by Joseph Zabinski, PhD, Senior Director, AI & Precision Medicine, OM1
Sponsored content brought to you by
Personalizing medicine using population-level insights can seem like a contradiction. On one hand, the goal is to better fit treatment to the individual patient, moving away from generalized approaches toward a “one size fits one” model. On the other, we have to draw knowledge and information to improve medicine from groups, rather than individuals. At least with current technology, there are real limits to what can be reliably learned from any single case and separating signals from noise requires studying a sufficiently large number of patients.
The gap between the aspiration of precision and the realities of population-based learning can be bridged through a combination of real-world data (RWD) and artificial intelligence (AI). RWD fills in the space left around clinical trial data, reflecting everyday lived realities and offering insight into vastly larger, more diverse populations and circumstances than can be reasonably captured through randomized trials alone. AI offers a suite of tools that can reliably gather insights from these data, learn common patterns, and apply that information to make strong predictions at the individual level.
The more we learn about these patterns, the more complex they’re revealed to be. Symptoms, test results, and behaviors can all be uninformative—or even misleading—in isolation, but very valuable if seen as parts of an interacting pattern picture. It’s very difficult to prespecify what these patterns will look like, making their investigation through traditional epidemiologic methods and study design more challenging. Yet, at some level, we know they exist: Patients with certain conditions, or subgroups of patients within larger diseases, can exhibit characteristics that don’t immediately point to a diagnosis or defined subtype but still appear when we look.
Identifying these patterns using RWD and AI can be especially helpful in rare diseases. While many of these conditions have clear-cut distinguishing characteristics, many more only manifest through a constellation of seemingly unrelated signs, symptoms, and behavioral patterns that can vary from patient to patient. Fabry disease, for example, is a rare condition that can affect patients’ hearts, kidneys, digestion, and mental health, among other areas. Many patients experience highly diverse sets of symptoms, and almost none point unequivocally to the actual underlying disease. Yet common threads—which tests have been run, which symptoms have been reported, in what order—can help develop a larger RWD pattern of this disease that can then be found in other patients’ data records.
This RWD+AI approach is not limited to rare diseases. For example, in more common conditions with many millions of patients—think rheumatoid arthritis, or major depressive disorder—there exist different treatments, all of which work for some patients, but none of which will work for all. Physicians have certain tools to help select (“personalize”) the best course of treatment, but beyond a point, they need to rely on a degree of trial and error to find an effective option. Yet even in these cases, there is often some intuition that a patient’s chances of improvement are greatest with one treatment, rather than others available. This intuition is based in physicians’ deep experience, picking up on patterning from cases studied and seen before. Some evidence for those common patterns, associated with response and outcomes, can exist in RWD and be detected and translated using AI—at a scale that permits both population-level insight and personalized recommendations.
These tools are by no means fully mature; we’re just beginning to appreciate what is possible using RWD and AI together for personalization in clinical practice. Finally, though, we have the data, technology, and proof points to begin moving past the hype and toward real-world impact.
Find out more about OM1: www.om1.com
