Fig. 4

Performance of selected biomarkers to predict the presence of HBD + DIC and 90-day mortality during sepsis. A AUC for each of the 26 biomarkers to predict the presence of HBD + DIC in patients with sepsis, organized from highest to lowest. B ROC curve representing the model using 26 biomarkers for predicting the HBD + DIC phenotype in patients with sepsis. C Violin plots showing the distribution of predicted probabilities for the presence of HBD + DIC in sepsis. The model performed well at classifying both sepsis controls and sepsis with HBD + DIC. The majority of sepsis controls had predicted probabilities of presenting with HBD + DIC less than 0.50 (median: 0.12, IQR: 0.03–0.28). By contrast, the cases had predicted probabilities that were overwhelmingly greater than 0.50 (median: 0.96, IQR: 0.65–0.99). D ROC curve representing the model using 26 biomarkers for predicting 90-day mortality among the HBD + DIC subset of patients with sepsis. E Violin plots showing the distribution of predicted probabilities for mortality in the cases. The model performed well at distinguishing between survivors (median: 0.13, IQR: 0.01–0.32) and non-survivors (median: 0.97, IQR: 0.84–0.99) among those with HBD + DIC