An NBC News story on OpenEvidence marks a real inflection point in clinical AI. According to company representatives cited by NBC, nearly two-thirds of US physicians, roughly 650,000 doctors, now actively use the tool. In April alone, they reported nearly 27 million reported clinical consultations.
This is one version of what responsible AI in medicine can look like. At the point of care, a physician with a complex question gets a synthesized answer grounded in the peer-reviewed literature, including randomized controlled trials, in seconds rather than hours. Daniel Nadler and his team deserve enormous credit for what they’ve built and for putting it in the hands of clinicians who need it most.
And yet, every clinician reading this already knows the next sentence.
The patient sitting across from us is rarely the average patient represented in the evidence base.
Randomized controlled trials (RCTs) are, and will remain, the gold standard for establishing causal effects of therapies. But to achieve internal validity, eligibility criteria frequently exclude older patients and patients living with multiple comorbidities. And historically under-represented populations remain under-enrolled for a different and more complex set of structural reasons: access, geography, historical trust, and social determinants.
The published literature is, of course, richer than RCTs alone. It includes observational studies, registries, and a growing body of real-world evidence reports. But even the broader published literature is constrained by what has been studied, analyzed, written up, peer reviewed, and published, often with meaningful lag. By the time a finding is in print, the point-of-care question has often become more specific than the published evidence can answer.
The published literature tells us how carefully studied cohorts responded under the conditions of those studies. It tells us much less about how the 72-year-old patient in clinic this morning, living with diabetes, chronic kidney disease, atrial fibrillation, and a new oncologic diagnosis, is likely to respond to the therapy we are considering.
From my years in clinic, I know what physicians need at the point of care: a synthesis of the published evidence on the question being asked, alongside the real-world experience of patients in routine care who resemble the one being treated, putting that evidence in context. Both, integrated into one workflow.
As a clinician, I want a tool like OpenEvidence integrated with what Truveta Intelligence delivers. This isn’t competition. It’s completion.
The future of AI-enabled clinical decision support is not published evidence or real-world evidence. It’s both. In one workflow, at the point of care. The teams that get that combination right will help define how the next decade of medicine is practiced.
