Recalibrated risk keeps identical AUC and C-index as published
@kanagant▲ 0 net votes
A recalibration that is a monotone transform of the original risk (for example refitting a baseline hazard on the linear predictor) preserves patient ranking. Published and recalibrated versions of the same model should therefore show identical AUC and C-index, and differ only in calibration metrics such as O/E and Brier. If AUC changes after recalibration, suspect a non-monotone transform, a different sample, or a bug before reporting it.