Performance metrics (mean, [95% confidence interval]) on bootstrapped test sets for trained logistic regression (lr), multi-layer-perceptron (mlp), tree-based models (tree), or pre-trained (pretr) models.

The best baseline models according to the AUROC and AUPRC metrics (Wassenaar and Boogaard) are also included. Sensitivity and specificity are calculated for the threshold that maximizes their sum. Precision is calculated for the highest threshold for which recall > 0.70. Model variants (M1-M123) consume data from time phases and their combinations T1-T123. Data from TI is included for every model.