Against fixed percentage-of-1RM loading, variable resistance with bands or chains (SMD 0.62, 95% CI 0.29 to 0.95), APRE (0.59, 0.16 to 1.02) and velocity-based training (0.41, 0.10 to 0.73) raised back-squat 1RM more, while RPE/reps-in-reserve loading did not differ (0.12, −0.31 to 0.56); for the countermovement jump APRE (1.08) and velocity-based training (0.62) beat percentages. A network meta-analysis of 27 small RCTs (694 healthy, active people aged 13-35; 4 trials at high risk of bias; no certainty rating), in which the top three methods did not differ significantly from each other. Read in full
Main caveat: The authors' own: few studies and small samples for several comparisons; comparisons among the top methods partly rest on indirect evidence.
Bao and colleagues asked which way of prescribing lower-body resistance training best raises maximal strength and jump power: fixed percentages of a tested 1RM (PBT), the three autoregulation methods (APRE, where the reps achieved in a set taken to volitional exhaustion set the next load; VBT, where bar velocity sets load and velocity loss ends sets; RPE-AT, where perceived exertion or reps in reserve set the load), or variable resistance (VRT), where bands or chains add load toward lockout.
They searched six databases to November 2025 and kept 27 RCTs with 694 participants aged 13-35. In the back-squat network (25 studies), VRT, APRE and VBT each beat PBT by small-to-moderate standardised margins, RPE-AT did not, and none of VRT, APRE and VBT differed significantly from each other; SUCRA put VRT first (83.1%) and PBT last (7.4%).
In the jump network (15 studies), APRE (SMD 1.08) and VBT (0.62) beat PBT, VRT did not significantly, and RPE-AT ranked last (5.7%), below PBT. Heterogeneity was 25.9% for the squat and 49.9% for the jump; no inconsistency between direct and indirect evidence and no small-study effect were detected, and age, sex mix, duration and frequency did not moderate the effects.
The authors read the rankings as probabilities, not proof of superiority. Our reading: autoregulated and band-or-chain loading look at least as good as fixed percentages for leg strength, with modest margins; perceived-effort loading did not separate from percentages; the rankings among the better methods are not settled.
Participants
27 RCTs, 694 participants; trial sizes 14-57; ages 13-35 years; older adults (over 60) and injured or ill people excluded.
Populations by Table 2: Division I and III collegiate football, academy rugby league, semi-professional rugby union, collegiate wrestlers, basketball (collegiate men and women, youth female), soccer (adolescent girls and young men), taekwondo, badminton, boxers, hockey, highly trained rowers (Held 2021), competitive female cyclists (Montalvo-Pérez 2021), resistance-trained men and women, untrained women (Arazi 2020) and recreationally active novice lifters (Shoepe 2011).
Sex: both; several all-female and all-male samples; sex not reported in four trials.
Protocol
Search: six databases from inception to 1 November 2025, plus reference lists; 13,138 records, 49 sought, 46 obtained (2 RPE-AT and 1 VRT not retrieved), 27 included.
Inclusion (PICOS): healthy, physically active people with a sport or lifting background, from novices to competitive athletes; APRE, VBT, RPE-AT or VRT as the main lower-body intervention; compared with PBT or another of these methods; back-squat 1RM or CMJ height before and after; RCTs. Concurrent or multi-component programmes excluded.
Methods as used: APRE with 10RM, 6RM or 3RM protocols, the third set to volitional exhaustion setting the fourth set's load; VBT mostly with 10-20% velocity-loss thresholds; RPE-AT at RPE 7-9; VRT with bands or chains adding about 20-40% of the load; PBT at fixed percentages until the next 1RM test. Duration 3-24 weeks, mostly 2-3 sessions a week.
Analysis: pairwise and network meta-analyses in R (metafor, netmeta); SMD (Hedges' g) with 95% CI; random-effects consistency model after node-splitting, loop and design-by-treatment tests; SUCRA ranking; funnel plots and Egger's tests; network meta-regression on female proportion, mean age, duration and frequency; leave-one-out sensitivity. One three-arm trial's control group was split.
Risk of bias: RoB 2, two reviewers. No GRADE or CINeMA certainty rating.
Key Results
| Back-squat 1RM (25 studies) vs PBT | SMD (95% CI) | SUCRA |
|---|---|---|
| VRT (bands or chains) | 0.62 (0.29 to 0.95) | 83.1% |
| APRE | 0.59 (0.16 to 1.02) | 79.1% |
| VBT | 0.41 (0.10 to 0.73) | 57.1% |
| RPE-AT | 0.12 (−0.31 to 0.56), not significant | 23.4% |
| PBT | reference | 7.4% |
Among the others (squat): VRT vs VBT 0.21 (−0.24 to 0.66), APRE vs VBT 0.18 (−0.28 to 0.64), APRE vs VRT −0.03 (−0.57 to 0.51), VRT vs RPE-AT 0.50 (−0.05 to 1.04), APRE vs RPE-AT 0.47 (−0.13 to 1.06), VBT vs RPE-AT 0.29 (−0.20 to 0.77); none significant.
Squat heterogeneity Q = 29.70, p = 0.126, I² 25.9%; global inconsistency p = 0.150.
| Countermovement jump (15 studies) vs PBT | SMD (95% CI) | SUCRA |
|---|---|---|
| APRE | 1.08 (0.30 to 1.87) | 96.6% |
| VBT | 0.62 (0.23 to 1.02) | 74.4% |
| VRT | 0.27 (−0.14 to 0.69), not significant | 49.7% |
| PBT | reference | 23.6% |
| RPE-AT | −0.38 (−1.13 to 0.36) against PBT, not significant | 5.7% |
Among the others (jump): APRE vs RPE-AT 1.47 (0.48 to 2.45) and VBT vs RPE-AT 1.01 (0.29 to 1.72), both significant; VRT vs RPE-AT 0.66 (−0.20 to 1.51), APRE vs VBT 0.46 (−0.22 to 1.14), VBT vs VRT 0.35 (−0.22 to 0.92), APRE vs VRT 0.81 (−0.08 to 1.70), not significant.
Jump heterogeneity Q = 21.96, p = 0.025, I² 49.9%; global inconsistency p = 0.157.
Meta-regression: no association of female proportion, mean age, duration or frequency with the effects in either network (all p > 0.05; frequency for the squat p = 0.092).
Leave-one-out: direction and significance held in 84.6-100% of re-runs (squat) and 93.3-100% (jump).
Risk of bias: 4 trials high risk (Helms 2018, Graham & Cleather 2021, Held 2021, Shattock & Tee 2022); 11 trials raised concerns about blinding of participants and staff, 5 about randomisation; the authors rate most trials as "high quality".
Direct comparisons by our count from Table 2: APRE was tested against PBT in 3 trials and against VBT in 3; VBT against PBT in 8; RPE-AT against PBT in 4 and against VBT in 1; VRT against PBT in 8. In the jump network APRE rests on 3 trials, one of them against PBT.
Practical Application
Fixed percentages of a tested 1RM ranked last for squat strength here (though not separable from RPE-based loading); adjusting load to the day by a rep test (APRE) or bar velocity, or adding bands or chains, gave modestly larger squat gains (SMD 0.41-0.62). The three better methods did not differ significantly from each other (wide intervals), so no order among them is shown; Inference, not tested here: the choice between them can rest on kit and habit; the gap to fixed percentages is small to moderate and comes from small trials.
Loading by perceived effort or reps in reserve did not beat fixed percentages for the squat (0.12, CI −0.31 to 0.56) and ranked last for the jump (−0.38 against PBT, not significant; clearly below APRE and VBT). Where it differs from the guide: SYNTH-101 sets the next number from the last set's reps and RPE, and cites REF-685 for autoregulation ranking above fixed percentages.This analysis splits that family: the RPE/RIR trials matched but did not beat fixed percentages, while APRE (load set from the reps completed in a set taken to exhaustion) and VBT did.It did not test the guide's rule: RPE used to decide hold or step within a double progression (add reps at the target RPE, then load), in the guide's setting of no percentage of 1RM and no velocity device, in endurance athletes. Nor does it show RPE loading doing worse than percentages for squat strength (interval −0.31 to 0.56).The RPE/RIR evidence is thin: five squat trials and only two jump trials (Arede 2020, Shattock & Tee 2022); three of the five (Helms 2018, Graham & Cleather 2021, Shattock & Tee 2022) are among the four rated high risk of bias; and targets ran from RPE 7-9 to reps in reserve of 0 to 10.So it weakens the support SYNTH-101 draws from REF-685 for RPE loading specifically, without testing or overturning the guide's rule.
APRE takes one set to volitional exhaustion each session. That conflicts with the short-of-failure rule for endurance athletes in SYNTH-43 and SYNTH-44. Inference, not tested here: the transferable part for a cyclist or runner is the logic of setting the next load from what was actually lifted, not the set to failure.
For endurance athletes the evidence is thin. One included trial was in competitive female cyclists (VBT at 55-75% 1RM against PBT at 80-90%) and one in highly trained rowers; the review gives no separate estimate for them, and no endurance performance was measured.
It disagrees in detail with the base's earlier network meta-analysis REF-685, which found no moderate or large squat differences between methods and ranked APRE first and RPE-based loading second, above velocity-based training, for the squat; this review finds clear squat margins for VRT, APRE and VBT over PBT and puts RPE-AT fourth, below VBT. Both rank fixed percentages last.
Limitations
The authors' own: few studies and small samples for several comparisons; comparisons among the top methods partly rest on indirect evidence; the jump network is small with moderate heterogeneity; methods vary within each category (band or chain load, velocity-loss threshold); samples range from novices to adolescent and competitive athletes; results do not extend to clinical, sedentary or older people.
No certainty-of-evidence rating, so the strength of each comparison is judged only from the intervals and the risk-of-bias figure.
The arms often differed in more than the method: in some VBT trials the VBT group lifted lighter (Montalvo-Pérez 2021, 55-75% against 80-90% 1RM; Fernandez Ortega 2020, 65% against 80%), so a method effect cannot be separated from a load or volume difference.
Two APRE trials come from the same first author in Division I football players (Mann 2010 and Mann 2011); the review does not say whether their samples overlap.
Three of the four high-risk trials are RPE-AT trials, so the weak showing of perceived-effort loading rests mostly on the least reliable part of the network.
RPE-AT protocols varied widely, and one RPE-AT trial (Helms 2018) is listed at RIR 5-10, which is far from the RPE 7-9 described for the category.
SUCRA ranks are probabilities of being best, not effect sizes.
English-language studies only; three eligible studies could not be obtained.
Funding: none received; the authors declare no conflicts of interest.
How this entry was checked and changed
Why this level
a network meta-analysis of RCTs whose risk of bias we read in the full text (4 of 27 trials high risk, the rest low risk or some concerns), held below A because the trials are small (14-57 participants each), no certainty rating (GRADE or CINeMA) was made, the jump network has 15 studies with moderate heterogeneity (I² 49.9%), several comparisons rest on few or indirect trials, and the compared arms often differed in load as well as in method; no marker step, since squat 1RM and jump height are performance measures