Unlocking Clinical Meaning from Patient-Generated Health Data


By Simon MacGibbon, CEO & Co-Founder, Myia Labs

It was supposed to be the revolution in healthcare, and a no-brainer.

The potential of patient-generated health data (PGHD) to help clinicians anticipate, diagnose, and treat disease more effectively has been percolating for several years now. However, despite a proliferation in devices and a compelling promise, PGHD has not made the leap across the consumer-clinician divide. Its intuitive potential remains unfulfilled in healthcare, despite an abundance of inputs. Consider:

  • The ‘quantified self’ movement began in 2007;

  • 76 million Fitbits have been sold since 2009, and by some estimates, 46 million Apple Watches since April 2015;

  • Gartner forecasts that 153 million smart watches, wristbands and other “fitness monitors” will be shipped in 2018.

Taken together, these devices generate an astronomical amount of healthcare data. In fact, the International Data Corporation (IDC) estimates that all sources combined will produce over 2,300 exabytes (one exabyte = one billion gigabytes) in 2020 (including Electronic Medical Record (EMR) data). While that may seem to be an incomprehensible number, it amounts to an astonishing 48% annual growth rate since 2013. Some experts believe that by 2020, we will be generating the majority of clinically-relevant data outside of clinical settings.

Our faith in the potential of real-world patient data to change healthcare was not just Silicon Valley exuberance. In 2013, the Office of the National Coordinator for Health Information Technology estimated that remote patient monitoring would save an impressive $36 billion globally through 2018. In the absence of good facts, we cannot say whether that projection became reality, but we would pass the ‘red face test’ in asserting that the impact has been be a lot less.

Either way, while expectations of fostering a more preventative approach to healthcare may be ill-timed, they are not misplaced. PGHD from the real-world, in its modern form, will be an unlock that drives the prevention evolution. The Centers for Medicare and Medicaid Services (CMS) are also signaling their belief in capturing real-world patient data by releasing several new reimbursement codes related to remote physiological monitoring, effective in January 2019.

It will only take hold if collected, synthesized, and surfaced to clinicians correctly.

Following the recent Apple Watch 4 announcement, a thoughtful piece entitled “Should Your Watch Monitor Your Heart?” appeared in “The Atlantic.” The authors of the article, Alexis Madrigal and Lolade Fadulu, referred to a 2016 review of mobile-health technology in the Journal of the American Heart Association. The concluding statement lucidly defines the real challenge:

“If these previously uncollected streams of continuous data can be translated into actionable information and presented at meaningful decision points, patients and their providers may be better able to achieve health goals and manage chronic conditions.”


A Wealth of New Data, Waiting to Provide Value

Within the four walls of an outpatient clinic, people with chronic conditions typically get 12–15 minutes with a doctor who does his or her best to piece together a picture of the patient’s health status. There are obvious flaws with this reality. For example, some of the decision-driving biometric measurements collected in the clinical setting are prone to erroneous results. “White coat hypertension”, a well-published example, whereby blood pressure presents as abnormally high for many people — a simple result of anxiety related to being in and/or getting to the clinic. Other measures such as resting heart rate are inherently subject to contextual contamination; and some measures, despite their relevance for some conditions, are not collected at all in the clinical setting, e.g., respiratory rate and objective sleep position in Heart Failure patients.

People with chronic conditions spend somewhere between 99.97–99.99% of their time living their lives as they do in the non-clinical world. Regardless of the skill and passion of clinicians, 12–15 mins every one to three months presents an insurmountable challenge to inquire and understand the contours of the patient’s situation, and not nearly enough to have a fulfilling conversation — especially with stoic patients unwilling to reveal the reality of their symptoms. Our doctors and nurses simply don’t have enough time to ask for and take in all the relevant data.

Outside the clinical setting, we humans live lives rich with nuance. The data that our various connected devices, such as Fitbits and Apple Watches, can collect, transmit, and monitor is likewise stunning in its diversity and sheer size, and it can tell our clinicians a lot about us that cannot be ascertained in a short visit:

  • Biometrics, e.g., heart rate, heart rate variability, blood pressure, respiratory rate, O2 saturation, weight, temperature, galvanic skin response, collected blood tests

  • Activity and physical markers, e.g., steps walked, distance from home, mobility, gait

  • Behavioral markers, e.g., sleep patterns, activities of daily life, food consumption, medication adherence

  • Engagement, e.g., responsiveness to outreach, social engagement, responsiveness to messaging

  • Environmental context, e.g., access to transportation, proximity to nearest pharmacy or hospital, mean household income, weather, temperature, air quality

We can even think of PGHD also extending to data collected from implantable devices for patients across the care spectrum from heart disease to orthopedics.

It’s not just about “prediction”

Making this data available to clinicians is going to have a significant payoff. We have spent countless hours over the last two years observing patient-clinician interactions, interviewing patients in their homes, researching healthcare’s outcome and economic pain-points, and iteratively prototyping our solution with practicing doctors and nurses. PGHD, collected and synthesized in the right way, can (and will) have a material positive impact by facilitating more “individualized” and timely care. For example:

  • Facilitate getting patients to their most effective medication regimens and doses, adhering to professional societies’ Guideline-Directed Medical Therapy (GDMT); A recent peer-reviewed research report on Heart Failure medication management found that only 1% of patients are on the expected combination of medications and respective dosage. Today, cardiovascular physicians express frustration at flying half-blind when trying to titrate Heart Failure patients’ medication without seeing the physiological impact

  • Equip clinicians to stage preventative and rescue interventions with at-risk chronic and post-acute patients. This includes spotting longitudinal health status changes that catch patients who should be considered for more advanced therapies, before the “point of no return.”

  • Facilitate richer patient-clinician dialogue and care-planning during their scarce time together, which may be in-person or virtual. Most patients we have spoken with express hope at the thought of having a fact-based voice to explain how they have been feeling.

The paradigm shift will pay-off in terms of improved patient confidence and quality-of-life, fewer unplanned hospital visits, and perhaps even re-energized clinicians at an all-time risk of burnout.

Why no impact yet? The missing links:

Passive and Clinically Relevant Data

It is no secret that wearables suffer from significant user decay, e.g., according to some insiders, only 30% of Fitbit users are still active after their first 30 days. Reasons for this user loss include obvious frictions such as battery life and garden variety behavior change challenges that present themselves when trying to get people to adopt something new. Needless to say, this kind of data loss will not ‘cut it’ if meaningful healthcare benefits are the goal.

Even if doctors are asking their patients to contribute their data, a PGHD ‘system’ has to include data collection methods that are largely passive. For example, some bed sensors are capable of generating reliable physiological data such as heart rate and respiratory rate, with little-to-no patient effort once installed. “Set-and-forget” should be the gold standard where possible. However, some data, such as blood pressure, does need to be “actively” collected ‘today’ as continuous and passive sensors are still in development. …but that will soon change.

Naturally, the location of measurement on the body and the quality of the device sensor itself are also important considerations, e.g., a ring-based PPG sensor that taps into near-direct arterial flow in the finger, will produce a higher-quality pulse waveform than a PPG sensor measuring from the top of the wrist via a smartwatch. For now, let us leave aside elaborate perspectives on this and simply appreciate that not all ‘wearables’ are created equal.

It should go without saying that the combination of parameters must be both clinically relevant as well as comprehensive to the patient’s condition. Currently, there are remote monitoring solutions deployed in clinical settings today that measure the same things for multiple patient types and miss critical physiological parameters — this needs to change.

Context and Meaning

Capturing the missing 99.99% of patient data brings with it both opportunity and challenge. With greater volume we can better model the noise and interrogate only the high-confidence regions of the most relevant signals. At the same time, what may be considered ‘noise’ at first glance, may prove to provide valuable insights in the right context of longitudinal modeling.

Needless to say, understanding the context in which data are acquired is essential to creating a foundation of trust in the value of the information being communicated. A “resting heart rate” of 89 has little meaning without understanding the context surrounding the measurement — was the patient truly at rest, and if so, what was the time of day? Were they active prior to being at rest, and if so, how quickly did the heart rate return to its resting baseline? …you get the picture.

In essence, having a robust understanding of what apples-to-apples normallooks like for a patient is an essential foundation before applying machine learning — even sophisticated deep learning models are not immune to the “junk-in, junk-out” universal rule. This is to say, the power of machine learning (and other AI techniques) can be unleashed only after the hard and unglamorous work described above is done.

Bringing meaning to PGHD is a truly wonderful application of machine learning in healthcare — one that does not threaten people’s jobs, but rather, augments the powers of frontline clinicians with levels of patient insight that were previously inaccessible. Machine learning is well suited to the inhuman task of making sense of myriad data inputs and surfacing what really matters at a population, segment, and individual level. Without advanced analytics, there are no answers to the “so what?” questions that are often surfaced from multivariate PGHD, beyond the intrinsic value of getting access to discrete measures that clinicians know and can act upon today.

The Clinical Last Mile

Needless to say, the best data and algorithms in the world will not count for anything if clinicians do not value and take action on them. This ability, to get clinicians to actually adopt PGHD in everyday practice, at scale, is the summation of the concept of “closing the last clinical mile” and enabling better care delivery, to everyone.

The recent New Yorker article by Dr. Atul Gawande titled “Why Doctors Hate Their Computers” has a particularly vivid articulation of this challenge. A “massive monster of incomprehensibility” is how one physician described their EHR system. This unfortunate healthcare tech history cannot be ignored. The bar for successfully putting PGHD in the hands of clinicians is unreasonably high because of it. Drawing on some of the realities elucidated in Dr. Gawande’s article, let us consider the principles that new technology ought to embody if it is to have a chance at closing the clinical last mile — a system would:

  • Effectively give time back to doctors and nurses. Specifically, it would have to shift the overall ratio of screen time to patient time (about 2:1 today)

  • Do work behind the scenes, perhaps triaging patient symptom information that was previously elusive given the sheer constraints of clinician time

  • Produce something that clinicians and patients can jointly review. It would need to add value to their interaction, not detract from it (50:50 computer v. patient in a visit today)

  • Draw attention to what matters, enabling clinicians to learn something from it quickly, without the need to hunt for and piece together valuable insights

Ultimately, the challenge is to introduce technology that actually fosters and enhances the “human connection” between clinician and patient. It would allow clinicians to be better clinicians and people to be better patients.

Easier said than done, perhaps. There are other complications beyond the EHR technology hangover. For example, much of this new PGHD, even if revealed by machine learning to be relevant, cannot be acted on by clinicians if their medical training and existing guidelines do not encompass it. Sleep patterns are a good example. An increase in sleep disruption and respiratory rate may be strong predictive markers of patient decompensation, but there is not a playbook for how to titrate medication based directly on them.

We can use advanced computational techniques to make sense of multivariate real-world data, but the outputs ought not be a ‘black box’ should clinicians want to understand the underlying drivers of a recommendation or risk assessment.

Despite the unreasonably high bar and complexity, we are energized by clinicians’ curiosity and belief that there is a better way. We are working with leading providers and clinicians who are excited to uncover and apply new markers from PGHD, understand the potential for machine learning to surface actionable insights they do not have today, and are dedicating their time and expertise to shaping what the platform looks like for them and how it integrates into daily workflow to realize a new standard of patient care. They realize it is a start of a journey and not something that will be omniscient overnight.

This is what we live for at Myia.

Putting Insight into Action: Myia

Myia is an intelligent health monitoring platform. We capture and transform streams of personal, vital health data into timely and actionable clinical insights.

We started Myia with the goal of bringing all of the above to life: creating the most frictionless experience for patients, bringing extraordinary context and meaning to patient data, and closing the clinical last mile with a platform that enables clinicians to deliver even higher standards of patient care. To do this, we address the way the data is captured, analyzed, and packaged.

Being device-agnostic, we partner with device companies to curate the most contextually-relevant PGHD from the highest-fidelity and low compliance sources and inputs, then with the help of machine intelligence and AI tools, transform it into information that clinicians can actually employ. We are supporting a shift to prevention and more individualized treatment plans for patients living in the real world, especially those with chronic conditions, like heart failure.

To do this, we are supported by a dream-team of clinical partners, academic teams, investors, and advisors. For example, as an early investor and advisor, the American College of Cardiology also believes that PGHD can catalyze a more preventative approach to care, and that technology should be designed in a more clinician-centric way. Our philosophy and approach from the beginning has been built on collaboration — this is not an opportunity that can be realized outside-in. It goes without saying that we have a wonderful cross-disciplinary team with proven success in the fields of design, engineering, product, advanced analytics, and medicine.

The scale of this challenge seems achievable when working with such thoughtful and committed partners, investors, and team members. Advocating every day for the goals of patients and their clinical care teams makes it thrilling and purposeful.

For more information on how we are addressing this, please visit: Myiahealth.com.

sam rawstorne