Dr. Fernanda Schumacher, College of Public Health, Division of Biostatistics

Dr. Kellie Archer, College of Public Health, Division of Biostatistics

SUMMARY

Epigenetic “clock” algorithms are increasingly used to quantify biological aging and examine population-level variation in health, mortality, and social determinants of health. However, nearly all studies treat predicted epigenetic ages as error-free measurements, despite their generation from high-dimensional machine learning models that introduce substantial uncertainty. Ignoring prediction uncertainty can distort subgroup comparisons and inflate false-positive findings, undermining the reliability of research on population-health disparities. This project develops Bayesian methods to quantify individual-level prediction uncertainty for DNA-methylation (DNAm)–based aging measures and to propagate this uncertainty into population-level subgroup comparisons. We will first establish feasibility using the Hannum whole-blood methylation dataset and then demonstrate uncertainty propagated subgroup comparisons in a population-health dataset that integrates DNAm with psychosocial, demographic, or stress-related characteristics. The specific demonstration dataset and application area will be selected in partnership with IPR mentors to ensure strong alignment with NICHD Population Dynamics Branch priorities. This seed project provides the necessary proof-of-concept and preliminary results to support a future R01 that will extend the Bayesian framework to more complex biological aging metrics, including multivariate aging measures (e.g., components of GrimAge and PhenoAge) and longitudinal indicators such as Pace of Aging, to advance population-health inference from DNAm-based aging tools.