Peripheral scope (exercise-programming) — admitted on the same evidence bar as any exposure, kept low in the ranking because attention is an anti-signal and this is a heavily-discussed, mostly small-between-option domain. It earns a page only because the evidence is gold (a Bayesian network meta-analysis, 178 strength / 119 hypertrophy RCTs) and it settles one genuinely decision-relevant thing: the resistance-training dose is not one dial — strength and hypertrophy are driven by different variables, and the gap between any-RT and no-RT dwarfs the gap between prescriptions.
Evidence-tier note (confidence: low). Single gold NMA (Currier 2023). The surrogates (1RM
strength, muscle size) are RCT-grade; the health-outcome transmission is not established by this source
(see the surrogate boundary below), and the primary trials are unblindable (moderate–high risk of bias).
Held low on total web support pending an independent line — and Currier is not independent of the
staged ACSM 2026 stand or of Morton’s protein RCT-MA (shared Phillips/McMaster team), so a second
same-lineage source would not raise it. (inferred from Currier et al., 2023)
The big rock: any prescription beats none — the between-prescription differences are second-order
Currier compared 12 prescriptions (load H ≥80% 1RM / L <80%; sets M multiset / S single; frequency ≥3 / 2 / 1 per week) against non-exercise control (CTRL). Every one beat CTRL for both strength (SMD 0.75–1.60 vs CTRL) and hypertrophy (SMD 0.10–0.66). But once you are training, the choice of prescription barely separates: «The 95% CrI contained zero for a striking 91% (101/111) of all between-RTx comparisons.» So the decision that carries the effect is train vs not-train, not which protocol — which is why Currier’s own conclusion is that «adults should engage in RT, even if they cannot meet existing recommendations», not that they hit an optimal scheme. This is the Layer-1 ranking made concrete: the first dollar (start RT) buys almost everything; optimizing the prescription is the long tail.
The decomposition (the value): strength is load-driven, hypertrophy is volume-driven
The one place prescription does matter splits by outcome — and the two do not track together, so the resistance-training dose is a terminological conflation of two different curves:
| Outcome (surrogate) | What drives it | Top-ranked RTx | Effect vs CTRL (SMD, 95% CrI) | Between-RTx separation |
|---|---|---|---|---|
| Strength (1RM) | load (≥80% 1RM), multiset | HM3 (heavy, multiset, ≥3×/wk) | «1.60 (1.38 to 1.82)» | 9 of 10 non-zero comparisons were HM2/HM3 vs a lower-load RTx |
| Hypertrophy (muscle size) | volume (sets), load ~irrelevant | HM2 (heavy, multiset, 2×/wk) | «0.66 (0.47 to 0.85)» | only 1 of 45 comparisons excluded zero |
- Strength: «higher-load, multiset programmes caused the largest strength gains» — the only variable that reliably separated prescriptions. Robust under sensitivity analysis and threshold analysis (HM3->HM2 the only revision).
- Hypertrophy: «All RT prescriptions may comparably promote muscle hypertrophy, and the influence of load was less apparent» — sets/volume, not load, rank the top prescriptions. Training to failure did not explain the hypertrophic response in these (mostly untrained) participants (network meta-regression for ‘failure’ didn’t improve fit); Currier flags failure «may… be increasingly important for trained individuals».
Why this is a real distinction, not a fake tension (parameter-table check): the two SMD columns are different outcomes measured on different instruments (1RM force vs cross-sectional area / lean mass), so load matters for one and not the other is not a contradiction to reconcile — it is two curves. Filing it as a tension would compare non-commensurable quantities.
Minimal effective dose — a floor, not a located knee
- «There was a 95% probability that RT with at least two sets or two sessions per week increased strength … and training with at least two sets and two sessions per week resulted in hypertrophy.»
- Lower-CrI floor across prescriptions: «at least a moderate (SMD>0.47)» strength and «small (SMD>0.16)» mass increase — i.e. even the leanest prescription is not trivial. (Currier et al., 2023)
The curve’s shape is under-determined here, and that is a G-gap, not a plateau. Currier coded load / sets / frequency categorically (H/L, M/S, 1/2/3), not continuously, so a true knee within load or volume cannot be located from this analysis — Currier says so and calls for continuous, model-based dose-response NMA. So the honest reading is: a large step from zero, then a broad flat region across prescriptions (hypertrophy) or a modest load-gradient (strength) — with the minimum effective dose a region (~2 sets, ~2×/week) rather than a point. (inferred from Currier et al., 2023) — the categorical-coding limit and its shape-under-determined consequence are Currier’s stated limitation read against the wiki’s dose-response vocabulary (a threshold quoted from categorical data marks the edge of the evidence, not a feature of the curve). This is another instance of The Underivable Optimum — a broad flat region (its Route 1) plus categorical/measurement under-determination (its Route 3), the same under-identification the protein ~1.62 g/kg knee carries: hold the RT dose numbers loosely too, and read the ~2 sets / ~2x per week as a floor, not an optimum.
Effect modifiers — mostly absent (route-b is quiet here)
Network meta-regression found no obvious modifying effect on relative RTx effects from age, training status, proportion female, duration, volitional fatigue, relative weekly volume load, measurement tool / region, or publication year — data-sparse nodes reduced precision. So there is little positive evidence that the relative ranking of prescriptions changes by stratum: personalization of the protocol rests on preference and constraint (Route e), not on demonstrated effect modification (Route b). Baseline (untrained) status still governs absolute gain — the big step is largest for the untrained. (Currier et al., 2023)
Sex is not a meaningful effect modifier — one prescription for both (route-b null)
The most-asked stratification of RT — should women train differently? — has a direct answer, and it is a well-bounded null on relative gains. Roberts pooled male-vs-female RELATIVE adaptation to the SAME protocol across 50 studies (ages 18-50, >=5 weeks; supplements/HRT excluded), splitting by outcome. The effect size is male-group ES minus female-group ES, so a negative value favors females; every ES is a within-group, baseline-normalized (relative) change, NOT absolute kg/cm:
| Outcome (relative gain) | k (outcomes / studies) | Pooled ES (male-minus-female) | 95% CI | I2 | Verdict |
|---|---|---|---|---|---|
| Hypertrophy | 12 / 10 | 0.07 | -0.09 to 0.23 | 0 | No sex difference (tight null) |
| Upper-body strength | 19 / 17 | -0.60 | -0.93 to -0.26 | 72.1 | Favors females (moderate) |
| Lower-body strength | 23 / 23 | -0.21 | -0.54 to 0.12 | 74.7 | No sex difference |
| (Roberts et al., 2020) |
- Headline: «males and females adapted to RT with similar effect sizes for hypertrophy and lower-body strength, but females had a larger effect size for relative upper-body strength.» (Roberts et al., 2020) The one non-null runs toward women, not away — so nothing here motivates a lighter/different female prescription; if anything untrained women gain upper-body strength at least as fast relative to baseline.
- The absolute-vs-relative trap, named by the source: «Although it is true that absolute hypertrophy and gains in strength are larger in males after RT, it seems that relative increases in both muscularity and lower-body strength are similar between the sexes, and relative gains in upper-body strength may be larger in females.» (Roberts et al., 2020) Men gain more absolute size/strength (higher baseline mass, more upper-body androgen receptors); the relative response curve is the same. Reading the absolute gap as a different response is the error — it is a different starting point (a Route-a baseline fact, not a Route-b effect modification).
- The upper-body female signal is plausibly an artifact, not biology — the authors flag it: high heterogeneity (I2 ~72%) unreduced by covariates, mostly untrained short trials, and «This could cause a ceiling effect for motor skills that may explain differences in upper-body strength because the studies were conducted in mostly untrained subjects.» (Roberts et al., 2020) Men are often more familiar with upper-body movements (e.g. bench press), leaving women more short-run motor-learning headroom. So even the one non-null may not survive longer training or trained populations — it does not upgrade to a prescription difference.
- Lifting makes women bulky — refuted on BOTH axes. Relative hypertrophy is equal, not greater, in women (ES 0.07; CI -0.09 to 0.23; I2 = 0 — an unusually clean null), and absolute muscle gain is smaller in women. The same training does not build more muscle on a woman than on a man; the testosterone gap that was once invoked to predict blunted female hypertrophy did not produce it. (Roberts et al., 2020)
Convergence with Currier’s covariate null — but a different parameter. Currier’s prescription NMA found «no» modifying effect of proportion female on the relative ranking of PRESCRIPTIONS (a meta-regression covariate); Roberts is a direct male-vs-female contrast of the response itself. These are different quantities (a between-RTx-ranking covariate vs a pooled within-protocol sex-difference ES), so this is two independent designs/teams converging on the same question — sex is not a route-(b) modifier here — not the same measurement re-pooled. (Currier et al., 2023; inferred from Roberts et al., 2020)
Decision: prescribe RT the same for both sexes (load for strength, volume for size — as above). Any sex-tailoring rests on preference/constraint (Route e) or absolute-baseline scaling (Route a), NOT on demonstrated effect modification. The source is explicit that the direct trials do not settle a prescription difference either way: «it is currently difficult to know if exercise prescription should be different between sexes.» (Roberts et al., 2020) The surrogate boundary below applies unchanged — these are 1RM/size gains, not a health outcome.
Limits (Roberts): mostly untrained subjects, short trials, high strength heterogeneity unexplained
by measured covariates, and no formal risk-of-bias scoring (the primary trials cannot blind
exercise, so the authors judged standard quality scales unusable) — a high-tier MA resting on
unblindable primaries, same design ceiling as Currier.
(Roberts et al., 2020)
The surrogate → outcome boundary — the load-bearing honesty
Strength and muscle size are surrogates (Surrogate Outcomes), and Currier is unusually explicit that the health-outcome link is not in this analysis: «We do not know how these RTx affect relevant health outcomes» and «The effects on health outcomes of various RTx remain largely unknown.» (Currier et al., 2023) The transmission differs by surrogate, and the ranking of surrogates matters more than the ranking of prescriptions:
- Strength → moderate transmission. Strength (esp. grip) and muscle-strengthening activity track lower mortality observationally -> Grip Strength and Mortality, Muscle-Strengthening Activity and Mortality (any MSA vs none: all-cause mortality RR 0.85; MSA + aerobic RR 0.60). But that is activity/strength predicting death, not this NMA’s 1RM gains reducing death — no RCT closes it.
- Strength → injury reduction is the one patient-important outcome established at CAUSAL (RCT-MA) grade — see Exercise Interventions and Sports Injury Prevention (Lauersen: strength-training RR 0.315, injuries cut to <1/3; the standout intervention, stretching null). This is the closest the resistance-training case gets to a hard endpoint on interventional rather than observational evidence. Exposure-identity caveat: Lauersen’s strength arms are eccentric / sport-specific injury-prevention protocols in young athletes, NOT the hypertrophy-oriented general RT this page prescribes — so the transfer to a recreational/older gym trainer is a transportability gap, not a settled property of “RT”. (inferred from Lauersen et al., 2013)
- Hypertrophy → weakest transmission. Low muscle mass predicts mortality (Low Muscle Mass and Mortality), but that raising size via training lowers mortality is unproven — hypertrophy is largely a surrogate for a surrogate. Do not read the 0.66 hypertrophy SMD as a health effect. (inferred from Currier et al., 2023)
- The closest-to-patient-important signal here is physical function in older adults: LM2/LM3/HM3 improved mobility and gait speed, HM3 improved balance (few studies, ≥55y) — function is on the outcome menu directly, but the evidence is thin. (Currier et al., 2023)
- Why the older-adult stratum needs this dose at all — the mechanism is Anabolic Resistance. Aging blunts the muscle-protein-synthesis response to a given protein dose; resistance training is the non-nutritional lever that partially restores that sensitivity, so the training stimulus and the higher per-meal protein target (Protein Intake for Older Adults: ~1.2 g/kg/day for the active older adult vs ~1.0 sedentary) are complementary, not substitutes — RT raises protein needs, and protein is what the restored response acts on.
Decision relevance
- The big rock is doing any resistance training at all. Prescription choice is a second-order refinement — 91% of between-protocol comparisons were indistinguishable. Someone not currently training should not wait for the right scheme.
- If the goal is strength: bias toward heavier loads (>80% 1RM), multiple sets, ~2–3×/week. Load is the one variable that reliably buys more.
- If the goal is size/hypertrophy: chase volume (sets); load is flexible — lighter loads work if the sets are there, and training to failure is not required (untrained). Pairs with the protein lever (~1.6 g/kg/day) on Protein and Resistance Training for Muscle and Strength — the other input to the same adaptation.
- Minimal effective dose: roughly 2 sets, 2×/week captures most of the available strength and size gain; more is a modestly steeper strength curve, not a different category.
- Adherence and preference win the ties. With prescriptions near-equivalent, the sustainable protocol beats the theoretically-optimal-but-abandoned one — Currier frames the whole result as licensing choice.
- Do not oversell the endpoint. These are surrogate gains; the mortality/function payoff is inferred from separate observational lines, strongest for strength, weakest for pure hypertrophy.
(inferred from Currier et al., 2023)
Limits
- Surrogates only — 1RM and muscle size; no mortality/disease endpoint (Currier states this outright).
- Categorical coding (H/L, M/S, 1/2/3) — cannot locate a continuous knee; periodized programmes, rest intervals, tempo, time-under-tension excluded/under-reported.
- Unblindable primary trials — moderate–high risk of bias (strength 22% high; hypertrophy 18% high); gold design, but the underlying RCTs cannot double-blind exercise.
- Healthy adults only — athletes, comorbidities, frail excluded; older-adult function data sparse.
- Single source, shared lineage — not independent of ACSM 2026 or Morton 2018 (Phillips/McMaster);
confidence: lowuntil an independent line lands.
(inferred from Currier et al., 2023)