AI can now filter clinical trials by who they actually studied, not just statistical power

A new system uses AI to find relevant clinical trials and weight them by whether their patient populations match the question being asked, rather than just by statistical precision. This means drug safety estimates and treatment comparisons can now account for real-world patient differences instead than treating all studies as equally valid regardless of who was enrolled.

Clinical meta-analyses have always had a hidden problem: they weight studies by how many patients they enrolled and how tight their confidence intervals are, but ignore whether those patients looked anything like the people the drug will actually be used on. A trial of olaparib in young, fit cancer patients gets the same statistical weight as one in older patients with comorbidities, even though eligibility criteria tell you they studied different populations. This system makes that mismatch visible and corrects for it. In the olaparib example, accounting for eligibility alignment shifted the pooled risk estimate from 2.18 down to 1.97 — a meaningful change that conventional meta-analysis would have missed. The practical effect: drug safety signals and treatment comparisons become more honest about which patients they actually apply to.