Table 3.

Average performances (std.) of the optimal predictive model based on the held-out testing sets in the initial cohort. Raw annotated labels referring to initial labels manually entered by clinicians before adjustment. Those labels were later adjusted to meet current clinical guidelines for whole exome sequencing and whole genomic sequencing test ordering.

Model	Recall	Precision	F1	Accuracy	AUROC	AUPRC
Random Forest (raw annotated labels)	0.842 (0.022)	0.728 (0.016)	0.780 (0.011)	0.731 (0.015)	0.781 (0.024)	0.811 (0.022)
Random Forest (After label adjustment)	0.941 (0.024)	0.801 (0.021)	0.865 (0.019)	0.793 (0.029)	0.822 (0.045)	0.918 (0.023)