Maenner et al. [2009] used a case-only study design and the random forests (RF) algorithm, a type of machine learning, to identify SNPs that may be involved in gene×smoking interactions related to the age at onset of CHD. They used data from the Original and Offspring Cohorts. After ranking the covariate importance score in each of four runs of RF using 500 trees each, one SNP (rs2011345) ranked as the most important SNP and was within the top ten of all ranked covariates in three of the four runs. Using generalized estimating equations to adjust for sex and account for familial correlation, there was significant evidence of a main effect for both the SNP and smoking status, as well as significant evidence of an interaction between the two.
