Our power analysis also indicates that these results have important implications for the conduct of GWAS with respect to sample size. Considerably, smaller sample sizes may be sufficient to detect robust genome-wide associations if these use better outcome measures—there is a trade-off between sample size and phenotype quality and/or precision. It may not always be practical or financially possible to collect such phenotypes in large samples. An initial GWAS based on a preliminary phenotype, which is easy to collect (such as cigarette consumption), can be followed up by high-quality/precision phenotyping (such as cotinine measurements) in a sample selected by genotype on which the stored biological specimens are available. This combination could lead to considerable increases in statistical power and efficiency. In conventional observational epidemiology, where associations between self-reported smoking behavior and outcomes underestimate the actual etiological associations that exist (52), more detailed phenotyping reveals stronger associations (46). Residual associations between other exposures and smoking-related outcomes, such as lung cancer, which persist after statistical control for self-reported cigarette consumption, should therefore be treated with caution (53).
