Table 3. Results of the reader study which shows the performance of physicians predicting infections within the next 90 days—without (MD) and with (MD+AI) decision support, in comparison to the automatic risk prediction (AI).

Evaluation is carried out using ROC (Area under Receiver Operator Curve) and PRC (Area under Precision-Recall Curve), as well as SEN (sensitivity; recall), SPEC (specificity; true negative rate), and PPV (positive predictive value; precision) defined by different thresholds (thrs). The threshold defines the cutoff value to map from regression to binary classification.

	ROC	PRC	SEN	SPEC	PPV	thrs
AI	0.719	0.567	0.754	0.718	0.264	F2
			0.580	0.844	0.333	F1
			0.329	0.940	0.424	F0.5
MD	0.630	0.474	0.448	0.689	0.406	>50%
MD+AI	0.622	0.469	0.368	0.829	0.500	>50%