From trial data to personalized medicine: a validated framework with an application to Crohn's disease

Clinical practice is currently guided by studies that average over patient outcomes. This may not be the best approach, as different patients may have different treatment responses. Here we extend a method for simulating clinical trials to identify optimal treatments for each patient, and we illustrate this approach in the context of Crohn's disease. Using the data from 15 randomized trials (N = 5703), we used statistical hypothesis testing to identify seven subgroups with distinct responses to three different drug classes. The largest subgroup consisted of patients with equivocal responses to all drug classes, whereas the second largest showed superiority with anti-TNFs. We also identified a subgroup of women over 50 with superior responses to anti-IL-12/23s. Interestingly, this group appeared under-represented in the trials (2%) compared to patients at the University of California (25%). Overall, these results underscore the importance of studying personalized medicine, demonstrate the value of clinical trial data, and provide a roadmap for applying this method broadly across diseases. These results also highlight the importance of diverse and representative recruitment into clinical trials.