BMC ARTIFICIAL INTELLIGENCE SUPPLEMENTARY PACKAGE
Collection: AI in chronic disease management

Manuscript:
Benchmarking Machine Learning Models for Cardiovascular Disease Classification in Population Health Surveys: Discrimination, Calibration, Interpretability, and Transportability

PACKAGE CONTENTS

1. Additional_File_1_Supplementary_Tables.xlsx
   Submission-ready supplementary workbook containing:
   - S1 country eligibility audit
   - S2 country-specific external discrimination
   - S3 subregional external performance
   - S4 subgroup external performance
   - S5 overall model performance
   - S6 detection-adjustment changes
   - S7 temporal benchmarking
   - S8 model settings
   - S9 predictor definitions
   - S10 full data dictionary

2. Additional_File_2_Reproducible_R_analysis.R
   Reproducible R implementation of cohort construction, fold-specific preprocessing,
   survey and class weighting, country-grouped benchmarking, Platt recalibration,
   threshold selection, performance estimation, temporal benchmarking, SHAP feature
   importance, and figure generation. Code sections S01 to S15 map directly to the
   Methods components described in the manuscript.

3. Additional_File_3_CVD_analysis_dataset.csv.gz
   De-identified analysis-ready dataset containing the 187,791 records from the
   46-country binary-classification cohort.

   IMPORTANT: Do not upload this participant-level file to the journal until WHO STEPS
   and applicable country-specific data-use conditions have been checked and confirmed
   to allow redistribution. If redistribution is not permitted, retain this file in the
   private reproducibility archive and submit the code, dictionary, harmonization details,
   and aggregated outputs instead.

4. Additional_File_4_Reproducibility_information.txt
   Reproducibility seed, model settings, software requirements, file hashes, analytic
   cohort counts, temporal-pair definition, and package manifest.

ANALYTIC BENCHMARK
- Outcome: self-reported prevalent CVD history
- Primary cohort: 187,791 adults, 46 countries, 15 subregions, 16,305 CVD cases
- Geographic validation: five-fold country-grouped internal-external cross-validation
- Temporal validation: Mongolia 2013 to 2019; Uganda 2014 to 2023; Viet Nam 2015 to 2021
- Models: logistic regression, elastic net, random forest, XGBoost
- Class imbalance: class-weighted fitting and PR AUC co-primary with ROC AUC
- Detection-adjusted analysis: hypertension and diabetes removed

SUBMISSION NOTE
The workbook contains the main supplementary tables referenced by the manuscript.
The participant-level dataset should only be included in the journal submission if its
redistribution is permitted. The code and aggregated supplementary tables can be submitted
regardless, subject to the journal's normal file requirements.