...

After defending my Ph.D. and returning to finish clinical training, I was struck almost immediately by how little scientific information went into clinical decisions. Medicine felt so behind compared to our understanding of science, and it seemed unfair that our patients did not benefit from the past decades of molecular advances. At the time, the first draft of the human genome had just been completed and I (like many others at the time) was excited by the prospect of how genetics might revolutionize care. While in residency and fellowship, I did a postdoc with David Reich in statistical genetics and, after taking my first academic position at the University of California San Francisco, focused my outpatient clinical and laboratory efforts on inherited cardiovascular disease.

From this experience, I (and many other scientists and clinicians) realized just how modest and unpredictable genetic effects were in determining the likelihood of disease. Even for patients with autosomal dominant forms of a disease, where the mutation arguably should increase disease risk by 10 to 1000-fold, there was tremendous variation from one family member to the next. And the majority of genetic scores being proposed to date are far more modest in effect, perhaps raising risk by 1.5-2.5 fold compared to baseline.

These modest and unpredictable effect sizes are one of the main reasons genetics has been so slow to incorporate into clinical practice (though it’s perhaps a blessing if you’ve inherited some of these variants or mutations). The second reason is related and far more sobering regarding clinical innovation (whether it’s AI or wearables or ‘omics) and its likelihood of utility and adoption. What has emerged from our research and that of many others, is that most biological signals are at least partially redundant. This means that the information you get from genetic testing overlaps heavily with what’s computed from measuring labs, or ‘omics, or wearables, or imaging, or fancy AI models – to the point where it’s unclear if it’s adding anything new. The only way to know that with greater certainty is to measure all these things simultaneously in the same patients and follow them for years and years, which, for the most part has been unfeasible due to exorbitant cost, and the frustrating fact that new innovations keep popping up (I’m being a little facetious). So at present, a patient can spend a lot of money on this testing, and not really be confident they’ve learned anything beyond what was already captured in the rest of their care.

Nonetheless, there will certainly be value in many of these innovative data types, especially where consistently larger effect sizes exist. But there is considerable ambiguity. Patients will need someone thoughtful to help make decisions under uncertainty, guided by both deep scientific expertise and a commitment to understanding individual patient values. We look forward to working with our patients to assist them in this journey.

Rahul D.