Biomedical AI may be headed for a replication crisis. (This work below is not about AI-generated reports; it’s about studies of biomedicine that us…

A meta-analysis of 210 biomedical AI studies finds 97% used statistically invalid tests when comparing models under cross-validation, signaling a systemic replication crisis in ML-based biomedical research. Critical for practitioners publishing or consuming biomedical AI results, as most reported model comparisons may be statistically unsound.