For the interaction analysis of variants in the CHRNA5/A3/B4 region, both logistic regression and generalized multifactor dimensionality reduction (GMDR) analyses [36] were performed for binary traits (SI-1 and SC), whereas the cumulative logit model was applied to the ordinal phenotypes (SI-2 and SQ). The logistic regression model and cumulative logit model included the main effects of SNPs, interaction effects among the SNPs, as well as adjusting covariates. The log-likelihood ratio test was performed for testing the joint significance of SNP main and interaction effects. The GMDR analysis was carried out only for the binary traits, as GMDR analysis is not applicable to ordinal traits. Age, sex, and recruiting area were considered as adjusting covariates for all interaction analyses. The best SNP combination for all two- to six-locus models was selected via the prediction rate.
