Design and model choices shape inference of age-varying genetic effects on complex traits

Understanding how genetic influences on complex traits change with age is a fundamental question in genetic epidemiology. Both cross-sectional (between-subject) and longitudinal (within-subject) approaches can contribute to answering this question but come with distinct strengths and limitations. Here we show that age-varying genetic effects obtained from the two designs were highly concordant in direction (84.21% of the 57 identified variants) but showed only moderate agreement in effect-size magnitude (Pearsonʼs $$r=$$ r = 0.51). Confounding by gene-by-birth year effects accounted for the largest proportion of variance in effect-size differences across single-nucleotide polymorphisms (SNPs) with age-varying effects between designs (70.8%). Participation bias accounted for an additional 11.6%, whereas unmodeled nonlinear age trajectories contributed minimally to these differences (4.2%). Overall, our results demonstrate that both cross-sectional and longitudinal designs can yield different estimates of age-varying genetic effects, principally due to cohort confounding and participation bias. As neither approach is immune to design-specific limitations, we recommend integrating both designs for robust inference, to help improve interpretability and more accurately characterize how genetic effects on complex traits change over the lifespan.