. Author manuscript; available in PMC: 2015 Jul 1.

Published in final edited form as: Addict Biol. 2013 Jan 30;19(4):708–721. doi: 10.1111/adb.12035

Table 6.

Results of SNP subsets from varying P-value thresholds

P-value threshold	Mean AUC	95% Confidence Interval		Median p-value for AUC
P-value threshold	Mean AUC	Lower	Upper	Median p-value for AUC
Pt < 0.50	0.565	0.562	0.568	1.37E-05
Pt < 0.40	0.565	0.562	0.568	1.42E-05
Pt < 0.30	0.564	0.561	0.567	1.82E-05
Pt < 0.20	0.564	0.561	0.567	2.62E-05
Pt < 0.10	0.562	0.559	0.565	4.81E-05
Pt < 0.05	0.559	0.556	0.562	1.04E-04
Pt < 0.01	0.549	0.546	0.552	0.00166
Pt < 0.001	0.528	0.526	0.531	0.0631
Pt < 0.0001	0.517	0.515	0.519	0.29

Summary statistics for 100 random 50% splits of the combined COGA-SAGE sample into discovery samples and validation samples. Sum scores were created based on SNPs meeting each p-value threshold, by adding minor alleles weighted by the log of the odds ratio for AD. Confidence intervals are based on 100 AUC estimates from 100 separate sum score calculations at each p-value threshold. Median p-value threshold was calculated because distributions of p-values were skewed.