Influence of race, ethnicity, and sex on the performance of epigenetic predictors of phenotypic traits.

Abstract

Background: DNA methylation-based predictors of phenotypic traits including leukocyte proportions, smoking activity, biological aging, and circulating levels of plasma proteins are widely used as biomarkers in public health research. However, limited racial and ethnic diversity of research participants is an ongoing issue for epigenetics research, and the potential downstream impacts of limited diversity in training samples on the performance of epigenetic predictors remains poorly understood. We examined the performance of epigenetic predictors of chronological age (also known as epigenetic clocks), telomere length, cell proportions, and plasma proteins within a diverse sample of adult NHANES participants during the 1999-2000 and 2001-2002 survey cycles, both overall and stratified by self-reported race/ethnicity and sex. We utilized correlation coefficients and median absolute errors (MAE) to judge predictor performance, and bootstrapping and multivariate regression to assess the significance of differences between groups.

Results: All epigenetic predictors were significantly associated with their corresponding phenotypic traits in the overall population, with particularly high correlations for the epigenetic clocks and cell proportion estimates. Several significant differences in performance were observed between racial/ethnic groups, particularly for the plasma protein predictors, with a reoccurring trend of lower correlation in Mexican American and non-Hispanic Black participants compared to non-Hispanic White participants. Sex-differences in performance for several predictors were also identified but were not as pronounced. Multivariate regression models indicated that disparities in epigenetic predictor performance persisted after accounting for overall differences in epigenetic predictions related to race/ethnicity and sex, as well as further adjustment for estimated cell proportions and SES variables.

Conclusions: We found evidence for substantial disparities in epigenetic predictor performance, with each predictor exhibiting at least one significant difference in correlation or MAE related to race, ethnicity, or sex.