A 72-Hour Machine Learning Prognostic Model for ARDS Patients Requiring Invasive Mechanical Ventilation

Code accompanying the manuscript:

Development and External Validation of a Machine Learning-Based 72-Hour Prognostic Model for Acute Respiratory Distress Syndrome Patients Undergoing Invasive Ventilation

Contents

• R/ — Boruta and LASSO feature selection on 69 candidate variables
• python/grid_search/ — stratified 10-fold cross-validated grid search for 8 classifiers, plus the extended XGBoost grid
• python/final_models/ — final model fitting with the selected hyperparameters
• python/resampling/ — bootstrap optimism with feature selection re-run inside every resample, nested cross-validation, frozen hyperparameters
• python/evaluation/ — discrimination, calibration and clinical-utility metrics, 10-fold cross-validation, two-parameter logistic recalibration

See README.md for the full pipeline, software requirements and run order. Cite the concept DOI (10.5281/zenodo.23938386) so the reference resolves to the newest release.

Data

Patient-level data are not included in this repository. De-identified data are available from the corresponding author on reasonable request.