview, marketing strategy, users, concept functionality, geographies, and customer engagements. We provided many examples, and we also established the mechanism (relying on our now trusted friend, Design Review) for moving the information across the separations between feasibility and design and development planning, and market planning.

Quarantining Work Products “In regulated medical development, One of the fundamental challenges early in the PLC is properly transitioning the artifacts developed outside of design controls (during Explore and Assess however, there must be a logical Feasibility phases) into the Design and Develop Product phase. Work firewall at the divide between the products of the early phases include lists of possible requirements, design alternatives, potential features, mockups, prototypes, etc. Because they were Assess Feasibility phase and the developed outside of design controls, they usually do not have carefully documented plans, objective evidence, and traceability. In other product groups Design and Develop Product at Intel, it is perfectly natural to continue the development and maturation phase.” of product prototypes through downstream PLC phases. In regulated medical development, however, there must be a logical firewall at the divide between the Assess Feasibility phase and the Design and Develop Product phase. We can’t just carry artifacts forward into the design and development process. Yet we need to capture the value of the work done during the early phases. We needed a mechanism to prevent excursions from design controls, without throwing away useful work.

In the Digital Health Group, we created a landing zone for those artifacts under a set of internal systems and processes using Subversion* (a software version control repository), which is a validated tool under Intel’s design control and quality management system.

“All of these reviews are done in an As concepts are promoted from Assess Feasibility into the Design and Develop Product phase, we perform a series of reviews with the originating effort to assess which assets might be research teams, the user experience team, the marketing team, the quality and regulatory team, and any external vendors who may have participated transferable to our quarantine area.” in analysis or prototyping efforts. All of these reviews are done in an effort to assess which assets might be transferable to our quarantine area within Subversion. Once in the quarantine area, they undergo further examination, modification, adaptation, and evaluation for reuse, as we develop the functional requirements, engineering specifications, patterns, and designs for the product.

206 | Developing Consumer and IT Platforms in a Regulated Medical Device Environment Intel® Technology Journal | Volume 13, Issue 3, 2009

Much like any technology, vendor, or software-of-unknown-provenance “How are we making sure that we (SOUP) component that might be selected for use in the design and development process, a work product crossing from Assess Feasibility to have thought about the product safety Design and Develop Product must be carefully selected and vetted to ensure risk, as we design and develop our that the proper hazard and risk analysis have been done and that the work is thoroughly documented and reviewed. Moreover, the V&V processes must product?” provide for the correct level of objective evidence and traceability to show that the eventual released product has been properly developed under the appropriate levels of design control.

Risk Management — Do it Early, Do it Often Risk Management is a prime example of a term that means one thing to Intel at large, and something very different to the FDA and ISO. To Intel, it means the risk to our brand identity, the risk that the CPU may malfunction, or the risk that the notebook may get too hot, because of our circuit design. To the FDA and ISO, it means the risk to the user(s) of the product we’re designing and manufacturing: will it harm them? How are we making sure that we have thought about the product safety risk, as we design and develop our product? The ISO places more emphasis on risk than does the CFR; we therefore work (A) Planning to the ISO standard. • Scope and Timing • Risk Analysis Team Selection • Communication/Reporting The purpose of implementing a Product Risk Management Plan is to establish and maintain a documented process for identifying hazards associated with a device (the potential source of harm), safety risks (combining the probability of occurrence of harm and the severity of that harm), estimating, evaluating, and (B) Risk Analysis controlling them, determining the acceptability of residual risk, and then • Intended Use identification • Hazard identification monitoring the effectiveness of the control during the PLC (see Figure 8). We • Risk estimation learned that the FDA and ISO expect product risk management to be implemented early and maintained throughout the entire PLC. We also realized that we could derive benefits from embedding product risk Assessment

(C) Risk Evaluation Risk management into our thinking. By establishing a metric for the severity of a • Risk acceptability decisions given risk, and establishing the likelihood of occurrence of the risk, we could quickly determine if the risk was acceptable, that is, as low as reasonably Management possible (ALARP), or intolerable. Intolerable risks clearly had to be addressed Risk and mitigated, but acceptable risks were already done. By integrating this (D) Risk Mitigation thinking into our SOPs and templates, design teams were able to work more • Option analysis • Implementation effectively, putting their efforts into those areas where risks were intolerable • Residual Risk Evaluation • Overall Risk acceptance rather than trying to mitigate risks that were acceptable.

(E) Post-SRA Risk Monitoring

• New hazards/failure modes • Existing risks no longer acceptable

Figure 8: Risk Management Flow Source: ANSI/AAMI/ISO 14971:2007

Developing Consumer and IT Platforms in a Regulated Medical Device Environment | 207 Intel® Technology Journal | Volume 13, Issue 3, 2009

Defined User Needs

Device and labeling must be easily read by visually impared user

Specification

PHS interfaces shall be visually equivalent to 14pt Verdana

Output (Component/Product)

Screens and Labeling materials in 14pt font

Figure 9: Verification Source: Digital Health Group, Intel Corporation

Validation and Verification — Not at the Same Time “Initially, our thinking was that they Verification and Validation (V&V) were not new concepts to our group, nor to Intel, but the execution of them in a medical device manufacturing were similar in intent, and could environment was wholly different. Initially, our thinking was that they were easily be run in parallel: we were similar in intent, and could easily be run in parallel: we were wrong. wrong.” What we had to learn was that validation, ensuring that our device conformed to the user needs, was insufficient if verification (ensuring that our design output met our design input requirements) was not completed. Therefore, if we did not verify and validate our design in sequence, we would end up redoing a lot of work. Granted, the more often we did V&V, the better we became, but it was an unnecessary step that we imposed upon ourselves, due to a lack of full comprehension of what was needed. We initially had a bad habit of pushing engineering testing “over the wall” to the verification teams. Instead of verifying that our design outputs met our inputs, the verification teams would do code analysis, mechanical debugging, and whatever else they could think of to measure. In truth, correctness to specification was what we needed to be measuring. This is when we understood the value of a clear, measurable specification, just as in the readability versus 14 pt font case we referred to earlier. We needed to shift our focus, from testing everything we thought of, to improving the specification so that it was measurable, and then defining the test that would substantiate the verification. Once defined, we had to execute the test and provide the objective evidence that it had been satisfied. If there were exceptions to the result, we needed to clearly understand and explain the “If there were exceptions to the result, reasons for the exceptions. We needed to assess whether we needed a different we needed to clearly understand and test, or whether the exception was acceptable, based on our specifications and our design. All of this needed to be clearly documented and approved, before explain the reasons for the exceptions.” we could consider the design to be verified.

208 | Developing Consumer and IT Platforms in a Regulated Medical Device Environment Intel® Technology Journal | Volume 13, Issue 3, 2009

Defined User Needs

Device and labeling must be easily read by visually impared user

Specification

PHS interfaces shall be visually equivalent to 14pt Verdana

Output (Component/Product)

Screens and Labeling materials in 14pt font

Figure 10: Validation Source: Digital Health Group, Intel Corporation

We also had adopt a definition of validation that required a user (or someone “The question we had to ask ourselves who represented our target user group) to sit down with the device and use it. The question we had to ask ourselves was “does our finished device correctly was ‘does our finished device correctly satisfy the intended use?” satisfy the intended use?’ ” We had to accept that our users were more concerned about the usability of the user interface than they were about disk access speed. Our validation efforts needed to be focused on the real need, not our comfort zone.

Were the instructions in the manuals sufficient to enable the user to actually understand and use the device? Or, had we assumed too much, due to our own intimate familiarity with the device? This was the true test of all our research, design, and development: did we really understand our users and had we met their needs? We needed to go beyond simulation and get true feedback: can our device be effective with our users? Validation testing is about closing the loop. This kind of validation process was substantially more involved than what we had initially planned and scheduled.

Complete Traceability — No Requirement Left Behind Design traceability, to the FDA, means that inputs correlate to user needs “Traceability means that objective and risks. Outputs match inputs, and test cases support the V&V of a device. Perhaps most importantly, traceability means that objective evidence exists evidence exists for all of these linkages, for all of these linkages, and it has been appropriately created or obtained and properly reviewed and dispositioned. We have discussed traceability in and it has been appropriately created previous sections — how it starts with clear, unambiguous and non-conflicting or obtained and properly reviewed and requirements. We addressed the need to define our outputs clearly, the importance of writing a good specification, and of crafting appropriate plans dispositioned.” for the execution and development of the design. Even with all the work we had done on inputs and outputs, we really did not completely understand the concept of traceability until we needed to document it. We found that

Developing Consumer and IT Platforms in a Regulated Medical Device Environment | 209 Intel® Technology Journal | Volume 13, Issue 3, 2009

even with all our testing, there were still some things that had been missed, things we needed to resolve. We discovered that typos can stymie the best of teams. We discovered that requirements or specifications that we thought had been changed were not changed in our system of record. We discovered that any manual processes for tracking were always subject to human error — not exactly an earth-shattering revelation — but one that cost us time and resources to correct.

Objective Evidence Standard 21 CFR 820, Subpart D pertains to Document Controls. “Each manufacturer shall establish and maintain procedures to control all documents that are required by this part.” It speaks to document approval and distribution, as well as to document changes.

“We had taken the time to establish a Our challenges in this area were twofold: implementing the right documentation and the right archival process. Our first task was to ask controlled document repository, which ourselves whether we were creating the right documentation. We were had been validated for its intended certainly creating a lot of documentation, but getting the teams aligned on the appropriate documentation was challenging. Where was the objective use.” evidence that we had the right documentation? E-mails, meeting minutes, or off-line conversations were not acceptable: these could not be presented to an auditor. Helping people to understand “if it’s not documented, correctly, it didn’t happen” was a big shift in our engineers’ thinking. Again, our templates and team training sessions (some formal, some ad hoc) were the key to getting this message out. Our second challenge was the archival process. Were we following the right procedures when seeking approval of a document and were we effectively archiving it? We had taken the time to establish a controlled document repository, which had been validated for its intended use (software tool validation is a different topic, and is not covered in this article). However, we did not have an electronic signature tool that had been specifically validated for that purpose: we were initially able to use hard-copy signatures only. This was a real mind shift for a geographically diverse business group that thrives on e-mails. We learned how to delegate to administrative assistants the tasks of obtaining physical signatures and sending faxes. We learned that any employee, no matter what his or her title, can always carry a few signed documents back to the document control office for archiving. We learned that no one goes out of town without putting a signature delegation authority in place, so as not to impede our progress. We learned most of these best-known-methods (BKMs) the hard way, as we balanced project deadlines with QSR imperatives. We also learned we really wanted electronic signatures and that we needed to make the investment in money, time, and people to establish a validated system that supported this capability. “We learned most of these best-known- methods (BKMs) the hard way, as we balanced project deadlines with QSR imperatives.”

210 | Developing Consumer and IT Platforms in a Regulated Medical Device Environment Intel® Technology Journal | Volume 13, Issue 3, 2009

Adapting to New Ways of Working From a software engineering perspective, there are many different “The medical device industry has methodologies for designing and developing new software platforms. Under regulatory requirements, the medical device industry has operated for the most operated for the most part by using part by using a waterfall or modified waterfall development model — also a waterfall or modified waterfall known as the Big Design Up Front model. development model — also known as In a waterfall model, things happen in a cascaded and sequential fashion. Progress may be slower as finish to start relationships pace the progress from the Big Design Up Front model.” phase release to phase release. Modified models allow for sequential overlap (Sashimi model) between design and implementation. Sashimi models add some additional risk because they may require further levels of branch prediction with regard to which phases of the design can be coded first and how forgiving the program plan is of mid-stream changes downstream.

Regardless of the development model, regulated design and development requires clear design inputs (product requirements, product architecture, engineering specifications, lower-level subsystem design, functional engineering specifications, and V&V plans) which lead to a series of formal design output reviews, in order to provide actual design requirements for downstream code development. Code development works from a planned release process through “Completely documenting all the a series of software milestone releases and formal V&V testing gates, until a objective evidence against 100 percent final Release Candidate can be produced. Completely documenting all the objective evidence against 100 percent of the originating design inputs is a of the originating design inputs is complex and resource-consuming process. a complex and resource-consuming Our software development process in the Digital Health Group improved process.” dramatically when we developed a clear understanding of what is necessary for the initial Design Input review:

1. A clear understanding of the product or platform architecture, roles, actors, constraints, assumptions, risks, hazards, and security threat matrices. 2. A clear set of product-level requirements, defined and approved. 3. A thorough, completed analysis of each of the subsystem designs and including the unit test, code inspection, and functional engineering specification needed to properly verify the product. 4. All design artifacts, decomposed in such a way that the V&V teams can create viable test plans to present the clear and objective test evidence necessary to support the original product requirements. If any link in this chain breaks, then you run the risk of a process loop. Some process loops are worse than others, and in the first development cycles of the Intel Health Guide, we had several. “If any link in this chain breaks, then you run the risk of a process loop.”

Developing Consumer and IT Platforms in a Regulated Medical Device Environment | 211 Intel® Technology Journal | Volume 13, Issue 3, 2009

Early in the V&V of the Alpha and Beta releases, it became clear that we had missed requirements. We also had many interpretive requirements, in which testers felt that the application should or should not function a specific way, regardless of how we had designed it. Yet there were still other cases where we misinterpreted the intent of a requirement.

“We approached each setback with All of these program interrupts caused a series of redesign efforts, as well as re-planning and re-engineering of the platform software subsystems and UI that final goal in mind, which was to that produced small program loops and resets. The Software Engineering produce a high-quality product that organization clearly understood the end-goal, and we approached each setback with that final goal in mind, which was to produce a high-quality product that was reliable, safe, and in compliance was reliable, safe, and in compliance with all regulations. with all regulations.” We left that initial platform development experience with a new attitude towards our own ability to control design specifications and a commitment to drive deeper clarity into the processes that fed into our work.

Lessons Learned and New Directions Fortunately, engineers are not generally gluttons for punishment. After the initial release of version 1.0 of the Intel Health Guide, we stepped back and analyzed several elements of the program that we believed we could directly control and could significantly improve. This would inevitably produce higher- quality software in a more controlled and predictable fashion. Our analysis of the process also drove us in several additional directions.

“We stepped back and analyzed several First, we had to further define the methods by which we would operate as a software development organization: elements of the program that we •• Formalize our software development plan structures. believed we could directly control and •• Develop our software subsystem designs and output processes. could significantly improve.” •• Perform more detailed task-level, peer code reviews. •• Implement more rigorous code version and release control. •• Follow secure coding standards. •• Have design effort scheduling that is risk and developer loaded by task, in order to produce better schedule quality and predictability. •• Rigorously adhere to the quality management system SOPs and SOIs that control the software engineering process. Second, we needed to separate the higher-level product requirements from the functional engineering specifications of the platform subsystems. The PRD, as stated earlier, is a component of the “What are we designing?” section of the design control process. Along with the initial hazard analysis and the Platform Architecture Document, these three pillars define what the platform will be.

212 | Developing Consumer and IT Platforms in a Regulated Medical Device Environment Intel® Technology Journal | Volume 13, Issue 3, 2009

Below the what level of design control, we realized that we needed to define “We needed to define the subsystem the subsystem and functional engineering specification outputs in such a way that no part of the design and development of the platform was left open to and functional engineering interpretation. This next-level design output would drive absolute traceability specification outputs in such a and transparency into the how portion of the design control process. In order to do this, we asked that each Platform Architecture Document specify the way that no part of the design and subsystems in the platform design which would need to be thoroughly vetted as individual lower-level design artifacts. development of the platform was left

As a result of this change in the PLC, the Software Engineering organization open to interpretation.” created both a Software Subsystem Design (SSD) artifact and a Systems Engineering Specification (SES) artifact.

Platform architecture for client-server enterprise applications, such as the Intel HCMS, might comprise a dozen or more major subsystems at the architecture and platform design level. These might be represented as a Security SSD, a Manageability SSD, a UI SSD, a Data Modeling SSD, a Peripheral Connectivity SSD, a Communications SSD, or an Infrastructure SSD.

Each SSD is a specific how implementation design document for a given subsystem. Not only does it deal with the next-layer design definitions, it also represents the constraints, assumptions, standards, and implementations of that subsystem. The SSD addresses the overall subsystem design: it defines library interfaces, code classes, algorithms, service methods, attributes and implementations, SOUP, component interfaces, and other functional elements of the subsystem design.

A table in each of the SSD documents specifies both the functional and non- “Each SSD is a specifichow functional engineering specifications for that subsystem. It also defines the test methods for each specification; whether they will be covered by unit test cases, implementation design document for a code inspections, or functional tests. given subsystem.” Any functional engineering specification requiring some level of functional test coverage is automatically promoted to the SES, once the subsystem design is completed and approved. The SES provides for a traceability link between the PRD, the Initial Hazard Analysis, and the lower-level subsystem design itself. It also forms the basis for the Verification Test Plan portion of the overall V&V plan.

The SES is of course iterative through the design and development phase of the PLC. Different subsystem designs are completed at different stages prior to the final design output review.

During that design and development process, the SES matures — along with the other design artifacts — providing the verification test engineers with a clear understanding of the design outputs they will use to build their automated and/or session-based tests. It is equally important for the verification test engineers to work closely with the design review teams for each subsystem design. Each SSD becomes an artifact that Test Engineering can refer back to for the development and execution of the final test plans for “During that design and development V&V. process, the SES matures.”

Developing Consumer and IT Platforms in a Regulated Medical Device Environment | 213 Intel® Technology Journal | Volume 13, Issue 3, 2009

Improving Quality, Predictability, and Design Assurance “We often heard our engineers ask, In version 1.0 of the Intel Healthcare Management Suite’s development, we often heard our engineers ask, “When is done, done?” This question came out ‘When is done, done?’ ” of many frustrations, as we collectively ground down through several internal release cycles and subsequent field trials of the product.

Engineering reviews of test data output with the platform V&V test teams exposed individual tester interpretations of poorly worded system requirements which in turn led to escalated Product Design and Systems Validation review meetings. In addition, implied requirements and even edge boundary case functional tests were being performed in formal V&V sessions, because of a lack of understanding between the development and test engineering teams. In the absence of a clear requirement, the test teams continued to push the envelope until the product ceased to function, and then it became a product defect. It became increasingly clear that we could not operate as an effective and efficient team if we did not have better control over all aspects of our “All of this led to a clearer definition design, development, and V&V processes. of the intended development work After the 1.0 release, we made a number of significant improvements in our ability to gather, define, and prioritize requirements and specifications in the by each engineer and a very clear hope that this would improve the overall design methodology, test framework expectation of what was to be tested implementation, test integration, and predictive control of the testing process. All of this led to a clearer definition of the intended development work by each and how.” engineer and a very clear expectation of what was to be tested and how.

Other improvements happened in the areas of release milestone task definition (what elements were to be developed and when). This drove a better understanding of development across the organization, including which requirements and which known defects would be satisfied in each release phase. It is essential that organizations gain agreement upfront and align to formal testing plans under a phased development release model, so that anyone testing any release phase of a platform will clearly understand what to expect at that point in the code development process.

Between better design definition and cross-organizational design reviews, more consistent code reviews, clearer software project plans, and release management phase controls, we became much better at predicting the actual schedule of each of the development phase gates and the overall quality of the software we delivered at each milestone.

Another major improvement area for us was the way we performed V&V testing cycles at certain release milestones. In version 1.0, every software release “We became much better at predicting phase immediately entered a formal test cycle. This caused lots of internal congestion and confusion, because testers who were running formal session- the actual schedule of each of the based testing on incomplete release code (Alpha for instance) expected to see development phase gates.” feature implementations and therefore logged lots of application defects.

214 | Developing Consumer and IT Platforms in a Regulated Medical Device Environment Intel® Technology Journal | Volume 13, Issue 3, 2009

In redefining our test processes for the next version of the software, we defined two additional test cycles in the Alpha and Beta phase of development known as a Developer Test (DT) cycle and a Characterization Test (CT) cycle. Developer Tests are run by developer peers within the team, checking each other’s code in an informal, integrated release runtime environment. The developers run the verification test cases on their own code as well, but in an informal manner, in order to find defects, issues, or implementation problems with the code. After the software engineers run the DT and complete “We defined two additional test their cleanup of known defects, the Alpha release (for example) is handed to cycles in the Alpha and Beta phase of Verification, Validation, Technical Marketing, and other teams in order to run CTs against that release. development.”

While the CTs are being run, the developers are free to continue onto the next development phase. As defects arise from the CTs, they are addressed as a part of the Beta development phase (in this example). The benefit here is that we make multiple informal test passes through the code-base during its development lifecycle so that by the time we reach actual Release Candidate status and perform formal V&V under our plan, all of the groups involved in the development of the platform clearly understand the expected features and the expected test-plan execution suite. As a result, this produces a much higher confidence level in the expected outcomes.

All of these steps lead to fewer Release Candidate builds. Our development “The quality of our software improved schedules improved dramatically and the quality of our software improved four-fold over the previous process.” four-fold over the previous process, with regard to the number and severity of the defects found.

Like building a house, the foundational elements are the most important thing. As the work progresses, checks and balances between the architect, the builder, various subcontractors, and building inspectors ensure a quality outcome.

Software development is no different. If you understand the elements of each phase of the design and review and plan for them accordingly, your developers will always know that they have successfully completed the product requirements with a high degree of design quality and assurance and will in fact know when done is actually done. “Your developers will always know that they have successfully completed the product requirements with a high degree of design quality.”

Developing Consumer and IT Platforms in a Regulated Medical Device Environment | 215 Intel® Technology Journal | Volume 13, Issue 3, 2009

Conclusion “Inflection points offer promises as well Implementing a regulated product development process was to say the least, challenging. Early assumptions and confidence in our own PLC, technical and as threats.” manufacturing capabilities gave way to uncertainty and frustration. Andrew S Grove, the former CEO and Chairman of Intel Corporation, wrote a book on the subject of organizational tension and reconstruction entitled Only the Paranoid Survive: How to Exploit the Crisis Points That Challenge Every Company, in which he discusses the concept that …“inflection points offer promises as well as threats (to your business survival). It is at such times of fundamental change that the cliché adapt or die takes on its true meaning.”

The Digital Health Group had a deep commitment to product quality from its inception. During our first product development cycle we pushed ahead, resetting our process several times, focusing on doing the right things right, and in the end, producing a high-quality, safe, and effective product that adhered to the regulatory requirements set forth by the CFR and other regulatory bodies. “Designing processes to meet the It became clear to all of us that we had no choice but to adapt our development requirements of the healthcare processes if we were to be an effective medical device manufacturer. Designing processes to meet the requirements of the healthcare industry presented a set industry presented a set of new of new challenges; we had to learn to operate within a much more constrained challenges.” and formalized development process, while still developing innovative products and working efficiently.

Before beginning our next product development cycle, we made a number of significant decisions and structural changes to our PLC, our phased development methodologies, our SOPs, SOIs, and development artifacts. We completely restructured the way we capture, measure, and implement requirements and specifications. We also integrated other structural development process improvements, such as Security Design and Development, Risk Management Design Assurance, Labeling and Documentation Workflows into the overall process, instead of making them parallel breakpoint activities.

We continue to practice user-centered design as we develop next generation products for home health. Our design and development processes have converged during our first development cycles to fully meet regulatory requirements. We are now adding increased focus on predictability and efficiency in our design and development activities. Our constant goal is the delivery of products that help to improve the health and quality of life of the people who use them, while lowering cost and improving access to healthcare. “We are now adding increased focus on predictability and efficiency in our design and development activities.”

216 | Developing Consumer and IT Platforms in a Regulated Medical Device Environment Intel® Technology Journal | Volume 13, Issue 3, 2009

References [1] The Intel Corporate Product Life Cycle (PLC), Intel Corporate Program Management office, Intel Corporation

[2] The World Health Organization. “How can chronic disease management programmes operate across care settings and providers?” In WHO European Ministerial Conference on Health Systems, Tallinn, Estonia, 2008 Debbie Singh. Available at www.euro.who.int

[3] 21 CFR Part 820–Title 21 Food and Drugs, Chapter 1, Food and Drug Administration, CFR (Code of Federal Regulations), Department of Health and Human Services, Subchapter H Medical Devices, Part 820, Quality System Regulation. Search for CFR Part 820. Available at www.accessdata.fda.gov

Authors’ Biographies Barbara Fullmer is Director of Product Development Controls at Intel and has been with company for eight years. She has been working in the semi-conductor industry for over 21 years. Barbara has spent 16 years in various planning functions, from wafer fab planning to multi-national capacity commits for fab, wafer sort, assembly, and test. Prior to joining the Digital Health Group in 2005, Barbara worked in the Technology and Manufacturing Group as the Strategic Capacity Planning Demand Manager. She has received various awards at Intel for process improvement, including re-engineering aspects of the long-range planning process. Prior to joining Intel, Barbara worked at Affymetrix, a bio-medical device manufacturer, as the Manufacturing Planning Manager. Her e-mail is barbara.h.fullmer at intel. com.

Alan Boucher is the Director for Software Architecture and Engineering in the Digital Health Group. He is a 21-year veteran of Intel where he has focused on the healthcare market since 1999. Throughout his career, Alan has led numerous software engineering and architecture teams for various Intel platform groups and has managed field systems and network engineering teams worldwide for more than half his career. He received the Intel Achievement Award (IAA) twice (2002 and 2005) for leading the Technical Solution Blueprint team and for technical contributions to the Mobilized Software Initiative. His e-mail is alan.boucher at intel.com.

Developing Consumer and IT Platforms in a Regulated Medical Device Environment | 217 In te l ®

tec hnology jo ur n a l | E N AB L I NG h ea l Intel® Technology Journal

SEPTEMBER 2009

Enabling Healthcare in the Home t h care i n t h e ho me

More information, including current and past issues of Intel Technology Journal, can be found at: http://developer.intel.com/technology/itj/index.htm vol 13 | issue 03 SEPTEMBER 2009

ISBN 978-1-934053-23-2

9 781934 053232 7735858 21143

$49.95 US

ITJ9-3_Cover_BFC_39spn_090409.indd 1 9/4/09 8:28:45 AM