For some diseases, fairly high AUC has already been observed [17], [19]. This does not conflict with the present work but reflects the presence of major gene effects, usually in the MHC, which depart from the quantitative model treated here. Similarly, some diseases have non-genetic risk factors that already admit clinically useful predictors. There the more relevant issue is the extent to which genetics improves established models [34]. Again the focus has tended to lie on identifying specific markers to improve prediction, rather than the sample size needed to accurately estimate their combined effects. The approach taken here could easily be extended to accommodate additional fixed effects.
