Taiwan measured 13,423 adults aged 65 and over on seven simple fitness tests, then followed them through the national death registry for a median of seven years. The fittest fifth died at a rate 61 percent below the least fit. What the study cannot say is whether getting fitter is what makes the difference.
Rise from a chair, walk eight feet around a cone, turn, come back, sit down. Between 11 January 2015 and 25 November 2016, 13,423 Taiwanese adults aged 65 and over did exactly that at fitness stations set up across 22 cities and counties, twice each, with the better of the two times recorded. Six other measurements went alongside it. Then the researchers linked every participant to the national death registry and waited. By the end of 2022, 1,631 of them had died, 12.2 percent of the group, over a median of seven years.
The paper reports what happened next in the standard currency of survival research, the hazard ratio, which is the rate at which deaths occurred in one group divided by the rate in another over the same stretch of time. A ratio of 1 means no difference between the two. Below 1 means fewer deaths in the first group. Each ratio arrives with a 95 percent confidence interval, the range of values the data leave plausible, so a narrow interval marks an estimate that is well pinned down and a wide one an estimate that is not.
On that chair and cone test, the fifth of participants with the fastest times had an adjusted hazard ratio of 0.41 against the slowest fifth, confidence interval 0.33 to 0.51. Combine all seven measurements into a single index and the top fifth sat at 0.39, interval 0.32 to 0.48: a death rate 61 percent below the bottom fifth over the same seven years. The adjustment behind those numbers is not thin. It covers sex, age, body mass index, whether the person reported at least 150 minutes of moderate activity a week, income, education, employment, living alone, region, how urban their township was, ten named conditions from high blood pressure to heart failure, and a summary comorbidity score on top of the individual diagnoses.
How we know
The study is a cohort study, the design that follows groups defined by an exposure people already have rather than one the researchers hand out, and this one was built by stitching two Taiwanese national systems together. The National Physical Fitness Survey recruits participants each year at 46 testing stations across 22 cities and counties, using quota sampling with age and sex targets set to match the national population. It tested 18,173 people aged 65 or over between 11 January 2015 and 25 November 2016, and 13,423 of them are the group studied here. The 4,750 who fell out did not fall out at random: 1,508 could not be linked to the national insurance registry and 9 were missing fitness data, and the researchers then set aside everyone over 90, everyone with dementia, a stroke or a prior limb amputation, and everyone holding a Catastrophic Illness Certificate, which covers primarily cancer, autoimmune disease and chronic mental health conditions. What is left is community dwelling adults over 65, meaning people living at home rather than in a nursing home or other care institution, minus the sickest of them: mean age 72.9 years, and 8,394 of them, 62.5 percent, women. Their records were followed for deaths in the National Health Insurance Research Database through 31 December 2022, a median follow-up of 7.0 years with an interquartile range of 6.7 to 7.1. The work was funded by Taiwan's National Science and Technology Council.
The seven measurements are the Senior Fitness Test battery, and they need almost nothing: a chair, a stopwatch, a cone, a tape measure. A 2-minute step test for cardiorespiratory fitness. A 30-second arm curl and a 30-second chair stand for muscular strength. A back scratch and a chair sit-and-reach for flexibility. A 1-leg stance with eyes open, held up to 30 seconds, and the 8-foot up-and-go for balance and agility. The composite index was built by converting each person's result on each test into a sex-specific percentile rank, a score saying what proportion of same-sex participants they beat, then summing all seven with equal weight and cutting the total into fifths.
The seven tests, ranked
The tests did not perform alike. Six of the seven are reported individually in the main results, each comparing the top fifth with the bottom fifth: the 8-foot up-and-go came first at 0.41 (0.33 to 0.51), then the 1-leg stance at 0.50 (0.42 to 0.59), the 30-second chair stand at 0.55 (0.46 to 0.65), the 2-minute step test at 0.58 (0.49 to 0.68), the 30-second arm curl at 0.63 (0.53 to 0.75), and the chair sit-and-reach last at 0.79 (0.67 to 0.93). The back scratch test carries no separate figure there, though it sits inside the combined index like the rest. Balance, agility, lower body strength and cardiorespiratory fitness sat at one end; upper body strength and flexibility at the other. The gradient across the intervening fifths was monotonic: each better-performing group had a lower rate than the one below it, rather than everything hinging on a single cutoff.
Split by sex, the picture shifts a little. Among men, balance and agility carried the lowest ratios, followed by cardiorespiratory fitness. Among women, lower body strength sat alongside balance and agility rather than behind them.
What the design cannot do
The authors are direct about the limit: an observational design precludes causal inference, and despite the adjustment, residual confounding and reverse causation cannot be excluded. Confounding is the distortion that arises when a third factor is tied both to the thing being measured and to the outcome, so an association shows up for a reason other than the one under study; the residual kind is whatever survives the adjustment, including every factor nobody thought to record. Reverse causation is the one that bites hardest here. Cancer that has not yet been found, heart failure that has not yet been diagnosed, the early years of a neurological condition: each of these can slow a person out of a chair and shorten their life, and the arrow would run from the illness to both, not from the fitness to the survival. The standard defence is a landmark analysis, which throws away the deaths in the first year or two on the grounds that those are the ones most likely to reflect illness already present at testing. This paper does not report one.
Three further limits sit in the paper's own list. Fitness was measured once, so nobody's trajectory was captured. The requirement to complete seven physical tests selects for healthier, ambulatory volunteers, which is a selection bias the authors name themselves. And smoking status was not captured at all, leaving one of the largest influences on mortality outside the adjustment entirely.
Why it matters
The question the paper set itself was narrower than the headline figure suggests: whether objectively measured fitness is associated with death rates over and above what people report about their own activity. Self-reported exercise was in the adjustment, and the gradient survived it. That is the claim, and the authors take it to a practical place, suggesting that objective fitness assessment folded into routine geriatric and internal medicine care might sharpen how risk is judged in older patients. Assessment, not prescription: what these seven measurements did in Taiwan was sort a population, not treat one.
Whether that sorting adds anything to the risk scores clinicians already use is the open question, and this study did not formally test it. A cohort study that measured fitness twice, or one that set aside its earliest deaths, would say considerably more about which way the arrow points. Until then, what stands is an association across 13,423 people and seven years, larger and better adjusted than most, and still an association.
Nobody was assigned to be fit or unfit: the researchers measured people as they found them, and the authors say plainly that such a design cannot establish cause. Two possibilities survive the adjustment. Differences nobody measured, or nobody thought to record, could be doing the work. And the arrow could run backwards, from illness to both a slow chair stand and an early death: an illness nobody has diagnosed yet can do exactly that, producing this pattern with fitness playing no part in it. That second possibility is the one that bites here, and the paper reports no landmark analysis setting aside deaths in the first year or two, which is the usual way of testing it. Fitness was measured once, at a single visit, so nobody's improvement or decline was tracked. The sample is self-selected toward healthier, ambulatory people, since turning up to a fitness station and completing seven tests is itself a filter, and it is filtered again by design: of 18,173 people aged 65 or over who were tested, 13,423 were studied, after the researchers set aside everyone over 90, everyone with dementia, a stroke or a prior limb amputation, and everyone holding a Catastrophic Illness Certificate. That exclusion runs the same way, since it removes the people most likely to be both unfit and close to death. Smoking status was not recorded and could not be adjusted for. And it is one country, one testing system, one age group.
Whether becoming fitter changes anyone's risk of dying. The study measured fitness once and counted deaths afterwards; it never tested an intervention, so a person who improves their chair stand time is outside what these data can speak to. It also did not examine causes of death, so the association is with dying, not with dying of anything in particular. And it did not formally test whether adding these measurements to existing clinical risk models predicts better than those models already do.
Does this show that getting fitter helps you live longer?
No, and the authors say so. The study measured fitness once and then counted deaths, which shows that better performers died at lower rates but not that the performance caused the difference. Nobody in it was asked to change anything.
What does an adjusted hazard ratio of 0.39 mean?
It means the top fifth on the combined fitness index died at 39 percent of the rate of the bottom fifth over the same follow-up, once the listed differences between the groups had been accounted for. At the 95 percent level, the range the data leave plausible for that figure runs from 0.32 to 0.48. It is a comparison between two groups, not a probability for any individual.
Why does reverse causation matter so much here?
Because an illness that has not yet been diagnosed can make someone slower out of a chair and also shorten their life, producing this exact pattern with fitness playing no causal part. The usual check is to discard deaths in the first year or two of follow-up and see whether the association holds, and this paper reports no such analysis.
Which test came out strongest?
The 8-foot up-and-go: the fastest fifth died at 41 percent of the rate of the slowest. Balance, agility, lower body strength and cardiorespiratory fitness all sat well ahead of flexibility, where the same comparison gave 0.79. The paper did not test whether any of these adds to the risk models clinicians already use.
This is research reporting, not medical advice. We describe what studies found. We never tell you what to do. Talk to a doctor before changing anything about your health. Read our full position.
The evidence, read properly, every three days.
We read the paper so you do not have to. Free, and no supplement ads, ever.
Powered by Buttondown