AI in Healthcare AI in Healthcare
Clinical Evidence Tracker

1,357 AI Devices Cleared.
3 Tested on Patients.

AI is proving itself in landmark clinical trials. But 99.8% of FDA-cleared devices have never been tested on whether they actually help patients.
99.8%
Untested on
Outcomes
67%
AI Triage
Accuracy
+39%
More Aneurysms
Detected
0.2%
3 of 1,357 devices
FDA Outcome Testing Rate
Of 1,357 AI/ML devices cleared by the FDA, only 3 have been tested on patient-centered outcomes like mortality and morbidity. The 510(k) pathway requires only “substantial equivalence” — not clinical proof.
67%
vs. 50–55% for physicians
Harvard ER Triage Trial
OpenAI’s o1 model matched the exact diagnosis in 67% of 76 ER cases at Beth Israel Deaconess — outperforming two attending physicians. On treatment plans: AI scored 89% vs. 34% for 46 doctors.
+39%
55 cases radiologists missed
Brain Aneurysm Detection
AI detected 39% more aneurysms across 3,856 CT angiograms at Northwell Health. Those 55 additional cases led to 40 consultations, 12 angiograms, and 2 preventative surgeries.
The 510(k) Evidence Chain Problem
99.8% No Outcome Data 43% No Validation Data At All Radiology: Most Cleared 510(k): Most Used Pathway
Under 510(k), new devices clear by showing “substantial equivalence” to existing devices — which themselves may lack validation. Each generation inherits the evidence gap of the last.
Regulatory approval has outpaced clinical validation, creating an ecosystem where innovation advances without accountability.
— PLOS Digital Health, August 2026

Explore the Live Data Behind This Analysis

Updated daily from 57+ sources across regulatory filings, deal trackers, clinical trials, and trend monitors