Aging is a complex and multifactorial process characterized by various phenotypic changes, including epigenetic modifications, inflammation, and metabolic dysregulation. In recent years, DNA methylation (DNAm) clocks have emerged as valuable tools for estimating biological age and predicting age-associated outcomes based on DNAm profiles.
However, most existing DNAm clocks have relied on linear models, which may not adequately capture the heterogeneity and non-linearity of survival-related DNAm signatures. To address this limitation, the authors of this study constructed a heterogeneous stacked ensemble survival model using DNAm data from the Framingham Heart Study.
The researchers identified 190 CpG loci through elastic net Cox regression and developed a survival prediction model by integrating five complementary survival models using a neural network meta-learner. This innovative approach aimed to enhance the prediction power of mortality risk based on biological age.
The performance of the survival model was rigorously evaluated in an external validation cohort, demonstrating strong predictive capabilities for all-cause mortality. Notably, the model’s performance significantly surpassed that of PhenoAge and was statistically comparable to GrimAge, highlighting its potential utility in assessing mortality risk.
It is important to note that the performance estimates derived from this study were based on cohorts of European ancestry and were externally validated in postmenopausal women aged 50-79 years. Therefore, the applicability of these findings may be limited to demographically similar populations.
What this study does not show
This study does not establish the clinical utility of the DNA methylation clock for predicting health outcomes beyond mortality risk, nor does it provide evidence for its effectiveness across diverse populations.
Evidence scorecard
- Analytic validity: Established
- Clinical validity: Partial
- Clinical utility: Not established
The evidence is based on primary research involving human participants.