What sample size would permit estimation of a score with AUC at a clinical useful level, or otherwise close to its maximum value? The answer depends on , the proportion of null markers in the panel, because if this is high then the individual marker effects will also be high and a low P-value threshold will eliminate much sampling error from the estimated score. Figure 3 shows AUC as a function of sample size for Crohn's disease, which has a high heritability of 76%, and breast cancer, which has low heritability of 44% [21], based on a panel of 100,000 independent markers. This is a similar number to current genotyping products, and results are given under a scenario in which the panel explains half the heritability [30]. For each sample size and , the P-value threshold is applied that leads to the highest AUC. An AUC of 0.75 is generally regarded as the minimum useful level for screening subjects already considered at risk, whereas AUC of 0.99 is sufficient for screening the population at large [31]. For these two diseases
