An analysis of 691 FDA-cleared artificial intelligence and machine learning devices (1995–2023) found that fewer than 2% cited randomized clinical trial data. Most were class II devices cleared via the 510(k) pathway, concentrated in radiology (76.9%), cardiology (10.1%), and neurology (2.9%). Premarket reporting was often incomplete, with few devices disclosing study design, sample size, patient outcomes, or bias assessments. Postmarket data revealed 489 adverse events and 113 recalls, most related to software issues, underscoring the need for more rigorous regulatory standards and surveillance to ensure patient safety.

Source: JAMA Health Forum