Three of 1,357 FDA-authorised AI medical devices were tested on patient outcomes
A PLOS Digital Health analysis found only 34 of the cleared devices were linked to a registered clinical trial, and most studies excluded pregnant women and adults over 75.
Original cover art, generated for this story. THE VISSION does not republish third-party press imagery.
- Of 1,357 AI-based medical devices authorised by the US Food and Drug Administration for patient care, only three were evaluated on patient-centred outcomes such as mortality, stroke, hospitalisation or quality of life.
- Only 34 of the 1,357 devices were linked to a registered clinical trial at all, according to the analysis published in PLOS Digital Health on August 19.
- Most of the evaluations that did exist were run in well-resourced health systems and excluded groups including pregnant women, adults over 75 and non-English speakers.
- The authors, led by Rawan Abulibdeh at the University of Toronto, attribute the gap to financial and logistical incentives that reward speed of clearance over evidence of benefit.
An analysis published in PLOS Digital Health on August 19 examined all 1,357 AI-based medical devices the US Food and Drug Administration has authorised for use in patient care and found that three had been tested on whether they improve patient outcomes — measures such as mortality, stroke, hospitalisation or quality of life. Only 34 were linked to a registered clinical trial of any kind.
The evidence that does exist is also narrow. The authors, led by Rawan Abulibdeh at the University of Toronto, found most studies were conducted inside highly resourced health systems and excluded patient groups that are common in practice, including pregnant women, adults over 75 and people who do not speak English. A device validated only on the population easiest to study is a device whose behaviour outside that population is unknown.
The paper attributes the gap to structural incentives rather than negligence: outcome trials are slow and expensive, clearance does not require them, and the commercial reward goes to whoever reaches market first. The authors argue for more FDA oversight tying authorisation to demonstrated patient benefit.
Regulatory clearance is routinely read as evidence a device works, and this puts a number on the distance between those two things — 1,354 of 1,357 cleared devices have no outcome evidence behind them. For anyone building clinical AI, that is both the competitive opening and the coming risk: outcome data is currently a differentiator nobody is required to produce, which is exactly the kind of gap regulators close retroactively.