In a review of 130 FDA-cleared AI tools, researchers found that the vast majority of them relied on retrospective studies. Most of the developers also didn’t share information on how many sites the algorithms had been tested in, or how they performed in different patient demographics.
