VO2max and lifespan: how large the association is, and in whom

The link between cardiorespiratory fitness and lifespan is one of the most durable findings in four decades of epidemiology. But "durable" and "large" are different words, and "large" and "applies to me" are further apart still. Here is how big the association is, in whom it is steeper, and where observation ends and the thing nobody has shown begins.

If you are new to the measure itself, start with the basics — what VO2max is and how it is measured. This piece is about the size of the association and its limits.

What was actually measured

Almost everything known about fitness and mortality comes from observational cohorts: fitness was measured, and then researchers watched for years or decades to see who died. That is a fundamentally different design from a trial in which fitness is deliberately changed.

The three cohorts cited most often:

• The Aerobics Center Longitudinal Study (Blair et al., 1989) — 10,224 men and 3,120 women who completed a maximal treadmill test at the Cooper Clinic in Dallas, followed for roughly eight years. One of the first works to show a graded "more fitness, less mortality" relationship in a largely healthy population.
• Cleveland Clinic (Mandsager et al., 2018) — a retrospective cohort of 122,007 patients referred for exercise treadmill testing on clinical grounds between 1991 and 2014. This is not a sample of healthy volunteers: a physician sent them.
• The US veterans cohort (Kokkinos et al., 2022) — 750,302 people aged 30 to 95 who completed a standardised treadmill test within the VA system.

The difference in populations is not a formality. In a cohort of clinically referred patients, low fitness can be a consequence of the very disease that later kills; in a cohort of healthy volunteers it more often reflects how someone lives. The same hazard ratio does not mean the same thing in those two samples.

🔴 And one more thing almost always lost in retelling: in none of these three cohorts was VO2max measured by gas analysis. What was recorded were peak estimated METs — derived from treadmill speed and grade, or from time on protocol. That matters further down, when watch readings come up.

How large the association is

Fitness is usually expressed in metabolic equivalents (METs): by convention 1 MET = 3.5 ml of oxygen per kilogram of body weight per minute, roughly resting expenditure. The convention is known to be approximate — measured resting expenditure in large samples averages lower — but it is the standard conversion for reading cohort METs.

The most quoted figure is the extreme-group comparison. In the Cleveland Clinic cohort, mortality among people in the top percentiles of fitness was several times lower than in the bottom group: the hazard ratio between the least fit and elite groups was roughly five, after adjustment for age, sex and major risk factors. The authors noted separately that the magnitude of the association for the least fit group was comparable to — or greater than — smoking, diabetes and coronary artery disease.

That is a comparison of extremes, and it should be read as one: it describes the difference between two groups of people, not what will happen to one person who takes up running.

A more usable form of the same finding is the per-MET gradient. In a meta-analysis of 33 cohorts (Kodama et al., 2009, roughly 103,000 participants), each additional MET of fitness went with about 13% lower all-cause mortality over follow-up. That is an averaged slope across very different samples — an observed association, not an effect — and its magnitude varies between individual studies depending on what was adjusted for.

Is there a ceiling

Two different questions get conflated here, and they have different answers.

For measured fitness, no threshold has been found: in both the Cleveland Clinic and the veterans cohorts the relationship stays graded into the highest percentiles, with no plateau and no upturn in risk at extreme values. The authors of both papers say so explicitly.

For self-reported activity volume, the picture differs. In large cohorts that asked people how much they move, the maximum benefit falls at 3–5 times the recommended minimum, after which the curve plateaus (Arem et al., 2015, over 660,000 participants); in a Danish study of jogging (Schnohr et al., 2015) the most strenuous group had confidence intervals so wide it cannot be statistically separated from sedentary people. These data do not show harm rising at very high volumes — they show the increment shrinking, and too few people in the extreme group.

So "too much is harmful" is not currently supported, but "more is always better without limit" is a claim about measured fitness, not about hours of training.

The no-ceiling finding also carries a general limitation. People with elite fitness in such a cohort are not a random sample: reaching and holding that level requires baseline health. Observation here cannot separate "fitness protects" from "the healthy are able to be fit".

In whom the association matters more

The practically useful part of the data is not the cohort average but where the gradient is steepest. Observations converge on the bottom of the scale. In the Cleveland Clinic cohort the hazard ratio between the least fit and elite groups is about five, while between two adjacent middle groups it is about one and a half. Further along, the curve flattens: the association persists, but each additional MET buys less.

The association has also been examined:

• in older adults — where low fitness is entangled with sarcopenia, gait speed and grip strength, and observation cannot apportion their contributions;
• in people with chronic heart failure and transplant candidates — where gas-analysis exercise testing is standard practice and feeds into clinical algorithms;
• across sex and age — in the veterans cohort the graded relationship held in every age band including the oldest, but the absolute fitness value corresponding to a given percentile differs by sex and by age.

How much it differs is visible in the FRIEND registry reference tables (Kaminsky et al., 2015): on a treadmill, the fiftieth percentile for men aged 20–29 is around 48 ml/kg/min and for men aged 70–79 around 24; for women, roughly 38 and 18. Which leads to something easy to miss: your number is meaningful against a reference for your own sex and age, not against a figure someone posted online.

The age trajectory

VO2max declines with age, and not linearly. In the Baltimore Longitudinal Study of Aging (Fleg et al., 2005), 435 men and 375 women aged 21 to 87 were measured repeatedly — here by direct gas analysis. The rate of decline accelerated with each decade: on the order of 3–6% per ten years in the twenties and thirties, and over 20% per ten years after seventy.

🔴 And a detail worth reading the paper for: the rate of decline was similar across all quartiles of self-reported physical activity. More active people sat higher in absolute terms at every age, but their curve fell at the same steepness. This is an observation, not a test of training: people were not randomised, they were asked how much they move.

The practical meaning: your own trajectory is more informative than a single reading. One number says where you are; a series across years says where you are heading. The same holds for the other measures collected in a biomarker map.

What the trials did not show

Here is the most important part, and the one the field usually skips.

Observation says: people with high fitness die at lower rates. It does not say that raising fitness lowers mortality. These are different claims, and the second does not follow from the first.

Only a randomised trial can test the second. Such trials exist, and their results are more modest than expectation:

• Generation 100 (Stensvold et al., 2020) — a Norwegian randomised trial in 1,567 people aged 70–77, followed five years, with all-cause mortality as the primary endpoint. Three arms: high-intensity intervals (3.0% died), moderate training (5.9%), and a control group given national activity recommendations (4.7%). For the main comparison — supervised training against control — no difference was found, and no pairwise comparison reached significance. An important detail: the control group did follow the recommendations, and performed more high-intensity work than the moderate-training arm. This was not movement versus immobility.
• Look AHEAD (2013) — an intensive lifestyle intervention in 5,145 people with type 2 diabetes and excess weight, median follow-up 9.6 years. Estimated fitness in the first year rose considerably more than in the support arm, but the primary endpoint was not mortality — it was a composite of cardiovascular events, and on that no difference was found; the trial was stopped early for futility.
• LIFE (Pahor et al., 2014) — a physical activity programme in 1,635 sedentary people aged 70–89. Its primary endpoint was not mortality but persistent loss of mobility — inability to walk 400 metres — and on that endpoint the intervention arm did better.

The conclusion is not "exercise does not work". It is more precise and duller: there is at present no randomised trial in which a rise in fitness led to lower all-cause mortality. One trial put mortality directly as its primary endpoint and found no difference against ordinary activity recommendations. The others measured a different outcome or accrued too few events: mortality in these groups is low, and catching a difference would take samples and timescales nobody has.

How trainable the number itself is

A separate question, usually treated as settled: train, and VO2max will rise. On average, yes — but the spread between individuals is wide.

In the HERITAGE Family Study (Bouchard et al., 1999), 481 sedentary participants from 98 families went through the same supervised twenty-week aerobic programme, with VO2max measured directly before and after. The mean gain was around 400 ml/min, while individual responses ranged from near zero to more than a litre under identical training. Variance between families was substantially larger than within them: part of the spread reflects heritability of the training response, not merely diligence.

🔴 But "does not respond" is a property of the dose, not of the person. In work by Montero and Lundby (2017), 78 young men trained from one to five times a week: the proportion of non-responders fell as volume rose, and when those who had not responded were given a further two hours per week for another six weeks, no non-responders remained. The population is narrow — young healthy men — but the authors' conclusion is unambiguous: non-response is abolished by increasing the dose.

Two things follow. First, a number that fails to rise on an ordinary programme is a described phenomenon, not proof that someone is doing it wrong, and not proof of a biological ceiling. Second, the benefit of movement does not reduce to the VO2max figure: blood pressure, insulin sensitivity, muscle mass and sleep change by their own routes and are measured separately.

Does risk move when the number moves

The intermediate link between observation and trial is cohorts with repeat measurements. There the analysis is of change rather than level.

In work on Aerobics Center Longitudinal Study data (Blair et al., 1995), men were examined twice about five years apart and then followed for roughly five more. Among those who stayed in the unfit group, mortality was 122 per 10,000 person-years; among those who moved from unfit to fit, 67.7; among those who stayed fit, 39.6. Each additional minute on the treadmill between the two tests went with roughly 8% lower risk.

That is closer to a causal statement than a single reading, but it remains observational: people whose fitness improves differ from everyone else in more than fitness.

How it is measured, and why that matters for reading numbers

The reference method is cardiopulmonary exercise testing (CPET): a stepwise increasing workload with breath-by-breath gas analysis, oxygen uptake measured directly. That is how the numbers in Fleg and in HERITAGE were obtained. In the three large mortality cohorts this piece opened with, there was no gas analysis — peak METs were calculated from the protocol. Calculation diverges systematically from direct measurement, which is one reason absolute thresholds from different papers do not line up.

A wrist device produces its estimate differently again: from the relationship between heart rate and pace, adjusted for age and weight. That is a model, not a measurement, and its divergence from CPET has been quantified: some devices underestimate by around 6 ml/kg/min on average, others show a comparable shift upward, with wide limits of agreement. A device can hold a given person several units above or below their true value.

Therefore:

• comparing a watch estimate against percentiles from clinical cohorts is not valid: the numbers have different origins;
• comparing your estimate against someone else's on a different device is not valid either;
• cautiously comparing your number against your own earlier number on the same device is reasonable — provided the device's bias does not drift over time. Direct evidence on how stable that bias is remains thin, so treat this as a sensible reading rule rather than a proven property of the instrument.

What to do with all this

An honest summary:

• Fitness is among the strongest observational predictors of mortality, shown across very large and varied samples.
• The association is largest at the bottom of the scale — where a person moves out of the least fit group into the next one; higher up, each increment buys less.
• The association is observational. There is no trial in which raising fitness lowered all-cause mortality; the one trial with that primary endpoint found no difference.
• The number responds to training unevenly between people, but non-response tracks the dose rather than being a verdict, and it does not cancel the other effects of movement.
• Your trajectory across years is more informative than one reading, and the meaningful comparisons are against a reference for your sex and age, and against your own earlier result on the same instrument.

On the strength of this body of data, the American Heart Association proposed treating fitness as a clinical vital sign (Ross et al., 2016). What is actually recommended: at minimum, an annual non-exercise estimate at a routine health check; ideally, periodic maximal testing where it is available — with the age of a first test and the interval between tests left unspecified in the document. It is a measure to be watched, not a procedure at every visit.

What a particular value means for you is a question for a clinician who sees the rest of the picture: blood pressure, resting heart rate and variability, body composition, history.

---

This material is informational and does not replace a consultation with a doctor. Decisions about training load, testing and treatment are made together with a clinician who knows your situation.