Reproducibility code — AAA mortality-prediction manuscript

All numbers in main.tex are produced by these scripts (SEED=42, deterministic).

File	Purpose
run_analysis.py	Core pipeline: load registry, build features/targets, nested-CV models (LR/RF/XGBoost), AUROC/bootstrap-CI, DeLong test, SHAP. Emits metrics.json.
run_experiments.py	Additional analyses EXP-1..7 (temporal validation, decision-curve, VQI/VSGNE recalibration, per-arm, competing-risks, multiple imputation, operating points). Imports run_analysis.py. Emits experiments.json.
ANALYSIS_PLAN.md	Data dictionary, column names, missing-data handling, evaluation plan.
metrics.json	Primary-analysis results (5-yr prediction, 1-yr benchmark, SHAP).
experiments.json	EXP-1..7 results.

Run

pip install pandas numpy scikit-learn xgboost shap openpyxl lifelines matplotlib
python run_analysis.py        # -> metrics.json + fig1..4
python run_experiments.py     # -> experiments.json + fig5_dca

Note: the registry .xlsx is NOT included (held at Asan Medical Center; access subject to institutional/IRB approval). Scripts expect it under 복부대동맥류 (AAA)/ per ANALYSIS_PLAN.md.

Every manuscript number is traceable to metrics.json / experiments.json. No fabricated values; underpowered/negative results are labelled honestly.