Fig. 5

Model optimization for aetiology, disease severity and outcome, starting from the maximum number of parameters resulting from the elastic net regression models. The most optimal model was chosen based on the minimum decline of 5% in area under the ROC curve (AUROC) and sensitivity with the lowest number of proteins in the model. Aetiology: viral vs. bacterial sepsis, the most optimal model had an AUROC of 94% and sensitivity of 86% with six proteins in the model. Disease severity: SOFA score > 4 vs. SOFA score < 2, the most optimal model had an AUROC of 80% and a sensitivity of 83% with six proteins in the model. Outcome: worse vs. less severe outcome, the most optimal model had an AUROC of 76% and a sensitivity of 68% with ten proteins in the model. The most optimal models for aetiology and outcome were chosen after SMOTE procedure. AUROC, PR-AUC, sensitivity and specificity are shown