REF-1363 · Schoenmakers et al., 2026 · BMC Sports Science, Medicine and Rehabilitation · Paper · PubMed · Free full text
Do long intervals keep me near VO2max longer than 30/15?
Not head-to-head: no average difference in 21 within-study comparisons, and 30/15 matched or beat long reps in trained cyclists. One-session proxy, not training (B+).
Main caveat: Acute surrogate. Every pooled number is time near VO2max in one session.
Schoenmakers et al. pooled every peer-reviewed English study that reported time at or near VO2max (tVO2max) during an interval session in healthy people: 86 articles, 239 protocols, any sport, any training status.
They defined tVO2max broadly as time at or above 80% VO2max, prioritised time at or above 90% and 95% for analysis, and sorted protocols by work-interval length into short (30 s or less, SI-HIIT), moderate (more than 30 s to under 2 min, MI-HIIT) and long (2 min or more, LI-HIIT).
Across protocols, long-interval protocols accumulated more time at or above 90% VO2max (unadjusted group means 555 vs 336 and 322 s, from sessions that were also longer: 2058 vs 1390 and 1431 s) and at or above 95% (319 vs 162 and 137 s), and the difference held after adjustment for cumulative session duration; they also carried about twice the work-interval time (1296 vs 659 and 694 s), which was not adjusted for.
Session length and interval category together explained only 6.4% and 7.2% of the variance. In the within-study meta-analysis, moderate beat short (SMD 0.66) and long beat moderate (SMD 1.03), but short and long compared head to head did not differ (SMD 0.06, 95% CI −0.32 to 0.44, 21 comparisons; Fig. 4), a result the text does not mention.
The paper counts a 30/15 set as short intervals (30-s work bouts); most 30/15-type comparisons in Fig. 4 favoured the short format.
Variable-intensity work intervals (fast start, over-under, isoeffort) gave more time near VO2max than even-paced ones (SMD 0.80, 0.42 to 1.18 in Fig. 6, 20 comparisons), and active versus passive recovery made no difference (SMD −0.15, −0.42 to 0.13, 11 comparisons, I-squared 0%).
Our reading: the review supports "accumulate time and shape the rep" for a single session, does not show that long reps beat a 30/15 set when work or mean power is matched, and says nothing direct about adaptation.
Protocol
Design: systematic review following PRISMA 2020; PubMed and Scopus searched on 23 September 2025 (terms for time near/above/at VO2max or VO2peak AND interval training), reference lists screened (9 extra records); peer-reviewed English articles only. Not registered (PRISMA checklist, item 24a); certainty of evidence (item 22) and sensitivity analyses (items 13f, 20d) marked "na".
Eligibility: tVO2max reported during an interval session in apparently healthy people; no limit on age, training status, modality, session length, number of reps or work intensity; studies with supplements or hypoxia included when a control interval session could be extracted. Excluded: continuous exercise only, clinical populations.
Outcome and thresholds: tVO2max = time at or above 80% VO2max or higher (the authors' reason: the heavy/severe boundary varies and is not fixed at 90%). Time at or above 90% and 95% VO2max were the quantitative outcomes; 80%, 85% and 100% descriptive only. In the meta-analyses time at or above 90% was used when reported, otherwise the closest threshold (85% or 95%).A few studies used VO2max minus 1 SD (Turnes 2016) or minus 2.1 (Billat 2000; units not given), or time above VT2 or the respiratory compensation point; "VO2max" includes VO2peak.
Categories: SI-HIIT work intervals of 30 s or less; MI-HIIT more than 30 s to under 2 min; LI-HIIT 2 min or more (after Wen et al. 2019). Series of short on/off bouts (for example 11 × 30 s on / 15 s off, which some authors treat as one 7 min 45 s interval) were kept as SI-HIIT.
n: n = 86 articles, 239 protocols; time at or above 90% VO2max in 183 protocols, at or above 95% in 117. Table 1 lists 94 participant rows summing to 1104 participants (our sum; at least two studies are listed twice), 115 women, 15 of unspecified sex; work intervals 6 to 480 s; about 60% of protocols had a fixed number of reps, and most of the rest went to exhaustion.
Analyses: (1) ANCOVA of time at or above 90% and 95% across SI, MI and LI protocols with cumulative session duration (work plus recovery) as the only covariate, weighted by sample size, Tukey post hoc; (2) linear mixed models of time at or above 90% and 95% on cumulative duration with a random intercept per study, weighted by sample size, testing modality, interval category and their interaction by likelihood-ratio tests; (3) random-effects (inverse variance) meta-analyses in RevMan 5.3 of within-study SMDs for interval length, active vs passive recovery and variable vs even-paced work, with sample sizes divided when one study fed several comparisons; I-squared for heterogeneity; no prediction intervals, funnel plot or sensitivity analysis.
What was controlled in the long-vs-short comparison: cumulative session duration only (ANCOVA covariate and LMM predictor). Work-interval time, work intensity, work:rest ratio, recovery intensity and mean power were not adjusted for or used to form subgroups; LI protocols had about twice the work time. The within-study comparisons hold participants constant but not protocol matching, which varied by study.
Quality and risk of bias: a PEDro scale shortened to 7 items (blinding items dropped): 5.6 ± 0.7. RoB-2 for crossover trials: overall low in 76, some concerns in 7, high in 5 of the 88 study rows of Supplementary Table S3 (our count); the period and carryover domain was the weakest (31 some concerns, 2 high, 13 not applicable). Screening, extraction and quality rating by KM and JW independently after PS's extraction.
Funding and interests (as printed): "no financial support was received for the research, authorship, and/or publication of this article"; "The authors declare no competing interests." Our observation: the review's authors wrote five of the included studies (Bossi, four; Schoenmakers, one), including the variable-intensity crossover the introduction uses as its lead example.
Key Results
| Analysis | Comparison | Result | Protocols or comparisons | Heterogeneity |
|---|---|---|---|---|
| ANCOVA, time at or above 90% VO2max | LI vs SI vs MI (unadjusted group means ± SD from Table 3; only the F test was adjusted, for cumulative duration) | 555 ± 234 vs 336 ± 273 vs 322 ± 174 s; F(2,179) = 8.038, p < .001; LI > SI (p < .01, g 0.87) and > MI (p < .001, g 1.06) | 86, 62 and 35 protocols | not applicable |
| same | work-interval time; cumulative duration | 1296 ± 566 vs 659 ± 586 vs 694 ± 355 s; 2058 ± 799 vs 1390 ± 838 vs 1431 ± 887 s | ||
| ANCOVA, time at or above 95% VO2max | LI vs SI vs MI | 319 ± 132 vs 162 ± 140 vs 137 ± 94 s; F(2,113) = 22.459, p < .001; g 1.15 and 1.52 | 44, 46 and 27 protocols | not applicable |
| same | work-interval time; cumulative duration | 1084 ± 371 vs 673 ± 434 vs 762 ± 483 s; 1749 ± 584 vs 1476 ± 736 vs 1497 ± 985 s (duration not different) | ||
| Mixed model, at or above 90% | time vs cumulative duration (s per s) | SI 135.4 + 0.15 × duration (p < .001); MI 309.6 + 0.04 × (flatter by −0.112, p = 0.027); LI 254.7 + 0.13 × (slope not different from SI, p = 0.483); modality did not improve fit (p = 0.117) | 183 protocols | marginal R-squared 0.064, conditional 0.162, ICC 0.104 |
| Mixed model, at or above 95% | same | SI 223.5 − 0.040 × (not associated, p = 0.117); MI 165.4 − 0.004 ×; LI 126.4 + 0.107 × (steeper than SI, p < .001) | 117 protocols | marginal 0.072, conditional 0.138, ICC 0.071 |
| Meta-analysis, interval length (Fig. 4) | all within-study comparisons, longer vs shorter | SMD 0.45 (0.15 to 0.75) | 46 | I-squared 76%, Tau-squared 0.79 |
| same | MI vs SI | 0.66 (0.02 to 1.31) | 14 | 80% |
| same | LI vs MI | 1.03 (0.42 to 1.64) | 11 | 72% |
| same | LI vs SI | 0.06 (−0.32 to 0.44), p = 0.74 | 21 | 73% |
| Meta-analysis, recovery mode (Fig. 5) | active vs passive | −0.15 (−0.42 to 0.13), p = 0.30 | 11 | 0% |
| Exhaustive protocols (5 running studies, two in children or adolescents) | time at high intensity, passive vs active recovery | 909 ± 443 vs 582 ± 433 s, g 0.67 | 5 studies | not applicable |
| Meta-analysis, pacing (Fig. 6) | variable vs even-paced work | 0.80 (0.42 to 1.18 in Fig. 6; 0.40 to 1.18 in the text), p < .01; mean difference 165 s (97 to 223) | 20 | 68%, Tau-squared 0.49 |
| same | decreasing (fast start) | 0.76 (0.38 to 1.14) | 12 | 42% |
| same | increasing (fast finish) | 0.28 (−0.80 to 1.36) | 1 | not applicable |
| same | variable (over-under, surges, isoeffort) | 0.94 (0.06 to 1.83); subgroup difference χ² 0.90, p = 0.64 | 7 | 85% |
Which protocols counted as variable (Discussion and Fig. 6): decreasing (fast-start) work intervals — de Aguiar 2013, Zadow 2015, Lisbôa 2015 and 2019, Beltrami 2021 (cycling and running), Wommer 2021, Miller 2023, Rønnestad 2020 (REF-194) and Rønnestad 2022 (REF-700); increasing — Wommer 2021; variable — Bossi 2020 (three 30-s surges in 5-min reps at matched mean power), Rønnestad 2022 speed variation (REF-700), Urianstad 2024 (REF-046) and Odden 2024 (REF-041) variable 8-min reps, and Appelhans 2025, whose two "variable" comparisons are isoeffort (self-paced maximal) against fixed power (SMD 2.94 and 3.27).The decreasing-duration ladder (HIDIT, REF-130) is not in this analysis.
Long vs short in the comparisons designed to match intensity, mean power or effort (our reading of Table 2; SMDs as printed in Fig. 4, negative = more time with the short format): 30/15 against 5-min reps at about the same mean power (about 83% vs 84% of MAP, our arithmetic), Appelhans 2025 fixed: −1.09 (−1.96 to −0.22); 30/15 against 8-min reps, Urianstad 2024 (REF-046): −0.08, −0.15, −0.54 and −0.52 (each CI crossing zero after the sample was split), Odden 2024 (REF-041): 0.15 and 0.02; at the same work power ridden to exhaustion, Rønnestad & Hansen 2016 (REF-1051): −1.60 (−2.67 to −0.52) and −1.34 (−2.37 to −0.31); effort-matched in elite cyclists, Almquist 2020: −0.84 (−1.87 to 0.20).With 1:1 work:rest the sign flips: 30 s/30 s against 3 min/3 min at the same work time and session time, Ksoll 2021 (same intensity, 80% MAP): 0.79 (0.20 to 1.38); Fleckenstein 2025 (short reps harder, at vVO2max vs 95% vVO2max): 0.57 (−0.25 to 1.39).
Our re-pooling (DerSimonian-Laird, standard errors from the printed CIs; it reproduces all six printed pooled results exactly; comparisons that share an arm are counted twice, as in the paper): short vs long, 30/15-type sets (2:1 work:rest; 11 comparisons from 5 studies) −0.52 (−0.87 to −0.17), I-squared 33%; other short formats (1:1 or 3:2; 10 comparisons) 0.71 (0.21 to 1.21), I-squared 69%.Variable vs even-paced without Appelhans 2025's isoeffort-vs-fixed pair: 0.54 (0.26 to 0.82), 18 comparisons, I-squared 38%; restricted to the 12 comparisons in which both conditions share the same listed work intensity: 0.21 (−0.04 to 0.45), I-squared 0%; adding Zadow 2015's two all-out-vs-constant pairs, whose matching Table 2 does not state: 0.33 (0.10 to 0.56), I-squared 0%, 14 comparisons. 95% prediction intervals (not reported by the authors): short vs long −1.55 to 1.68; variable vs even-paced −0.72 to 2.33; active vs passive −0.46 to 0.17.
Descriptive: 133 of 239 protocols reached 5 min or more at tVO2max and 48 reached 10 min or more. Reliability of tVO2max: ICCs 0.57 to 0.90 across three studies, with large coefficients of variation (authors: poor to moderate).
Training studies discussed (not pooled): the introduction states, citing four training meta-analyses that are not in the base (Wen 2019, Rosenblat 2022, Yang 2025 and Wiesinger 2024, the last two in athletes and elite athletes by their titles), that interventions with 2-min-plus work intervals gave larger VO2max gains than shorter ones (second-hand, not checked against those papers); in the authors' reading, VO2max rose in training studies whose sessions gave under 5 min (Turnes 2016), 5-10 min and 10 min or more near VO2max; the authors write that Turnes 2016 found larger VO2max gains with short intervals (80-s reps, +6.3%) than with long intervals near critical power (5-min reps, +3.3%) despite markedly lower time near VO2max, but their own Supplementary Table S2 lists more time for the 80-s sessions (115 ± 122 and 127 ± 113 s vs 19 ± 19 and 26 ± 39 s at VO2max minus 1 SD), so this paper cannot settle which group spent less time near VO2max; Ben Abderrahman 2013 found no VO2max difference between active and passive recovery despite different tVO2max; Odden 2024 (REF-041) found the fraction of VO2max predicted gains (adjusted r-squared 0.54 for VO2max).
Numbers that do not add up: (1) Table 5 labels its predictions "time at or above 90% VO2max", but its values (152, 158 and 319 s for 30-min SI, MI and LI sessions) are what the at-or-above-95% equations give; the 90% equations give about 405, 382 and 489 s (our arithmetic), so long intervals lead short ones by about a fifth and moderate ones by about a quarter, not the doubling Table 5 shows.(2) The pacing CI is 0.40-1.18 in the text and 0.42-1.18 in Fig. 6; the overall interval-length p is "<.001" in the text and 0.003 in Fig. 4. (3) The 165-s mean difference has an asymmetric CI (97 to 223 s).(4) Zafeiridis 2015 and Patterson 1997 (2:1), both 2 min vs 30 s, sit in the SI-vs-MI subgroup although 2 min is long by the paper's definition; Germano 2019's two MI-vs-LI rows are its passive-vs-active recovery pairs, with the same numbers as in Fig. 5. (5) Arms with SDs of 0.01 or 0.1 s (Rozenek 2007, Bogdanis 2022, Patterson 1997) yield SMDs up to 4.68.(6) Lisbôa 2015 and 2019 enter both the interval-length and the pacing analysis with identical numbers. (7) The abstract says the positive relation with session duration held "when longer work intervals were used"; at 90% the short-interval slope (0.15 s per s) was as steep as the long one, and only at 95% was it specific to long intervals.
Mechanisms
Measured: time near VO2max grew with cumulative session duration for short and long intervals at the 90% threshold and only for long intervals at the 95% threshold; together with interval category this explained 6-7% of the variance, 7-10% sat between studies, and the rest was unexplained.
Authors' interpretation: work intervals must last long enough for VO2 to rise to and stay near its maximum; short intervals still produce high VO2 when work is hard and recoveries brief, because VO2 then oscillates around VO2max. A fast start speeds the VO2 rise, may recruit type II fibres earlier and raise the oxygen cost of breathing.Active recovery is psychologically demanding and shortened time at high intensity in exhaustive sessions; starting a rep from a lower metabolic rate slows the VO2 rise but enlarges its amplitude, so time near VO2max ends up similar.A priming warm-up may help the first reps (time at or above 90% VO2max was 59.7 s in the first rep vs 87.8 and 98.6 s in the second and third in McKee 2021).
Inference: at the 90% threshold the cross-protocol long-interval advantage largely tracks work time: per second of work interval, the crude group means give 0.43 s at or above 90% VO2max for long, 0.51 for short and 0.46 for moderate protocols (our arithmetic from Table 3). At 95% it does not: 0.29 s for long vs 0.24 for short and 0.18 for moderate (our arithmetic from Table 4). The ANCOVA adjusted for session duration, not for work.
Inference: in the head-to-head comparisons the sign follows the work:rest ratio: 30-s bouts with 15-s recoveries matched or beat long reps, 30-s bouts with 30-s (or 20-s) recoveries lost to them. The paper did not test the ratio, and the split is our post hoc re-pooling.It is confounded: all 11 two-to-one comparisons are crossovers in well-trained or elite cyclists (VO2max 62-81) from one Norwegian group (four of the five papers list Rønnestad as an author), while the ten others include runners, a rowing study and adolescents (VO2max about 48-60). Ratio, sport, training status and laboratory cannot be separated.
Practical Application
Do long intervals keep me near VO2max longer than 30/15? Not head-to-head: no average difference in 21 within-study comparisons, and 30/15 matched or beat long reps in trained cyclists. One-session proxy, not training (B+).
The minutes come first: more cumulative session time went with more time at or above 90% VO2max (about 0.13-0.15 s per second of session for short and long formats); short reps with equal recoveries (1:1) gave less time than long reps (our post hoc re-pooling of the paper's head-to-head comparisons, not a finding of the paper; the 1:1 and 2:1 studies also differ in sport, training status and laboratory; acute).
Shaping a long rep helps a little, not a lot: variable-intensity reps (fast start, over-under, surges) beat even-paced ones by SMD 0.80 overall, but where both conditions list the same work intensity the pooled gain was small, 0.21 (−0.04 to 0.45), or 0.33 (0.10 to 0.56) if Zadow 2015's all-out pairs count as matched (our re-pooling); the large effects came from isoeffort-vs-fixed pairs (2.94 and 3.27) and from fast-start pairs not listed at the same intensity or rep length (de Aguiar, Zadow, Lisbôa: 0.97 to 2.04).Fits REF-700 and REF-194 (B, skiers).
Recovery between reps: passive or easy is fine. Active recovery added no time near VO2max (SMD −0.15, 11 comparisons, I-squared 0%), and in five running studies taken to exhaustion (two in children or adolescents) passive recovery allowed more work at high intensity (909 vs 582 s). Duration of the recovery is the lever (SYNTH-28).
Do not chase "5 or 10 minutes in the red zone": in the authors' reading of the training studies, VO2max rose with less than 5 min per session, and tVO2max itself repeats only poorly to moderately (ICC 0.57-0.90) — one session's reading is a weak verdict.
Limitations
Acute surrogate. Every pooled number is time near VO2max in one session; no adaptation or performance outcome was pooled, and the training studies the authors discuss give mixed dose-response results (their lead example, Turnes 2016, is described as more VO2max gain with less time near VO2max, but their own Supplementary Table S2 lists more time near VO2max for the 80-s group that gained more: 115-127 vs 19-26 s).
The headline is a between-protocol comparison. "Long beats short" rests on an ANCOVA across protocols from different studies, people and laboratories, adjusted only for session duration; long protocols carried twice the work time. Within studies, short vs long was null (SMD 0.06), and the text reports only the moderate-vs-short and long-vs-moderate subgroups.
Classification. 30/15 sets count as short intervals although their series last about 8 min; the authors say this obscures the true comparisons. Within categories, 3-min and 8-min reps are treated alike.
Heterogeneity and method. I-squared 68-80% for interval length and pacing (85% in the variable-intensity subgroup), no prediction intervals, no certainty rating, no sensitivity or publication-bias analysis, not registered; pooled effects mix protocols matched and unmatched for mean power (isoeffort vs fixed counted as pacing), and some comparisons enter two analyses.
Measurement. VO2 averaging (breath-by-breath to 30 s) and the VO2max method differ between studies (Supplementary Table S4); short averaging inflates VO2max and shifts the thresholds. tVO2max repeats poorly to moderately (ICC 0.57-0.90), and many studies measured one session per protocol.
Who. About nine in ten participants are men; cycling and running dominate; children and adolescents are included; training status was not classified.
Reporting. Table 5's "90%" predictions come from the 95% equations; small CI and p-value mismatches between text and figures; see Key Results.
Interests. No funding and no competing interests declared; five included studies are the review authors' own, including the lead example for variable-intensity reps (our observation, not a declared conflict).
How this entry was checked and changed
Why this level
a meta-analysis of mostly small randomised crossovers would sit at A- (serious heterogeneity: I-squared 68-80% in the main interval-length and pacing analyses, 85% in the variable-intensity subgroup, 0% only for recovery mode); held one step lower because the outcome is an acute surrogate (time near VO2max in one session, not adaptation or performance) and because the headline long-versus-short difference comes from a between-protocol ANCOVA adjusted only for session duration, while the within-study comparison of short and long reps was null. Read from the full text, the five supplements and the forest plots (open access, CC BY 4.0)
Full text (checked 2026-10-28)
PubMed and Scopus searched on the 23rd of September 2025; tVO2max defined as time at or above 80% VO2max; 86 studies, 239 HIIT protocols (full text)
Work interval categories: short (≤ 30 s), moderate (> 30 s to < 2 min), long (≥ 2 min); 11 repetitions of 30 s on/15 s off kept as SI-HIIT (full text)
Time ≥ 90% VO2max: SI-HIIT 336 ± 273 s, MI-HIIT 322 ± 174 s, LI-HIIT 555 ± 234 s, F(2,179) = 8.038; total work interval duration 659 ± 586, 694 ± 355 and 1296 ± 566 s (full text)
Time ≥ 95% VO2max: 162 ± 140, 137 ± 94 and 319 ± 132 s; Hedges g 1.15 and 1.52; for ≥ 90%, g 0.87 and 1.06 (full text)
Mixed models: SI-HIIT time ≥ 90% = 135.4 + 0.15 × cumulative duration, LI-HIIT slope not different (p = 0.483), marginal R-squared 0.064; LI-HIIT time ≥ 95% = 126.4 + 0.107 × cumulative duration, marginal 0.072 (full text)
Interval length, all comparisons: SMD = 0.45 [95% CI: 0.15–0.75], I-squared 76% (full text)
Fig. 4, SI-HIIT vs LI-HIIT subgroup: 0.06 [-0.32, 0.44], Z = 0.33 (P = 0.74), I-squared 73%, df = 20, transcribed from the figure (full text)
Active vs passive recovery: SMD = −0.15 [95% CI: −0.42 to 0.13], Z = 1.04; exhaustive protocols 909 ± 443 s vs. 582 ± 433 s, Hedges g = 0.67 (full text)
Variable vs even-paced work intervals: SMD = 0.80 [95% CI: 0.40–1.18], I-squared 68%; mean difference 165 s [95% CI: 97–223 s] (full text)
Table 5, 1800-s sessions: predicted 152, 158 and 319 s for SI-, MI- and LI-HIIT (full text)
133 HIIT protocols accumulated tVO2max ≥ 5 min, of which 48 ≥ 10 min; ICCs ranged between 0.57 and 0.90 (full text)
Turnes et al.: VO2max +6.3% with short intervals vs +3.3% with long intervals near critical power (full text)
PEDro rating 5.6 ± 0.7; no financial support received; the authors declare no competing interests (full text)