Most artificial intelligence medical devices cleared by the US Food and Drug Administration reach the market without independent evidence of safety or effectiveness, according to a commentary in Annals of Internal Medicine. The authors noted that 97% of AI and machine learning devices are cleared through the FDA’s 510(k) pathway, which requires substantial equivalence to existing devices rather than direct proof of clinical benefit. Premarket studies are rarely randomized, prospective, multisite, or representative of diverse patient populations. The authors also highlighted major weaknesses in postmarket surveillance and adverse event reporting, warning that clinicians and health systems may misunderstand what FDA clearance actually guarantees about device performance.

Source: Annals of Internal Medicine