Fig. 4

Prediction of ICU/death outcome in out-of-sample patients. A Violin plots for the set of seven cardiometabolic protein biomarkers that were included in the best model with biomarkers for both logistic regression and random forest. The figure depicts the distribution and box plot of these seven biomarkers, stratified by the ICU/death outcome, in the in-sample patient population. The P values shown for each biomarker are based on the rank-inverse normalized data, while the odds ratios (OR) are based on the data standardized to have a mean of 0 and standard deviation of 1. B The predictive performance of the best models with and without biomarkers in the out-of-sample patients. The figure shows the receiver operating characteristic curve and corresponding area under the curve (AUC) for the best logistic regression (left) and random forest (right) models with biomarkers (gold) and without biomarkers (bronze) in the out-of-sample patients. The best model with biomarkers, for both the logistic regression and random forest, included the same set of seven biomarker, shown in (A), along with two hospital labs: procalcitonin and LDH. All models were developed and trained using only the in-sample data. Thrombotic thrombocytopenic purpura, TTP; acute respiratory distress syndrome, ARDS