Going from near-zero to modestly active is one of the largest, most certain health levers there is — the dose-response curve is steepest at the least-active end, so the first hour a week buys the most. Three payoffs stack, roughly independent: aerobic activity and fitness (mortality, cardiometabolic), strength training (mortality, function), and — later in life — balance and functional work (falls, independence). The dose-response is broadly monotone over the studied range; more is better with diminishing returns, and guideline thresholds mostly mark where the data thin out, not a real edge. Fitness, grip, and muscle mass predict death powerfully but are mostly unproven as targets — train the activity, not the number. Exercise is a weak weight-loss tool: the body compensates, and diet owns the calorie deficit; its benefit runs through fitness and cardiometabolic pathways, not the scale. How the levers rank depends on where you start: for the inactive, just start; for the already-active and the old, strength and balance.
Getting off the couch is the biggest lever you have
The verdict’s claim that the first hour a week buys the most is not motivational rounding — it is the measured shape of the curve. Device-measured total activity maps onto all-cause mortality as a line that falls steeply, then flattens, and it is steepest at the least-active end. In Ekelund’s harmonised meta-analysis of eight accelerometry cohorts (n=36,383, 2149 deaths), moving out of the least-active quartile roughly halves mortality: total-activity hazard ratio (HR) 0.48 (95% CI 0.43-0.54) for the second quartile against the least-active referent, then 0.34 (0.26-0.45) and 0.27 (0.23-0.32) across the top two (Ekelund et al., 2019). The single biggest drop is the first step off the floor.
And the increment that buys that halving is tiny — and mostly not exercise. Ekelund places the greatest risk reduction «when the second quarter was compared with the referent, for all activity intensities» (Ekelund et al., 2019), and the gap between those quartiles amounted to roughly 5 min/day of moderate-to-vigorous physical activity (MVPA — brisk walking and harder) over the referent, the rest made up of light, incidental movement (NEAT — the energy of everyday non-exercise motion: standing, strolling, chores) (Ekelund et al., 2019). So the near-sedentary person’s largest lever reads as move more at any intensity, and sit less — not “take up aerobic exercise.” Light movement carries most of that first, decisive drop.
Why the effect is this large is itself the finding. Accelerometry roughly doubles the association questionnaires show: the effect sizes are «about twice as large compared with those previously reported in studies assessing physical activity by self report» (Ekelund et al., 2019). Self-report drags the true gradient toward the null and blurs the light-intensity end of the curve; the device recovers both. A device-measured hazard ratio is therefore not comparable to a self-reported one for the same behaviour — the self-report plateau was partly a measurement artefact. The un-attenuated curve flattens above roughly 24 min/day of MVPA (or ~375 min/day of light activity); read 24 minutes as the centre of a broad flattening zone (~20-40 min/day), not a precise minimum, since near a plateau the exact figure barely moves the decision.
Steps are the dose most people actually carry on their wrist. Paluch’s harmonised meta-analysis (15 cohorts, n=47,471, 3013 deaths; quartile medians ~3,500 / 5,800 / 7,800 / 10,900 steps/day) finds the same shape: against the lowest quartile, mortality HR 0.60 (95% CI 0.51-0.71), 0.55 (0.49-0.62) and 0.47 (0.39-0.57) across ascending quartiles (Paluch et al., 2022). Risk falls steadily, then plateaus — and the plateau moves with age (interaction p=0.012): in adults 60 and older it arrives at ~6,000-8,000 steps/day, in those under 60 at ~8,000-10,000 (Paluch et al., 2022). These are the studied ranges, not extrapolations — the curve was mapped from roughly 3,500 to 11,000 steps/day, and above that there is little data.
The familiar 10,000-step target is a marketing number, not a threshold to clear: it traces to a 1960s Japanese pedometer campaign, and the benefit plateaus below it for most adults and well below it for the old (Paluch et al., 2022). That is the general caution about any activity cut-point. A step or minute threshold most likely marks where the data thin out, or where a round number was chosen — not a knee in the curve. For all-cause mortality the data locate no knee: the curve is smoothly steepest at the bottom and flattens to a plateau across the studied range, with no U or J shape and no harmful upper arm at any achievable dose. And total volume, not pace, carries the signal — cadence added little beyond step count (Paluch et al., 2022).
One threshold is worth stating here, because a desk worker will ask it: can activity cancel the harm of long sitting? Ekelund’s million-adult pooled analysis says a high dose can — «about 60 to 75 minutes per day» of moderate activity «appear to eliminate the increased risk of death associated with high sitting time» (Ekelund et al., 2016). Note the asymmetry: 60-75 min/day is far above the ~24 min/day that banks the mortality plateau, so enough activity to live longer is not automatically enough to offset heavy sitting — the offset asks more. (Sitting is a separate exposure with its own dose-response, treated in its own section; the estimate is self-reported and observational.)
All of this is measured activity — what a person does. The same accelerometry that maps the dose also feeds a single physiological output that predicts death better than almost any behaviour you can put in: cardiorespiratory fitness (CRF — the VO2max, or maximal oxygen uptake, the activity produces). It is a marker — a predictor, not yet a proven lever the way the dose above is. And where self-reported activity dose flattens early, objectively-measured fitness keeps paying, with no plateau in sight -> Cardiorespiratory Fitness and Mortality.
Fitness predicts death, but you train the activity, not the number
That single number is cardiorespiratory fitness (CRF) — peak oxygen uptake, VO2max, the most oxygen your body can use per minute, read in METs (1 MET = 3.5 mL/kg/min). It is one of the strongest mortality predictors in medicine, and it grades cleanly. Kodama’s meta-analysis puts each 1-MET-higher stratum at 13% lower all-cause mortality (RR 0.87, 95% CI 0.84-0.90) and 15% lower risk of CHD/CVD events (RR 0.85, 95% CI 0.82-0.88), a gradient that holds across the studied categorical range with no knee located (Kodama, 2009). Mandsager’s 122,000-patient cohort finds the same slope run further: fitness is inversely tied to mortality «with no observed upper limit of benefit» (Mandsager et al., 2018) — the elite band still beats the merely-high (adjusted HR 0.77, 95% CI 0.63-0.95).
How big is low fitness? Bigger than the classic risk factors. In Mandsager’s own adjusted model, the least-fit face 5.04x the mortality of the elite (95% CI 4.10-6.20) — a larger signal than smoking (HR 1.41), diabetes (1.40), or coronary artery disease (1.29) carry in the same model (Mandsager et al., 2018). Neither source is graded (GRADE is not applied); both are observational, and Kodama carries publication bias (Egger p=.002, moderately attenuated by trim-and-fill). So the certainty is “strong, consistent association,” not proven cause.
And that gap is the whole point: CRF is a marker, not yet a proven lever. Both authors say so plainly — the association «does not prove causation», and Kodama can only «suggest… a clinical trial to determine whether an intervention that improves CRF by exercise reduces the risk» (Mandsager et al., 2018). A high VO2max is a superb risk-stratifier and the trackable outcome of training, but a surrogate earns the status of a target only insofar as raising it is evidenced to move a patient-important outcome. Read the between-person gradient correctly: it is front-loaded (Kodama low-vs-high RR 1.70, 95% CI 1.51-1.92; most of the benefit is in escaping the bottom), it bundles heritable capacity (~50% of VO2max is genetic) and reverse causation (subclinical illness lowers fitness), so it cannot be compounded into a personal promise — 5 METs gained does not deliver 0.87^5. Train the activity, not the number.
So how far does training move CRF, and does moving it help? Meeting standard activity guidelines buys roughly a 10% CRF rise in sedentary adults, and intensity moves it more than duration — with the effort needed scaling to baseline (~50% of heart-rate reserve below 10 METs, 65-85% at 10-14 METs, >85% above) (Ross et al., 2016). High- intensity interval training (HIIT — short hard bursts) edges out moderate-intensity continuous training (MICT — steady moderate work): Poon’s gold umbrella (24 reviews) finds HIIT gives «similar or greater» CRF gains than MICT, between-group SMD 0.18-0.99 (WMD 0.52-3.76 mL/kg/min) (Poon et al., 2024). The head-to-head edge is real but modest and smallest where fitness is already normal (SMD 0.04-0.64 in healthy adults; sprint-interval-vs-MICT a wash at 0.04-0.18), and its constituent reviews are low-certainty (mostly moderate-to-critically-low AMSTAR-2). So the genuine draw of intervals is time-efficiency, not a large fitness advantage.
Does raising it lower risk? The evidence upgrades the marker toward a target — partially. Within-person change (a stronger design than the cross-sectional gradient) tracks outcomes: men who went from unfit to fit between exams cut mortality risk 44% vs those who stayed unfit (Blair), and the one randomized signal, HF-ACTION, links every 6% CRF rise to a 4% lower risk of CV death or hospitalization (Ross et al., 2016). This narrows the reverse-causation worry without closing it — the evidence stays overwhelmingly observational, and the one trial is in heart-failure patients on a composite endpoint. Net: VO2max is part lever, part marker; the proven lever underneath is the activity, and CRF is how you measure whether it worked.
But aerobic capacity is only half the engine. Raising CRF is an aerobic job — resistance training barely touches it, and carries its own, largely independent mortality payoff through a different channel: muscle and strength.
Add strength training for a payoff aerobic work does not give
That separate, independent payoff is the reason to program strength deliberately rather than assume aerobic minutes cover it. Muscle-strengthening activity (resistance training, RT) lowers mortality and major disease independently of aerobic activity — the cohorts adjust for aerobic exposure and the association survives (Momma et al., 2022). Any RT versus none, from one meta-analysis of prospective cohorts:
| Outcome | RR (95% CI) | GRADE |
|---|---|---|
| All-cause mortality | 0.85 (0.79 to 0.93), −15% | very low |
| Cardiovascular disease | 0.83 (0.73 to 0.93), −17% | very low |
| Total cancer | 0.88 (0.80 to 0.97), −12% | very low |
| Diabetes | 0.83 (0.77 to 0.89), −17% | low |
Independence makes it additive, not redundant. Because these are adjusted for aerobic activity, RT and aerobic work stack: RT plus aerobic gives all-cause mortality RR 0.60 (0.54 to 0.67) — roughly a 40% reduction versus neither, the best-evidenced activity state (Momma et al., 2022). But weight the warrant honestly: the evidence here is observational, self-reported, very-low-GRADE — a directionally-robust association, held one tier below the RCT-grade blood-pressure and aerobic levers, not above them Muscle-Strengthening Activity and Mortality.
The dose is small, and a little does most of it. Momma reports J-shaped associations for all-cause mortality, CVD and total cancer, with the maximum risk reduction (~10–20%) at ~30–60 min/week (Momma et al., 2022); the pooled nadir sits near 40 min/week. Read that as a wide, imprecise ~30–80+ min/week region (a constituent study puts the mortality nadir at 82 min/week), not a 40-minute target — on sparse self-reported data the single point is false precision. This is total weekly minutes, not frequency: the curve says nothing about session spacing, so the standard “at least 2 days/week” sits inside the effective region but is not derived from it.
Do not chase the upper arm. Above ~130–140 min/week the mortality hazard rises past 1.0 in this data, but that arm is likely artifact, not dose: the authors disown it, the evidence is very-low-GRADE at high volumes, and the diabetes curve for the same exposure is L-shaped — monotone decline to ~60 min/week over the studied range, no upturn — with a clean muscle-glucose mechanism (Momma et al., 2022). The outcome whose mechanism is clear shows no turn; the outcomes whose turn lacks a mechanism show one — the signature of an artifactual arm The U-Shaped Association Artifact. What survives is the lower arm: a modest weekly volume reaches most of the benefit, and more is not to be feared or chased for mortality.
Programming, only as far as the outcome evidence reaches. A gold network meta-analysis (178 strength / 119 hypertrophy RCTs) settles one decision-relevant thing: any prescription beats none, and the gap between protocols is second-order — the 95% CrI contained zero for 91% (101/111) of between-protocol comparisons, so the effect rides on train vs not-train (Currier et al., 2023). The authors’ own conclusion: adults should engage in RT even if they cannot meet existing recommendations (Currier et al., 2023). Where prescription does matter, it splits by goal and the two do not track together:
- Strength is load-driven — heavier loads (>=80% 1RM), multiple sets rank highest (top SMD 1.60 vs no-exercise); load is the one variable that reliably separates protocols.
- Hypertrophy is volume-driven — sets, not load, rank the top protocols (top SMD 0.66); training to failure is not required in untrained people.
Minimal effective dose: roughly 2 sets, ~2x/week captures most of the available gain — a floor, not a located optimum, because Currier coded load and volume categorically and cannot place a knee within either (Currier et al., 2023) The Underivable Optimum. Two honesty caveats bound how far this licenses detailed programming. First, these are surrogates (1RM force, muscle size), and the source states outright that their transmission to health outcomes is not in the analysis — strength/MSA carries the mortality signal above, pure hypertrophy is the weakest-evidenced Surrogate Outcomes. Second, a higher protein target acts on the training stimulus in older adults, so RT and protein are complementary inputs to the same adaptation Protein Intake for Older Adults — that is the extent of the cross-lever, not a nutrition sub-section. Beyond load-for-strength and volume-for-size, protocol choice is preference and adherence, which win the ties.
How much you move — aerobic and strength — is one axis. How much you sit is a separate axis with its own signal, and where along the activity range you start changes what a given increment buys — the next two questions.
Sitting is its own lever, and leisure activity is not occupational activity
Two facts complicate “move more.” How much a person sits is a separate axis from how much they move, and where they are active — the job or their own time — can flip the sign of the effect. Neither is captured by a daily step or minute target, and each changes a decision.
Sitting is an independent mortality axis, not the inverse of activity. Sedentary time and activity are close to uncorrelated: Willett notes «there is little correlation between sedentary behaviors and physical activity … suggesting that sedentary behaviors are not simply the opposite of physical activity» (Willett, 2012). A person can be high on both. And the sitting-mortality association is adjusted for activity — «Independent of PA, total sitting and TV viewing time are associated with greater risk for several major chronic disease outcomes» (World Health Organization, 2020). So “sit less” and “move more” are two levers, not one, and clearing a step target does not by itself address prolonged sitting. -> Physical Activity Dose and Mortality
The sitting dose-response: flat, then rising above a threshold. Pooled cohorts locate an inflection, not a straight line — «For all-cause and CVD mortality, a threshold of 6-8 h/day of total sitting and 3-4 h/day of TV viewing was identified, above which the risk is increased» (World Health Organization, 2020). The shape is threshold-then-monotone-increasing over the studied range — no clear excess below the threshold, rising risk above it, and no protective lower arm (this is not a U). Two facts the number needs: no confidence interval is reported for the threshold, so read it as a region rather than a point; and it is self-reported sitting. Objective accelerometry (Ekelund 2019) puts the inflection higher, near 9.5 h/day (Ekelund et al., 2019) — the same self-report/device gap that roughly doubles the measured activity effect. TV viewing crosses its threshold earlier (3-4 h/day) and, unlike total sitting, is only attenuated, not erased, by activity.
Activity offsets sitting — but the offsetting dose is large. About 60-75 min/day of moderate-to-vigorous physical activity (MVPA — anything from a brisk walk upward) eliminates the mortality risk of prolonged sitting — an established figure «about 60 to 75 minutes per day … appear to eliminate the increased risk of death associated with high sitting time» (Ekelund et al., 2016), derived in full elsewhere in this document. Two things to keep straight. That offsetting dose sits far above the ~24 min/day MVPA where the mortality curve itself plateaus — enough activity to bank the mortality benefit is not automatically enough to cancel heavy sitting. And the movement that works against sitting is any-intensity incidental movement — NEAT (non-exercise activity thermogenesis, the energy of everyday standing, fidgeting, and walking about) — not structured aerobic training specifically; the heaviest TV viewers kept excess risk even at high activity.
Occupational activity runs the opposite way — the physical-activity paradox. Activity performed at work associates with higher mortality, not lower. Coenen 2018 (gold SR+MA; 17 studies, 193,696 participants, mean follow-up ~19.9 y) found «men with high level occupational physical activity had an 18% increased risk of early mortality compared with those engaging in low level occupational physical activity (HR 1.18, 95% CI 1.05 to 1.34). No such association was observed among women, for whom instead a tendency for an inverse association was found (HR 0.90, 95% CI 0.80 to 1.01)» (Coenen et al., 2018). The contrast is high-vs-low occupational category, on all-cause mortality, in men; no absolute-risk translation is given, so the excess scales with the worker’s baseline risk. Confidence is LOW, and the reasons cut one way: every included study measured exposure by self-report; heterogeneity was high (I2 = 76%); the result is sex-discordant; and the harm looked stronger in healthier samples — the healthy-worker-selection signature, where fitter people select into strenuous jobs. The association nonetheless persisted after adjusting for leisure activity, so it is not merely that manual workers skip the gym. -> The Physical Activity Paradox
The decision this changes: demanding work does not substitute for leisure exercise. A warehouse worker’s all-day exertion and a desk worker’s evening run are not the same exposure. Occupational activity — >=40 h/week, static or repetitive, little recovery — does not bank the leisure-activity mortality benefit and, in men, may carry cardiovascular risk of its own. Someone whose movement is mostly at work should not read “I’m on my feet all day” as exercise already done; leisure-time moderate activity with recovery remains the evidenced lever. The evidence does not support doing less at work — only not counting it as one’s exercise.
These levers compound with age. The sitting threshold and the occupational caveat both bite harder on an older, higher-baseline-risk worker, where the same relative effect buys a larger absolute change. But past roughly 70 the outcome that dominates the ledger shifts: the question is no longer mainly whether activity postpones death, but whether it preserves the function, muscle, and balance that keep a person independent — which is where the next section turns.
After about 70, the payoff is staying on your feet
The outcome that pays now is not another year of life but another year of doing your own shopping. This section matters most after about 70, when the big longevity levers are largely pulled and the exposure that moves a patient-important outcome is the one that keeps an 80-year-old off the floor and out of the fracture ward. Falls and lost function are not surrogates standing in for something the person values later — they are what the person values. That reframes the whole ranking: the number to move is a fall averted, not a lab value.
Exercise cuts the rate of falls, and this is one of the few HIGH-certainty results in the whole activity evidence. A gold-tier Cochrane review — 108 RCTs, 23,407 community-dwelling older adults, mean age 76 — puts the rate of falls at RaR 0.77 (95% CI 0.71-0.83), HIGH certainty (GRADE), 59 RCTs: roughly a 23% lower fall rate (Sherrington et al., 2019). In absolute terms, at an illustrative control rate of 850 falls per 1000 people per year, that is about 195 (144-246) fewer falls per 1000 per year. Falls are directly observed and hard to game, exercise-versus-control is randomisable, and dropping the high-risk-of-bias trials barely moved the estimate — which is why non-blinding did not cost it a GRADE level (Sherrington et al., 2019). The feared downstream endpoint moves too: exercise reduces fall-related fractures RR 0.73 (95% CI 0.56-0.95), low certainty, 10 RCTs — promising but thinner than the falls-rate result, because fractures are rarer events that need larger trials, not because the effect is absent (Sherrington et al., 2019).
The active ingredient is balance, not volume. Balance-and-functional training carries the effect on its own — RaR 0.76 (0.70-0.81), 39 RCTs, HIGH certainty — while resistance training alone does not reduce falls (RaR 1.14, 0.67-1.97, point estimate above 1, very low certainty) (Sherrington et al., 2019). In the source’s own words: «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).» (Sherrington et al., 2019) The decision object is therefore what kind of exercise, not how much — and “walk more” is not fall-prevention advice (walking programmes have insufficient evidence here), even though walking is the intuitive default (inferred from Sherrington et al., 2019).
A network meta-analysis sharpens the dose — supervised and sustained, not a one-off class. Sherrington ranks exercise types; Pillay’s network meta-analysis for the Canadian Task Force (219 RCTs, 167,864 participants) ranks the whole fall-prevention menu — vision treatment, home-hazard assessment, multifactorial programs and the rest — and balance work still lands on top: «Fourteen of the 21 (67%) interventions with some mod- erate certainty evidence for benefit had a focus on exer- cise», the top tier being supervised, long-duration balance/resistance and group tai chi (Pillay et al., 2024). The dose is the upgrade Sherrington could not make: supervised means «> two sessions, not including ini- tial instruction» and long-duration «> 3 months» — the effective program is supervised and sustained, not a one-off class (Pillay et al., 2024). Pillay «included 125 of 283 studies included in the previous review» (Pillay et al., 2024), so it is a type-F refinement of Sherrington, not independent (type-E) corroboration.
Frailty predicts death — and, dosed right, it moves. Frailty is the syndrome of depleted physiological reserve, and it is a potent prognostic marker: across 31 prospective studies (158,764 adults >=65) the frail die at more than double the rate of the robust (pooled OR 2.34 [1.77-3.09]; HR 1.83 [1.68-1.98]), with parallel signals for disability, hospitalisation, and institutionalisation — all observational, heterogeneous (I2 95-98% for mortality), so prognosis, not proven cause (Vermeiren et al., 2016). The lever is multicomponent physical activity: in 24 RCTs of 8,022 identified prefrail/frail adults, it improves mobility (SMD 0.60 [0.37-0.83]), activities of daily living (SMD 0.50 [0.15-0.84]) and frailty status itself (SMD -1.29 [-2.22 to -0.36]; RR 0.58 [0.36-0.93]), at moderate certainty (Racey et al., 2021). Two caveats bind: the frailty-reversal figure rests on only 4 of 23 studies measuring frailty as an outcome, and the general activity dose must be dosed DOWN for the frail to avoid provoking the falls it aims to prevent.
A boundary worth stating: in the already-frail, the falls signal goes uncertain. The HIGH-certainty fall reduction above is for the general older adult; in the identified frail stratum, activity does not significantly cut falls (RR 0.80 [0.51-1.26], very low certainty, 7 studies) even while it still improves mobility and ADLs (Racey et al., 2021). Read it as sequence, not contradiction: fall-prevention exercise is best-evidenced before deep frailty, and once frailty is established the payoff shifts to function and independence Frailty.
Grip strength and muscle mass are cheap, strong risk METRICS — and that is all they are proven to be. Low grip strength predicts mortality at scale: in UK Biobank (n=502,293, ages 40-69), each 5 kg lower grip carried all-cause HR 1.20 (1.17-1.23) in women and 1.16 (1.15-1.17) in men, fully adjusted, with events in the first two years excluded (Celis-Morales et al., 2018). Over the studied grip range the gradient is monotone — no knee or plateau located (spline-judged, so weak evidence of true linearity; power thins at high strength) (Celis-Morales et al., 2018). Low muscle mass also tracks death — those who died carried about 0.18 SD less appendicular mass (SMD -0.18, 95% CI -0.23 to -0.12) across 9 cohorts of non-frail adults >=65 — a small effect, and a between-group difference that yields no absolute risk or dose curve (de Santana et al., 2021).
Grip’s single-cohort signal now upgrades to a 48-study dose-response meta-analysis (~3.1 million adults, 40+ countries): all-cause mortality falls close-to-linear over the 26-50 kg grip range, while cancer and cardiovascular mortality trace a flattened U — a significant risk reduction over 16-33 kg and 24-40 kg respectively (López-Bueno et al., 2022). Read the cause-specific U as measurement, not biology: the authors self-diagnose it — «The inversion of the right part of the dose-response curves in this study likely reflect the sparsity of data/events rather than a genuine lack of beneficial association at higher levels of handgrip strength» (López-Bueno et al., 2022) — the range-edge sparsity that manufactures a U with no confounder The U-Shaped Association Artifact. This refines Celis-Morales rather than corroborating it independently: that UK Biobank cohort sits inside the 48-study review, so it is the same evidence at higher resolution (a type-F upgrade), not a second route to the finding. And it moves the marker no nearer a target — Lopez-Bueno leaves reverse causation unaddressed and over-reaches toward exercise prescription — so grip stays a strong predictor, not a proven number to train up.
Mass and strength are different quantities, and strength tends to out-predict mass — grip is a hazard ratio per kg of force, ASMI a mean-difference in DXA-measured quantity; they are not one “muscle” number (Celis-Morales et al., 2018). But the marker-versus-lever line holds for all of them: grip, muscle mass, and VO2max are strong PREDICTORS, not proven treatment TARGETS. Grip is partly heritable (~52%) and lowered by occult illness before death, so a low value places a person in a higher-risk stratum (route-(a) baseline risk) without proving that squeezing harder buys survival — no RCT shows that raising grip, or mass, lowers mortality (Celis-Morales et al., 2018). The rule is train the activity, not the number: the surrogate-to-outcome link is unclosed for these metrics, whereas falls and function are the real outcomes to steer by Surrogate Outcomes.
What to measure, then. The EWGSOP2 consensus screens strength and function first, demoting mass to a confirmatory role — «muscle strength comes to the forefront, as it is recognised that strength is better than mass in predicting adverse outcomes» (Cruz-Jentoft et al., 2018). Probable sarcopenia is flagged on low grip (calibrated dynamometer; <27 kg men, <16 kg women) or a slow five-rise chair-stand (>15 s), enough to act before any imaging; DXA/BIA only confirms; gait speed (<=0.8 m/s) grades severity (Cruz-Jentoft et al., 2018). These cut-offs are normative — set at roughly -2 SD against a healthy-young reference, not validated against outcomes — so a threshold here marks the edge of a reference distribution, not a knee in a risk curve (Cruz-Jentoft et al., 2018).
So the older-adult prescription writes itself in kind but not in quantity: balance-and-functional work is the evidenced fall lever, strengthening defends the muscle behind it, and the metrics tell you where you stand rather than what to chase. Which raises the real question for anyone with limited time — if balance, strength, and aerobic work each earn a place, how should a fixed weekly budget be split among them?
Splitting a fixed weekly budget between cardio, strength, and walking
So take the person the last section left standing: time-limited, with a fixed weekly budget — say 90 minutes — and the question of how to split it between cardio, resistance training (RT, working muscles against load), and walking. The organizing rule is benefit-per-minute x adherence, judged against the realistic alternative — not the ideal program, but the minutes actually done. A larger dose abandoned loses to a smaller one sustained, so the split that gets done beats the split that optimizes on paper.
The single most useful allocation fact: aerobic work and strength work pay off partly independently, so a mix banks both. Muscle-strengthening activity lowers all-cause mortality adjusted for aerobic activity — RR 0.85 (0.79-0.93), very-low certainty, observational and self-reported (Momma et al., 2022) — and the two together reach RR 0.60 (0.54-0.67), well below either alone (Momma et al., 2022). Two partly separate payoffs mean concentrating the whole budget on one modality leaves the other’s benefit on the table. So the default for a mixed-goal reader is a mix, not a single modality maximized — a modest RT dose (a wide ~30-80 min/week region carries most of the mortality signal, and RT is the lever for strength, glucose disposal, and — with balance work — falls Muscle-Strengthening Activity and Mortality, Exercise for Preventing Falls in Older Adults) alongside aerobic activity, whose own mortality benefit is steepest at the least-active end (total-PA Q1->Q2 HR 0.48 (0.43-0.54), device-measured Physical Activity Dose and Mortality).
Within the cardio slice, higher intensity buys fitness in less time — direction only. For a fixed budget, interval work (HIIT) delivers comparable or slightly greater cardiorespiratory fitness (CRF) than longer moderate-continuous training (MICT) for fewer total minutes; the head-to-head edge is real but modest and smallest where fitness is already normal, with sprint intervals versus MICT essentially a wash (Poon et al., 2024) -> Measuring and Raising Cardiorespiratory Fitness. The practical case is time-efficiency, not a large fitness advantage — which is exactly what a fixed budget rewards. Two limits bound the enthusiasm. The per-unit-time CRF gain of interval-vs-continuous is a named gap: the fabric holds a direction and a standardized between-group difference, not a clean per-minute magnitude, so “how much fitness per spare minute” cannot be quoted. And intensity carries an injury / recovery counterweight — more load per session raises musculoskeletal-injury and inadequate-recovery risk, patient-important outcomes the fabric does not yet quantify (a named gap, not a settled null). Push intensity for time-efficiency, but not past what recovers.
Walking is the cheap, high-adherence contributor that largely lands outside the dedicated budget. Much of it arrives as NEAT — non-exercise activity thermogenesis, the movement of daily living (walking to transit, stairs, standing, errands) that no one schedules. Because the mortality curve is steepest at the bottom, incidental steps are worth the most for someone starting near-sedentary, and they cost almost no willpower — high adherence by construction; the step curve plateaus around 7000-9000/day (Q4 HR 0.47 (0.39-0.57), harmonised cohorts (Paluch et al., 2022)) Physical Activity Dose and Mortality. So walking rarely competes for the 90 minutes at all: it is the free base layer under the split, and raising it is often the highest benefit-per-scheduled-minute move available, because the minutes are not scheduled.
One caution the budget cannot escape, whatever the split: the body partly compensates for the energy that exercise burns — roughly 28% offset on average via reduced resting expenditure, rising with adiposity (Careau et al., 2021) -> Exercise Energy Compensation. That erodes the weight return without touching the fitness, mortality, or function returns the split is actually built on. Which points to the one thing no allocation of this budget will buy — weight loss.
What exercise does not do: melt the scale
The one thing the movement budget will not buy is weight loss, because the body treats a training-induced calorie deficit as something to close, not to keep. Careau’s doubly-labelled-water landmark (n=1,754 free-living adults) measures the leak directly: energy compensation averages ~28%, so only about 72% of the calories burned in extra activity actually leave the daily energy budget — «energy compensation by a typical human averages 28% due to reduced BEE; this suggests that only 72% of the extra calories we burn from additional activity translates into extra calories burned that day» (Careau et al., 2021). Part of that offset is not the fork at all but a quiet drop in basal energy expenditure — the body spends less at rest to pay for what it spent moving. The rest is behavioural: cut NEAT (non-exercise activity thermogenesis — the calories of fidgeting, standing, and unstructured daily movement) and eat somewhat more. No CI is reported on the 28% point figure; the existence of the effect is carried by the slope, not this single number.
The intervention literature agrees on direction and adds two twists that make the lever weaker still. Riou’s SR of 61 exercise trials puts the mean offset at 18% with a standard deviation of ±93% — the spread is five times the mean, so the average barely predicts any one person (Riou et al., 2015). Compensation worsens with duration (it «approached 84%» by ~80 weeks, though that point rests on thin long-trial data) and, in Careau’s between-person gradient, worsens with adiposity — from ~27.7% offset at the 10th BMI percentile to ~49.2% at the 90th (Careau et al., 2021), so the person carrying the most fat, who most wants to burn it off, compensates the most (an association, direction contested — Riou’s intervention interaction runs the other way short-term). The intuitive worry that higher intensity makes the offset worse is not supported — Riou found intensity a non-significant predictor — but that test was underpowered (intensity dichotomised at 60% VO2max), so read it as insufficient evidence, not refutation, not a licence to expect HIIT to “count” more (Riou et al., 2015).
So on the scale — the marker most people watch — exercise is a weak lever. But the scale is measuring the wrong thing, because the depot that carries the cardiometabolic risk moves on a different curve.
At a matched weekly energy deficit, exercise beats diet for visceral fat specifically. Recchia’s SR+MA of 40 RCTs (2,190 adults, gold-tier) is the first to compare the two while holding the weekly caloric deficit constant — which isolates the lever from the deficit size. Two facts define the curve. First, only exercise is dose-dependent: each extra 1,000 kcal/week of deficit spent on exercise removes measurably more visceral fat (ES −0.15, 95% CI −0.23 to −0.07, p<0.001), while diet’s per-deficit slope is flat and null (ES 0.03, −0.12 to 0.18, p=0.64) (Recchia et al., 2023). Second, the deficit-controlled head-to-head favours exercise (ES −0.18, −0.33 to −0.04, p=0.012), at moderate GRADE certainty (each arm downgraded one level for bias and heterogeneity) (Recchia et al., 2023). Three caveats keep this honest: these are standardized effect sizes, not cm² of fat (no absolute volume is recoverable); the slope is a single meta-regression coefficient over the studied range, so it shows a direction, not a located knee or a proven-linear curve; and diet’s flat slope is more likely a power artifact (16 vs 46 effects) than a true zero — do not read it as “diet dose is irrelevant” (Recchia et al., 2023).
The two findings compose rather than conflict. Compensation says exercise under-delivers on the energy deficit (so it disappoints the scale); Recchia says exercise over-delivers on visceral fat per unit prescribed deficit — because a large part of its fat effect is weight-independent, the compensated calories blunt the weight benefit without erasing the depot benefit. Recchia’s vivid version, reporting Verheggen’s earlier pool: «In the absence of weight loss, exercise produced a 6.1% reduc- tion in visceral fat, whereas hypocaloric diets showed essen- tially no change» (Recchia et al., 2023). Exercise reshapes the metabolically active depot even when body weight barely moves.
The weak weight-loss effect is not a weak health effect — that is the pivot the section is built to protect. Exercise’s mortality and cardiometabolic benefit runs through cardiorespiratory fitness, cardiometabolic function, and visceral-fat reduction — pathways that are largely independent of the number on the scale. Compensation offsets the energy balance that weight loss depends on; it does not touch the fitness and visceral-depot channels that carry the outcome benefit. Judging exercise by scale weight therefore imports a category error — it grades a fitness-and-depot intervention on a body-mass outcome it was never the strong lever for. The scale is the wrong scoreboard. The real reach of movement is not into body mass at all but into pain and joint function, cancer risk, and the brain — which is where the next sections go.
Movement is safe, and it reaches pain, joints, cancer, and the brain
The mortality and cardiometabolic case is the core, but movement’s reach does not stop at the heart. Four other patient-important outcomes — chronic pain, joint function, cancer risk, and cognition — carry evidence worth stating plainly, and the honest verdict differs sharply across them.
Chronic pain — safe, probably helps a little, evidence genuinely weak
For chronic non-cancer pain, the load-bearing finding is safety: a gold-tier Cochrane umbrella (21 reviews, 381 studies, ~37,000 people) found «none of the physical and activity interventions assessed appeared to cause harm to the participants» (Geneen et al., 2017) — adverse events were mostly transient soreness, and no review worsened pain. On benefit, physical function is the most robust signal (significantly improved in 14 reviews, though only small-to-moderate effect sizes), while self-reported pain moved inconsistently (low certainty, tiny underpowered trials). The decision this changes is the fear-avoidance one: a patient told for decades to rest, afraid that movement will damage them, has that fear directly countered — it is safe, it may help function, start. Certainty stays low because the outcome (self-reported pain) is measured worst; this is a probable-small-benefit, not a demonstrated one.
Knee osteoarthritis — weight loss is the dominant lever, exercise a genuine adjunct
For the overweight or obese person with knee OA, weight loss is the primary modifiable lever and exercise is a real but secondary one — present them in that order. The IDEA RCT (n=454, 18 months) split the mechanisms cleanly: diet-driven weight loss lowers knee compressive load (D vs exercise-only -200 N, 95% CI 55-345) and systemic inflammation, while exercise converts that into function and symptom relief. Diet+exercise beat exercise alone on pain (mean 1.02 on WOMAC 0-20, 95% CI 0.33-1.71) with 38% reporting little/no pain versus ~21%; the >=10% weight-loss target sits on a monotone dose-response over the studied 0-32% range (no knee located) (Messier et al., 2013). Exercise is nonetheless additive and safe across arthritis broadly (EULAR MA, 49 RCTs: moderate gains in fitness SMD 0.56 and strength SMD 0.54; «No detrimental effects of PA were reported in any study» (Rausch Osthoff et al., 2018)) — so lose weight AND exercise, not either/or. Scope note: this is the risk-factor / function lever, not clinical management of established disease; the weight-loss lever also presupposes overweight and does not transport to the lean patient -> Knee Osteoarthritis and Modifiable Levers.
Cancer — activity associates with lower risk of several cancers
Physical activity is graded by WCRF/AICR (2018, the gold-standard continuously-updated synthesis) as reducing risk of colon cancer (Convincing) and endometrial and postmenopausal breast cancer (Probable) — placing it, alongside body fatness and alcohol, among the leading diet-adjacent cancer levers after tobacco. WCRF reads the gradient as «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). Two cautions travel with it: the evidence base is observational cohorts (few lifetime-diet RCTs are feasible), and these are population-level causal grades, not per-person effects — no source here shows that one person’s taking up activity lowers their cancer, and WCRF does not quantify a per-person magnitude (World Cancer Research Fund & American Institute for Cancer Research, 2018). So: associates with lower risk, strong grade on breadth of sites, but read as risk not prevention -> Diet Physical Activity and Cancer Prevention.
Cognition and dementia — one of several levers, modest and uncertain
Physical inactivity is one of the 2024 Lancet Commission’s 14 modifiable dementia risk factors (Livingston et al., 2024). Treated as an isolated lever, physical activity carries a strong protective association: Iso-Markku’s 58-study meta-analysis reports all-cause dementia RR 0.80 (95% CI 0.77-0.84, n=257,983), Alzheimer’s 0.86 (0.80-0.93), and vascular dementia 0.79 (0.66-0.95) (Iso-Markku et al., 2022). What lifts this above the usual observational-dementia caveat is that the association holds even in follow-ups of 20 years or more for all-cause dementia and Alzheimer’s — the check that meets the field’s central worry, that short-follow-up studies inflate the effect because incipient dementia lowers activity years before diagnosis (reverse causation) (Iso-Markku et al., 2022). It is an association, not a trial effect. (The individual-participant analysis reporting the association absent beyond 10-year follow-up, which these >=20-year data rebut, is not yet held.)
That is the observational arm. The randomized tests that could turn it into a trial effect bundle exercise with other levers, and they are mixed: FINGER moved a cognitive-composite surrogate by a small margin (Cohen’s d 0.13); MAPT was null on the same kind of surrogate at population level; and preDIVA found no effect on the hard endpoint, clinical dementia incidence (HR 0.92, 95% CI 0.71-1.19) (Ngandu et al., 2015). Pooling the two hard-endpoint trials at participant level (Coley 2025) found no dementia-incidence benefit overall and no responder subgroup — including no benefit in the higher-risk strata where the surrogate signal had concentrated (Coley et al., 2025). None of these isolates exercise from the bundle.
The honest state is a split verdict: the single-lever observational signal is strong and reverse-causation-checked, while the interventional evidence — which never isolates exercise — stays weak. So exercise is a plausible, modest cognitive lever worth pulling because it is a cardiometabolic big rock already, not because a randomized dementia payoff is demonstrated -> Dementia Prevention and Modifiable Risk Factors, Multidomain Lifestyle Intervention and Cognitive Decline.
With movement’s reach across these outcomes mapped, the last question is not whether it helps but how the levers rank for a given person — and where, once the big rocks are pulled, the ceiling on further gains sits.
How the levers rank, and what is still open
Those outcomes do not rank themselves, and where the ceiling sits depends entirely on who is asking. No universal ordering of exercise levers exists. Layer 1 ranks by expected effect size x certainty and asks one question — what is the largest remediable gap in this stratum? — so the order changes with the person, not with the evidence Layer 1 - Ranking Interventions for a Stratum.
- For the inactive/sedentary person, “just start” dominates every refinement. The dose-response curve is steepest at the least-active end, so the first hour a week is the largest lever on the whole menu (device-measured total activity, HR 0.34 [0.27-0.43]). Its warrant is a large, consistent observational association across harmonised accelerometry cohorts — not a GRADE-HIGH grade (the one HIGH-certainty result in this deliverable is the Cochrane falls estimate); what makes it the safest move is effect size x consistency, not a certified certainty level Physical Activity Dose and Mortality. While near-total inactivity stands — a named big rock — no question about intensity, split, or modality competes with it; every other exercise decision is second-order until it is pulled.
- For the already-active midlife adult, that big lever is spent, so the marginal questions shift. They move to strength training — an independent mortality-and-function payoff that aerobic minutes do not buy Muscle-Strengthening Activity and Mortality — and to intensity and allocation. These are real but smaller levers, now competing among themselves rather than against inactivity.
- For the older or frail adult (after ~70), the ranking changes again. Balance and strength work that keeps a person on their feet moves to the top, because the dominant outcome has shifted from long-run mortality to falls, function and independence (exercise cuts falls -23%, HIGH certainty, Cochrane) Exercise for Preventing Falls in Older Adults.
The ceiling is itself a finding. Once the big rock is pulled — once someone is active — the curve flattens: maximal risk reduction is reached near ~24 min/day of MVPA (a plateau located over the studied range), so remaining gains are small by construction Physical Activity Dose and Mortality. Reporting your remaining levers here are small is a real result, not a failure to find one — it licenses an already-active person to stop optimizing their weekly split rather than implying they should push harder for a gain the evidence does not show.
What the evidence does not settle — named as absence, not as a direction:
- The per-unit-time fitness magnitude for higher-intensity vs moderate work is not held. That intervals raise cardiorespiratory fitness somewhat faster per minute is directional; the size of that per-minute advantage, and whether it carries into a hard outcome, the deliverable does not have. Direction only, no magnitude.
- “Functional fitness” is not held as a distinct construct. The falls, frailty and sarcopenia pages already carry what the term would name — mobility, staying upright, independence; there is no separate evidence base for it, and treating it as one would double-count.
- Overtraining and CNS fatigue are consensus and coaching lore, not patient-outcome evidence. The wiki holds no appraisal of either against a measured health outcome. That is a named gap, not a finding in either direction.
- General adverse events of exercise are systematically under-reported in trials. Outside the well-studied osteoporosis/fracture case, trials record benefits far more completely than harms, so an apparently clean safety profile reflects in part what was measured, not what occurred — a streetlight gap. The deliverable names it and does not fill it.
Where this nets out
Start from where you are — that, not the perfect program, is what sets the order. The one move that matters most is the one you have not made. If you are inactive, start: any regular movement, at any intensity, is the largest and most certain lever on this page, and nothing about an ideal routine competes with it. If you are already active, the further gains are smaller and split roughly three ways — add strength training for the independent payoff, keep the aerobic base, and after about 70 add the balance and functional work that keeps you on your feet. Do not recruit exercise to move the scale: its benefit runs through fitness, cardiometabolic and visceral-fat pathways, largely off body weight.
And know when to stop optimizing. Past the big rock the curve is flat, so the distance between a good weekly routine and a theoretically optimal one is small and largely unmeasured. The open questions are real — how much faster intervals build fitness per minute, where overtraining actually bites, how often exercise harms rather than helps — but each sits below the threshold where it would change what an active person does next. The largest decision here stays binary: move, or don’t. Everything after that is refinement.
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 30 sources — 18 gold, 9 high, 2 moderate Overall certainty High (see Rating Certainty of Evidence) Source-selection note 2 source(s) below the gold evidence bar feed this page: Kodama (meta-analysis, moderate); Ross (narrative review, moderate). Each labelled by tier; none load-bearing for the core claims. Last updated 2026-08-31 · Independently reviewed: No · Full edit history