Use of Correct and Incorrect Methods of Accounting for Age in Studies of Epigenetic Accelerated Aging: Implications and Recommendations for Best Practices

Abstract

Motivated by our conduct of a literature review on social exposures and accelerated aging as measured by a growing number of epigenetic “clocks” (which estimate age via DNA methylation patterns (DNAm)), we report on three different approaches – 1 incorrect and 2 correct – in the epidemiologic literature on treatment of age in these and other studies using other common exposures (i.e., body mass index and alcohol consumption). Among the 50 empirical articles reviewed, the majority (n = 29; 58%) used the incorrect method of analyzing accelerated aging detrended for age as the outcome and did not control for age as a covariate. By contrast, only 42% used the correct methods, which are either to analyze accelerated aging detrended for age as the outcome and control for age as a covariate (n = 16; 32%), or to analyze raw DNAm age as the outcome and control for age as a covariate (n = 5; 10%). In accord with prior demonstrations of bias introduced by the incorrect approach, we provide simulation analyses and additional empirical analyses to illustrate how the incorrect method can lead to bias to the null, and we discuss implications for extant research and recommendations for best practices.

Maternal childhood adversity and infant epigenetic aging: Moderation by restless sleep during pregnancy

“…Study cohort was not included as a covariate since the two cohorts in this study reflected largely separate racial and ethnic groups; cohort was moderately correlated with the race and ethnicity variables (i.e., r > .30), and we did not want to adjust for additional proxy variables of race or ethnicity. Lastly, we adjusted for gestational age at birth in each model, following best practices when predicting epigenetic age deviation (Krieger et al., 2023).…”

Negative Life Events and Epigenetic Ageing: a Study in the Netherlands Twin Register

“…For each of these analyses, we considered three models. In the primary model, we corrected for chronological age (as previously recommended by Krieger et al, 2023), sex, and technical covariates only. Additional models were fitted to examine whether the associations between negative life events and epigenetic aging might be driven by lifestyle or blood cell composition.…”

Depressive symptoms are associated with DNA methylation age acceleration in a cross-sectional analysis of adults over age 50 in the United States

“…GrimAge is trained on healthrelated and disease-related indicators like sex, education, race, and smoking and is therefore likely to be a stronger and more sensitive predictor for all-causes mortality, age-related health status and biological aging in psychiatric conditions [13][14][15] . To create the age acceleration values, we regressed the DNA methylation age estimate against chronological age and captured the residuals from this model 13,23 . We modelled these residuals as continuous in linear regressions and for bivariate analyses and we dichotomized the residuals into accelerated aging (residuals > 0) or no age acceleration (residuals ≤ 0).…”