People who move least gain the most from moving more, and almost any movement counts, light walking and chores included. The first step off the couch carries the biggest gain in overall survival; past a modest daily amount the curve flattens, so people already active at that level have little left to gain there (Physical Activity Dose and Mortality). That flattening is a result in its own right: an active person can stop optimizing for a longer life.
Other outcomes follow other curves. Fitness keeps paying, with shrinking returns (Cardiorespiratory Fitness and Mortality). For some cancers, the World Cancer Research Fund judges that more activity is likely better still (Diet Physical Activity and Cancer Prevention). Older adults avoid falls through balance and functional training, not just more activity (Exercise for Preventing Falls in Older Adults). Within a given goal, the type of exercise matters less than doing it, so pick the kind you will keep doing (Is the Exercise Modality Doing Any Work). Blood pressure is a partial exception: in trials of people starting at 140 mmHg or higher, combined endurance and resistance training lowered it more than any single type, on indirect, provisional evidence. Exercise trims belly fat but does little for the scale.
Most of this evidence comes from observational studies of people who chose how much to move, so it shows that activity goes with longer life more firmly than that it causes it. Falls prevention is the outcome trials prove best.
The biggest gain comes from the first step off the couch
The least active people gain the most, and almost any movement counts. Ekelund pooled eight cohorts that measured activity with waist-worn accelerometers (hip or lower back) (36,383 adults aged 40 and over, median follow-up 5.8 years, 2,149 deaths). Ranked into quarters by total activity, mortality fell from 1.00 in the least active quarter to 0.48 (95% CI 0.43-0.54) in the second, 0.34 (0.26-0.45) in the third and 0.27 (0.23-0.32) in the most active. The steepest drop is the first one. In the authors’ words, «the greatest risk reduction for mortality was observed when the second quarter was compared with the referent, for all activity intensities» (Ekelund et al., 2019) -> Physical Activity Dose and Mortality.
These are associations in observational cohorts, not the effect of a trial. The associations are large, but no trial randomizes people to a lifetime of activity and counts deaths -> Cardiometabolic Interventions and Hard CV Outcomes in Low-Risk People.
The first step is mostly light movement, not a workout
Moving from the bottom quarter to the second took little in exercise terms. For each measure the step was «broadly equal to 60 min/day of light intensity physical activity», or about 5 minutes a day of moderate-to-vigorous activity (MVPA) (Ekelund et al., 2019). So move more at any intensity and sit less — standing, walking, chores — not to start a training programme. Light activity followed the same shape as MVPA: quarter 2 sat at 0.60 (0.54-0.68) for light activity and 0.64 (0.55-0.74) for MVPA, each against its own least-active quarter (Ekelund et al., 2019). Sitting is a separate exposure with its own curve, covered below.
The curve flattens early, inside a studied range
Above a modest dose the mortality curve stops falling. Ekelund found maximal risk reduction at about 24 minutes a day of MVPA (HR 0.39, 0.26-0.59) or about 375 minutes of light activity, and «No further risk reductions occurred with higher levels of activity except for low light intensity physical activity where the risk appeared to decrease further» (Ekelund et al., 2019). The curve falls and then levels off over the studied range.
Read 24 minutes as roughly where the curve stops falling, not as a target -> The Underivable Optimum. Few people logged much MVPA, so «wide confidence intervals were observed at this part of the dose-response curve». The total-activity optimum of about 300 counts per minute «mirrors the median total physical activity for the third quarter across studies». The curve was fitted to each study’s quarter medians, so this optimum sits on the third of four exposure points and only the most active quarter informs the curve beyond it; that quarter’s hazard ratio (0.27) was still lower than the third quarter’s (0.34) (Ekelund et al., 2019). So the flattening beyond about 300 counts per minute rests on a single exposure point, the most active quarter, and that quarter’s own hazard ratio kept falling. Where the curve flattens, and how fully, is uncertain.
Steps give the same shape, with a lower plateau for older adults
Paluch pooled 15 cohorts that counted steps (47,471 adults, mean age 65, 3,013 deaths). Against the lowest quartile (median 3,553 steps a day), mortality was 0.60 (0.51-0.71), 0.55 (0.49-0.62) and 0.47 (0.39-0.57) in the higher three. Risk stopped falling at about 6,000-8,000 steps a day in adults aged 60 and over, and at about 8,000-10,000 in younger adults (Paluch et al., 2022). The age difference was a formal interaction (p=0.012) (Paluch et al., 2022). An older adult can therefore reasonably aim lower than a younger one, because the effect itself differs by age, not just the baseline risk -> Physical Activity Dose and Mortality.
10,000 steps is not a feature of the curve. Paluch notes the goal «is not based on evidence, but instead originates from a marketing campaign in Japan». Pace adds little: the authors «found inconsistent evidence that step intensity had an association with mortality beyond total volume of steps» (Paluch et al., 2022). For overall survival, count the steps rather than chase the pace. Fitness, discussed next, follows a different rule.
Steps, minutes and counts are not interchangeable
Total activity in counts per minute, MVPA in minutes a day and steps a day each measure movement, but on different instruments, against different reference groups. A hazard ratio from one cannot be converted into another: the device total-activity top quarter (0.27) and the steps top quarter (0.47) are different contrasts, not a disagreement -> Physical Activity Dose and Mortality.
The device figures are also larger than older self-report studies suggested. Ekelund’s effect sizes were «about twice as large compared with those previously reported in studies assessing physical activity by self report» (Ekelund et al., 2019). Self-report covers few activity domains and rarely captures light activity, and its imprecision makes the association look smaller than it is (Ekelund et al., 2019) -> Measurement Error in Dietary Assessment. Accelerometers, by contrast, «capture more movement across multiple domains» (Ekelund et al., 2020), including movement at work, which may not behave like leisure activity -> The Physical Activity Paradox.
How far the numbers can be trusted
Reverse causation is checked only weakly. Sick and frail people move less, which inflates any activity-mortality gradient. Dropping deaths in the first two years left the hazard ratios «materially unchanged», but Ekelund adds that «bias from reverse causation might persist» (Ekelund et al., 2019). No genetic or randomized check exists -> The U-Shaped Association Artifact, Upgrading Observational Evidence.
Three gaps limit what a reader can do with these figures. The device cohorts were all aged 40 or older, and Ekelund states «it is unclear whether the estimates of the absolute physical activity levels associated with maximal risk reduction apply to younger people» (Ekelund et al., 2019). No trial tests activity against death. And these cohorts report hazard ratios only. They give no absolute risk reduction at a stated baseline risk, so no figure says how many deaths per 1,000 people moving more would prevent -> Layer 1 - Ranking Interventions for a Stratum, Baseline Risk and the Relative-Absolute Split.
Where this ranks depends on where you start
For an inactive person, moving more is probably one of the largest levers available, likely ahead of most dietary refinements, though no absolute figure exists to rank it by, and the first extra minutes carry the most. For someone already active at moderate levels, the mortality curve has largely flattened, though how fully is uncertain. More activity then becomes a small lever competing with other small levers. That is worth knowing in itself: it frees an active person to stop optimizing activity volume for overall survival -> Layer 1 - Ranking Interventions for a Stratum. The flattening is specific to all-cause mortality and to activity volume. The activity curve flattens early. Fitness, measured on a treadmill, does not flatten in the same place.
Fitness keeps paying after activity levels off
Fitness here means cardiorespiratory fitness (CRF): the peak aerobic capacity a treadmill test measures, expressed in METs (1 MET = 3.5 mL/kg/min of oxygen). It is a capacity, not a behaviour. Minutes of activity are what a person does; CRF is partly what that activity builds and partly what they were born with. The two curves have different shapes, and fitness is the one that keeps sloping down. Cardiorespiratory Fitness and Mortality
Each extra MET goes with lower mortality, with shrinking returns
A meta-analysis of cohorts found that each 1-MET-higher fitness level went with 13% lower all-cause mortality (RR 0.87, 95% CI 0.84-0.90) and 15% fewer heart-disease and cardiovascular events (RR 0.85, 0.82-0.88). Compared with people of high fitness, people of low fitness had an all-cause RR of 1.70 (1.51-1.92). (Kodama, 2009)
A Cleveland Clinic cohort of 122,007 patients referred for treadmill testing found no point where the curve stopped: «Cardiorespiratory fitness is inversely associated with long-term mortality with no observed upper limit of benefit.» Even the elite band (top 2.3% for age and sex) had lower mortality than the high band (HR 0.77, 0.63-0.95). The lowest performers had an adjusted HR of 5.04 (4.10-6.20) against the elite, and being unfit was a mortality signal comparable to or larger than smoking (HR 1.41) or diabetes (HR 1.40) in the same model; below-average against above-average fitness alone carried an HR of 1.41. (Mandsager et al., 2018)
An AHA scientific statement, drawing on earlier cohorts, puts it differently: «Most of the lower mortality risk associated with a higher CRF occurs by the time a CRF of 10 to 12 METs is achieved.» Above 12 METs, it says, the impact on all-cause and cardiovascular mortality is «relatively lower». (Ross et al., 2016)
The two readings fit together once you look at where the gains sit. The benefit is front-loaded: most of it comes from leaving the bottom of the fitness range, and each further MET buys less. The curve never goes flat within the range studied, but it is not a straight line either. Mandsager’s «no upper limit» describes the observed range in a referral population; it does not show that more fitness always pays the same. So a person who holds their training once past about 10 METs is making a defensible choice on mortality grounds, and a person pushing higher is buying a smaller return per MET. Cardiorespiratory Fitness and Mortality Measuring and Raising Cardiorespiratory Fitness
This shape also means the per-MET figure should not be compounded naively. Five extra METs do not multiply to 0.87^5, about half the risk. The gradient is steepest at the low end, so a MET gained from a sedentary start buys far more than one added near the top, and a single averaged coefficient hides that. Cardiorespiratory Fitness and Mortality
Where a given fitness level sits
Kodama’s analysis suggested «a minimal CRF of 7.9 METs may be important» for preventing death and cardiovascular events, with 7.9 METs referenced to a 50-year-old man. (Kodama, 2009) The analysis assumed women run about 2 METs below men and that capacity falls by about 0.1 MET per year of age, so the corresponding floor is lower for women and for older people. (Kodama, 2009)
Mandsager’s age-and-sex percentile grid places a reading in a band. For men aged 50-59, low is below 8.2 METs and elite is 14.0 or more; for women aged 50-59, low is below 7.0 and elite is 13.0 or more. These bands come from a referral population using estimated, not measured, METs (Mandsager et al., 2018), so a lab-measured VO2max placed against them is approximate. Cardiorespiratory Fitness and Mortality
Fitness is partly a lever and partly a marker
Every study above is observational. Mandsager says so directly: the association «does not prove causation», and how far high fitness preselects healthier patients rather than lowering their mortality «is not discernible». (Mandsager et al., 2018) Three things make the gap between fit and unfit people a poor guide to what raising your own fitness can do. Illness lowers fitness, so low CRF partly marks ill health rather than causing it. About half the variance in CRF is heritable (Ross et al., 2016), so part of the gradient belongs to people born fitter. And a trained gain probably buys less than the cross-sectional comparison implies. What holds up is smaller: training your way out of the low-fitness bottom helps, not the five-fold gap read as a personal promise. Cardiorespiratory Fitness and Mortality
Within-person change narrows that gap without closing it. Men who went from unfit to fit between two examinations had 44% lower mortality than men who stayed unfit. In HF-ACTION, a randomized exercise trial in heart-failure patients, every 6% rise in measured peak VO2 over three months went with a 4% lower risk of cardiovascular death or cardiovascular hospitalization. (Ross et al., 2016) HF-ACTION was a randomized trial, but this link between fitness change and outcome is an observational analysis within it, and its outcome includes hospitalizations. CRF is best treated as a modifiable target whose improvement tracks lower risk, and as the measurable output of the activity that raises it. Measuring and Raising Cardiorespiratory Fitness
Raising it: for fitness, intensity matters more than time
For fitness, intensity counts for more than minutes. Ross: «Increases in CRF appear more responsive to increases in intensity than increases in session duration or frequency.» Meeting the standard activity guidelines brings an improvement of about 10% in previously sedentary adults. (Ross et al., 2016) This is a statement about CRF only. For overall mortality, the activity data reward total movement at any intensity (previous section).
The fitter someone already is, the harder they must work to gain more:
- Below 10 METs: about 50% of heart-rate reserve is enough.
- 10 to 14 METs: 65-85% of heart-rate reserve is likely more effective.
- Above 14 METs: more than 85% may be needed for most people.
(Ross et al., 2016) Heart-rate reserve is maximum minus resting heart rate. At the top of the range the aim changes: «Among people with a CRF ≥13 METs, the primary goal of increasing CRF is generally more related to improving performance than health.» (Ross et al., 2016) Whether intervals beat steady exercise at a fixed weekly budget is covered with the other budget choices below. Measuring and Raising Cardiorespiratory Fitness
Measuring it, and what a measurement adds
Three tiers exist. A cardiopulmonary exercise test (CPX) measures peak VO2 directly and is the reference standard. A maximal treadmill or cycle test without gas analysis estimates it from peak work rate (Ross et al., 2016). Non-exercise equations estimate it from routine clinical data with no test at all. Two of them (Jurca, Nes) predicted long-term mortality about as well as measured fitness, but Ross warns that in «most clinical patient subsets, nonexercise estimated CRF should not be viewed as a replacement for objective assessment of CRF.» (Ross et al., 2016)
A bike or walk estimate is a band, not a point. For peak watts from a maximal cycle test, the equation matters: in 5,100 adults with measured VO2max, the ACSM equation read 15.5% high on average (11% in men, 24% in women), the FRIEND equation 0.5%. The paper reports no standard error of the estimate or limits of agreement, so one person’s error is not stated (Kokkinos et al., 2018).
Without a bike, a fast 2-km walk estimates VO2max from walk time, finishing heart rate, age and BMI (Oja et al., 1991). In 410 Spanish adults aged 18-64, 95% of walk estimates fell within about ±12.5 ml·kg-1·min-1 of measured VO2max (limits −12.9 to +12.3), and in 95% of people two walks a week apart agreed within about 8 ml·kg-1·min-1. The authors judge the test valid and reliable across sex, age and activity level, a group-level verdict (Marín-Jiménez et al., 2023). For one person, the retest band is larger than the roughly 10% gain (about 3-4 ml·kg-1·min-1 at a VO2max of 35) that a previously sedentary adult meeting the guidelines can expect, so one before-and-after pair of walks cannot confirm a training gain. Measuring and Raising Cardiorespiratory Fitness
A fitness reading can change a risk estimate, not just correlate with risk. Adding CRF to age, blood pressure, cholesterol, diabetes and smoking correctly reclassified 12.1% of men for cardiovascular death at 10 years in one cohort. In stage-II hypertension, low versus high fitness meant a 30-year risk of cardiovascular death of 18.4% versus 10.1%, although the 10-year risks looked similar (2.3% versus 1.2%). (Ross et al., 2016) The 2013 AHA/ACC risk calculator left fitness out (Ross et al., 2016), so a fitness reading informs judgement rather than a computed score. Measuring and Raising Cardiorespiratory Fitness
The shape of the benefit depends on the outcome
Mortality is one outcome. For diabetes, blood pressure, cancer, dementia, mood and joints, the curve looks different, and so does the evidence behind it. A person who has collected most of the survival benefit from activity can still gain on other outcomes. Some outcomes have trial evidence and others only cohorts; some respond to a particular kind of exercise, and some fade over time. Each outcome below gets its own evidence state, and none inherits the all-cause plateau. -> Physical Activity Dose and Mortality
Diabetes: a lifestyle programme prevents it, and activity may help even with little weight loss
The Diabetes Prevention Program enrolled people with prediabetes (mean BMI 34). Its lifestyle arm aimed for a 7% weight loss and at least 150 minutes of activity a week. It cut diabetes incidence by 58% (95% CI 48 to 66) against placebo, from 11.0 to 4.8 cases per 100 person-years (Knowler, 2002).
That arm was a bundle, so it cannot isolate exercise. Da Qing did: it randomized clinics to diet, exercise, or both. After adjustment, risk fell 31% with diet, 46% with exercise and 42% with both. By intention to treat the three arms did not differ significantly, and the authors found «no additional benefit of combining the interventions» (Pan et al., 1997). Activity may help even with little weight loss: in a non-randomized analysis of DPP’s lifestyle arm, participants who missed the weight goal but met the activity goal had 44% lower incidence, independent of their small weight loss (about 2.9 kg) (Hamman et al., 2006). Evidence state: benefit, from trials, for the lifestyle bundle in people with prediabetes; the activity-alone signal is observational -> Lifestyle vs Metformin for Diabetes Prevention.
Blood pressure: exercise matches a drug only in an indirect comparison
A network meta-analysis of 391 trials put exercise and blood-pressure drugs in one comparison. Across all trials, exercise lowered systolic pressure by 4.84 mmHg and drugs by 8.80 mmHg. But the exercise trials started at a mean of 132 mmHg and the drug trials above 150, and a higher starting pressure leaves more to lower. Restricted to people with hypertension (140 mmHg or more), exercise lowered systolic pressure by 8.96 mmHg (95% CrI 7.64 to 10.27), and the gap to drugs was 0.18 mmHg (95% CrI −1.35 to 1.68) (Naci et al., 2018).
No trial compared the two directly; every figure here is inferred across separate trials, and the exercise trials were smaller and unblinded. The authors say the effects «appear similar to that of commonly used antihypertensive medications, but this is tempered by the observed differences in study population characteristics» (Naci et al., 2018) -> Blood Pressure Lowering and Cardiovascular Events.
In trials of people with high blood pressure, combining endurance and resistance training lowered it most among exercise types. In Naci’s network, endurance, dynamic resistance and isometric training lowered systolic pressure by similar amounts, but the combination beat dynamic resistance alone by 2.98 mmHg (95% CrI −5.04 to −0.93). In trials starting at 140 mmHg or more, the combination beat every single type and lowered systolic pressure by 13.51 mmHg (95% CrI −16.55 to −10.45) against control (Naci et al., 2018).
The edge is provisional. Only 30 of the 197 exercise trials compared one type with another, so most type contrasts are network estimates. Combination trials started higher than resistance trials, at a mean 135 against 125 mmHg, and endurance trials at 134 (Naci et al., 2018). A higher start leaves more to lower, so part of the all-populations edge over resistance may be baseline; the edge over endurance in the 140-plus trials is less exposed to that.
Naci says «only a small subset of studies» tested the combination; these were «often effective, especially in hypertensive populations», but their frequency and duration were not consistently reported (Naci et al., 2018). So part of the edge may also be more total training. Among single types, the one a person will keep doing is a reasonable choice; for a hypertensive person who can do both, indirect evidence leans toward the combination. -> Is the Exercise Modality Doing Any Work
Blood pressure is a surrogate: Naci counted no heart attacks or strokes, and whether an exercise-driven drop prevents events at the rate a drug-driven one does is assumed, not shown -> Surrogate Outcomes. A cheap, safe drug can stand in for exercise on blood pressure, but not on fitness, glucose or weight, so its existence lowers exercise’s rank only for the blood-pressure channel. -> Layer 1 - Ranking Interventions for a Stratum
Cancer: no plateau in sight
The World Cancer Research Fund grades physical activity as protective against colon cancer (convincing) and against endometrial and postmenopausal breast cancer (probable) (World Cancer Research Fund & American Institute for Cancer Research, 2018). Its reading of the dose is the opposite of the mortality curve: «For cancer prevention, it is likely that the greater the amount of physical activity, the greater the benefit» (World Cancer Research Fund & American Institute for Cancer Research, 2018). Someone exercising to lower cancer risk cannot treat the survival plateau as a stopping point.
Three cautions apply. The grades rest largely on observational studies, so reverse causation (illness lowers activity) is not ruled out by design. WCRF uses its own grading system, not GRADE. And the shape is the panel’s judgement, not a fitted curve: this guide holds the grade, not a pooled effect size. -> Diet Physical Activity and Cancer Prevention, Physical Activity Dose and Mortality
Dementia: two strong analyses disagree
A 58-study meta-analysis links higher activity to lower dementia risk, RR 0.80 (95% CI 0.77 to 0.84), and the association held in studies with 20 or more years of follow-up, RR 0.79 (0.71 to 0.87) (Iso-Markku et al., 2022). The 2024 Lancet Commission cites that figure in its section on physical activity, one of its modifiable risk factors (Livingston et al., 2024).
An individual-participant analysis of about 404,840 people reached the opposite answer. Inactivity was tied to dementia when measured less than 10 years before diagnosis (HR 1.40, 1.24 to 1.59), but not when measured earlier (HR 1.01, 0.89 to 1.14), while its links to diabetes, heart disease and stroke held in both periods (Kivimäki et al., 2019). The live explanation is reverse causation: early dementia makes people less active. The meta-analysis’s own cleanest subset, only three studies, gave the same point estimate but was not significant, RR 0.79 (0.62 to 1.01) (Iso-Markku et al., 2022). The disagreement stays open -> Does Physical Activity Protect Against Dementia Beyond Ten Years, Dementia Prevention and Modifiable Risk Factors.
The trials measure thinking scores, not dementia. Bundled programmes that include exercise moved a cognitive test composite in FINGER, but preDIVA, nurse-led vascular risk care that included activity, found no effect on dementia incidence, HR 0.92 (0.71 to 1.19) (van Charante et al., 2016). A network meta-analysis of 109 trials found exercise alone improved global cognition in cognitively unimpaired older adults by a small SMD of 0.14 (0.05 to 0.22) against health education (Mendes et al., 2025); an umbrella review put exercise’s overall effect at d = 0.22 (Blomstrand et al., 2023).
Tai chi and yoga look larger (d = 0.48) (Blomstrand et al., 2023), but that comparison sets separate reviews side by side, mostly against inactive controls, so the edge is most likely a comparison artefact -> Multidomain Lifestyle Intervention and Cognitive Decline, Mind-Body Exercise and Cognition. Evidence state: insufficient for a direct effect on dementia; the route through heart disease and diabetes is not in dispute.
Mood and chronic pain: probably helps, on weak evidence
In a network meta-analysis of 218 trials in people with major depression, walking or jogging reduced symptoms beyond active control (Hedges’ g −0.63, −0.80 to −0.46); yoga (−0.55) and strength training (−0.49) also worked, against −0.26 for SSRIs in the same network (Noetel et al., 2024).
Here intensity tracked the effect, while the weekly-volume curve was unclear: a different curve again. But «confidence in accordance with CINeMA was low for walking or jogging and very low for other treatments», mostly because few trials blinded participants and staff, so expectancy effects may contribute (Noetel et al., 2024) -> Depression and Modifiable Exposures. For chronic pain, a Cochrane overview found function improved in 14 of 21 reviews, by small to moderate amounts, while pain scores moved inconsistently; the evidence is low quality, but «none of the physical and activity interventions assessed appeared to cause harm to the participants» (Geneen et al., 2017) -> Chronic Pain and Physical Activity.
Joints: the benefit fades within a year or so
Across 77 trials in knee or hip osteoarthritis, exercise beat usual care at about 8 weeks on pain (SMD 0.56, 0.44 to 0.68) and function (0.50, 0.38 to 0.63). The gain then fades: «The effects were reduced gradually after 2 months and became no better than the usual care group at 9 to 18 months depending on the outcome» (Goh et al., 2019). Here the curve runs over time, not dose, so exercise for arthritis is better treated as a habit to keep than a course to finish -> Exercise Therapy for Osteoarthritis Pain and Function.
The type of exercise matters little: stationary cycling was neither better nor worse than other exercise (Luan et al., 2020) -> Stationary Cycling for Knee Osteoarthritis. For overweight people with knee osteoarthritis, weight loss adds to exercise rather than replacing it. In an 18-month trial, diet plus exercise left people with less pain than exercise alone or diet alone. In a post hoc analysis, 38% of the diet-plus-exercise group reported little or no pain at 18 months, against 20% with diet alone and 22% with exercise alone (Messier et al., 2013).
Exercise does not harm an arthritic joint: in the systematic review behind the 2018 EULAR activity recommendations, covering rheumatoid arthritis, spondyloarthritis and hip or knee osteoarthritis, «No detrimental effects of PA were reported in any study» (Rausch Osthoff et al., 2018). Fear of damaging the joint is not a reason to stay still -> Knee Osteoarthritis and Modifiable Levers. Choosing drugs, injections or surgery lies outside this appraisal.
Smaller and weaker signals
- Parkinson’s disease: highest vs lowest activity RR 0.79 (0.68 to 0.91) in cohort studies, null in women (Chen et al., 2021) -> Parkinsons Disease Modifiable Risk Factors.
- Fatty liver: exercise reduces liver fat even without weight loss; evidence on liver scarring is far thinner -> Fatty Liver MASLD and Weight Loss.
- Cataract: RR 0.90 (0.86 to 0.94), graded probable; macular degeneration: RR 0.92 (0.84 to 1.01), insufficient (Aune et al., 2026) -> Cataract and Modifiable Exposures, Age-Related Macular Degeneration and Modifiable Exposures.
- Kidney disease: RR 0.91 (0.85 to 0.97), low to very low certainty (Seidu et al., 2023) -> Chronic Kidney Disease and Modifiable Exposures.
What none of these studies measures
Every result above is an event count, a test score, or a symptom scale read at a fixed time. None tracks the shape of decline: how long a person stays able before function falls away, which is the outcome many people care about most. A lower dementia or cancer risk says nothing direct about that trajectory, and the function data here stop at 8 weeks to 18 months. This is a named gap, not a null. -> Surrogate Outcomes
Sitting is a separate risk from not exercising
All of the evidence so far counts what people do. Sitting is a separate thing people do, and it carries its own risk. Hours of sitting and minutes of activity are two different exposures, not one scale read in two directions. Willett’s review of how activity is measured notes «little correlation between sedentary behaviors and physical activity», so a person can hit a step count and still sit for most of the day. (Willett, 2012) In practice, a move more target does not on its own deal with long sitting. Studies of one say little about the other. -> Sedentary Behaviour and Chronic Disease Risk
Total sitting and TV viewing follow different curves
The main dose-response evidence is a meta-analysis of 1,331,468 participants in prospective cohorts that separates total sitting from TV viewing and adjusts for physical activity. (Patterson et al., 2018) Both curves stay nearly flat at first, then steepen. The harm per extra hour grows past a threshold, the opposite of the activity curve, where returns shrink. Per extra hour a day, adjusted for activity:
| Exposure -> outcome | Below the knot, RR (95% CI) | Above the knot, RR (95% CI) | Knot (self-reported) |
|---|---|---|---|
| Total sitting -> all-cause death | 1.01 (1.00-1.01) | 1.04 (1.03-1.05) | ~8 h/day |
| Total sitting -> CVD death | 1.01 (0.99-1.02) | 1.04 (1.03-1.04) | ~6 h/day |
| TV viewing -> all-cause death | 1.03 (1.01-1.04) | 1.06 (1.05-1.08) | ~3.5 h/day |
| TV viewing -> CVD death | 1.02 (0.99-1.04) | 1.08 (1.05-1.12) | ~4 h/day |
Read the knots as rough regions, not targets. They are spline bends with no confidence interval on their position, and nearly all of the studies relied on people reporting their own sitting (Patterson et al., 2018). Three in four people in the reference population reported under 4 hours of TV a day, so the high-TV arm rests on thinner data (Patterson et al., 2018). With accelerometers the sitting-mortality curve bends higher: the excess was «more pronounced at greater than 9.5 hours» a day. (Ekelund et al., 2019) The self-report knots and the device bend measure different things and are not interchangeable. -> Physical Activity Dose and Mortality
TV viewing carries the larger risk, partly through what comes with it
TV viewing is associated with more risk than total sitting on every outcome, and most strongly with type 2 diabetes: RR 1.09 (1.07-1.12) per hour a day, with no threshold. (Patterson et al., 2018) The authors estimate that 29% (26-32%) of diabetes cases are «related to TV-viewing», but they also note that this figure «rests on the assumption of causality» and on unbiased estimates with no measurement error. (Patterson et al., 2018) Neither assumption is secure here, and undiagnosed diabetes may already have raised sitting time before anyone measured it. Snacking and evening eating tend to come with TV, so part of this association may come from diet rather than from sitting itself. -> Sedentary Behaviour and Chronic Disease Risk
Activity blunts the sitting risk, and the dose depends on how it is measured
Being active offsets much of the sitting risk, but the two main estimates use different instruments and should not be merged. With self-reported activity, about 60-75 minutes a day of moderate activity «appear to eliminate» the excess risk of death from long sitting; the same dose «attenuates, but does not eliminate» the excess risk from heavy TV viewing. (Ekelund et al., 2016) Among the most active, more than 5 hours of TV a day still carried HR 1.16 (1.05-1.28). (Ekelund et al., 2016)
With accelerometers the dose is lower, and the verb is weaker. In nine device cohorts (44 370 adults followed 4 to 14.5 years), «about 30–40 min (median of medians=34 min; IQR: 26–48 min) of MVPA per day» attenuated the risk of death in the most conservative model. (Ekelund et al., 2020) In that top third of activity, the most sedentary group did not differ significantly from the reference group, though its interval was wide (0.87-2.26) (Ekelund et al., 2020). In the least active third, who averaged about 2 minutes of MVPA a day, every level of sitting carried excess risk. The authors attribute part of the drop from 60-75 to ~34 minutes to the instrument: devices pick up movement that questionnaires miss. (Ekelund et al., 2020)
On a device, then, the dose that blunts the sitting risk (~34 min/day) is in the same range as the dose where the activity-mortality curve flattens. The same half hour or so of activity does both jobs, though it attenuates the sitting risk rather than erasing it. All of these cohorts are observational and mostly aged 40 and over. -> Physical Activity Dose and Mortality
Breaking up sitting improves blood sugar; nobody has shown it prevents disease
The common advice to get up every half hour has been tested only on short-term markers. In 37 lab crossover trials, breaking up a long sit with brief activity lowered post-meal glucose (SMD -0.54, 95% CI -0.70 to -0.37) and insulin (SMD -0.56, -0.74 to -0.38), with a smaller effect on triglycerides. (Loh et al., 2019) These are surrogates (Surrogate Outcomes). For hard outcomes, the US Physical Activity Guidelines Advisory Committee found «insufficient evidence that bouts or breaks in SB are important factors» for death, cardiovascular disease, cancer or type 2 diabetes. (Loh et al., 2019) So the cohort evidence supports sitting less in total; whether breaking up the same total changes a patient-important outcome is still unknown.
What replaces the sitting matters. Patterson’s group notes that «greater reductions in risk» may come when sitting is replaced with strenuous exercise than with walking for pleasure. (Patterson et al., 2018) The lever matters most for people who sit past the knots and for anyone concerned about diabetes, where cutting TV time is the sharper target. Aerobic movement and less sitting are two levers. Strength training is a third, and it works through a different channel.
Strength training adds benefit, mostly by keeping people able
People who do some strength training die at modestly lower rates than people who do none, even after accounting for their aerobic activity. That mortality evidence is weak, though. The stronger case for strength training runs through function — the strength, power and muscle that let an older adult rise from a chair, climb stairs and keep living independently. This ordering is this guide’s reading across Muscle-Strengthening Activity and Mortality and Power Training and Physical Function in Older Adults.
Mortality: a modest association, weakly warranted
Pooling cohort studies, people reporting any strength training had 15% lower all-cause mortality than those reporting none (RR 0.85, 95% CI 0.79-0.93), adjusted for aerobic activity; GRADE certainty was very low. Cardiovascular disease (RR 0.83), total cancer (0.88) and diabetes (0.83) ran the same way. People doing both strength and aerobic activity had 40% lower mortality than people doing neither (RR 0.60, 0.54-0.67). That 0.60 compares both against neither. It is not the 0.85 added to an aerobic benefit, so the two figures should not be stacked. (Momma et al., 2022) The meta-analyses report relative risks only, so no absolute risk difference can be stated.
Three meta-analyses agree, but they re-pool largely the same self-reported US cohorts, so the agreement does not raise certainty. Shailendra’s pool repeats the 15% figure (RR 0.85, 0.77-0.93), and every one of its 10 studies carried a high or very high risk of bias in how the exposure was measured. (Shailendra et al., 2022) This guide holds this lever at low confidence, below the aerobic-activity evidence (Muscle-Strengthening Activity and Mortality).
The dose: a region of roughly 30-80 minutes a week, not a session count
The mortality curves are measured in total minutes per week, and the benefit arrives at a small volume. Momma’s lowest risk sat at about 40 min/week (RR 0.83, 0.79-0.86), inside an authors’ headline range of about 30-60 min/week; one large earlier cohort, cited in Momma’s discussion, reported its lowest risk at 82 min/week. (Momma et al., 2022) Shailendra’s largest reduction came at about 60 min/week (RR 0.74, 0.64-0.86), but only 4 studies reported minutes: «because this analysis only included 4 studies, it should be viewed with caution.» (Shailendra et al., 2022) Read the dose as a ~30-80+ min/week region, not 40 or 60 minutes. Momma could not separate frequency from total time (Momma et al., 2022), so the curves say nothing about twice versus three times a week. The Underivable Optimum
Both curves rise again past their low point; Momma’s estimate stays below that of no strength training only up to about 140 minutes a week. That upper arm is not evidence that more strength training is harmful. Shailendra’s authors describe volumes above 60 minutes a week as giving «smaller or no mortality risk reductions», not added risk, and urge caution because only 4 studies reported minutes (Shailendra et al., 2022). The rise begins near the typical dose, between Shailendra’s 30- and 150-minute spline knots, so the main caveats are the 4-study base and self-reported minutes rather than a thin tail of data — the pattern The U-Shaped Association Artifact describes.
For diabetes, the same exposure shows no upturn: «the risk of diabetes sharply decreasing until up to 60 min/week of muscle-strengthening activities followed by a gradual decrease.» (Momma et al., 2022) That contrast fits an artifact in the mortality curve but does not prove one. The diabetes curve rests on 3 studies (Momma et al., 2022), and a real cost at high volume is not ruled out: Shailendra cites earlier clinical studies in which heavy, high-volume training «may also result in adverse effects on blood pressure, heart rate, and arterial compliance» (Shailendra et al., 2022).
This mortality curve and the muscle-growth findings below are different outcomes. Growth keeps rising with training volume while the mortality association flattens near an hour a week; the two do not contradict each other.
Any programme beats none; the choice of programme matters far less
Across a network meta-analysis of resistance-training trials, starting to train mattered much more than which programme. All load, set and frequency combinations raised strength and muscle size against no training, and comparisons between programmes mostly could not be told apart: «The 95% CrI contained zero for a striking 91% (101/111) of all between-RTx comparisons». (Currier et al., 2023) For a few sparsely studied schemes the interval against no training also crossed zero, which the authors attribute to imprecision. Currier’s ordering is therefore no ranking to follow.
The floor is low: «There was a 95% probability that RT with at least two sets or two sessions per week increased strength», and at least two sets twice a week produced muscle growth (Currier et al., 2023). Read that as a minimum region, not a located knee -> Resistance Training Prescription - Load Sets and Frequency.
Where programme did separate, strength followed load and muscle size followed volume. Heavy (at least 80% of one-repetition maximum), multi-set training two or three times a week gave the largest strength gain; higher set volume ranked highest for hypertrophy. Training to failure did not explain muscle growth in these mostly untrained participants. (Currier et al., 2023) All of these are surrogates, and the authors say so: «We do not know how these RTx affect relevant health outcomes.» (Currier et al., 2023) (Resistance Training Prescription - Load Sets and Frequency, Surrogate Outcomes)
For older adults: train for speed, work around heavy loads, and expect gains to survive a break
Training to move fast beats training to lift heavy on function tests in older adults. Against conventional strength training, power training improved activity tests — walk tests, chair rises, balance — by SMD 0.43 (0.23-0.62), at high GRADE certainty, in 15 trials of healthy adults with a mean age over 65. The endpoint stops at clinic tests: «None of the included studies used the level of physical activity in daily life as an outcome», and none counted falls. (el Hadouchi et al., 2022) (Power Training and Physical Function in Older Adults)
Blood-flow restriction is a workaround for people who cannot load heavily. In older adults, light training with a limb cuff built muscle about as well as heavy training (ES 0.21, -0.14 to 0.56) but gave less strength (ES -0.42, -0.70 to -0.14). The base is thin: 11 studies, 238 people. (Centner et al., 2018) The review counted no harms of its own, and its authors «recommend a thorough screening and physical examination of all trainees before commencing this training regimen» (Centner et al., 2018): the conditions that rule out heavy loads are the ones least studied under a cuff. (Blood Flow Restriction Training)
A planned break does not erase the gains. After a completed resistance or multicomponent programme, older adults who then stopped for 4-36 weeks (mean about 11) still scored higher on functional-capacity tests than non-exercising controls (ES 0.88, 0.47-1.29; I2 81%). That figure is a between-group effect size, not the share of gains kept, and the advantage was smaller in studies of older participants. Every trial studied a planned stop, with participants told «to carry on with their normal daily activities». (Buendía‐Romero et al., 2025) A layoff forced by illness, a fall or a hospital stay was not studied, so these results do not cover it. Lasting benefit still needs continued training. (Detraining and Residual Effects in Older Adults)
Grip strength and muscle mass are markers to measure, not targets to train toward
Weaker grip predicts earlier death. In UK Biobank adults aged 40-69, each 5 kg lower grip strength was associated with higher all-cause mortality (HR 1.20 in women, 1.16 in men), after excluding deaths in the first two years. But only 13,322 of 502,293 participants died over about seven years, so at this age the absolute difference is small. (Celis-Morales et al., 2018) Grip is cheap to measure and adds information: it improved the prediction of an office-based risk score built from age, sex, diabetes, BMI, systolic blood pressure and smoking, with no blood test (C-index gain 0.013 for all-cause mortality) (Celis-Morales et al., 2018). The absolute stakes rise with age and baseline risk (Baseline Risk and the Relative-Absolute Split).
Grip is about half heritable and drops with hidden illness, so it works as a marker that places a person in a risk group. No trial shows that training grip strength lowers mortality. Grip Strength and Mortality Low muscle mass tells the same story more weakly: older adults who died had slightly less appendicular muscle than survivors (SMD -0.18, -0.23 to -0.12). (de Santana et al., 2021) That is why the sarcopenia definition now screens on strength first — «strength is better than mass in predicting adverse outcomes». (Cruz-Jentoft et al., 2018) The consensus cut-offs for low grip strength are under 27 kg in men and under 16 kg in women (Cruz-Jentoft et al., 2018). (Low Muscle Mass and Mortality, Sarcopenia Definition and Diagnosis)
Strength training is also the main defence of muscle during weight loss, on GLP-1 drugs and through menopause, and protein is its partner input. Those questions live on their own pages: Protein and Resistance Training for Muscle and Strength, Resistance Training During Caloric Restriction, Resistance Training for Sarcopenic Obesity, GLP-1 and Lean Mass, Anabolic Resistance, Menopause and the Shifting Levers.
For older adults, the specific kind of exercise starts to matter, because the outcome that dominates is a fall.
For older adults, balance training is the best-proven exercise
Exercise prevents falls in older adults, and the trials behind that claim are strong. Every other outcome in this appraisal rests mostly on cohort data; falls rest on randomized trials rated high certainty. The work that does it is balance and functional training. Strength training on its own did not cut falls. Exercise for Preventing Falls in Older Adults
Falls: the one high-certainty trial result
A Cochrane review pooled 108 trials with 23,407 community-dwelling older adults, mean age 76 (Sherrington et al., 2019). Two counts need keeping apart: the falls rate (how many falls happen) and fallers (how many people fall at least once).
- Falls rate: rate ratio 0.77 (95% CI 0.71-0.83), 59 trials, high certainty. At a control rate of 850 falls per 1,000 people a year, exercise prevents 195 falls per 1,000 a year (between 144 and 246 fewer).
- Fallers: risk ratio 0.85 (0.81-0.89), 63 trials, high certainty — 72 fewer people falling per 1,000 a year (between 52 and 91 fewer), from a baseline of 480.
Falls are counted directly and are hard to game, and dropping the trials at high risk of bias left the result essentially unchanged. That is why this outcome, unlike the mortality curves earlier, can carry a causal verb. Exercise for Preventing Falls in Older Adults
Balance and functional training does the work; strength alone does not
The type of exercise matters here (subgroup test P=0.004) (Sherrington et al., 2019):
| Exercise type | Falls rate, rate ratio (95% CI) | Certainty |
|---|---|---|
| Balance + functional (gait, balance, task training) | 0.76 (0.70-0.81), 39 trials | High |
| Multiple types (usually balance + strength) | 0.66 (0.50-0.88), 11 trials | Moderate |
| Tai chi | 0.81 (0.67-0.99), 7 trials | Low |
| Strength training alone | 1.14 (0.67-1.97), 5 trials | Very low |
| Walking or dance alone | uncertain | Very low (Sherrington et al., 2019) |
The review’s authors put it in one line: «Exercise programmes that reduce falls primarily involve balance and functional exercises, while programmes that probably reduce falls include multiple exercise categories (typically balance and functional exercises plus resistance exercises). Tai Chi may also prevent falls but we are uncertain of the effect of resistance exercise (without balance and functional exercises), dance, or walking on the rate of falls.» (Sherrington et al., 2019)
Strength still earns its place inside the combined programmes, but not as the falls lever on its own. Falls are a balance and gait problem; strength training’s case runs through muscle and function, a different outcome. Power training improves the chair-rise and gait tests that predict falls, but no review in this appraisal has measured fall counts for it — a named gap. Power Training and Physical Function in Older Adults
A network meta-analysis of 219 trials set exercise against the rest of the falls-prevention menu (vision treatment, home-hazard checks, multifactorial programmes, education). Most exercise options with moderate-certainty benefit were supervised and lasted over three months, and balance and resistance programmes and group tai chi showed benefit on the most outcomes. Stacking other components on top of exercise added little: «Adding other interventions to exercise does not appear to substantially increase benefits. Overall, effects appear most applicable to those with elevated fall risk.» (Pillay et al., 2024) None of the interventions with moderate-certainty benefit focused on walking (Pillay et al., 2024). Both reviews draw largely on the same body of falls trials, so the network ranking refines the Cochrane result rather than confirming it independently.
Fractures matter more than falls, and trials measure them less well
People who have lived through these events rate a fracture as far worse than a fall. A year on, a hip fracture carries a disutility of 0.16 (0.12-0.20, high certainty) on a scale where 1 equals death; a fall carries 0.09 (moderate certainty that fracture is the more important outcome) (Pillay et al., 2024). Yet the fracture evidence is thinner. Exercise may reduce fall-related fractures: risk ratio 0.73 (0.56-0.95), 10 trials, low certainty — 47 instead of 64 per 1,000, about 17 fewer. Quality of life showed little difference (SMD -0.03, low certainty) (Sherrington et al., 2019). Fractures are rarer than falls and need far larger trials, so low certainty here reflects what trials could detect, not evidence of no effect.
Exercise also reaches bone by a second route: loading raises bone mineral density. In postmenopausal women the gain is small, SMD 0.27 to 0.41 across spine and hip, on a surrogate (Mohebbi et al., 2023). A fracture meta-analysis from the same research group found exercise cut major osteoporotic fractures (rate ratio 0.75, 0.59-0.94) (Hoffmann et al., 2022). Because one group produced both, the second does not independently confirm the first. Bone density cannot see the falls route, so it may understate what exercise does for fractures. Exercise and Bone Mineral Density
The pill most people reach for instead does not hold up in this group. In community-dwelling adults without deficiency, osteoporosis or a prior fracture, vitamin D with or without calcium showed no fracture benefit (Kahwati et al., 2018). Exercise for Preventing Falls in Older Adults
Higher-risk people gain more per person, but everyone at risk gains
The relative effect did not change with baseline fall risk, age (75 and over vs under 75) or group vs individual delivery. Absolute benefit scales with baseline risk, so people already prone to falling avert more falls each, while those at ordinary community risk still benefit (Sherrington et al., 2019). No subgroup claim is needed for this; it follows from arithmetic. Baseline Risk and the Relative-Absolute Split
Frailty predicts decline and responds to exercise
Frailty — reduced reserve across several body systems — predicts bad outcomes. Across 31 studies of adults 65 and over, frail and prefrail people together had higher hazards of early death (1.83, 1.68-1.98), institutional care (1.65) and falls (1.24) than robust people; frailty alone carried more risk than prefrailty (death RR 2.01 vs 1.47) (Vermeiren et al., 2016). These are associations, not effects of frailty itself.
Frailty also moves when people exercise. In 26 studies, mostly randomized trials, of identified prefrail or frail adults, activity programmes improved mobility (SMD 0.60, 0.37-0.83), daily-living tasks (SMD 0.50, 0.15-0.84) and frailty status (risk ratio 0.58, 0.36-0.93), all moderate certainty (Racey et al., 2021). Only 4 of the trials reported frailty after the intervention, so read the 0.58 loosely. Frailty
Falls are the exception in this group: activity did not clearly reduce them (risk ratio 0.80, 0.51-1.26, very low certainty, 7 studies). The review was prompted partly by the worry that standard advice «may be too advanced or too intense for a frailer population, put- ting them at risk for falls and injuries» (Racey et al., 2021). Balance training has its strongest evidence before deep frailty sets in; once frailty is established, a falls benefit is unproven rather than ruled out, since Racey’s interval (0.51-1.26) spans the benefit seen in community-dwelling older adults (rate ratio 0.77 for falls, risk ratio 0.85 for fallers). That the payoff shifts toward mobility and independence is a reasonable guess, not a tested finding. Frailty
Who supervises, and where, matters less than doing it
For older adults in general, home exercise keeps most of the benefit. A review of 34 studies from 30 head-to-head trials (adults 60 and over) found supervised programmes ahead on only one function outcome that survived sensitivity checks: knee-extension strength, SMD 0.18 (0.07-0.30). That edge is small, carries a nominally significant Begg’s test (p=0.037) that hints at small-study bias, and held in trials that gave both arms the same programme (Gómez-Redondo et al., 2024). Pooled attendance was similar, 81% in both arms, but the authors note that unsupervised attendance mostly came from participants’ diaries and that 21 of the 34 studies built some supervision into the unsupervised arm (Gómez-Redondo et al., 2024). Whether people actually did the prescribed exercise was measured in only two studies (Gómez-Redondo et al., 2024). Supervised vs Unsupervised Exercise
Two groups are different. For the frail, the trials tested structured, mostly instructor-led programmes. And in the falls network meta-analysis most exercise options with moderate-certainty benefit were supervised, so the supervision trials, which measured function rather than falls, do not settle the question for falls. Unsupervised is a sufficient default for the general older adult, not a proven equal for falls in the frail.
The balance work can be done in water. Both aquatic and land-based exercise improved balance. After the authors’ sensitivity check, water kept an edge on one overall-balance scale (the Berg) but not on three other tests, and the authors read the result as favouring water (Deng et al., 2024). Water matters for people whose knees, hips or weight make land training painful — it lets them do the work at all. Aquatic Exercise vs Land-Based Exercise for Balance in Older Adults
Osteoporosis does not rule exercise out
Serious harm from structured exercise in people with osteoporosis or osteopenia was rare, and the excess was minor aches. The one movement flagged is repetitive, loaded spinal forward flexion in a programme; the advice is to favour spinal extension instead. Ordinary daily bending is a different matter: the authors note there is no evidence that single forward-flexion movements in daily life cause vertebral fractures (Kunutsor et al., 2018).
The reassurance is weaker than it sounds. In the falls-trial update, 55 of 107 trials (51.4%) did not report adverse events at all (Kunutsor et al., 2018). The fair summary is no sign of danger, not proven safe. Exercise Safety in Osteoporosis
Broader function: a small signal on a soft measure
Researchers have tried to measure healthspan as intrinsic capacity, a WHO composite of cognition, locomotion, vitality, psychological and sensory function. In a review of 15 trials, exercise, alone or within multidomain programmes, gave the most consistent signal on it and on quality of life (11 of 13 studies), but «the effect sizes were generally small» and much of the data was self-reported in unblinded trials (Zheng et al., 2026). Intrinsic Capacity and Multidimensional Healthspan
No evidence reviewed here treats functional fitness as a lever in its own right; what exists folds into falls, frailty and sarcopenia. That is a named gap, not a finding.
Most people are not choosing between balance and nothing. They have a few hours a week and must decide how to spend them.
With a fixed weekly budget, pick the exercise you will keep doing
Once a person knows roughly how much to do, the next choice is how to spend a few hours a week. Within a single goal, the type of exercise barely changes the result; across goals it does. That turns most which exercise is best? questions into a different one: which option will you sustain, and which is least likely to hurt you? An exercise you stop doing has no effect, so adherence is part of the effect, not a footnote to it. -> Is the Exercise Modality Doing Any Work
Within one goal, the mode is mostly a wash
Head-to-head trials mostly find small or null differences between options aimed at the same outcome: stationary cycling against other exercise for knee osteoarthritis (joints section above), supervised against home exercise for older adults in general (older-adults section above), and one type of exercise against another for weight loss in obesity (weight section below). Water against land training for balance is a partial exception: water kept an edge on one balance scale (older-adults section above).
The flat result has two limits. Across goals, modes do not substitute: strength training builds strength and muscle, aerobic work builds fitness (CRF). And a bundled programme is not automatically better. In one network meta-analysis of hip and knee osteoarthritis, the mixed category ranked last on pain. (Goh, Persson, Stocks, Hou, Welton, et al., 2019) -> Is the Exercise Modality Doing Any Work. The evidence behind all of this is mostly low-to-moderate certainty: small trial pools, effects that often vanish on sensitivity analysis, and short follow-up.
HIIT buys time, not much extra fitness
High-intensity interval training (HIIT) is the most argued-over choice within aerobic exercise. Three separate lines of evidence bear on it, and no review joins them into one estimate.
On fitness, HIIT edges out moderate continuous exercise, modestly. An umbrella review found most reviews reported «similar or greater» CRF gains with HIIT, a between-group advantage of 0.52 to 3.76 mL/kg/min. Sprint-interval training against moderate continuous work showed only a small edge (SMD 0.04-0.18). Most constituent reviews scored moderate to critically low on AMSTAR-2, a quality rating for systematic reviews. Safety concerns did not appear greater than with moderate exercise, though only 7 of 24 reviews reported on safety at all. (Poon et al., 2024) -> Measuring and Raising Cardiorespiratory Fitness
In type 2 diabetes, a meta-analysis of 13 trials found HIIT raised VO2peak by 3.37 mL/kg/min more than moderate continuous training (1.88-4.87), at moderate certainty. Its HbA1c advantage (-0.37%) is low-certainty, and an earlier meta-analysis found no significant difference. These trials likely overlap with the umbrella’s, so the two do not corroborate each other independently. (Liu et al., 2018) -> HIIT vs Continuous Training for Type 2 Diabetes
HIIT feels worse while you do it, and the intensity, not the interval format, is why. Pooling 15 of the review’s 33 studies, HIIT rated about a point lower than moderate continuous exercise on the Feeling Scale halfway through the bout (MD -1.07, -1.47 to -0.67). Against vigorous continuous exercise it was no different. In the authors’ words, «Affective responses to HIIE are similar to responses to VICE during and after exercise, perhaps suggesting that the intensity may explain the responses, rather than the interval or continuous nature of the activity.» (Niven et al., 2020)
Whether that discomfort makes people quit is unproven. In the one study cited on the link, in-task feeling predicted later exercise for moderate work but not for HIIT. (Niven et al., 2020) -> Affective Response to High-Intensity Interval Exercise. Nor does harder exercise trigger more energy compensation, the unconscious offset through eating more or moving less elsewhere. Intensity was not a significant predictor, though the analysis split it crudely into two groups, which may have cost statistical power. (Riou et al., 2015) -> Exercise Energy Compensation
Put together, HIIT is a time-saving option, not a superior one, and its edge is on fitness, a surrogate, not on events. Even the time saving is plausible rather than demonstrated. No source reviewed here says how much benefit HIIT delivers per minute at a fixed weekly budget; that is a named gap.
Concentrated or spread: the weekly total carries the signal
People who pack their activity into one or two sessions a week look much like people who spread it out. A pooled analysis of four cohorts found no difference between this weekend-warrior pattern and regular activity for all-cause or cardiovascular mortality (GRADE very low). In Mexico City, all-cause HRs were 0.88 for both patterns. (Kunutsor et al., 2022) (O’Donovan et al., 2024)
A wrist-accelerometer study in UK Biobank compared the patterns within deciles of weekly MVPA (moderate-to-vigorous activity), a coarse volume match. It found no detectable difference for atrial fibrillation, heart attack, heart failure or stroke (HRs 0.92-1.12, every CI spanning 1). (Khurshid et al., 2023) These are observational nulls, not proof the patterns are equivalent. In the Mexico City cohort, weekend-warrior sessions under 30 minutes showed no significant association with mortality (HR 1.03, 0.94-1.14). That is a self-reported category, not a tested threshold. (O’Donovan et al., 2024) -> Weekend Warrior Activity Pattern and Mortality, Temporal Distribution vs Total Dose
Exercise snacks are a different question with a different outcome. These are short vigorous bouts (2 minutes or less in most trials), such as stair climbs or short sprints on a bike, spread through the day. Over 4-12 weeks they raised VO2max (SMD 1.43, 0.61-2.25, very low certainty, significant only after dropping one high-risk study) and lowered LDL. No trial measured events or mortality. (Wan et al., 2025) (Wan et al., 2025) The authors themselves frame snacks as an add-on or fallback: «ExSn should be considered a complementary or al- ternative strategy for improving health in individuals unable to meet traditional physical activity guidelines, rather than a re- placement for these recommendations.» (Wan et al., 2025) -> Exercise Snacks and Cardiometabolic Health
Snacks and short weekend sessions do not contradict each other. One moves fitness and lipid markers over weeks, the other is a mortality association over years.
Walking counts
For mortality, total steps carry the signal, at any pace (see the first section). Device studies point the same way for light activity, the incidental movement of daily life (sometimes called NEAT). -> Physical Activity Dose and Mortality
One commuting study looks like a partial exception. In UK Biobank, cycling to work was associated with lower all-cause mortality (HR 0.59, 0.42-0.83). Walking to work was not associated with lower all-cause mortality, though it was with lower cardiovascular incidence and mortality; in the distance analysis, lower cardiovascular incidence appeared only above about six miles a week. (Celis-Morales et al., 2017) That is one way of measuring walking, in one cohort, where cyclists were fitter at the start. It shows that a short walking commute may fall below the dose that matters, not that walking does not count. -> Physical Activity Dose and Mortality
A tracker adds a few hundred steps, for a while
Activity trackers and apps raise daily steps by roughly 600 to 750 over a few months. One meta-analysis found 627 steps/day (417-862, GRADE low). Another, of trials averaging 13 weeks, found 753 (440-971) in trials that measured steps directly. (Brickwood et al., 2019) (Laranjo et al., 2020) The often-quoted 1,850 steps is a mixed-outcome effect converted into steps, not a step count (Laranjo et al., 2020). That gain is about a fifth to a third of the gap between neighbouring quartiles in the step cohort, and no one has shown that it lasts.
Treat a tracker as a short-term aid, not a durable lever; the two trials that ran 12 months or longer reported lower adherence. (Brickwood et al., 2019) -> Wearable Activity Trackers and Physical Activity
Work activity is not exercise
Do not count a physically demanding job as your exercise. In a meta-analysis of 17 studies, men with high occupational activity had higher all-cause mortality than men with low occupational activity (HR 1.18, 1.05-1.34). Women showed no such association (HR 0.90, 0.80-1.01). The result held after adjustment for leisure activity, but exposure was self-reported and the causal reading is conditional. (Coenen et al., 2018) -> The Physical Activity Paradox
This sits awkwardly with the device studies, whose activity counts include movement at work. No source reviewed here reconciles the two.
Ramp up slowly; the injury that ends the attempt is the real cost
Strength training is the best-supported way to prevent sports injury, and stretching does not. Across 25 RCTs, strength training cut sports injuries to under a third (RR 0.315, 0.207-0.480). Stretching had no effect (RR 0.963, 0.846-1.095). The trials were in sport and military settings, nearly half of them in adolescents, so the transfer to a recreational or older exerciser is unproven. (Lauersen et al., 2013) -> Exercise Interventions and Sports Injury Prevention
The popular 10% a week rule has no evidence behind it. The one study that tested it found no difference between 10% and 24% weekly increases. Runners who added more than 30% a week had a raised but non-significant injury hazard against those under 10% (HR 1.59, 0.96-2.66). (Damsted et al., 2018) -> Training Load Progression and Running Injury
Tendon adapts more slowly than muscle, so a beginner’s muscles can generate forces their tendons are not yet conditioned for. (Bohm et al., 2015) -> Tendon Adaptation to Mechanical Loading The low-regret default is to raise load gradually, especially for a deconditioned or heavier beginner facing high-impact running. That is where a too-fast start most often ends in the injury that stops the attempt. For running-specific risk factors, see Stress Fracture Risk Factors in Runners and Biomechanical Risk Factors for Running Injury.
Two harms stay largely unmeasured. Adverse events in supervised-versus-unsupervised exercise trials in older adults may be undercounted, partly because supervised and unsupervised arms were not dose-matched. (Gómez-Redondo et al., 2024) Overtraining has no evidence in this appraisal at all. Both are named gaps, not reassurance.
One budget question stays: will the exercise take weight off?
Exercise trims belly fat but barely moves the scale
Exercise on its own takes off little weight. In Franz’s pooled weight-loss trials, people in exercise-only arms lost a mean 2.4 kg at 6 months and kept about 1.0 kg off at 24 months. Diet plus exercise reached 7.9 kg at 6 months and held 3.9 kg at 36 and 48 months (Franz et al., 2007). These are simple pooled means among people who finished the trials, so the kept-off figures are probably optimistic (Weight-Loss Maintenance and Metabolic Adaptation).
No type of exercise changes that. A network meta-analysis of 45 RCTs in 3,566 adults with a BMI of 30 or more found that exercise, «irrespective of which prescription, induced at best, a small reduction in weight loss (mean weight loss ranged from −0.05 to −1.01 kg).» (O’Donoghue et al., 2020) Combined aerobic and strength training ranked top for waist, body-fat percentage and fitness, though no programme significantly improved fitness; low-to-moderate-load strength training alone ranked lowest. More than 70% of the trials carried moderate-to-high risk of bias (O’Donoghue et al., 2020) (Exercise Modality for Body Composition in Obesity).
Visceral (belly) fat responds where the scale does not. Recchia pooled 40 RCTs in 2,190 adults with overweight or obesity and modelled the size of the weekly energy deficit alongside the type of intervention. Exercise had a stronger dose-response effect on visceral fat than eating less (standardized effect −0.18, 95% CI −0.33 to −0.04), and only exercise showed a dose-response: −0.15 (−0.23 to −0.07) per 1,000 kcal/week of deficit. The diet slope was flat, which the authors attribute partly to fewer diet trials (Recchia et al., 2023). Unmatched for deficit, calorie restriction looked larger (−0.53 vs −0.28), possibly because those trials ran bigger deficits.
These effects are standardized, not square centimetres of fat, and the authors call the exercise effect small. Waist circumference is the only absolute anchor: down a mean 3.15 cm with exercise and 4.67 cm with calorie restriction, again not matched for deficit (Recchia et al., 2023). Visceral fat is a surrogate; none of these trials counted heart attacks or deaths (Exercise vs Caloric Restriction for Visceral Fat, Surrogate Outcomes).
Part of the scale’s stubbornness is energy compensation. People eat back or move less to offset some of what exercise burns: «The overall mean energy compensation was 18% ± 93%.» (Riou et al., 2015) The ±93% is a standard deviation across study groups, so compensation ranged from none, or even negative, to more than was burned. In the longest studies, about 80 weeks, compensation approached 84%, on thin data, and part of the measured offset may be measurement error rather than biology (Exercise Energy Compensation).
A doubly-labelled-water analysis of 1,754 free-living adults found a typical offset of 28%, through lower basal expenditure, and a larger one in people with more body fat: 27.7% of activity calories at the 10th BMI percentile against 49.2% at the 90th (Careau et al., 2021). That gradient is a between-person association whose direction of cause the authors leave open, and Riou’s intervention data point the other way on body fat, so it is unsettled rather than a rule that exercise counts less for heavier people. Compensation may erode the weight loss more than the visceral-fat loss, because exercise moves visceral fat partly independent of weight.
Training at FATmax, the so-called fat-burning zone, is a way to set intensity, not a better way to burn fat. The target is measured per person; in people with obesity it typically falls at about 42-49% of VO2max, though not universally, which makes it light-to-moderate aerobic exercise (Jiang et al., 2026). A meta-analysis of 19 RCTs (644 people) found −3.66 kg against non-exercise or diet-only controls (Jiang et al., 2026).
That figure is not comparable with the ~1 kg above: the comparator was no exercise, and some trials added diet, which may account for part of the loss. Jiang tested diet restriction as a moderator; its discussion reports larger VO2max gains in trials with diet restriction (Jiang et al., 2026), but diet was not among the moderators its results named for body composition (Jiang et al., 2026), so how much diet added to the weight loss is unknown.
No trial in this meta-analysis tested FATmax against ordinary moderate aerobic exercise, so whether the individualised intensity adds anything is an open question (FatMax Training for Body Composition and Cardiometabolic Health).
Exercise is not the weight-loss lever, and that does not make it a weak health lever. Diet carries the deficit; exercise earns its place through fitness, visceral fat, muscle and the mortality associations set out above (Physical Activity Dose and Mortality).
Put together, the decisions come down to a short list.
What to do with this
- If you are inactive, start with any movement. The first extra minutes carry the largest gain, and light activity counts (Physical Activity Dose and Mortality).
- If you are already moderately active, the survival returns are small. More fitness still helps, with diminishing returns; that is a licence to stop optimizing, not a failure (Cardiorespiratory Fitness and Mortality).
- If you sit for long hours, add movement and sit less in total. In device-measured data, about half an hour a day of activity blunted the extra risk of long sitting, though it did not erase it; breaking up sitting improves blood sugar but has no hard-outcome evidence (Sedentary Behaviour and Chronic Disease Risk).
- Add strength training, counted in minutes per week. The observational data point to a region of roughly 30-80 minutes a week rather than a set number of sessions; its clearest payoff is staying able to do things (Muscle-Strengthening Activity and Mortality).
- After about 60, do balance and functional exercise to prevent falls. Strength training alone did not cut falls in the trials (Exercise for Preventing Falls in Older Adults).
- If you want weight off, let diet carry the deficit. Exercise helps visceral fat, fitness and muscle, not the scale (Exercise vs Caloric Restriction for Visceral Fat).
- Ramp up gradually, especially if deconditioned. Tendons adapt more slowly than muscle (Tendon Adaptation to Mechanical Loading).
Three things remain unknown: the absolute risk reductions behind most of these ratios, whether HIIT beats moderate exercise minute for minute, and how often exercise causes harm, which trials rarely report. This guide grades the reasoning from the evidence; nothing here checks a choice against what later happened to the person who made it.
Evidence box
Question What is the effect of physical activity and structured exercise (by modality, dose, intensity) on each patient-important outcome, what is the dose-response shape for each, and how do the levers rank against each other for a given stratum? Evidence included 74 sources — 52 gold, 17 high, 4 moderate, 1 method Overall certainty Medium (see Rating Certainty of Evidence) Source-selection note 4 source(s) below the gold/high evidence bar feed this page: Kodama (meta-analysis, moderate); Ross (narrative review, moderate); Oja (method, moderate); Marin-Jimenez (method, moderate). Their roles and limitations are described in the text. Last updated 2026-10-07 · Independently reviewed: No · Full edit history