Independent Female-Cohort Recalibration of a 10-CpG Saliva Epigenetic Clock: Age-Adjusted Phenotype Associations

Background/Objectives: A previously reported saliva epigenetic clock identified 10 age-associated CpGs in a mixed-sex cohort. We evaluated the prespecified marker set in an independent female cohort with no participant overlap with the original study, recalibrated the coefficients, and tested questionnaire associations before and after direct chronological-age adjustment. Methods: The study included 898 females aged 12.15–85.62 years. Female recalibration was evaluated by shuffled five-fold cross-validation. Primary age-adjusted phenotype associations were estimated from joint HC3-robust regression models containing the phenotype and a prespecified cubic B-spline of chronological age; residual DNAm age was retained as a sensitivity analysis. Results: Across the original five folds, Pearson r was 0.908 (participant-bootstrap 95% CI 0.890–0.920), predictive R2 was 0.824 (0.791–0.847), MAE was 4.05 years (3.83–4.34) and RMSE was 5.27 years (4.95–5.71). Aggregate EpiDISH deconvolution was available for 896 participants; mean estimated epithelial, fibroblast/stromal and immune fractions were 24.2%, 2.7% and 73.1%, respectively, with weak linear correlations with chronological age (|r| ≤ 0.059). Among 23 broad questionnaire phenotypes, reported medical conditions and eyesight ratings were associated with unadjusted DNAm-predicted age after FDR correction, but no broad phenotype remained significant in the primary joint age-adjusted models. Sleeping disturbance/insomnia showed a nominal joint age-adjusted association of +1.41 years (95% CI 0.39–2.43; p = 0.0068; q = 0.240), which did not survive FDR correction. Conclusions: The previously selected marker set retains substantial chronological-age information in an independent female saliva cohort. Questionnaire associations with raw chronological-age clocks can be strongly confounded by chronological age; the present phenotype data provide insufficient evidence for robust associations after age adjustment. Cell-composition results remain descriptive rather than cell-adjusted clock validation.