08/27/2026
1,357 AI medical devices cleared by the FDA.
Only 3 were tested for whether they actually improve patient outcomes.
Let that sink in.
A new systematic review in PLOS Digital Health found that the vast majority of FDA-cleared AI tools were never evaluated for patient-centered outcomes โ not mortality, not morbidity, not quality of life. The 510(k) pathway allows "evidence gaps to propagate through chains of predicate devices," many of which themselves lack rigorous clinical validation.
And yet โ the tools are already in our workflows.
This week brought a striking contrast:
โ
The FDA authorized Vitestro's Aletta โ the first autonomous robotic blood draw device. Years of trial data. A 95% first-stick success rate. Equity requirements built into the regulatory framework, requiring performance data across different skin tones and vein access difficulty. A 1:3 human supervision ratio written into the authorization itself.
โ Meanwhile, 1,000+ AI tools are already deployed in clinical settings with no post-market surveillance, no outcome data, and no requirement to prove they work in the populations they're used on.
The FDA says GenAI regulatory guidance is "coming." ECRI just expanded its error reporting network to track AI failures in patient care. These are positive signals โ but they're reactive, not proactive.
Here's what nurse leaders can do RIGHT NOW:
1๏ธโฃ Ask for outcome data before adoption, not just accuracy metrics
2๏ธโฃ Demand post-deployment monitoring โ use ECRI's reporting network
3๏ธโฃ Comment on the FDA's GenAI discussion paper before ๐ข๐ฐ๐๐ผ๐ฏ๐ฒ๐ฟ ๐ญ๐ต
4๏ธโฃ Apply a structured evaluation framework โ because "AI-powered" isn't the same as "evidence-based"
The PIVOT framework exists because ad-hoc adoption isn't governance. Purpose. Intelligence. Values. Outcomes. Technology. In that order.
What AI tools are currently in your clinical workflow โ and when was the last time you asked whether they actually improved patient outcomes?
, , , , , , , , , , and
