Figure 1

Application of CART to the XPD*IL10 interaction. CART sequentially partitions the data into two homogeneous subsets: first using XPD-[Lys751Gln], {AA} versus {AC, CC}; and then the {AA} subset is split according to IL10-[G(-1082)A], {AA, AG} versus {GG}. The splitting variables leading to such groups are inherent main or interaction effects. For example, a low-risk subgroup is defined by the two SNPs: XPD-[Lys751Gln] and IL10-[G(-1082)A] and the tree suggests an interaction between these two SNPs. Multi-way interactions can be detected in a similar way. The terminal nodes can be classified as low- or high-risk subgroups (indicated by different color density) and their association with the outcome can be estimated (that is, the corresponding odds ratio is 0.63 with a P-value of 0.002). Therefore, investigating the tree terminal nodes provides a natural way to identify interactions and characterize high- or low-risk subgroups.