The case-only independence assumption: associations between genetic polymorphisms and smoking among controls in two...

The independence assumption for a case-only analysis of statistical interaction, i. e. that genetic (G) and environmental exposures (E) are not associated in the source population, is often checked in surrogate populations. Few studies have examined G-E association in empirical data, particularly in controls from population-based studies, the type of controls expected to provide the most valid surrogate estimates of G-E association. We used controls from two population-based case-control studies to evaluate G-E independence for 43 selected genetic polymorphisms and smoking behavior. The odds ratio (OR(z)) was used to estimate G-E association and, therefore, the magnitude of bias introduced into the case-only odds ratio (COR). Odds ratios of moderate magnitude [mmOR(z)], defined as OR(z)≤0.7 or OR(z)≥1.4, were found at least one of the six smoking measures (ever, former, current, cig/day, years smoked, pack-years) for 45% and 59% of the SNPs examined in the control groups of two independently conducted North Carolina studies, respectively. Consequently, case-only estimates of G-E interaction in the context of a multiplicative benchmark would be biased for these SNPs and smoking measures. MmOR(z)s were found more often for smoking amount than smoking status. We recommend that a stand-alone case-only study should only be conducted when G-E independence can be verified for each polymorphism and exposure metric with population-specific data. Our results suggest that OR(z) is specific to each underlying population rather than an estimate of a 'universal' OR(z) for that SNP and smoking measure. Further, misspecification of smoking is likely to introduce bias into the COR.