Alcohol consumption is associated with increased prediction error in a buccal swab-based DNA methylation age prediction model.
Keskin Nursena N, Ersoy Gokhan G, Sımsek Sumeyye Zulal SZ, Can Kadriye K et al.
DNA methylation (DNAm)-based age estimation is a promising forensic tool, but prediction accuracy may be affected by lifestyle factors such as alcohol consumption. This study evaluated a buccal swab DNAm age prediction model, developed in non-drinking individuals, when applied to individuals with varying drinking patterns. Buccal swabs were collected from 90 healthy non-drinkers (mean age 44.9 years; range 22-93) for model development and from 62 regular alcohol consumers (mean age 32.8 years; range 19-63), classified as light (n = 29), moderate (n = 21), or heavy (n = 12) drinkers per NIAAA criteria, for model evaluation. Stepwise multivariate linear regression across seven candidate age-associated CpG loci identified a 3-CpG model (SLC12A5, ASPA, FHL2) explaining 61% of age variance (adjusted R2 = 0.613; MAE = 7.7 years; r = 0.791, p < 0.001). When applied to the alcohol-consuming cohort, prediction error increased significantly (MAE = 9.9 years; p = 0.046), with light and heavy drinkers showing markedly elevated error while moderate drinkers did not differ from non-drinkers. The association remained significant after adjusting for smoking status and chronological age. An extended model incorporating alcohol consumption as a predictor, retrained on the combined cohort (n = 152), achieved comparable accuracy (adjusted R2 = 0.605; MAE = 7.4 years) and retained alcohol consumption as a significant negative predictor of age (B = -6.84, p < 0.001). At the locus level, ASPA's age association was attenuated in alcohol consumers, whereas FHL2 and SLC12A5 showed no significant alcohol interaction. As lifestyle data are usually unavailable in casework, alcohol-robust CpG sites should be prioritized in forensic model development.