To optimize statistical power, rather than conduct a traditional family-based association analysis on the related sets of individuals, we opted for a more flexible linear mixed effects model (Bates et al., 2014). This approach allowed us to model longitudinal measures of BMI (Waves II, III and IV) from all 1,886 individuals while accounting for biological relationships (if present) and within-individual variation in BMI as well as controlling for age, sex and MDS-derived components of ancestry. BMI measures from women who were pregnant were excluded from this analysis. Note that only 5 respondents did not have BMI measures across all three waves.
