Future-Information Leakage in 30-Day Readmission Prediction: An Empirical Audit and a Calibrated Electronic Health Record Model in Medicare-Insured Adults
En palabras de los autores
Objective: To develop and internally validate a calibrated discharge model of 30-day within-system readmission in Medicare-insured adults, and to quantify how a future-information leak and outcome artifacts distorted performance, timing, and subgroup conclusions. Materials and Methods: Retrospective study of 236,906 Medicare-insured MIMIC-IV v3.1 admissions (87,750 patients; prevalence 21.4%). Predictors were audited for discharge-time availability; history features were tested for future information and recomputed from raw timestamps. A consensus gradient-boosted (XGBoost) model was selected inside patient-grouped folds; CIs used patient-cluster bootstrap. Results: A feature documented as a prior 180-day length-of-stay trend used each patient's whole record, including later admissions; removing it and two derived terms lowered test area under the receiver operating characteristic curve (AUROC) from 0.774 to 0.706. The corrected procedure reached an out-of-fold AUROC of 0.7057 (95% CI 0.7016 to 0.7102; primary); the 44-feature model scored 0.7056 on the historically exposed test partition (calibration slope 0.99), above reconstructed LACE and HOSPITAL (both P<.001). A day-1 discrimination peak largely reflected 2,368 probable administrative continuations within 1 hour of discharge; merging these into the preceding stay gave test AUROC 0.698 (0.702 after refitting). The White minus Black AUROC difference was 0.003 (95% CI -0.018 to 0.024); specificity was lower for Black patients (0.61 vs 0.69). Discussion: Documentation audits and patient-grouped validation missed a leak that inflated discrimination and created spurious subgroup and calendar-period differences; simple empirical checks found it. Conclusion: Empirical leakage checks should accompany availability audits. The corrected model is calibrated, moderately discriminating, and needs external validation before clinical use.
Apareció: lunes, 16 de noviembre. medRxiv. Preprint, todavía sin revisión por pares.