The Underrepresentation of You in Medical Research
You probably think taking a baby aspirin daily will reduce your risk of getting a blood clot. You’ve likely heard that a thousand times. It’s one of those pieces of medical common knowledge. But here’s the thing: it doesn’t work in about a third of people, and doctors don’t know why. It’s a phenomenon called aspirin resistance, and people who have it don’t get the anti-blood-clot benefits from the drug. But for people who do respond to aspirin, it works pretty well.
At the individual level, aspirin resistance is really important, obviously. But big randomized controlled trials suppress these important individual factors, leading to generic conclusions that don’t apply to a large swath of patients. Let’s simplify things a bit in attempt to discern how this works, then we’ll consider some other individual factors that don’t get captured by broad research conclusions.
For the sake of a thought experiment, assume that aspirin resistance is binary: aspirin either works for you or it doesn’t. Now, assume researchers take a random sample of 100 people, divide them randomly into two groups of 50, and give one group a placebo and the other aspirin. After a year, there were 20% fewer blood clots in the aspirin group than in the placebo group. What conclusions do the researchers draw? Of course, that aspirin reduces blood clots by one-fifth relative to placebo.
But that’s not right.
Aspirin resistance would have been evenly distributed among the random groups so that the findings would represent “some” aspirin resistance. But if, at the individual level, aspirin resistance is binary, then the benefit would theoretically be either zero or some number much greater than 20%.
Although this thought experiment is very simplified, it serves to illustrate the problem of the underrepresentation of important individual characteristics in large-scale research conclusions.
What we’ve started to realize over the last few decades is just how unique each person really is. And we don’t have good systems in place for taking into account these individual factors in health research. Not long ago, scientists assumed that once we’d figured out the human genetic sequence, miraculous medical feats would become commonplace and amazing drugs that cured everybody would be rapidly developed. Well, we found the human genome sequence in 2001, and we still haven’t developed those miracle cures.
Part of the problem, scientists now realize, has to do with what we call epigenetics. Epigenetic processes regulate the real-world effects of our genes by modulating the rate at which genes are changed into proteins. Even identical twins, whose gene sequence is shared, often turn out quite different because of epigenetics. How we respond to a given medication has a lot to do with epigenetics too.
Many factors influence epigenetics — what you eat, your environment, what medications you take, etc. But there’s one factor that makes this issue more complicated: a person’s microbiome also influences his or her epigenetics. Your microbiome consists of all the non-human organisms in and on your body. You’re composed of at least as many non-human cells as you are human cells.
And here’s where it get’s bewilderingly complex: the cumulative genome of the microbiome, called the metagenome, is way, way bigger than the human genome, and it’s influenced by its own epigenetic processes.
What this all means is this: there are no two individuals on the planet whose genetics, epigenetics, microbiome, metagenome, and meta-epigenetics all match exactly. And all of these things influence how we respond to medications.
Let’s circle back to medical research now. Warfarin is an anticoagulant drug used to prevent blood clots. Doctors give it to people with heart rhythm abnormalities to prevent blood clots from forming in one of the chambers of their heart. The problem is that it kills some people by rupturing blood vessels in their brain. This gets translated as a small risk when patients are told about the potential side effects of the drug. The “risk,” they’re told, is less than 1%. But for the people who do blow out a blood vessel and die after starting warfarin, the risk isn’t 1% — it’s 100%. Again, it’s a binary outcome at the individual level. You can’t be 1% dead — you’re either dead, or not.
So what determines who warfarin kills? It’s impossible to say — some combination of all the long “-ome” words I mentioned above. Researchers try to adjust their study data to identify certain factors that might increase or decrease a person’s response to a given drug. However, there is no way we will ever be able to account for the uniqueness that arises from the interactions of each person’s “-omes.” No matter what the rate of efficacy and side effects observed in a large trial, if a bad outcome happens to you, your risk is 100%, and it’s impossible to predict.
Let’s consider some real-world examples of how individual factors beyond our understanding influence response to medications. Take the cholesterol-lowering statin drugs. They are some of the most frequently prescribed drugs on the planet. For most people, they lower LDL-cholesterol levels without causing any problems (at least in the short term). But for some, they cause horrible muscle cramping. Here again, the statisticians tell us the “risk” is somewhere around 5% that a given person will develop muscle cramps. That’s again wrong — you do or you don’t get cramps. The individual factors that determine who cramps and who doesn’t are beyond researchers’ understanding.
Another example is metformin, a first-line drug used to treat diabetes. It works well in most people, but some people develop a potentially fatal metabolic aberration called lactic acidosis when they take metformin. It’s a low frequency event across a study population, but each individual either does or does not develop the condition. Incidentally, metformin is also interesting because we’ve recently realized that, although it’s been around for a very long time, we don’t understand how it works. Recent research has shown that it is the microbiome in the gut that facilitates much of metformin’s benefits, though it’s not exactly clear how.
How different are we at the “-ome” level? Given the complexity, it’s tough to know for sure. However, it’s fair to say: very different. In looking only at the diversity of the microbiome among healthy individuals, researchers in 2016 discovered more than an order of magnitude of difference in the composition of the organisms that made up the microbiome.
That’s a major problem for conventional medical research that relies on ever-larger “gold-standard” randomized controlled clinical trials. By the very design of the randomized trial, we blur out the clearly important individual differences between people, and sometimes the results are that people die.
But there’s another concern that arises when one considers just how different we each are. If a medication does have a clear effect across large groups of randomly selected people, it has to be seriously powerful to overshadow the inter-individual “-ome” variability. It’s not surprising, then, that we so often discover that drugs we once considered safe and effective are actually really dangerous (cf. Vioxx). And drug approval trials are nowhere near long enough for us to observe long-term ill effects that arise due to inter-individual variability.
As more and more money gets pumped into funding ever-larger trials and the edifice of “evidence-based-medicine” produces more newly approved drugs than ever before, there may be people out there right now taking drugs that will lead to disastrous long-term consequences due to inter-individual variability.
But there is a benefit to be had in understanding the implications of our uniqueness in the context of medical research. The next time you see news headlines touting a 30% improvement in this or that with the latest drug, remember that number was obtained by stripping away layers of individual uniqueness and cramming study participants into a box shaped like a bell curve. Human beings are not absolutely compatible with medical statistics based on the normal distribution, and there will always be outliers for reasons we don’t fully understand.
Health and healthcare must be tailored to the individual. For the patient, that means self-awareness and a willingness to communicate with your doctors. For physicians, it means not being so quick to get frustrated with non-compliant or non-responsive patients and an openness to deviating from gold-standard guidelines that emerge out of randomized trial data.