Fig. 2

Overview of simulation study. Simulations were based on longitudinal ARIES cohort data [15] available at ages 7, 9, and 15–17. Epigenetic age (EA) was calculated using the Horvath clock [3] or GrimAge [9] in separate simulations. The original EA measure was then altered based on a simulated exposure. In each binary exposure simulation, a random n = 100 individuals had their original EA increased by 2 years (fixed effect), which accumulated by 0.1 year of EA per year of life (interaction effect). In each continuous exposure, all individuals were assigned a value (N(3.5, 0.5²)) which impacted their original EA by 0.1 years, times the level of exposure (fixed effect), and caused an interaction between the exposure and age by 0.02 (interaction effect). In the next step of our simulation, a series of methods was applied to model the simulated effects. Models are linear mixed effect models (LME), generalized estimating equations (GEE), and regression on difference between two epigenetic age (EA) measures (Δ aging). Outcome variables included are epigenetic age acceleration (EAA, residual from regressing EA on age), or EA itself. We ran n = 1000 simulations for each exposure type (binary vs. continuous), epigenetic clock (Horvath vs. GrimAge2), and different scopes of data (two measures: Age 7 and 15–17 vs. three measures: Age 7, 9, and 15–17)