In Reply Dr Munafò and colleagues suggest that survival bias may account for our findings of attenuated associations between genotype and phenotype among cohorts born earlier in the century.1 Although we agree this is mathematically possible, the magnitude of influence of selective survival in our study was probably minimal. Unlike the age-stratified analyses reported by Taylor and Munafò,1 our analyses relied on a mixed model estimated in data with successive enrollments of more recent birth cohorts. Because of the HRS design, we were able to control for age, and we estimated genotype-phenotype associations for members of different birth cohorts at the same age (65 years for white participants). This eliminates most of the potential for survival bias, but a bias could still occur under 2 causal structures. In one scenario (in which genotype influences BMI and both BMI and cohort influence survival, so that among survivors the genotype-BMI association is altered), the magnitude of this bias would depend on the differential survival to the same age of different birth cohorts, and whether that differential was due to BMI.
Glymour MM, Walter S, Mejía-Guevara I. Association of a Genetic Risk Score With Body Mass Index—Reply. JAMA. 2016;316(17):1826-1827. doi:10.1001/jama.2016.14939