Table 1.

Comparison of the performance of PRS models developed using genome-wide approach in AA data: results in the validation set

P-value cut-offa	SNPs entering model (n)	SNPs selected (n)		OR (95% CI)b	AUC (95% CI)b
Overall breast cancer
<10−5	288	62		1.04 (0.99–1.10)	0.509 (0.495–0.524)
<10−4	2053	428		1.03 (0.98–1.09)	0.506 (0.489–0.522)
<10−3	19 067	2351		1.07 (1.01–1.13)	0.521 (0.507–0.535)
<10−2	175 161	10 647		1.12 (1.06–1.18)	0.535 (0.519–0.551)
<0.05	829 335	29 569		1.13 (1.07–1.19)	0.535 (0.519–0.551)
<0.1	1 615 762	46 854		1.15 (1.09–1.22)	0.541 (0.527–0.556)
ER-positive
<10−5	201	79		1.06 (0.99–1.13)	0.517 (0.499–0.536)
<10−4	2026	408		1.04 (0.97–1.12)	0.512 (0.491–0.534)
<10−3	20 186	2339		1.10 (1.03–1.18)	0.529 (0.508–0.550)
<10−2	178 697	10 493		1.19 (1.10–1.27)	0.543 (0.523–0.562)
<0.05	832 622	29 004		1.22 (1.13–1.31)	0.546 (0.527–0.566)
<0.1	1 624 378	45 997		1.22 (1.13–1.31)	0.546 (0.527–0.565)
ER-negative
<10−5	209	50		1.13 (1.04–1.22)	0.531 (0.508–0.554)
<10−4	1872	419		1.08 (0.99–1.17)	0.528 (0.506–0.550)
<10−3	16 751	2230		1.03 (0.95–1.11)	0.506 (0.482–0.531)
<10−2	160 097	10 138		1.14 (1.05–1.23)	0.535 (0.510–0.559)
<0.05	784 928	28 100		1.20 (1.11–1.31)	0.548 (0.525–0.572)
<0.1	1 552 045	44 889		1.23 (1.13–1.33)	0.551 (0.527–0.575)

^aThe P-value cut-off used for selecting SNPs based on their marginal associations with cancer risk and then in stepwise regression in the training set.

^bOR per 1 SD for the PRS. OR for association with breast cancer in the validation set was derived using logistic regression adjusting for age, consortium/study and 10 PCs. AUC of PRSs was calculated under the covariate-adjusted ROC model adjusting for age, consortium/study and 10 PCs of genotype data. PRS models highlighted were used for further analysis.