Telomeres and longevity: the numbers from 121,749 people, and what they do not measure

A 2018 meta-analysis of 25 studies, 121,749 individuals and 21,763 deaths: one standard-deviation shorter leukocyte telomere length carried a hazard ratio of 1.09 for all-cause mortality, and the shortest quarter 26 per cent higher than the longest. A 2017 individual-participant analysis of 12,199 adults found 1.23 for the shortest quintile. The association weakens above 80 years.

goodWell supported, with the usual caveats about effect size.Revised 28 Oct 2026

A single small plain unlabelled clear glass vial with a cork stopper standing upright on a pale grey stone surface beside a folded white cotton cloth and a plain white ceramic cup

Telomere length is the most replicated biomarker in aging research and one of the least actionable. The effect sizes are small, the direction is consistent, and the mechanism does not follow.

Telomere length is probably the most replicated finding in aging research, and it is also one of the ones most often over-read. Here is what the largest analyses actually report, and then what none of them measured.

The 2018 meta-analysis: 121,749 people, 21,763 deaths

Ageing Research Reviews, 2018. A meta-analysis of published cohort studies plus unpublished data from the Swedish Twin Registry, 25 studies in total: four STR cohorts (12,083 individuals, 2,517 deaths) and 21 published studies.

In the twin registry cohorts, one standard-deviation decrement of leukocyte telomere length corresponded to a 13 per cent increased all-cause mortality risk (95 % CI 7 to 19 per cent), and individuals in the shortest quartile had a 44 per cent higher hazard (95 % CI 27 to 63 per cent) than those in the longest.

Pooling everything eligible — 121,749 individuals with 21,763 deaths — the hazard ratio per standard-deviation decrement was 1.09 (95 % CI 1.06 to 1.13). Shortest quarter versus longest: 26 per cent higher hazard (95 % CI 15 to 38 per cent), with between-study heterogeneity observed.

Then the sentence that should travel with every telomere headline: the age-stratified analysis showed a smaller hazard ratio in individuals over 80 years old. The association is weakest in exactly the population whose telomeres it is supposed to explain.

The 2017 individual-participant analysis: 12,199 adults, two cohorts

American Journal of Epidemiology, 2017, pooling individual participant data rather than study-level estimates. 12,199 adults from two population-based prospective cohorts: ESTHER in Europe (blood samples 2000 to 2002) and the Nurses’ Health Study in the United States (1989 to 1990). Leukocyte telomere length measured by quantitative PCR, standardized across cohorts as z-scores. 2,882 deaths recorded during follow-up (Nurses’ Health Study 1989 to 2010; ESTHER 2000 to 2015).

Results: telomere length was inversely associated with age in both cohorts, and after adjustment for age a significant inverse trend with all-cause mortality was observed in both. In random-effects meta-analysis, age-adjusted hazard ratios for the shortest quintile versus the longest: 1.23 (1.04 to 1.46) for all-cause mortality, 1.29 (0.83 to 2.00) for cardiovascular mortality, and 1.10 (0.88 to 1.37) for cancer mortality.

Look at those last two intervals. Cardiovascular: lower bound 0.83, so the data are compatible with anything from a 17 per cent reduction to a doubling. Cancer: 0.88 to 1.37. Neither is a finding. The authors’ conclusion is appropriately narrow: in a population aged 43 to 75, leukocyte telomere length predicts all-cause mortality beyond its association with age.

The 2022 heart-failure analysis, with the most useful number in the field

Heart, 2022, and this one goes the other way: instead of asking whether telomeres predict death in the general population, it asks whether they add anything in people who already have a life-threatening illness.

BIOSTAT-CHF Index (n = 2,260) and BIOSTAT-CHF Tayside (n = 1,413), Cox models in each cohort, estimates combined by meta-analysis. In age- and sex-adjusted analyses, shorter telomere length was associated with higher all-cause mortality in both cohorts individually and combined: HR 1.16 per standard-deviation decrease (95 % CI 1.08 to 1.24), p = 2.66 × 10⁻⁵.

And here is the number that makes this paper worth an hour: the authors convert the effect and conclude it is equivalent to being four years older. They also checked whether it survives the obvious adjustment, and it does — after adjustment for the BIOSTAT-CHF clinical risk score, 1.12 (1.05 to 1.20), p = 1.04 × 10⁻³. Shorter telomeres predicted both cardiovascular deaths (1.09; 1.00 to 1.19) and non-cardiovascular deaths (1.18; 1.05 to 1.32).

The negative result is in the same paper: no association with heart failure hospitalisation (0.99; 0.92 to 1.07; p = 0.855). A biomarker that tracks death but not the hospital admissions that cause death in that population is telling you something about what it is a marker of.

What all of this adds up to

Three datasets, three designs, a consistent direction and small effect sizes: roughly a 9 to 16 per cent increase in all-cause mortality hazard per standard deviation of shorter telomere. That is a real, replicated, population-level signal. It is also a marker, not a mechanism — the analysis in Heart controls for clinical risk score and the signal survives, which is the strongest thing that can be said for it, and it is still an association in observational data.

The age-stratified result from 2018 is the one that keeps the honest reading honest: the association attenuates above 80, and the cardiovascular and cancer subtypes in the 2017 analysis have confidence intervals that include no effect.

The biology that connects telomeres to replicative senescence is described on the epithalon page, which is a different question: not what a telomere predicts, but what a telomere-targeting compound was measured to do.

What these studies do not show

They do not show that lengthening telomeres extends life. Nothing here intervened on telomeres; all three are observational, and no intervention arm exists in any of them. They do not show that telomere length beats age, calendar age or any of the routine clinical scores as a predictor — the 2018 analysis explicitly attributes part of the heterogeneity to measurement technique and age at measurement. The cardiac and cancer subtypes are not established, with intervals that include no effect. The 2017 cohorts are both women-heavy and mid-life at baseline, so the 43-to-75 age range is not the whole life course. And leukocyte telomere length is a blood measurement standing in for a claim about every tissue in the body, which is a substitution no paper here justifies.

This article summarises published meta-analyses and cohort analyses. It is not medical advice and not a recommendation for any individual.