Manuscript analyses validating reprodICU harmonization across seven ICU databases using mortality prediction models and cohort comparisons.
- 1_cohort.py — Build harmonized cohorts (all 7 databases)
- 1_cohort_from_original_data.py — Build same cohorts from source SQL based on public SQL queries from
mimic-code: https://github.com/MIT-LCP/mimic-codeeicu-code: https://github.com/MIT-LCP/eicu-codeAmsterdamUMCdb: https://github.com/AmsterdamUMC/AmsterdamUMCdb
- 2_models.py — Train LightGBM models; output coefficients and performance metrics
- 3_compare_cohorts.py — Leave-one-database-out cross-validation
- 1_plots.py — Cohort characteristics and missingness patterns
- 2_sofa_subscores.py — SOFA component prevalence analysis