9. Evan A. Boyle, Yang I. Li, and Jonathan K. Pritchard, “An Expanded View of Complex Traits: From Polygenic to Omnigenic,” Cell 169, no. 7 (June 2017): 1177–86, https://doi.org/10.1016/j.cell.2017.05.038.
2
Inside the Equations
To think effectively about these new types of data and the value they can add to our disease models, we need to start by talking about “biomarkers” and “biospecimens.” As with so many things in science and medicine, they're just fancy words for relatively simple things. We usually think about biomarkers and biospecimens in the context of traditional medical measurements, our genes, and physical tissue samples.
A biomarker, according to the International Programme on Chemical Safety, led by the World Health Organization, is “any substance, structure, or process that can be measured in the body or its products and influence or predict the incidence of outcome or disease.”1 In slightly simpler terms, it's something we can measure that tells us something about a person's disease. If someone has cancer, for example, we can go in and take a sample of their tumor—a biopsy. The tissue sample from that biopsy—a biospecimen—will be used by pathologists and evaluated in laboratories. Microscopes and assays are used to examine the physical and biochemical properties of that tumor, looking for useful biomarkers. We can search for specific sequences of DNA, such as a particular mutation that allows us to predict how aggressive the cancer might be. Or we can examine the shape of the cells or the presence of estrogen or progesterone receptors. All of this measuring and evaluating is typically done in real time, in the best cases helping to decide on the right course of therapy or determining whether a patient is responding to a therapy that he or she is currently on.
But we can also take that biopsy and preserve it for future use—often by freezing it. Now, we have a biospecimen that researchers can thaw at some later point in time to look for biomarkers that may have been missed or misunderstood.
We shouldn't associate this idea of biomarkers and biospecimens only with a chunk of tissue. Biomarkers can be found in liquids just as easily. My first research project out of college happened to do precisely that kind of searching. Most men over 40 are (or at least they were, until recent changes to the standard guidelines) tested for PSA, or prostate‐specific antigen. The presence of this protein—how much of it is produced by the body, and then how much of it leaks out of a patient's prostate and into their bloodstream—is easy to measure, and can be useful in diagnosing and determining the progression of prostate cancer or other more benign prostate disorders. All it takes is a tube of blood, and the presence of the PSA protein—a biomarker—can be measured in a lab.
In the mid‐1990s, I was fortunate to get to work on another dimension of what we could measure in the blood of prostate cancer patients.2 Every cell in our body has largely the same DNA sequence (the exceptions being in cases of disease where DNA is mutated, and in red blood cells, which don't carry copies of our DNA at all). But as our tissues differentiate, and in the course of normal physiological function, various genes get turned on or off. In the case of a theoretical prostate cancer patient, if we only looked at the DNA of the cells in their blood, there wouldn't be much more useful information that we could find about their cancer. However, the prostate cells that are actually making PSA—and only those cells—should have the gene that encodes it “on,” and that, we found in my research, could tell us something interesting.
We looked for PSA RNA in the body—the essential piece of the PSA production process that should have only been able to be found within cells inside the prostate. We had the ability not only to look for the PSA protein itself leaking out of the prostate, but we could also determine if there were prostate cells themselves that had escaped from the prostate and were swirling around the patient's body in their bloodstream. This is how cancer metastases occur. Cancer cells migrate from the original tumor and find places—in lymph nodes, bone, and other organs—to establish new tumors. The presence of RNA that was a precursor to the production of PSA was a biomarker associated with the potential for metastatic cancer. Like peeling back the layers of an onion, we were looking one level deeper into the molecular nature of cancer in a given patient, and using that data to help physiologically locate whether that cancer might be in their blood.
This research is an example of how we can find new biomarkers, and also illustrates how and why biospecimens are preserved. Imagine if there was blood stored from prostate cancer patients around the world. We could go back and look at whether there were prostate cells present, and then we could compare that biomarker with what we retrospectively know happened in the course of their treatment. Was the presence of circulating prostate cells associated with more aggressive cancers? What interventions—surgical or pharmaceutical—were associated with the best outcomes for patients where we could detect those cells?
Instead of proactively having to find patients for that research (which kept a lab researcher—namely, me—frantically running around the Columbia Presbyterian Medical Center 25 years ago), we could thaw those biospecimens and investigate them.
Of course, PSA RNA is just one possible biomarker. Since I worked on that particular research project, the ability to sequence single‐cell DNA to look for specific mutations in cancer has evolved spectacularly. More and more layers of useful measurements in determining the course of a condition can be checked, as long as we have those specimens preserved, our patients' permission to use them for research, and the ability to connect those specimens with the eventual outcomes related to the therapeutic courses given.
We now know that biomarkers can be found in our genes themselves, such as mutations in p53 that are associated with increased susceptibility to certain types of cancer, and that they can come from any of the phenotypic manifestations of what defines “us”—what genes are on or off, what proteins are present, where certain cells can be found in the body, and more. Looking particularly at the presence and function of proteins—proteomics—will tell us far more than we know today about what's going on inside our bodies. Researchers are discovering biomarkers with the potential to diagnose cancers, Alzheimer's disease, and more, well before we see symptoms.3 “Proteins can confirm an illness is underway, and they often appear in our blood long before we feel sick—months or years before symptoms, when many diseases are still curable,” wrote Michael Behar for the New York Times in 2018, explaining the potential of proteomic analysis to change the way we diagnose nearly everything. And I certainly agree.
However, it's important to look beyond just the physical biomarkers. The same way as we can look at cognition and behavior as measurements of health, they are also potentially useful as biomarkers in the way we think about a PSA level. In the patient equation, all “inputs” are really biomarkers. Perhaps the number of steps taken or the average daily territory a patient covers are useful predictors of cancer progression. Perhaps alone they're not, but instead they tell us something useful when combined with other inputs—in a multivariate patient equation. Keep this in the back of your mind as we continue this discussion.
Along the same lines, as I've said already, biospecimens don't just have to be physical samples of tissue or blood. We can log someone's step count today and go back a year from now to take another look. We can log everyone's step count today, and then, for the people who develop cancer in the future, go back and see if there were any hidden indicators. Perhaps it doesn't matter if someone is taking 12,000 steps a day or 10,000 steps—but maybe the rate of change in their pattern of movement is useful. If someone took 10,000 steps every day last year, but just 5,000 this year, perhaps that indicates that something is going on in their body—and perhaps it could indicate it even before a CT scan picks up evidence of a tumor. Perhaps that alone isn't enough, but if we combined it with their PSA, could we detect a pattern that is associated with prostate cancer? The more data we have—the more biospecimens—the more we can replay the past and test the patient equations of the future.
There is the potential for, say, Alzheimer's disease research to approach problems this very same way. We have traditional data about how patients are progressing with the disease, and we also have activity data, these digital biospecimens about their daily behavior, and their quality of life. We can look, perhaps, at how many times someone checks the calendar on their smartphone, and map that against traditional measures of disease progression. If I normally check my calendar three times a day, and then suddenly I'm checking it 8, 10, 12 times—does that mean my memory is getting worse? And does that data add to our understanding of my disease progression, or is it simply noise?
Like Lions of the Serengeti
There is potential for this type of data to be useful in diagnosis, prognosis, and even as a way to measure the value of therapies. Can we catch something earlier with cognitive or behavioral data than we can with traditional measures? Can we quantitatively and objectively detect changes in behavior that might give additional insight into what is happening at the molecular or cellular level in a patient—like, is their tumor burden growing or shrinking? Can we use a measurement—again, an objective and quantitative one that is free from the biases of a patient's self‐assessment or the limitation that a health care professional can't observe patients 24/7—as a proxy for quality of life or socioeconomic engagement?
The definitive answers to these questions are being researched right now, but two examples across behavior and cognition can frame the idea. And I am certainly hopeful that at least one of them, if not both, will prove to be worthy of inclusion in our patient equations.
Imagine a patient diagnosed with cancer. Assume for this exercise that there are two drugs available on the market—both indicated for the patient's condition, with no contraindications on their respective drug labels—that can extend life for what amounts to equal durations, based on all available research with previous patients. Let's say this extension of life is two years. Assume as well that the drugs have similar safety profiles—similar side effects and chances of, say, cardiotoxicity that could weaken the patient's heart.
So which drug does the patient take? Drug A or Drug B? Since there is no difference in predicted outcomes, there's no good way to choose. The patient and physician might as well flip a coin.
Now, consider the same choice if the patient and physician knew that previous patients taking Drug A spent three‐quarters of their survival period—a year and a half of the two years—bedridden, whether in the hospital or their home. But patients on Drug B were mobile and able to travel, work, and spend time with their families and friends for the vast majority of those two years. The choice becomes easy. You take Drug B.
Let's assume that a government is paying for the drug. If the patient takes Drug B, they will be buying plane and bus tickets, consuming food in restaurants, and producing products or intellectual property at work—they will be socioeconomically engaged, producing and consuming more outside of their health care than someone on Drug A. So the payer wants them to take (and would rather pay for) Drug B as well.
But how could we possibly know this information about Drug A and Drug B? That's where the idea of “patient territory” comes in.4 If we can measure how much a patient moves around, not necessarily with steps, but in terms of their home range—just like we would observe a pride of lions to see how far they roam—that should be a good proxy for socioeconomic engagement.
The patients on Drug A will be living in 100‐square‐meter areas on an average day—in a bed, perhaps moving to a bathroom or occasionally to a clinic or another part of a hospital. But the patients on Drug B, depending on how far their travels take them, may be roaming around thousands or tens of thousands of square meters in that same typical day.
We don't need to track people or rely on self‐reporting to get this measurement of territory. Virtually all of us—trending toward absolutely all of us, on a global scale—keep our mobile phones within a meter of us all day, every day. We keep them charged the whole time so we are ready for texts, emails, and social media. (And before you start to worry that the idea of patient territory is based on some Big Brother–esque tracking of our precise locations, there are other algorithms that can be used to compute territory based on anonymous positioning data.)
By taking location data at multiple timepoints throughout the day—kept within a phone, never shared on the cloud or with another party—we can assemble a series of vectors. Each vector has a direction and a magnitude. Two meters to the north, two meters to the east, two meters to the south…and you can see how a four‐square‐meter area is circumscribed. We don't need to know where. We just need to know the total.
If we had collected anonymous territory data from those patients previously taking Drugs A and B, the clear choice of Drug B could be made—with hard, numerical data. Playing with this thought experiment across different dimensions is another interesting exercise. Perhaps Drug A, although resulting in patients being largely bedridden, has an average of two years' additional survival, but Drug B only 12 months. If the patient is more interested in living to see, for instance, a family member's graduation 18 months from now than they are in travel, Drug A is the obvious choice. The point is not to create rules based on the territory data, but to make that data available in an equation (illustrating the tradeoffs between longevity and quality of life, in this example) that patients, physicians, and even payers can use to ensure that desired outcomes are achieved as much as possible.
If you look at the territory example as a behavioral marker that can be a proxy for socioeconomic engagement and quality of life, you can look at a similar example, but in the digital world, for cognition—say, a simple measurement of bandwidth usage. Everyone with an online existence, whether through texts, emails, and social media or the generation or downloading of large‐format audio or video files, uses bits and bytes of bandwidth. Again, the idea isn't to peer into what those bits and bytes are—download the media that you like, it's not anyone else's business—but if we were to merely look at the total amount of bandwidth you are using every day, is it a reasonable hypothesis that it will trend down with neurodegeneration and up with socioeconomic engagement? I think the answer is clearly that it will.
The Layer Cake
Imagine all of these different measurements—these biomarkers—stacked up on top of one another. There's temperature, weight, blood chemistry, scan results, genes, proteins, step count, territory, mood, what we're eating, how much sleep we're getting, and, say, the amount of pollution in the environment around us. Every time we discover something new that we can measure—or a way to measure it with a higher resolution—layers get added. We can measure more of these layers now than we could just a few years ago, and a few years from now we'll be measuring even more.
What we need to determine is which layers are helpful, and which layers add information that we can use to improve diagnosis, treatment, or quality of life. We need to figure out which layers indicate shared characteristics that lead to similar outcomes for certain conditions, and the same treatments working effectively. We need to see which layers glow most brightly in terms of helping us predict successful outcomes before we treat, or, once we do treat, showing us as quickly as possible whether or not the treatment is working. We need to figure out which layers can be useful inputs that indicate who we are and how best to treat us, which layers can be outputs that tell us whether and how much a treatment is working, and which layers can serve as both. And we need to figure out how different layers might combine and interact to produce surprising and useful answers about how to improve treatments, products, and lives.
I've already mentioned some of my favorite layers of the cake. I believe the rate of change in our step counts will turn out to have value, as we collect and analyze more and more of that data. I believe territory covered in an average day is a useful way to measure quality of life. I believe resting heart rate, as a continuous stream, can tell us something useful about our bodies—which is why I'm
wearing the patch on my chest. I believe that the patterns of our posts on social media—the word clouds of what we're thinking about, and what we're sharing—can tell us something about our mood, our happiness, and our overall health. I believe that what we now call “junk DNA”—the introns that we now discard because they don't encode proteins—might actually contain useful information. I believe our sweat matters—whether we can use it to sound an alarm when we're becoming dehydrated, or when we need the next dose of a particular medication, or whether there's even more information we can gather from it. I believe our socioeconomic production and consumption can serve as a proxy for how we feel—how much we are able to contribute or consume in a given day, with an active, healthy person able to engage far more in the world than one who is suffering.
But it doesn't matter what I believe. We have the computing capability and the potential data to determine whether I'm right or wrong about the usefulness of these measures, and to figure out in what ways they can help us.
Patients Like You, Patients Like Me
Anecdotally, people have been talking about the layers of their layer cake for a long time. We don't need sensors to know that some days we feel good, and some days we don't. Some treatments seem to affect us one way, and some treatments affect us quite differently. Sensors can make the measurements more objective, bring scientific rigor to our anecdotal evidence, and uncover patterns our brains haven't been able to recognize—but underneath it all, the information itself has been there forever.
PatientsLikeMe, purchased in 2019 by UnitedHealth, is an online network of over half a million members tracking their diseases and conditions, sharing treatments that work for them, and connecting with others on similar health journeys. The site has collected a huge amount of self‐reported data from patients dealing with virtually every illness or condition there is. The company's research team has published over a hundred studies, attempting to identify better treatments for patients, particularly in cases where rigorous clinical trials have not yet yielded useful answers. The company's co‐founder, Jamie Heywood, believes that the greatest impact of these new data‐collecting sensors will be to enable us to finally examine variables that have previously been ignored by the medical community—including hormones, social factors, air quality, metals and toxins in the environment, stress, metabolism, sleep, nutrition, activity and exercise, and more.5 If you want to know what layers of the layer cake we should consider looking at—and how we should look at them—Jamie is a great place to start.
The Patient Equation Page 4