Does a lifetime of activity slow ageing and death?
In 22,750 twins, active classes had at most 7% lower death risk than sedentary after adjustment; in 1,153 of them, the top class's clock read 1.2-1.6 y older than the two middle classes. Self-report, cohort.
Main caveat: Self-report, with changed questions in 1990 and many participants missing at least one survey.
Four classes came from activity in three surveys: sedentary, moderately active, active and highly active. Follow-up started in 1990, so everyone was alive then. Share of participants who died ran 38.8%, 30.8%, 29.0% and 25.4% across the classes. Adjusting for education, smoking and alcohol cut the advantage to at most 9%, and BMI to at most 7%.
The authors report the largest association near the guideline minimum, with no added gain above it. Since clocks showed no beneficial association, biological ageing was not modelled as a mediator.
Protocol
Exposure: MET index in h/day from questionnaires in 1975, 1981 and 1990, with changed wording in 1990. The guideline minimum corresponds to about 1.1 MET h/day.
Mortality models: discrete-time survival; short term 1990-2011, long term 2012-2020. Confounders came from 1981.
Clocks: PC-based GrimAge and DunedinPACE from blood taken 1993-2020, mean age 63.5.
Family control: within-pair models for all, monozygotic and dizygotic pairs.
Funding: Research Council of Finland, EC, NIH and Finnish foundations; no conflicts declared.
Key Results
Deaths: 6,949 (30.6%), 1991-2020.
Mortality, active classes vs sedentary: 16-24% lower with minimal adjustment; at most 9%, then at most 7%, with further adjustment. Exact class estimates are in figures only.
Short vs long term: associations were more consistent in the shorter term; high activity went with lower risk only there.
Within pairs, all participants: moderately active and active classes stayed lower than sedentary after lifestyle adjustment, including within monozygotic pairs. Excluding prevalent cardiovascular disease: associations weakened further. Within pairs, differences were no longer significant after adjustment, except moderately active vs sedentary within monozygotic pairs, which stayed significant.
Clocks: U-shaped before lifestyle adjustment. After adjustment, on GrimAge only, the highly active class stayed 1.2 years (95% CI 0.2 to 2.2) above moderately active and 1.6 years (0.6 to 2.7) above active.
Clock against death: per SD, GrimAge acceleration hOR 1.49 (1.29 to 1.74), DunedinPACE 1.31 (1.13 to 1.52).
Abstract against full text: the abstract concludes genetic confounding and reverse causality; the paper's conclusion leans on healthy phenotype and overall lifestyle, and within-monozygotic differences did not vanish entirely. Within-pair figures are not printed in the text read, so the claim on genes cannot be sized here.
Reporting: after adjustment for activity, DunedinPACE is printed as hOR 1.31, CI 1.35-1.52, which excludes its own estimate and is likely a slip for 1.13. Subsample smoking counts add to 1,085 against a stated 1,153 (our arithmetic).
Mechanisms
Measured: clock acceleration; DNA-methylation surrogates of cystatin C and beta-2-microglobulin were higher in the sedentary and highly active classes.
Authors' interpretation: the highly active class had higher methylation-estimated smoking despite adjustment for reported smoking, which may be under-reporting. GrimAge contains a smoking component, so residual smoking could explain part of the gap (inference).
Authors' speculation: a possible link with sudden cardiac death at high activity, which is untested here.
Practical Application
Does a lifetime of activity slow ageing and death? In 22,750 twins, active classes had at most 7% lower death risk than sedentary after adjustment; in 1,153 of them, the top class's clock read 1.2-1.6 y older than the two middle classes. Self-report, cohort.
Good for: showing why a large raw gap shrinks. Smoking measured in amount, alcohol, education and BMI accounted for most of it here.
Not good for: "high training ages you". A marker in a mostly female subsample with possible smoking under-reporting is no outcome, and the highly active had the lowest share of deaths.
Not good for: lifelong training claims. Activity stops being measured in 1990, and deaths run to 2020; the authors note high activity may need maintaining.
Limitations
Self-report, with changed questions in 1990 and many participants missing at least one survey.
Exposure ends in 1990, and confounders are from 1981; later changes are not captured.
Clock subsample: 80.9% women, fewer smokers, not a random slice.
Non-shared confounders: diet and occupation can differ inside pairs, as the authors note.
Over-adjustment possible: BMI may mediate rather than confound (inference; the paper raises mediation for lifestyle factors).
How this entry was checked and changed
Why this level
large observational cohort; within-pair models control family factors but remain observational, and the clock outcome is a marker