Key takeaway
In a network meta-analysis of randomised trials in healthy adults (178 studies and 5097 people for strength, 119 studies and 3364 people for muscle size), nearly all resistance training prescriptions beat no exercise. Higher-load, multiset, thrice-weekly training ranked best for strength (standardised mean difference 1.60, 95% credible interval 1.38 to 1.82) and higher-load, multiset, twice-weekly training for muscle size (0.66, 0.47 to 0.85). Most prescriptions could not be statistically separated from each other (the 95% interval contained zero in 91% of the comparisons between them) and risk of bias was moderate to high, so the rankings are tentative.
Study design
Trial meta-analysis
Reading basis
Full text
Reading focus
Design & applicability
The research question
Skeletal muscle is central to physical function and metabolic health, and resistance training raises muscle strength and mass, improves physical performance and helps combat chronic disease risk. How much training works best is still contentious, because a programme combines several variables: the load lifted, the number of sets per exercise and the number of sessions per week. Earlier meta-analyses mostly examined one variable at a time, although the variables are never prescribed alone, and a standard pairwise meta-analysis can compare only two prescriptions. A network meta-analysis can compare many prescriptions at once, using direct and indirect evidence. The authors set out to compare 12 predefined combinations of load, sets and weekly frequency, plus a non-exercising control, for muscle strength, muscle size (hypertrophy) and, in older adults, physical function.
Who and how
This was a systematic review and Bayesian network meta-analysis of randomised trials, registered in PROSPERO and reported under PRISMA-NMA guidance. The authors searched MEDLINE, Embase, Emcare, SPORTDiscus, CINAHL and Web of Science up to 7 February 2022, screened 16,880 records after removing duplicates, assessed 1051 full texts and included 192 articles. Trials had to enrol healthy adults aged 18 or older, last at least 6 weeks and compare at least 2 of 13 predefined conditions: a non-exercise control and 12 prescriptions. Athletes, military personnel, people with comorbidities, explicitly unsupervised (for example home-based) training and add-on techniques such as blood flow restriction were excluded. Each trial arm was coded by load (higher means at least 80% of one-repetition maximum, lower means less), sets per exercise (single or multiple) and weekly frequency (1, 2 or at least 3 days). A three-character code gives the result: HM2 is higher load, multiple sets, twice a week; LS3 is lower load, single set, at least three days a week. The strength network had 178 studies and 5097 participants (45% women) and the hypertrophy network 119 studies and 3364 participants (47% women).
Strength was measured by one-repetition maximum or isometric or isokinetic maximum voluntary contraction, and muscle size by lean or fat-free mass, whole-muscle cross-sectional area, volume, thickness or fibre area. In trials with a mean age of 55 or older, mobility, balance and gait speed were also extracted. Because tools differed, effects were pooled as standardised mean differences corrected for small-sample bias. Models were fitted in a Bayesian framework (four chains of 50,000 iterations, the first 20,000 discarded), choosing the more parsimonious of fixed- and random-effects models. Prescriptions were ranked with the surface under the cumulative ranking curve and the probability of ranking in the top three; inconsistency was checked with node-splitting and an unrelated-mean-effects model; and threshold analysis asked how much the evidence would need to change before the top recommendation changed. Sensitivity analyses removed outlying and inconsistent studies and nodes with a single study, a network meta-regression tested covariates (age, training status, proportion of females, duration, volitional fatigue, relative weekly volume load and others), and risk of bias was assessed with the Cochrane Risk of Bias 2 tool.
What it found
- 1Nearly every prescription raised strength and muscle size compared with no exercise, but by different amounts. Standardised mean differences versus control ranged from 0.75 to 1.60 for strength, the largest for HM3 (higher load, multiple sets, at least three days a week: 1.60, 95% credible interval 1.38 to 1.82), and from 0.10 to 0.66 for muscle size, the largest for HM2 (0.66, 0.47 to 0.85). The credible interval crossed zero for two strength prescriptions (LS1 0.75, −0.16 to 1.68; HS1 0.79, −0.88 to 2.45) and three hypertrophy prescriptions (HS2 0.10, −0.57 to 0.80; HS3 0.34, −0.02 to 0.71; HM1 0.40, −0.35 to 1.17); the authors attribute this to imprecision in sparsely studied prescriptions (for example only 13 participants contributed to HS1 for strength), not to those programmes failing.
- 2For strength, heavier loads dominated the top of the ranking. The probability of ranking in the top three was 85.5% for HM3, 83.5% for HM2 and 60.5% for HM1, all higher-load multiset plans. Only 9 of 66 comparisons between prescriptions (13.6%) had credible intervals excluding zero: HM2 beat LS1, LS2, LS3, LM2 and LM3, and HM3 beat LS2, LS3, LM2 and LM3. Lower-load programmes still clearly helped against no exercise, for example LM2 (1.23, 1.01 to 1.46) and LM3 (1.07, 0.89 to 1.25).
- 3For muscle size, almost all prescriptions looked alike and multiple sets characterised the best. Top-three probabilities were 86.9% for HM2, 48.7% for LM1 and 48.3% for LM2. Only 1 of 45 comparisons between prescriptions excluded zero (HM2 over LS3), and lower-load multiset plans did well against control (LM2 0.56, 0.42 to 0.71; LM3 0.50, 0.39 to 0.61). Training to failure did not improve model fit or change the estimates, which the authors read as a sign that lifting to fatigue does not explain the muscle gains.
- 4The "higher load plus multiple sets" message held up under stress, the exact top plan for strength less so. For hypertrophy HM2 stayed first under threshold analysis, with only two comparisons sensitive to uncertainty (both from one multi-arm study, which switched the top spot to HM1). For strength, 65 comparisons could displace HM3, but the replacement was HM2 in 60 of them (92%) or HM1 in 5 (8%), still higher-load multiset plans. Excluding single-study nodes, every remaining strength prescription excluded zero against control (HM2 largest, 1.49, 1.29 to 1.70), and HM2 was largest for hypertrophy (0.59, 0.39 to 0.78).
- 5Evidence on physical function in older adults was thin. Few studies measured mobility (25 studies, 859 participants), gait speed (15 studies, 488 participants) or balance/flexibility (11 studies, 323 participants), all with a mean age of 68 years. Against control, there was a 95% probability that LM2, LM3 and HM3 improved mobility and gait speed, and only HM3 improved balance/flexibility; no differences were found between prescriptions for any function outcome.
- 6Tests for hidden problems and modifiers were largely reassuring. Node-splitting flagged only LM1 versus HM1 in the strength network (p<0.01) and LS2 versus control in the hypertrophy network (p<0.01). Age, training status, proportion of females, duration, volitional fatigue, relative weekly volume load, measurement tool and region, and publication year showed no obvious modifying effect, and relative weekly volume load worsened model fit.
Limits & uncertainty
- 1The authors note that risk of bias was moderate to high: 22% of strength studies and 18% of hypertrophy studies were rated high risk, mainly because participants knew their allocation, analyses did not estimate the effect of assignment properly, and randomisation, concealment and prespecified analyses were rarely reported. Double-blinding is unfeasible in training trials, but the other issues are common and fixable.
- 2The authors note that elite athletes, military personnel and people with chronic disease were excluded, so findings should be applied to them cautiously. The physical-function results rest on few studies of healthy older adults who were not frail, and the authors advise caution with them.
- 3The authors note that load, sets and frequency were coded as categories, so periodised programmes that overlapped categories (for example loads ranging across 60 to 90% of one-repetition maximum) were left out, and the true dose–response shape cannot be read from the network. Rest between sets, tempo, time under tension and repetition velocity were reported too rarely to model, and the meta-regression is correlational.
- 4The authors note that comparing outcome measures from different tools and body regions, and taking only one measurement per outcome per study, may strain the assumptions of network meta-analysis. Our observation: many nodes were small (for example 13 participants for HS1 in the strength network and 11 for HM1 in the hypertrophy network), so comparisons involving them lean heavily on indirect evidence.
- 5Our observation: the outcomes are strength and muscle size, not health outcomes such as disease or mortality, which the authors say remain largely unknown for different prescriptions. The review itself had no funding, but one author (SMP) reports grants or research contracts from industry and other bodies, including the US National Dairy Council, Dairy Farmers of Canada and Nestle Health Sciences.
What to take from it
For a healthy adult, the review supports starting with a resistance routine you can keep up, rather than searching for the perfect plan: nearly every combination of load, sets and weekly frequency improved strength and muscle size compared with no exercise, and most differences between plans were small enough to be indistinguishable. If your priority is raw strength, the higher-ranked plans used heavier loads (at least 80% of one-repetition maximum) with multiple sets; for muscle size, the best-ranked plans used multiple sets. The authors argue that regularly doing any prescription matters more than optimising it, and that people who cannot meet the usual advice of at least twice a week can still benefit.
The highest rank does not mean best for everyone: heavy loads may suit beginners, older adults or people with injuries poorly, and trials of explicitly unsupervised training were excluded, so unsupervised home training may differ. The review does not show that any plan changes disease risk or longevity, and it excluded athletes and people with chronic conditions, who should talk with a clinician or coach. Confidence is good that resistance training works; it is low about small differences between specific plans.
Reading your own data
Wearables record training days and heart rate but not load, sets or repetitions, so keep a simple log of exercise, weight, sets, repetitions and sessions per week, and review it over at least 6 weeks, the minimum length of these trials. To follow strength, compare the same exercise with similar technique and effort from month to month; to follow muscle size, use consistent tape measurements or body-composition scans, because body weight alone cannot tell muscle from fat or water. Smart-scale lean-mass estimates vary with hydration and device, so compare only your own readings under the same conditions.
How to read this evidence
Pooling trials can improve precision, but compare populations, controls, bias and uncertainty. Network rankings include indirect evidence and do not identify a universal best treatment.
Sources & attribution
Resistance training prescription for muscle strength and hypertrophy in healthy adults: a systematic review and Bayesian network meta-analysis
British Journal of Sports Medicine · 2023-09
DOI: 10.1136/bjsports-2023-106807 ↗PubMed · 37414459 ↗MoveVitals editorial interpretation · educational content · no individual clinical review recorded
Sources checked 2026-11-01
Practical guides
Read in context
ACSM review of 137 reviews: resistance training works, and few programme details matter
The American College of Sports Medicine pooled 137 systematic reviews (over 30,000 healthy adults) and concluded that resistance training beats no exercise for strength, muscle size, power, endurance and several function tests. Few programme details changed results consistently: heavier loads (80% or more of one-repetition maximum) and 2 to 3 sets favoured strength, and at least 10 sets a week favoured muscle size, while training to failure, exercise type and periodization did not. It summarises reviews rather than new trials, gives no pooled effect sizes, and drew mostly on untrained adults and overlapping source trials.
Medicine & Science in Sports & Exercise
Muscle-strengthening linked to lower risk of death, heart disease, cancer and diabetes
Pooling 16 prospective cohort studies of adults, people who did muscle-strengthening activity had a 15% lower risk of all-cause mortality (RR 0.85, 95% CI 0.79 to 0.93) and a 17% lower risk of both cardiovascular disease and diabetes (RR 0.83 for each) than people who did none. The pooled mortality curve was lowest at about 40 min/week (RR 0.83), but whether larger amounts help or hurt is unclear. All of this is observational, from mostly US self-reported data, and the authors rated the certainty of evidence as very low for most outcomes.
British Journal of Sports Medicine
Protein supplements and weight training: small extra gains, smaller in older adults
Pooling 49 randomised trials with 1,863 healthy adults who did resistance training for at least 6 weeks, protein supplements added 0.30 kg of fat-free mass (95% CI 0.09 to 0.52) and 2.49 kg of one-repetition-maximum strength (0.64 to 4.33) on top of training alone. These gains were small next to the effect of training itself, were weaker with older age, and the suggested plateau near 1.6 g/kg/day was not statistically significant (p=0.079). One author has received funding from the US National Dairy Council, which supported trials in the analysis.
British Journal of Sports Medicine
Weight loss in frail older adults: aerobic plus strength training improved function most
In a 6-month trial of 160 obese, mildly to moderately frail adults aged 65 or older, a weight-loss programme plus combined aerobic and resistance exercise raised the Physical Performance Test score by 5.5 points, against 3.9 points for either exercise type alone and 1.0 for controls. Every exercise group still lost lean mass, and all but the resistance group lost hip bone density, so the combined programme softened these losses rather than preventing them. The volunteers were screened, trained under supervision and were mostly white, well-educated women.
The New England Journal of Medicine

