A time-to-event heritability framework for inferring the genetic architecture of longitudinal traits

Biobanks with longitudinal measurements have advanced our understanding of time-to-event (TTE) traits including age-of-onset and disease progression. However, limited work has characterized the heritability of TTE traits, a key parameter for comparisons of total association and predictive power. Here, we present COXMM, a Cox proportional hazard mixed model for estimating TTE heritability. Simulations show our model achieves nearly unbiased results, whereas non-TTE approaches severely underestimate TTE heritability. COXMM estimates also predict the expected accuracy of polygenic scores in survival analyses, informing study design. We analyzed a wide variety of traits and observed heritability patterns reflecting a mixture of TTE and case-control architectures with supporting GWAS and polygenic risk score analyses. Evaluating 18 trait pairs, progression to a severe condition has consistently lower heritability than all-cause incidence, suggesting that cause-specific progression may have stronger environmental influences. COXMM offers a novel framework for analyzing longitudinal disease architecture with implications for disease prediction.