The abstract is read at the publisher; the summary is JClub's.
A future-information leak in Medicare-insured adults’ discharge readmission prediction inflated AUROC from 0.774 to 0.706 when removed, using MIMIC-IV v3.1 admissions.
- Why it matters: Availability audits and patient-grouped validation can miss outcome artifacts that distort discrimination, timing, and subgroup conclusions, undermining trust in 30-day readmission models.
- What they did: In 236,906 Medicare-insured admissions (87,750 patients; 21.4% prevalence), predictors were audited for discharge-time availability and recomputed from raw timestamps; an XGBoost model was validated with patient-cluster bootstrapped CIs.
- The result: The corrected 44-feature model achieved out-of-fold AUROC 0.7057 (95% CI 0.7016–0.7102) and calibration slope 0.99, with a day-1 peak linked to 2,368 probable administrative continuations.
The findingWhy it mattersWhat they didThe result
AI summary of the abstract · may contain errors · not medical advice · read the paper.