How much strength work goes with a longer life?
Any vs none: RR 0.85 (0.79-0.93); lowest near 40 min/week (RR 0.83). Self-reported cohorts, GRADE very low: a marker of who lifts, not proof, and not a ceiling.
Main caveat: Few studies (the authors' first limit): subgroup work was restricted and publication bias could not be tested, so estimates may be inflated.
Sixteen cohorts entered the analyses, from 2012-2020, with 3809 to 479 856 participants and a longest median follow-up of 25.2 years. Any versus none went with lower all-cause mortality, cardiovascular disease, total cancer, diabetes and lung cancer.
Dose was non-linear for three outcomes: the lowest point sat at 30-60 min/week, and estimates stayed below 1.00 up to about 130-140 min/week. Our reading: the rising arm of the curve shows less association at reported high doses, where data are thin; it does not show harm.
Protocol
Search: MEDLINE and Embase to 25 October 2020, updated June 2021; English only; minimum 2 years of follow-up; surrogate outcomes excluded.
Exposure: 13 studies by self-report and 3 by interview.
Analysis: random-effects pooling; hazard ratios treated as relative risks; a restricted cubic spline for dose. Open-ended dose categories were assumed as wide as the nearest category.
Quality: a modified Newcastle-Ottawa Scale, 8 stars possible; studies scored 4-7.
Funding and interests: Japan's Ministry of Health, Labour and Welfare; none declared.
Key Results
| Outcome | Any vs none, RR (95% CI) | I² | Per 10 min/week, RR (95% CI) | Lowest point | GRADE |
|---|---|---|---|---|---|
| All-cause mortality | 0.85 (0.79-0.93) | 83% | 0.99 (0.98-1.00) | 0.83 (0.79-0.86) at 40 min/week | very low |
| Cardiovascular disease | 0.83 (0.73-0.93) | 73% | 0.996 (0.99-1.003) | 0.82 (0.76-0.90) at 60 min/week | very low |
| Total cancer | 0.88 (0.80-0.97) | 76% | 0.99 (0.98-1.004) | 0.91 (0.85-0.97) at 30 min/week | very low |
| Diabetes | 0.83 (0.77-0.89) | 36% | 0.98 (0.97-0.99) | L-shaped, steep up to 60 min/week | low |
| Lung cancer | 0.90 (0.83-0.98) | 0% | 0.99 (0.98-1.00) | — | very low |
Studies per row, any vs none / dose: all-cause 7/6, cardiovascular 7/5, total cancer 6/4, diabetes 5/3, lung 2/2. No association confirmed: colon 0.96 (0.91-1.01), kidney 0.88 (0.76-1.02), bladder 0.94 (0.84-1.05), pancreatic 1.12 (0.98-1.28). Two studies each.
Strength plus aerobic vs neither (three studies each): all-cause 0.60 (0.54-0.67), cardiovascular mortality 0.54 (0.41-0.70), total cancer mortality 0.72 (0.53-0.98). These compare people doing both with people doing neither, so they are not the value of adding strength to aerobic training.
Heterogeneity: for cardiovascular disease it fell to 0% without one study (Liu et al); for all-cause mortality it stayed unexplained by sex, quality or exposure method.
Reporting. The characteristics paragraph lists 12 mixed, 2 men-only and 3 women-only studies, 17 against 16 included (our arithmetic).
Mechanisms
Measured: none.
Authors' interpretation: strength exercise raises or keeps muscle mass, the major tissue for glucose handling, which they use to explain the diabetes curve. They suggest, with caution, that optimal doses may exist.
Inference: people who lift may also differ in income, health and other habits, and residual confounding is the authors' own stated limit.
Practical Application
How much strength work goes with a longer life? Any vs none: RR 0.85 (0.79-0.93); lowest near 40 min/week (RR 0.83). Self-reported cohorts, GRADE very low: a marker of who lifts, not proof, and not a ceiling.
Good for: a modest default for someone training for health: an hour or less a week already sits at the lowest point of these curves. The authors say the at-least-2-days recommendation could be reasonable, that a higher volume may require caution, and that they could not test frequency.
Not good for: a cap. Estimates stay below 1.00 to about 130-140 min/week, the authors call the effect of higher volume unclear, and they note that most musculoskeletal programmes in elderly people run ≥2 days/week.
Limitations
Few studies (the authors' first limit): subgroup work was restricted and publication bias could not be tested, so estimates may be inflated.
Self-report or interview of strength exercise; frequency, load and proximity to failure were not analysed.
Residual confounding and reverse causation are not removed by adjustment, which varied widely between studies.
Two databases, English only.
Mostly US samples. The authors downgraded certainty mainly for this indirectness.
Dose at the high end is thin and partly assumed by category widths.
How this entry was checked and changed
Why this level
read from the full text. A pool of cohorts is not a meta-analysis of controlled trials, so it stays at the observational ceiling. GRADE by the authors: very low for all-cause mortality, cardiovascular disease and total cancer, low for diabetes