REF-1680 · Quittmann & Piehl, 2026 · European Journal of Applied Physiology, · Paper · PubMed · Free full text
The 3-min all-out cycling test repeats well but does not give one rider's critical power or W′: across 20 small studies (284 participants), end-test power was reliable within three weeks (ICC 0.89-0.99) yet sat above critical power (pooled individual data: mean +11.7 W, 95% limits −37.1 to +60.5 W; random-effects gz 0.408, 95% CI 0.196-0.620), and work above end-test power agreed poorly with W′ (r = 0.499; limits −14.1 to +8.0 kJ). The authors advise isokinetic testing at 100 rpm or more after a full familiarisation and treating both numbers as their own measures. Read in full
Main caveat: The authors': methods, statistics, cadence and resistance, encouragement, familiarisation and warm-ups differed between studies.
Quittmann and Piehl (German Sport University Cologne) collected every study that tested the reliability or the validity of the 3-min all-out cycling test, in which the mean power of the last 30 s (end-test power, EP) is meant to estimate critical power (CP) and the work done above it (WEP) to estimate W′.
They searched two databases from 2006 to June 2025 and included 20 studies with 284 participants, mostly on a Lode Excalibur Sport ergometer in linear mode. Repeat tests within three weeks agreed well: intraclass correlations of 0.89-0.99 for EP and 0.76-0.98 for WEP. Validity was another matter.
Pooling individual data from studies that measured both, EP correlated very closely with CP (r = 0.993) but sat 11.7 W above it on average, with 95% limits from 37.1 W below to 60.5 W above, and the random-effects meta-analysis confirmed a systematic overestimate (gz 0.408) whose prediction interval stayed above zero.
WEP correlated only moderately with W′ (r = 0.499) and tended to underestimate it, with a prediction interval that crossed zero.
The authors find that test mode and cadence, familiarisation and encouragement move the results, advise against using the test to set an individual's CP or W′, and recommend isokinetic mode at 100 rpm or higher after a full familiarisation, with strong encouragement and no elapsed-time feedback. They treat EP and WEP as measures in their own right.
Participants
20 studies, 284 participants (as stated in the abstract); per study 6 to 53 in the text (the abstract says 6-21).
Reliability studies: 6-53 participants, ages 14.5 ± 0.3 to 40.5 ± 10.7 years, 0-40% female, performance levels 2-4.
Validity studies: 6-50 participants, ages 22 ± 2 to 42 ± 10 years, 0-67% female, performance levels 1-5; three studies did not report sex.
Protocol
Search: PubMed and Web of Science on 20 June 2025, from 2006 (the year the test was introduced); 81 records, 53 after removing 28 duplicates, 51 full texts, 20 included.
Inclusion: peer-reviewed original English-language studies with empirical 3-min all-out cycling data, an appropriate protocol and reference test, and reliability or validity data. Excluded: other sports, comparisons of different conditions for reliability, modified 3-min tests as the reference, tests not done all-out, reviews.
Ergometers and modes: Lode Excalibur Sport in most studies, then CompuTrainer, smart trainers and SRM; linear mode in 13, isokinetic in 4, freely geared on a smart trainer in 3.
Reference standards: in 10 of 15 validity studies, CP and W′ from several time-to-exhaustion trials; otherwise a 16.1-km time trial, time to exhaustion at CP, maximal metabolic steady state from VO2 or lactate, muscle oxygenation, or maximal accumulated oxygen deficit.
Quality: a modified Brink and Louw tool, 9 of 13 items used, rated by two reviewers (66.9% immediate agreement over 148 items, the rest by consensus).
Analysis: ICCs read by Koo and Li bands; validity by R² and limits of agreement; individual data (some read from scatter plots) pooled into regression and Bland-Altman plots; Hedges' gz per effect, random-effects pooling with 95% prediction intervals, fixed-effect for comparison, common-language effect size; no subgroup analysis by method because of the few effects; funnel plots; GRADE domains for EP and WEP reliability and validity.
Funding and interests: open-access funding by Projekt DEAL; no conflicts declared.
Key Results
Reliability (test-retest within three weeks, 8 studies):
| Measure | Result |
|---|---|
| EP, ICC | 0.89-0.99 ('good' to 'excellent') |
| WEP, ICC | 0.76-0.98 |
| EP / WEP, Pearson r (3 studies) | 0.97-0.99 / 0.90-0.99 |
| Smart trainers, freely geared (53 riders) | EP CV 3.3%, WEP CV 17.9% |
| Lode, linear vs isokinetic (12 riders) | EP ICC 0.99 (CV 1.2%) vs 0.97 (CV 1.9%); WEP 0.98 (CV 5.4%) vs 0.94 (CV 8.4%) |
| Isokinetic 60 vs 100 rpm (14 riders) | EP ICC 0.91 (typical error 11.2 W) vs 0.95 (9.2 W); WEP 0.84 (1.8 kJ) vs 0.79 (1.3 kJ) |
Validity against CP and W′ from several time-to-exhaustion trials:
| Comparison | Result |
|---|---|
| EP vs CP, pooled individual data (N = 159) | r = 0.993, SEE 21 W; mean difference +11.7 W, 95% limits −37.1 to +60.5 W |
| WEP vs W′, pooled individual data | r = 0.499, SEE 5.4 kJ; mean difference −3.0 kJ, 95% limits −14.1 to +8.0 kJ |
| EP vs CP, random-effects (k = 15) | gz +0.408 (95% CI 0.196 to 0.620; prediction interval 0.024 to 0.792); fixed-effect 0.041 (0.003 to 0.079) |
| WEP vs W′, random-effects (k = 15) | gz −0.389 (95% CI −0.719 to −0.059; prediction interval −0.954 to 0.176); fixed-effect 0.048 (−0.047 to 0.143) |
| Chance a rider's EP exceeds CP / WEP exceeds W′ | 62.1 ± 18.5% / 35.6 ± 17.3% |
| R² across the 10 CP-validation studies | EP 27-98%; WEP 0.09-70.6% |
Certainty (GRADE domains): EP reliability raised no concern, WEP reliability moderate concern, and the validity of both high concern, mainly from inconsistency and imprecision, then risk of bias and publication bias.
Quality items met: 35 of 49 (71.4%) for reliability studies and 65 of 96 (67.7%) for validity studies.
Publication bias: the EP funnel plot was asymmetric, so a bias toward small EP-CP differences cannot be ruled out.
Methods that moved agreement (single studies): higher cadence lowered EP and improved its agreement with CP (best limits 11 ± 26 W at preferred cadence + 10 rpm); isokinetic mode agreed better than linear mode in one study (4.03 ± 29.68 W against 30.22 ± 46.75 W); CP from three efforts of 1, 4 and 10 min agreed better with EP than CP from five efforts; WEP depended on cadence.
Reporting discrepancies (our reading): the abstract prints the prediction intervals in brackets as if they were confidence intervals; the methods speak of 13 effects from 8 studies while the results pool 15; the abstract gives per-study samples of 6-21 where the tables reach 53.
Practical Application
A 3-min all-out test gives a repeatable number but not your critical power: end-test power sat on average about 12 W above CP, with 95% of individual differences between 37.1 W below and 60.5 W above, while work above end power does not estimate W′. Do not set CP-based zones from it as if it were CP; the authors treat end power as a measure in its own right, and its repeatability was shown only for retests within three weeks, so tracking it across a season is inference, not tested here.
How the guide uses this: since 2026-11-10 SYNTH-53 reads the 3-min end power as its own number, not CP — compared only with the athlete's own earlier end power, never a base for zones — on this review.Until then its one-session lab option read the last 30 s of the test as approximately CP, resting on small lab studies of 9-12 participants (habitually active adults where stated), in one of which end power did not differ significantly from CP only at constant cadence (REF-877).This review pooled 15 effects across modes and performance levels without a subgroup by mode, found end power above CP on average with individual 95% limits from −37.1 to +60.5 W, and the authors advise against using the test to approximate an individual's CP, while naming isokinetic mode at 100 rpm or more as their preferred protocol; in the one study comparing modes (REF-877, included here) isokinetic end power sat 4.03 W above CP on average, with limits of ±29.68 W.So zones set from end power as if it were CP may sit tens of watts off for one rider, even in constant-cadence mode. The guide's points that mode matters and that W′ from the test is not valid agree with this review; its multi-effort CP default is not tested here, though the authors note that the CP criterion itself moved with the number and length of the efforts.
If the test is used, run it the authors' way: a full familiarisation test first, isokinetic mode at 100 rpm or more where the ergometer allows, strong encouragement, no elapsed-time display, and retests within three weeks; on a fixed-resistance ergometer the authors prefer resistance set at 3-4.5% of body mass.For VO2peak they add a constant-power bout 10% above end-test power afterwards. These are the authors' recommendations from heterogeneous studies, not tested protocols.
On a smart trainer, freely geared: end power repeated with a CV of 3.3% in 53 riders, WEP with a CV of 17.9%, so W′-like numbers from such a test swing widely between attempts.
Inference, not tested here: a rider whose zones hang on CP is better served by several maximal efforts of different durations, as SYNTH-53 already sets as its default; the 3-min result can sit beside it as a separate number.
Limitations
The authors': methods, statistics, cadence and resistance, encouragement, familiarisation and warm-ups differed between studies; several samples were very small; sex was unreported in some studies and women were underrepresented; performance-level allocation was difficult; effects from different approaches within one study were weighted as independent datasets; publication bias cannot be excluded, especially for EP; CP and W′ from the two-parameter model are themselves debated as a criterion.
Two databases, English only, no registration reported; part of the individual data was read off published scatter plots.
The test's validity against race performance was examined in one study only.
We did not read the supplements (risk-of-bias detail, data extraction, calculation appendix, GRADE table, whose content the text only summarises).
Revision 2026-11-10 (decision): the line on SYNTH-53 now says how the guide reads the test since the decision (end power its own number, not CP, no zones from it) instead of describing the earlier 'approximately CP' reading as current.
How this entry was checked and changed
Why this level
a systematic review of test validation studies whose quality appraisal and certainty we read in the full text; held at A- because the pooled data are small within-person measurement studies (most samples 6-21), certainty for validity carries high concerns (inconsistency, imprecision, possible publication bias), and several effects from one study were pooled as independent; no marker step, because end-test power and W′ are the measured subject of this validation (rule 3 exception)