Well, in theory that is what pre-publication peer review and/or independent critical review for clinical application is supposed to do. Unfortunately, a lot of studies are distributed and reviewed as pre-prints, and it is rare for people to go back and verify that a peer reviewed release is still consistent. And, as noted, peer review often doesn’t really involve a detailed review of the statistical analyses because that is a lot of work (and often as not data is not provided or made generally available for papers in ‘closed’ journals). As @Qadgop_the_Mercotan notes, clinicians should be using “evidence-based expert sources to review the data and make recommendations”; unfortunately, the pool of people both skilled with statistical methods and sufficiently conversant with clinical application is not large nor growing. I suspect ‘the answer’ is going to be to start using “deep learning” AI systems for this application despite the essentially impossibility of actually validating these things to be reliable.
Stranger