When high AUROC does not travel: internal-to-external performance differences in machine-learning models for antimicrobial resistance

Abstract Background: Machine-learning studies addressing antimicrobial resistance (AMR) often report high discrimination, but performance estimates may change when models are evaluated in new hospitals, countries, time periods, or data-generating environments. Recent AMR reviews have highlighted limited external validation and sparse calibration reporting, but the magnitude of internal-to-external performance differences has not been consistently synthesized.