Can Epigenetic Clocks Flag Frailty Before It Starts?

An epigenetic clock may detect a faster path toward frailty years before someone looks frail in the clinic. That is the intriguing result from a Swedish twin study: DunedinPACE, a measure of aging speed, predicted later frailty-index change beginning at age 50. But the result comes from one cohort, and a 2025 meta-analysis did not reproduce DunedinPACE's longitudinal association when it pooled studies. The clock may be an early warning signal. It is not yet a clinical forecast.

Why the type of clock matters

"Epigenetic clock" is an umbrella term for tests that use DNA methylation patterns to estimate aspects of biological aging. They do not all measure the same thing.

First-generation clocks, such as HorvathAge and HannumAge, were trained to predict chronological age. A person can have a methylation age older or younger than expected, but that number mainly describes deviation from an age-based norm. Second-generation clocks add clinical information. GrimAge incorporates methylation surrogates for plasma proteins, including GDF-15, PAI-1, and adrenomedullin, plus a smoking-pack-year signature. It therefore captures signals closer to cumulative physiological damage.

DunedinPACE takes another approach. It was built from longitudinal change in 19 biomarkers spanning cardiovascular, metabolic, kidney, liver, immune, dental, pulmonary, and other systems among people of the same chronological age. Its output aims to estimate the current pace of biological change, rather than accumulated wear. That distinction makes it an appealing candidate for identifying a window when prevention might still matter.

The midlife signal comes from one unusually detailed cohort

In a 2024 study of the Swedish Adoption/Twin Study of Aging, researchers followed 524 monozygotic and dizygotic twins, ages 50 to 90, for as long as 20 years. They used bivariate dual change score models across two-year age bins. This modeling approach asks a more specific question than a simple correlation: does a clock value at one assessment predict the change in frailty at the next?

Only DunedinPACE showed that one-way temporal relationship. A higher DunedinPACE at one wave predicted greater subsequent change in the frailty index; prior frailty did not predict a later change in DunedinPACE. PC-corrected HorvathAge, HannumAge, PhenoAge, and GrimAge did not show equivalent dynamic coupling in this analysis.

The frailty index accelerated nonlinearly across ages 50 to 90, while the DunedinPACE coupling was detectable at the model's age-50 intercept. That places the proposed signal roughly 15 to 20 years before frailty commonly becomes clinically apparent in the seventh decade. It is a compelling hypothesis: aging speed may separate people onto different frailty trajectories during midlife, before disability makes the difference obvious.

The study's design strengthens the observation. Repeated measurements and a model of within-person change can reveal timing that a one-time blood draw cannot. Yet its scope also limits the claim. The cohort was small, exclusively European, and composed of twins. A twin design partly controls genetic variation, but it may not reflect environmental variation in the wider population. No comparable dual change score analysis has yet confirmed the DunedinPACE-to-frailty pattern in a diverse, non-twin cohort.

The broader literature supports clocks and frailty, but not a DunedinPACE verdict

A 2025 systematic review and meta-analysis of 24 studies involving 28,325 people provides the necessary reality check. Across cross-sectional analyses, second-generation clocks related to frailty more consistently than raw methylation age or simple age deviation. GrimAge age acceleration had the largest pooled association with frailty status: β=0.11 (95% CI 0.06 to 0.15; 10,371 participants). PhenoAge acceleration and Hannum age acceleration had smaller significant associations, β=0.07 and β=0.06, respectively.

DunedinPACE had a numerically similar cross-sectional estimate, β=0.10 (95% CI 0.01 to 0.19; 7,895 participants), but its heterogeneity was extreme: I²=91.0%. That means the studies differed so much that the pooled number is a shaky summary. It does not tell you what a DunedinPACE result means for a particular person or population.

Longitudinally, the tension becomes sharper. Across five studies totaling 6,143 participants, DunedinPACE was not significantly associated with frailty change. Only GrimAge age acceleration reached statistical significance, with β=0.02 (95% CI 0.00 to 0.05; p=0.0481). That result barely excludes zero and has modest clinical meaning even if it holds up. The authors concluded that no available epigenetic clock is adequately validated for frailty-specific clinical use.

Why might the Swedish study and the meta-analysis disagree? The Swedish analysis modeled wave-by-wave coupling from midlife through old age. The pooled studies used heterogeneous populations, follow-up periods, frailty measures, and analytical approaches. It is plausible that a dynamic midlife signal would disappear in shorter or later-life studies. It is equally plausible that the Swedish result is cohort-specific. No published head-to-head comparison has settled that question.

Other evidence keeps GrimAge in the conversation. An Irish cohort study found GrimAge outperformed HorvathAge, HannumAge, and PhenoAge for age-related clinical phenotypes and all-cause mortality. The ESTHER cohort also linked GrimAge and frailty measures to mortality in a German general-population sample. These studies do not prove GrimAge predicts an individual's future frailty, but they give it a broader replication record as a snapshot of accumulated risk.

Can lifestyle change the clock?

There is randomized evidence that DunedinPACE can move. In the two-year CALERIE randomized trial, healthy, non-obese adults assigned to 25% caloric restriction had significant DunedinPACE deceleration compared with ad libitum controls. This showed a dietary intervention can affect the biomarker. The trial did not test whether that change prevented frailty, disability, or other patient-important outcomes.

Physical activity also has a favorable observational signal. A 2025 Health and Retirement Study analysis found that higher habitual physical activity was associated with lower DunedinPACE, PhenoAge, and GrimAge in a large U.S. cohort. Association is not proof that exercise caused the clock differences; healthier people can be more able to exercise. Still, physical activity needs no epigenetic-sales pitch to justify itself. Its health benefits already stand on firmer ground than any commercial clock interpretation.

The 2025 MACRO randomized trial found that weight-loss dietary interventions affected epigenetic aging measures in obese adults, extending the general modifiability signal beyond CALERIE's healthy, non-obese participants. The available report does not provide DunedinPACE-specific effect sizes or frailty outcomes, so it cannot fill the gap between a changing biomarker and prevented frailty.

What a test result can—and cannot—tell you

Commercial testing has outrun clinical validation. This evidence base does not establish validated clinical reference ranges, test-retest reliability, minimum retest intervals, or inter-laboratory reproducibility for commercial DunedinPACE products. A single result is not a diagnosis, and it should not dictate treatment.

Baseline readings may also reflect life history as well as current physiology. A 2026 systematic review and meta-analysis found that adverse childhood experiences were associated with epigenetic age acceleration across the clocks studied. The available abstract does not confirm a DunedinPACE-specific result, but it raises a reasonable caution: early adversity may complicate interpretation of an elevated clock reading. Sex-specific frailty coupling is also unknown; the Swedish study adjusted for sex but did not report sex-stratified coupling estimates.

The practical verdict

GrimAge has the stronger cross-cohort record as a correlate of frailty and age-related risk. DunedinPACE has the more interesting midlife hypothesis and randomized evidence that caloric restriction can slow the biomarker. Neither fact establishes that changing a clock score prevents frailty.

If you test, treat DunedinPACE as optional context and a possible motivator to measure the basics that already matter: physical activity, strength, body composition, cardiometabolic risk, diet quality, sleep, and functional capacity. For people in their 40s through 60s, the Swedish finding gives a rational reason to watch aging-related risk before symptoms arrive. It does not justify treating an epigenetic result as a medical verdict. The next useful evidence would be long, diverse, non-twin studies that test the same dynamic model—and then show whether a slower clock predicts less frailty in real lives.