Additional file 1 of An algorithm to identify cases of pulmonary arterial hypertension from the electronic medical record

Additional file 1: Figure S1. Feature importance for XGBoost and Elastic Net. Figure S2. Random forest importance values for each variable and associated variation (strength, durability, and persistence). Table S1. Number of subjects with each variable and percentage of Synthetic Derivative (N = 2,278,297). Table S2. Test characteristics of all three algorithms for the Development Cohort. Training results show mean AUC and 95% confidence intervals for each model based on 3 times repeated 10-fold cross validation (PPV and NPV were not computed during training). Table S3. Test characteristics of all non-RF algorithms for the Final Cohort. Training results show mean AUC and 95% confidence intervals for each model based on 3 times repeated 10-fold cross validation.