Table 3.

Algorithm performance.

	Logistic	SVM	RF	XGBoost
Panel A: suicidal ideation (N = 3292)
Area under the curve	0.837	0.844	0.851	0.861
Sensitivity	0.808	0.811	0.850	0.853
Specificity	0.867	0.877	0.852	0.869
Positive predictive value	0.808	0.820	0.799	0.819
Negative predictive value	0.867	0.870	0.891	0.895
Accuracy	0.843	0.850	0.851	0.863
Panel B: suicide planning or attempt (N = 488)
Area under the curve	0.872	0.872	0.857	0.880
Sensitivity	0.861	0.861	0.814	0.861
Specificity	0.883	0.883	0.900	0.900
Positive predictive value	0.841	0.841	0.854	0.861
Negative predictive value	0.898	0.898	0.871	0.900
Accuracy	0.874	0.874	0.864	0.884

Notes: SVM, support vector machine; RF, random forest; XGBoost, extreme gradient boosting.