We evaluated the associations of FTO SNPs on BMI after stratification of the sample according to “high” and “low” physical activity strata. Because the daily number of physical activity accelerometer counts was generally lower in women than in men and decreased with age in both sexes (data not shown), this stratification was performed in a sex- and age-specific fashion. After logarithmic transformation of mean total physical activity counts, subjects were dichotomized into the high- or low-activity stratum depending on whether their age-, age2- and sex-specific residuals were greater than or less than 0. In addition to these stratified analyses, we determined whether the association of genotype on BMI was modified by physical activity levels by constructing a regression model that included the following independent variables: sex, age, age2, age×sex, age2×sex, SNP genotype, ln-transformed physical activity counts, and an interaction term (ln-transformed physical activity counts×genotype). The presence of an interaction between physical activity and SNP genotype on BMI was assessed by a likelihood ratio test, in which we compared a model with the interaction term (full model) with a model without
