Lifting and lifespan: the dose–response for resistance training

In short

Two independent meta-analyses of cohort studies reached the same number: doing any muscle-strengthening activity rather than none was associated with a 15% lower risk of death from any cause (RR 0.85, 95% CI 0.79 to 0.93, and RR 0.85, 0.77 to 0.93).
The dose–response curve is J-shaped. Risk was lowest at about 40 minutes a week (RR 0.83, 0.79 to 0.86), and the pooled estimate stayed below 1.00 only up to roughly 140 minutes.
Doing strength work and aerobic activity, versus neither, was associated with a 40% lower risk of death (RR 0.60, 0.54 to 0.67) — a far larger gap than either alone.
Nearly all of it is observational, the activity self-reported, and the largest review graded the all-cause evidence very low.

Four analyses, one number

Momma and colleagues searched MEDLINE and Embase to June 2021, found 16 eligible prospective cohort studies, and reported that any muscle-strengthening activity rather than none was associated with a 15% lower risk of all-cause mortality: relative risk 0.85 (95% CI 0.79 to 0.93), pooled across seven studies covering 42,133 deaths among 263,058 participants.¹ Shailendra and colleagues, working from a different search and a partly different study set, reached the identical estimate — RR 0.85 (0.77 to 0.93) across six studies.²

Two earlier analyses sit slightly lower. Saeidifard and colleagues pooled 11 studies and 370,256 participants followed for a mean 8.85 years, reporting a hazard ratio of 0.79 (0.69 to 0.91).³ Stamatakis and colleagues pooled the raw data of 11 British cohorts — 80,306 adults, 5,763 deaths — and found 0.77 (0.69 to 0.87) for any strength-promoting exercise.⁴ Note the mixed currency: Momma and Shailendra report relative risks, Saeidifard and Stamatakis hazard ratios — comparable at these event rates, but not the same quantity.

Figure 1. The dashed line marks 1.00, where there is no association. The four upper estimates compare people who do some strength work with people who do none; the two lower ones compare doing both strength and aerobic activity with doing neither — hence the gap.

The benefit peaks at a strikingly small dose

The "any versus none" comparison hides the more interesting result. Momma's group ran a dose–response analysis on the six studies that reported activity in minutes, covering 236,331 participants and 37,178 deaths. They found no clear linear association but a non-linear one: the lowest relative risk, 0.83 (0.79 to 0.86), fell at 40 minutes a week, and the pooled estimate stayed below 1.00 only up to roughly 140 minutes.¹ The same J-shape appeared for cardiovascular disease, bottoming out at 60 minutes a week (RR 0.82, 0.76 to 0.90), and for total cancer at 30 minutes (RR 0.91, 0.85 to 0.97).

Shailendra's group, using four studies and a separate search, found the same shape and a deeper trough: a maximum risk reduction of 27% at around 60 minutes a week (RR 0.74, 0.64 to 0.86), noting that mortality risk reductions diminished at higher volumes.² Coleman and colleagues, following 416,420 US adults, counted sessions rather than minutes and saw the same pattern — muscle-strengthening exercise added benefit at one session a week (HR 0.89, 0.81 to 0.97) but appeared no longer beneficial at 7 times/week (HR 0.99, 0.94 to 1.04).⁵

One outcome escapes the pattern. For type 2 diabetes, Momma's group found an L-shape rather than a J: risk falling sharply up to 60 minutes a week, then continuing to fall gradually with no reversal.¹ The divergence suggests the upturn elsewhere is not a generic artefact of the modelling.

Table 1. Non-linear dose–response results for muscle-strengthening activity. The final column gives the weekly dose beyond which the pooled estimate no longer sat below 1.00.
Outcome and source	Studies	Dose at lowest risk	Risk at that dose	Below 1.00 up to
All-cause mortality Momma 2022	6	40 min/week	RR 0.83 0.79 to 0.86	~140 min/week
Cardiovascular disease Momma 2022	5	60 min/week	RR 0.82 0.76 to 0.90	~130 min/week
Total cancer Momma 2022	4	30 min/week	RR 0.91 0.85 to 0.97	~130 min/week
All-cause mortality Shailendra 2022	4	~60 min/week	RR 0.74 0.64 to 0.86	not reported

Figure 2. Blue bars give the weekly dose at which risk was lowest. The grey bar is a different quantity, shown for scale: where Momma's pooled all-cause estimate climbed back to 1.00. Read the gap as the width of the useful range, not a proven ceiling.

Strength and aerobic activity are not interchangeable

The largest contrasts here are not between lifters and non-lifters but between people who do both kinds of activity and people who do neither. Momma's joint analysis, covering 581,194 participants and 68,637 deaths, put that comparison at RR 0.60 (0.54 to 0.67) for all-cause mortality, 0.54 (0.41 to 0.70) for cardiovascular disease and 0.72 (0.53 to 0.98) for cancer mortality.¹ Saeidifard's independent estimate matched it: HR 0.60 (0.49 to 0.72).³ This is not the marginal value of adding strength work to an aerobic habit — it is the distance between doing both and doing nothing.

Stamatakis and colleagues separated the endpoints. Meeting the strength guideline alone was associated with all-cause mortality (HR 0.79, 0.66 to 0.94) and cancer mortality (0.66, 0.48 to 0.92), but their abstract reports no corresponding association with cardiovascular death. Meeting the aerobic guideline alone gave the mirror image: all-cause 0.84 (0.78 to 0.90), cardiovascular 0.78 (0.68 to 0.90). Meeting both: 0.71 (0.57 to 0.87).⁴ Different endpoints answer to different exposures — one reason substituting one form of training for the other is a poor bet.

The second strand: strength itself

A larger body of work runs parallel to this, treating muscle strength as a measured trait rather than a behaviour. Celis-Morales and colleagues followed 502,293 UK Biobank participants through 13,322 deaths over a mean 7.1 years; every 5 kg of lower grip strength carried a hazard ratio of 1.20 (1.17 to 1.23) for all-cause mortality in women and 1.16 (1.15 to 1.17) in men, with cardiovascular mortality at 1.19 (1.13 to 1.25) and 1.22 (1.18 to 1.26).⁶ The PURE study replicated this across 139,691 adults in 17 countries — 1.16 (1.13 to 1.20) per 5 kg — and found grip strength a stronger predictor of all-cause and cardiovascular death than systolic blood pressure.⁷

That last comparison should be read carefully. Grip strength is a trait, not a behaviour: the UK Biobank authors put its heritability at 52%, and a weak grip is a plausible early marker of undiagnosed illness. Prediction is not modification. The PURE investigators closed their own paper by calling for research to test whether improvement in strength reduces mortality and cardiovascular disease.⁷

What none of this can settle

The evidence is almost entirely observational. Saeidifard's 11 studies included a single randomised trial alongside ten cohorts;³ every other pooled estimate above comes from cohorts only. A trial powered on deaths in the general population would demand a scale and duration nobody has attempted, which is why the causal question stays open.

Of Momma's 16 studies, 13 assessed the activity by self-reported questionnaire and three by interview. The authors could not test for publication bias — too few studies — heterogeneity for all-cause mortality was high (I² = 83%), and they graded the certainty of the evidence for all-cause mortality, cardiovascular disease and cancer as very low, downgrading mainly for indirectness because most cohorts were American.¹ The attenuation at high doses rests on the thinnest data of all: the non-linear dose–response models, built from just four to six studies.

The literature also genuinely disagrees in one place. Stamatakis found a strong association with cancer mortality (HR 0.69, 0.56 to 0.86); Momma reported RR 0.88 (0.80 to 0.97) for total cancer and Shailendra RR 0.86 (0.78 to 0.95) for cancer mortality; Saeidifard found none at all.³ They differ in exposure definitions and which cohorts were admitted; no basis exists for calling one right.

The causal caveats are the authors' own. Celis-Morales and colleagues wrote that reverse causality is possible in any observational study and that residual confounding is always possible and the associations observed may not imply causality.⁶ People who lift differ from people who do not in ways no covariate list fully captures.

The bottom line

If the association is causal — which this evidence cannot establish — most of the mortality benefit attached to lifting appears within the first half-hour to hour a week, a smaller dose than almost anyone assumes. Nothing here supports dropping aerobic exercise in its favour: the largest and most consistent contrasts belong to people who do both. And nothing here shows more than an hour is harmful, only that the cohorts thin out badly at the top of the range.

Weekly minutes are easy to overestimate from memory. Kettle records strength workouts alongside runs and rides on iPhone, keeps the data on the device, and can write it to Apple Health with your permission.