In thinking about the application of such a filter approach in real data, there are a couple of important points to consider. First, while in the simulations studies performed we knew the correct number of SNPs to be found (whether the real disease model was due to single locus effects or interactions), of course in real data this is unknown. Hopefully the results of the current study encourage testing for both single locus and interaction models. As computational capabilities advance, higher-order interactions may be computationally feasible as well. Second, in real data, when a two-locus model is found by MDR (ranked very highly), to really understand how these loci confer risk, post hoc analysis should be considered to better understand the model. A high rank of two loci could be because of two strong main effects, or by interactive effects. A high rank alone does not necessarily indicate an interaction effect, and the development of methods to help dissect the underlying etiology of complex genetic models is an active research area.