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