Telomere length and mortality: what the cohorts show

Twenty-five cohort studies, 121,749 people, 21,763 deaths: one standard-deviation shorter leukocyte telomere length carries a hazard ratio of 1.09 for all-cause mortality, and the shortest quartile 1.26. In the Swedish twin registry the same comparison reads 1.44. The measured gap between studies is measurement itself — two assays on the same blood correlated at r² of 0.34.

goodWell supported, with the usual caveats about effect size.Revised 27 Oct 2026

A laboratory slide rack holding a row of plain glass slides beside a small closed notebook and a pair of reading glasses on a pale grey bench

The single largest source of disagreement in this literature is not the biology. It is that two assays run on the same blood sample do not agree with each other very closely.

Telomere length is the most commercially successful biological measurement that has never produced an intervention. Every commercial test that reports it is selling you a number; the epidemiology tells you what that number predicts; the clinical literature tells you almost nothing about what happens if you move it. Those are three separate sentences and they are usually run together.

The headline association is real and it is modest

The best single summary is a 2018 meta-analysis that combined published cohort studies with unpublished data from the Swedish Twin Registry — 25 studies in total. Pooling everything eligible, 121,749 individuals with 21,763 deaths, a one standard-deviation decrement in leukocyte telomere length carried a hazard ratio of 1.09 (95 per cent CI 1.06-1.13) for all-cause mortality. Those in the shortest quartile had 26 per cent higher hazard (15-38 per cent) than the longest quartile. The authors note between-study heterogeneity, and they name its two sources: measurement technique and age at measurement.

That is the whole result in one line. A 9 per cent increase in hazard per standard deviation is a real signal that will replicate, and it is also small enough that no individual person can read their own number into a life expectancy.

The Swedish twin data are the strongest single estimate, and nobody knows why

Within the same paper, the four STR cohorts — 12,083 people with 2,517 deaths — give a noticeably larger number. One standard deviation shorter: 13 per cent increased all-cause mortality risk (7-19 per cent). Shortest quartile versus longest: 44 per cent higher hazard (27-63 per cent).

A hazard ratio of 1.44 and a hazard ratio of 1.09 coming out of the same paper, from overlapping kinds of data, is worth pausing on. Twin registries are unusual in ways that matter here: they are enriched for longevity, they are unusually well phenotyped, and their members have been followed for decades with a completeness few cohorts match. Whatever produced the difference, it is a difference in the study populations and the measurement chain, not in the biology of telomeres. The same paper says so.

There is a second, sharper detail in that meta-analysis. Stratified by age, the hazard ratio was smaller in people over 80. The association that looks like a clock gets weaker precisely where you would want a clock to work best. That is a real constraint on how much weight the “biological age” framing can bear, and it comes from the authors rather than from a critic.

The individual-participant analysis, and the two outcomes that failed

A 2017 individual-participant-data meta-analysis of 12,199 adults from two prospective cohorts — ESTHER in Germany, sampled 2000-2002, and the Nurses’ Health Study in the US, sampled 1989-1990 — used quantitative PCR with standardised z-scores so the two labs could be pooled. There were 2,882 deaths.

Shortest versus longest quintile, age-adjusted, in random-effects meta-analysis: 1.23 (1.04-1.46) for all-cause mortality. Then, for the two specific causes you would most want telomere length to predict: 1.29 (0.83-2.00) for cardiovascular mortality and 1.10 (0.88-1.37) for cancer mortality.

Both of those intervals include 1.00. Neither cardiovascular nor cancer mortality was significantly associated with telomere length in this analysis. The whole signal for the specific outcomes was carried by all-cause mortality, which is by definition the sum of the causes that did not clear significance. In an age range of 43 to 75 years, in two cohorts, this is the cleanest negative half of the picture.

The finding that got replicated and the finding that did not

The 2014 Leiden Longevity Study is worth reading because of what it found and of what it failed to find in the same breath.

The positive: shorter telomeres were associated with increased prospective mortality in middle age (30-80 years) and in very advanced age (90 and above). And the association survived adjustment — it could not be explained by the association of telomere length with the molar ratio of insulin-like growth factor 1 to its binding protein 3, with C-reactive protein, with interleukin 6, with cytomegalovirus serostatus, or with white blood cell counts. That rules out a whole family of confounders in one sentence, which is more than most telomere papers do.

The negative: in a prospective meta-analysis across multiple cohorts (n = 8,165), no association was found between telomere-associated genetic variants and mortality. That is a Mendelian-randomisation-style test, and it is the closest thing this field has to asking whether the association is causal rather than a marker. It came back null.

The same paper also found no difference in telomere length between the middle-aged offspring and their spouses — a useful calibration, since those two groups are similar in almost every way that matters socially and differ mostly in genetics.

So the honest reading of the whole evidence base: short telomeres are a marker of prospective mortality, and the genetic-instrument analysis does not support the claim that telomere variation itself contributes to the genetic propensity for longevity. The authors’ own conclusion is exactly that, and it uses “marker” deliberately.

Where genetics does show up: respiratory mortality

The same 2019 study that mapped the genetic architecture — 23,096 Singaporean Chinese samples identifying ten genome-wide loci, plus meta-analysis with 37,505 European individuals revealing six more — found a functional signal. In the Singaporean sample, longer leukocyte telomere length was associated with protection against respiratory disease mortality, hazard ratio 0.854 (0.804-0.906), P = 1.88 x 10-7. An intronic variant that reduces telomere length, rs7253490, was associated with respiratory infections at P = 7.44 × 10⁻⁴ — though the authors are careful to add that this effect may not be strongly mediated through telomere length itself.

Two things worth noticing. The locus at the telomerase reverse transcriptase locus behaves as expected; the immune-infection link is suggestive rather than closed. And this is one population in one country with a genetic architecture itself under study. It is an interesting finding and it is not a universal one.

The measurement problem, which is not a footnote

A 2020 study in the Annals of the New York Academy of Sciences prospectively compared the two main assays for leukocyte telomere length — monochrome multiplex quantitative PCR and flow-FISH — in 105 patients, all of whom had genetic testing, with mutations in telomere-maintenance genes found in 16.

Whole-leukocyte telomere length by MM-qPCR correlated with lymphocyte telomere length by flow-FISH at r² = 0.34, P < 0.0001. The paper also notes that the correct threshold to trigger genetic follow-up is still under debate.

Read that back into the 2018 meta-analysis, which listed measurement technique as one of the two sources of between-study heterogeneity, and you have the measurement story in two papers. If two methods on the same blood sample share roughly a third of their variance, then a hazard ratio expressed in “standard deviations” is partly a statement about which assay produced the number. The Swedish twin result at 1.44 and the pooled result at 1.09 are not necessarily in conflict — they may partly be measuring different quantities and calling them the same thing.

And then there is the weather

It is in this literature because it should be. A 2022 study of 5,864 participants from two independent cohorts, using a high-resolution spatiotemporal model to estimate daily air temperature at each residential address, found that between individuals an interquartile-range increase in daily air temperature was associated with shorter leukocyte telomere length: −2.96 per cent at lags of 0-1 days (95 per cent CI −4.46 to −1.43), −2.79 per cent at 2-6 days, −4.18 per cent at 0-6 days, and −6.69 per cent at 0-13 days.

Two things about that result. The effect accumulates with lag, which is consistent with a real biological response and not with a one-off measurement artefact. And it is observational, between-individual, and modest — a few per cent of a value that is itself measured with wide error bars. It is a reason to take environmental exposure seriously in general. It is not a reason to do anything specific about your telomeres.

What this literature does not show

It does not show that telomere length is modifiable by anything you can buy. No trial in this set intervened on telomere length and measured mortality. The cohorts measured; nobody treated.

It does not show that the association survives to the oldest ages in a useful way. The stratified analysis found the hazard ratio smaller above 80, and the causal-instrument analysis was null. A measure of “biological age” that weakens exactly where longevity decisions are made is a research measurement, not a personal one.

It does not show that cardiovascular or cancer mortality is predicted by telomere length. In the largest individual-participant analysis, both intervals crossed 1.00.

And it does not show that two labs measuring your blood will agree. r² of 0.34 between methods on the same sample is the load-bearing fact in this whole article, and it is the one most often left out of consumer-facing material about telomere testing.

How to read the number, if you have one

Two honest sentences. Telomere length is one of the better-replicated prognostic markers in the whole aging literature — a one-standard-deviation shorter value carries roughly a 9 per cent higher all-cause hazard in the largest pooled analysis, and the association is not explained by inflammation, CMV status or white cell counts.

And that is a statement about populations and probabilities, measured with an assay that shares about a third of its variance with the other main assay, in a body where the association weakens with age. Anyone who tells you a specific number about your own telomeres is showing you their lab’s error bar as if it were your prognosis.