As a final example of association testing, this time with a quantitative trait, Simonson et al studied the Framingham Risk Score for cardiovascular disease risk [8]. They also used 10-fold cross-validation of a single sample, giving training samples of 1575 subjects and testing samples of 175 subjects. They used a full set of 250,378 SNPs, which is here assumed to be similar to 100,000 independent SNPs. They first selected SNPs with P-values <0.1 into the score, then selected SNPs with 0.1<P<0.2, then 0.2<P<0.3 and so on, giving ten analyses. Even if the trait is fully heritable and explained by these markers, this analysis has 20% power for the SNPs with P<0.1, reducing with each P-value interval. For 0.4<P<0.5, in which the authors found nominal significance of the score, the power is 6.7%. If all SNPs are included in the score, the power would be 38% if the trait is fully explained by the markers, but under a more conservative model in which the explained genetic variance is 30%, the power is just 8% and increases to 13% under an even
