DrABC: deep learning accurately predicts germline pathogenic mutation status in breast cancer patients based on phenotype data

Table 2 The prediction accuracy of the algorithms and NCCN criteria in multi-center validation cohort

	DrABC	BRCAPRO^a	BOADICEA^a	Myriad^a	PENNII^b	NCCN	NCCN expansion^c
BRCA1/2
AUC (95%CI)	0.792 (0.735–0.848)	0.699 (0.635–0.763)	0.586 (0.521–0.651)	0.587 (0.537–0.637)	0.628 (0.560–0.697)	NA	NA
Sensitivity	82.1%	53.8%	15.4%	9.0%	61.5%	83.3%	100%
Specificity	63.1%	72.1%	90.2%	98.9%	61.6%	31.4%	2.5%
Youden Index^d	45.2%	25.9%	5.6%	7.9%	23.1%	14.7%	2.5%
All cancer predisposition genes
AUC (95%CI)	0.737 (0.687–0.787)	0.650 (0.589–0.711)	0.571 (0.510–0.631)	0.556 (0.508–0.603)	0.606 (0.543–0.668)	NA	NA
Sensitivity	83.8%	45.5%	15.2%	7.1%	58.6%	78.8%	100%
Specificity	51.3%	71.6%	90.3%	98.9%	61.9%	31.2%	2.5%
Youden Index^d	35.1%	17.1%	5.5%	6.0%	20.5%	10.0%	2.5%

^aThe cutoff values were set as 5%
^bThe cutoff values were set as 10%
^cExpansion of the NCCN criteria included all women diagnosed with breast cancer younger than 65 years of age
^dThe Youden index was calculated as J = sensitivity+specificity-1
Abbreviations: AUC area under the curve, CI confidence interval, NA not applicable

ISSN: 1756-994X