The Patient Equation

Home > Other > The Patient Equation > Page 6
The Patient Equation Page 6

by Glen de Vries


  However, in this study, the incorporation of the Fitbit device itself—which measures our activity and feeds it back to us, numerically—adds a fascinating cognitive dimension to consider. Can stimulating patient motivation to exercise through a wearable device reliably produce a cognitive effect that cascades down to cause a significant change in the survival probability given an otherwise‐fatal diagnosis? There is significant work to be done, but there is reason to think the logic may hold up.

  An entirely different kind of mind‐body connection emerges when we look at the placebo effect. In the world of research, we need to control for this very carefully. It isn't just that people believe that they feel better when they receive a sugar pill instead of an active drug. They actually do feel better. This makes the placebo effect confounding to researchers. It isn't just an effect on someone's mental state; it is a true physiological effect. Endocrines are released, and motivations are introduced. Patient engagement and patient involvement in treatment can have both psychological and actual physiological consequences that we need to make sure research studies account for.

  Given all of that, the placebo effect is often looked at as a problem right now in clinical trials, something that gets in the way. But maybe it can be harnessed for benefit instead. That's not to suggest by any means that we treat patients with ineffective treatments. However, we need to remember that it can be a means of manipulating our health as a force for good, not just a statistical complication. We need to figure out how to use that force in a positive way, to optimize improved outcomes.

  In this respect—and, truly, in a lot of respects—we can learn from companies like Amazon. While they're not trying to optimize health outcomes, they are trying to optimize consumer habits, and they're doing so in a way that fits very much with the idea of patient equations. They're doing A/B testing, which is basically randomized research in the same mode as clinical trials. They're showing two different groups of people a different set of recommendations and figuring out which works better. The algorithm learns, and then it keeps refining to get the best outcome.

  What's interesting—and may or may not turn out to be relevant here—is that Amazon and companies like it have figured out that their predictive engines suffer if they're too good. People don't want their recommendations to be perfect—they get put off, creeped out, and they buy less. Imagine a dial that can be set from 1 to 10 with regard to how good recommendations are, based on previous purchases and browsing history. They've collected data to show that 10 is too high, and maybe people are most comfortable at an 8 (in reality, I'm sure it's much more complicated and whatever “consumer equations” e‐commerce companies have created are highly guarded trade secrets). Where this may have implications for us is that maybe it turns out that people's health care isn't fully optimized when we show them all the information we have about a disease or condition. Maybe people do better when they have slightly less information or worse recommendations—when they think things are more random than they actually are, perhaps, or when they believe they have more control.

  The “why” doesn't necessarily matter, but if the data were to tell us that people have better outcomes when we show them 80% of the relevant information we have rather than 100%, then maybe that is what we ought to be doing. The FDA certainly seemed to think that there were potential negative consequences from the genetic diagnostics that the company 23andMe was making broadly available to consumers.10 Too much information—particularly misinterpretable information—can lead to unnecessary overtreatment and negative effects on quality of life and medical outcomes.

  There is a company called Mindstrong that is attempting to translate what we do on our smartphones to actual data about our health, tracking how quickly we scroll, how frequently we check our phones at night, what we post and who we call, and seeing how that data correlates with disease symptoms.11 As Dr. Steve Steinhubl, director of digital medicine at the Scripps Translational Science Institute in San Diego, quoted in the New York Times, said, “if a sociable person suddenly stopped texting friends…it might indicate that he or she had become depressed…[or] it could mean that somebody's just going on a camping trip and has changed their normal behavior.”

  With this perspective, the world of cognitive data starts to resemble that of activity, whether we mean steps, territory, or a raw accelerometer stream digital biospecimen: we are collecting a rich, multilayered set of potential biomarkers, inputs to our patient equations. Within these biomarkers are useful measurements that, alone or more likely in combination, can predict, track, or act as proxies for disease progression.

  Better Measurements at Virtually No Cost

  I've already talked about the de minimis cost of storage for digital biospecimens. It's also increasingly the case that taking these measurements in the first place costs almost nothing. As I said in the Introduction, I often ask rooms full of people, “How many of you are wearing an activity tracker or medical device?” Inevitably not everyone in the room will raise their hand, but of course every smartphone in the room contains what is effectively a step counter, an altimeter, and a territory tracker all in one. Furthermore, with the proliferation of watches, rings, and a range of smart devices that we'll discuss, all around us every day, our phones can become an incredibly convenient hub for the acquisition of a huge range of behavioral, cognitive, and physiological data.

  Contrast the cost, and, more importantly, the quality, of the same medical concept—your cardiac fitness, as the example here—measured three different ways. You have perhaps been asked by your doctor, “How many flights of stairs can you climb before you are out of breath?” It's a pretty easy question for a health care practitioner to ask, and pretty easy to answer. But if you think of the number of times each day that question is asked in the world, and the number of total minutes it takes to get written down in a medical chart or recorded by a patient in a diary as part of a research project, the cost is certainly non‐zero. Moreover, our expectation of accuracy—based entirely on the biased recall of a party very much interested in presenting a certain picture to his or her doctor—should be low.

  At the other extreme, if we want higher accuracy, we can use a cardiac stress test and actually put patients, wired to the combination of an incline treadmill and an ECG, in a position that will give us an objective, quantitative measurement. However, this option comes at tremendous cost, thinking about time for the patient, the provider, and the cost of the device itself. It is also measuring just one moment in time. That measurement could change based on the amount and quality of sleep the patient got the night before, for instance. The result can also change over time—for example, with a dramatic shift in diet and exercise right after a clinic visit. This makes the single measure of limited use, and suggests the need for continued, expensive measurements made over time.

  As a third option, we can use the smartphones that are already, in the background of our patients' lives, doing the measuring we need. Already, most phones have an altimeter in addition to an accelerometer. More and more people are wearing a supplemental device checking heart rate. Instead of asking the question (or spending time and money on the stress test), we can get objective, quantitative measurements of how many flights of stairs we actually climbed and exactly how out of breath (or not) we were while climbing them. All day, every day.

  Consider another example: the 6‐minute walk test. Patients are timed as they walk back and forth down a hallway of a known length. The distance they travel over six minutes is their score, and it is used as an endpoint in clinical trials to demonstrate the therapeutic response for conditions that impede mobility, such as muscular dystrophy. The measurement works, in many cases. Patient response to treatments correlates with the test results. However, the measurement is also flawed, in some ways deeply.

  In the case of Duchenne muscular dystrophy, the 6‐minute walk test is often used as an endpoint for submission to the FDA. But there are some patients who cannot walk at all.12 Clearly the
test in these cases is meaningless, and in cases where the patient is simply having a “bad day”—and often these are children, given the life expectancy for the disease—the test may result in patients being excluded from a study simply because they weren't able to set an acceptable baseline against which progress could be shown later.

  This is an unfortunate consequence of the test being used for a single critical decision—whether a child will be allowed access to an experimental therapy, in this instance—as well as a proof point regarding the statistical weakness of point‐in‐time observations in conditions associated with intermittent symptoms. Consider a series of “good days” and “bad days.” The average score can be very different from the score at one point in time. Yet, right now, tests like the 6‐minute walk are used in a range of conditions, and the data is critical to get new drugs approved. With digital devices, measuring continuous activity in the background, we can do far better than the 6‐minute walk test and truly establish meaningful baselines for all patients, through good days and bad.

  Imagine a patient with a condition that flares up. There are days they get, for example, a migraine, and days they don't. In the calendar month depicted in Figure 2.1, days that are shaded are days when flares occurred. The patient was seen by a physician and interviewed on the last day of the month, marked by the double‐ringed circle. The days the patient can clearly remember (the previous week) are circled with one ring. Either at a glance, or by computing the frequency of problem days versus the total days measured, you can see that there are inconsistencies between the direct observation of the health care professional, the patient's recall, and the actual nature of the condition.

  Ten years from now, there won't be a clinical trial that isn't measuring potential biomarkers continuously rather than at discrete points in time, whenever they can. This can be with a biochemical or physical replacement for something done in clinic, or with the integration of digital biomarkers that cause no additional cost or pain—to the patient or anyone involved—using phones, watches, or other sensing infrastructure that is already there. This isn't simply a nice‐to‐have new set of tools. It's a paradigm change from needing physical interaction with a patient at a particular moment in time to being able to conduct trials remotely at a fraction of the cost with more reliable results. We are breaking the need for labor‐intensive, time‐intensive physical access to the body and allowing measurements at scale, irrespective of “good days” and “bad days,” instantly transmitted to the cloud and analyzed by a powerful algorithm.

  Figure 2.1 A month in the life of a patient

  From Hypothesis Confirmation to Hypothesis Generation

  Today, the world of medical data is about hypothesis confirmation—we suspect something, and we send off a test to see if we're right. In the future, these tools can move us toward hypothesis generation instead. The data tells us something first—and then we go off to investigate. Decide based on a moment versus predict based on stream—it's a huge difference, powered by digital data. We already know that not every moment in a patient's life is the same, and a prediction based on a larger sample of that life is going to be far more accurate.

  Once we have the entire stream of information, we need to take all of that data and translate it into medical evidence, useful conclusions about how we should proceed—in the doctor's office, in the lab, or in the marketplace.

  I mentioned Alzheimer's disease at the beginning of this chapter, and that perhaps knowing how many times someone checks the calendar on his phone each day is an indicator that a problem is starting to emerge. That may not be the measure. But the data may lead us somewhere useful. And what that will do is let us make smarter predictions and narrow the range of possibilities. Instead of not knowing if you're on a trend line toward needing intervention sometime in the future, maybe we can say there's a 75% chance you'll hit that point where we need to treat. Maybe we'll be able to tell 30% of the population that they're likely to die of something else long before dementia is an issue. Maybe these improved models of disease allow doctors to point their concern toward the right patients, payers to stop treating the wrong people for diseases they're never going to get, pharmaceutical companies to create better‐tailored drugs, and, ultimately, patients to have better outcomes.

  The smartest biotech entrepreneurs are thinking about these things right now—narrowing the cone of possibilities, improving our decisions about who to treat, for what, when, and how. And, of course, there are other entrepreneurs simply looking to capitalize on trends, employ catchy buzzwords, and release products out into the world that don't actually deliver on the promise of truly transformative patient equations. In the next chapter, we'll look across the landscape of products—technologies, drugs, devices, wearables—out in the marketplace today, and develop a toolkit for separating the hype from the worthwhile, learning from the failures, and, as we think about translating these ideas into practical business reality, figuring out what's most important to consider.

  Notes

  1. Kyle Strimbu and Jorge A. Tavel, “What Are Biomarkers?,” Current Opinion in HIV and AIDS 5, no. 6 (November 2010): 463–66, https://doi.org/10.1097/coh.0b013e32833ed177.

  2. Carl A. Olsson, Glen M. de Vries, Ralph Buttyan, and Aaron E. Katz, “Reverse Transcriptase‐Polymerase Chain Reaction Assays for Prostate Cancer,” Urologic Clinics of North America 24, no. 2 (May 1997): 367–78, https://doi.org/10.1016/s0094-0143(05)70383‐9.

  3. Michael Behar, “Proteomics Might Have Saved My Mother's Life. And It May Yet Save Mine.,” New York Times, November 15, 2018, https://www.nytimes.com/interactive/2018/11/15/magazine/tech-design-proteomics.html.

  4. Glen de Vries and Barbara Elashoff, Mobile health device and method for determining patient territory as a digital biomarker while preserving patient privacy. United States Patent 9,439,584, issued September 13, 2016.

  5. Jamie Heywood, interview for The Patient Equation, interview by Glen de Vries and Jeremy Blachman, February 27, 2017.

  6. Courtesy of Barbara Elashoff.

  7. Uzair Amir, “7 Times Apple Watch Saved Lives,” HackRead, April 27, 2019, https://www.hackread.com/7-times-apple-watch-saved-lives/.

  8. José R. Banegas et al., “Relationship between Clinic and Ambulatory Blood‐Pressure Measurements and Mortality,” New England Journal of Medicine 378, no. 16 (April 19, 2018): 1509–20, https://doi.org/10.1056/nejmoa1712231.

  9. Gillian Gresham et al., “Wearable Activity Monitors to Assess Performance Status and Predict Clinical Outcomes in Advanced Cancer Patients,” npj Digital Medicine 1, no. 1 (July 5, 2018), https://doi.org/10.1038/s41746-018-0032-6.

  10. “In Warning Letter, FDA Orders 23andMe to Stop Selling Saliva Kit,” GEN: Genetic Engineering and Biotechnology News, November 25, 2013, https://www.genengnews.com/news/in-warning-letter-fda-orders-23andme-to-stop-selling-saliva-kit/.

  11. Natasha Singer, “How Companies Scour Our Digital Lives for Clues to Our Health,” New York Times, February 25, 2018, https://www.nytimes.com/2018/02/25/technology/smartphones-mental-health.html.

  12. Craig M. McDonald et al., “The 6‐Minute Walk Test and Other Endpoints in Duchenne Muscular Dystrophy: Longitudinal Natural History Observations over 48 Weeks from a Multicenter Study,” Muscle & Nerve 48, no. 3 (2013): 343–56, https://doi.org/10.1002/mus.23902.

  3

  Fitbits, Smart Toilets, and a Bluetooth‐enabled Self‐driving ECG

  We looked in the previous chapter at the medical and life sciences implications of digital data. And yet, when we examine the world right now, it seems like there's a lot of hype about what we might call the nonclinical applications of sensor technology—or, perhaps even more to the point, the aspiring‐to‐be‐clinically‐useful‐but‐not‐quite‐there‐yet applications. Full of promise and potential—and clever marketing—but, at the end of the day, as surgeon and writer Dr. Atul Gawande told economist and author Tyler Cowen in an interview, “it's a dump of a ton of data that a clinician is supposed to use and
know how to integrate into practice, [but the information] hasn't been able to be used in such a way [that it's] actually demonstrating major improvements in people's outcomes.”1

  We're at this in‐between stage in the technology. The wearables market is undeniably huge—forecasted by one advisory firm to be over $34 billion by the time this book is in print2—and yet…we're still very much at the beginning of fulfilling the promise of multivariate patient equations changing the way we diagnose and treat most conditions.

  In addition, there have been high‐profile mistakes (Theranos, IBM's Watson, and Verily's glucose‐sensing contact lens, to name three that we'll talk about more in just a bit) that make it far too easy to write off the entire category of highly‐hyped “digital breakthroughs” in medicine. In this chapter, I want to explore what's out there, but I want to do it with a particular point of view hovering in the background: that is, looking at wearables for their own sake seriously misses the forest for the trees. No matter what the companies behind these devices say, the truth is that it takes much more than a wristband to power meaningful change.

  The true difference‐making is going to come from inside the life sciences industry rather than from the world of tech, when those wristbands become inputs to patient equations and part of feedback loops to continually hone them. The biggest impacts are going to involve using these new technologies not on their own but in concert with the kinds of things the life sciences industry is already doing and already understands—namely, rigorously scientific drug and device development and a legitimate understanding of medicine. The real work to make these technologies meaningful is expensive, and it's going to be pharmaceutical companies, medical device companies, and patient advocacy groups that have both the money to invest in building these equations and the long‐term business interest in taking advantage of the information that emerges. The smartest and most lasting business success in health care and life sciences will be when systems incorporate these technologies in meaningful and measurable ways with our foundational biology, physiology, cognition, and behavior.

 

‹ Prev