- Fewer than 2% of FDA-cleared AI/ML devices cited randomized trial data; only 7.7% included prospective studies.
- Most devices targeted radiology (76.9%), followed by cardiology (10.1%) and neurology (2.9%).
- Nearly half lacked study design details, over half omitted sample size, and less than 1% reported patient outcomes.
- Four hundred and eighty-nine adverse events were linked to 36 devices, with 40 devices recalled (mostly for software issues).
- Findings highlight inadequate efficacy and safety reporting, reinforcing the need for stronger regulatory pathways and postmarket surveillance.
Source: JAMA Health Forum
