Weak Correlation Between DNA Methylation Clocks and Frailty

It is with some excitement that I started reading the @The Lancet Healthy Longevity article "Biological age measured by DNA methylation clocks and frailty: a systematic review and meta-analysis" by a National University of Singapore team led by @Andrea B. Maier last week (link)00092-3/fulltext?rss=yes). If methylation clocks correlate convincingly with indices such as frailty, which measures susceptibility to adverse outcomes and deficit accumulation across multiple health domains, there is a potential to predict, prevent and manage decline with easily accessible tests.

Alas, the correlations are statistically significant but clinically modest. GrimAge (and to a lesser extent PhenoAge and pace-of-aging clocks) clearly track frailty on average, but the effect sizes are small enough that, right now, they’re not ready to be used as stand-alone clinical biomarkers to diagnose or screen for frailty in individual patients. They’re “promising research biomarkers,” not “order this test in clinic” biomarkers yet.

Standardized β coefficients for cross-sectional associations, frailty vs epigenetic age acceleration (EAA):

Hannum EAA: β ≈ 0.06

PhenoAge EAA: β ≈ 0.07

GrimAge EAA: β ≈ 0.11

Pace of aging (DunedinPoAm / DunedinPACE): β ≈ 0.10

A 1-SD increase in EAA is associated with roughly 0.06–0.11 SD higher frailty. In plain language: people whose clocks say they’re biologically “older than expected” tend to be a bit frailer, but the relationship is very small. There’s also substantial heterogeneity (I2 70–90%), and evidence of publication bias, which both suggest the true average association might be smaller. For change in frailty over time, or longitudinal associations, the numbers are even smaller, with GrimAge EAA the only clock (barely) below the significance threshold.

A clinically useful biomarker for frailty would ideally strongly discriminate frail vs non-frail or robustly predict who will become frail, beyond age, sex, co-morbidities, and simple clinical measures (gait speed, grip strength, FI), and do so consistently across cohorts with clear thresholds that clinicians can use. A β ≈ 0.1 means that even a big shift in epigenetic age might only move their frailty by a tiny amount. At population level that’s real, but at the individual level it’s not huge. In terms of individual-level clinical decision-making (e.g., “Should I classify this person as frail / pre-frail based on their GrimAge?”), these effect sizes are too small and too noisy to rely on alone. But to track risk across populations, detect group-level differences, or response to interventions, it is a useful research biomarker.