A research letter in JAMA Network Open, published on November 5, 2026, examined 163 AI and machine learning devices first cleared by the FDA between 2015 and 2021. Of these, 56 devices (about 34%) received at least one algorithm update, with a median time to the first update of 1.8 years.<\/p>\n
About three quarters of the updated devices added features or changed their input data, and roughly 16% were performance-only changes. Around 86% of the updated devices reported no clinical or organizational benefit, and many left out details such as data sources, sample size or performance metrics. The authors note limits: a small sample, reliance on public FDA records, and no visibility into updates made under predetermined change control plans. The study does not show that any device got worse after an update.<\/p>\n
Why it matters:<\/strong> as more medical equipment gains software that learns or gets updated, buyers and engineers will need to re-verify performance after changes, not only at purchase.<\/p>\n