Predicting 30-Day Hospital Readmission in Medicare Patients: An Interpretable Gradient-Boosting Model on MIMIC-IV v3.1
Year: 2026
Type: software
Abstract: Frozen software snapshot for the JMIR AI manuscript "Predicting 30-Day Hospital Readmission in Medicare Patients: An Interpretable Gradient-Boosting Model on MIMIC-IV v3.1" (Bandeira, Gonzalez, Poellabauer, Mondal). Contains the leakage-safe, development-only staged recursive feature elimination pipeline and the final consensus-31 XGBoost model (src/pipeline_v5_v6), canonical machine-readable results (results_v6/final_model_v6.json and companions), publication figures, the reproducibility report, the value-level consistency report, and the 207-predictor audit workbook. Three evidence tiers: development-data out-of-fold AUROC of the complete selection procedure 0.7748 (95% CI 0.7715 to 0.7782, primary); grouped development cross-validation of the fixed consensus-31 set 0.7756 (secondary); a single evaluation on the historically exposed test partition 0.7738 (95% CI 0.7671 to 0.7804, tertiary). Same-cohort baselines: adapted LACE 0.605 and 12-month HOSPITAL 0.649. MIMIC-IV row-level data are not included, per the PhysioNet Credentialed Health Data Use Agreement.
Institution: Florida International University
Language: English
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W7203662565