Figure 2.

Performance of the final model on the testing set within the development set. (a) AUROC. (b) AUPR. Solid curves were computed on the total set. Dashed curves were computed with a bootstrap procedure with 100 iterations, where, in each iteration, 50% of the testing set was sampled with replacement. (c) Calibration plot for the relationship between the predicted and observed probabilities for COVID-19 deterioration. The dashed diagonal line represents an ideal calibration. The purple line represents the actual model performance in five discretized bins. The blue histogram is the distribution of the risk predictions.