In their Research Article “Illusory generalizability of clinical prediction models” (12 January, p. 164), A. M. Chekroud et al. demonstrate that making robust and generalizable predictions in clinical contexts is difficult. Their results reveal flaws in the current strategy of applying artificial intelligence to prediction tasks. The reuse of existing data from randomized controlled trials and the lack of iterative model development hinder the ability of machine learning models to contribute in clinical settings.
