Fig. 4.

Illustration of the deep learning system for recognising sepsis. Our deep learning system is illustrated for one sample patient (of an unseen testing database) together with a subset of vital and laboratory parameters that were used for prediction. In the top two rows, the sepsis label is shown decomposed into its components, the suspected infection (SI) window (consisting of antibiotics [ABX] administration coinciding with body fluid sampling), and an acute increase in SOFA (ΔSOFA) of two or more points. The third row illustrates the hourly predictions as probability of sepsis. The last two rows show laboratory and vital parameters (Z-scored units for joint visualisation). Red dotted lines indicate the point at which the SOFA criterion is fulfilled. A decision threshold based on 80% sensitivity is indicated by the black horizontal dashed line. The displayed model was trained on eICU and here applied to a patient of the AUMC dataset.