Table 1.

CardioBoost outperforms existing genome-wide tools for the classification of holdout test variants.

%	Cardiomyopathies			Arrhythmias
	CardioBoost	M-CAP	REVEL	CardioBoost	M-CAP	REVEL
Overall accuracy	63.3a	28.4	17.4	81.2a	30.5	37
Proportion of variants classified with high confidence	70.2a	33.9	22	88.3a	33.8	40.3
Accuracy of high-confidence classifications	90.2	83.8	79.2	91.9	90.4	91.9
Proportion of variants with indeterminate classification	29.8a	66.1	78	11.7a	66.2	59.7
TPR	69.5a	41.5	28	83.3a	48.8	65.5
PPV	86.3	81.7	76.7	90.9	91.1	91.7
TNR	56a	13	5	78.6a	8.6	2.9
NPV	96.6	92.9	100	93.2	85.7	100

The performance of each tool is reported using the clinically relevant variant classification thresholds: high-confidence disease-causing (Pr ≥ 0.9), high-confidence benign (Pr ≤ 0.1), and indeterminate. For each predictive performance measure (see Supplementary Methods for details) the best algorithm is highlighted in bold. Permutation tests were performed to evaluate whether the performance of CardioBoost was significantly different from the best value obtained by M-CAP or REVEL.

NPV negative predictive value, PPV positive predictive value, TNR true negative rate, TPR true positive rate.

^aP value ≤ 0.001.