Training and Validation Cohorts
Patients were randomly split into training (80%) and validation (20%)
cohorts, ensuring equal distribution of the primary outcome in each
cohort. This method of stratifying by primary outcome before randomly
assigning patients to a cohort ensures an equal distribution of
mortality between training and validation datasets to avoid biasing
final model performance. The training cohort was used for feature
selection, dimensionality reduction, and machine learning model
development, keeping the validation cohort entirely separate and unseen
until assessment of the final model performance.