REF-1671 · Ramos-Campo et al., 2026 · Sports Medicine · Paper · PubMed

No imposed pacing strategy beat self-selected pacing in set-distance time trials of 400 m to 20 km

Against self-selected pacing a fast start gave SMD 0.02 (95% CI −0.22 to 0.26), even pacing 0.03 (−0.30 to 0.35) and a slow start 0.19 (−0.05 to 0.44), positive values favouring self-selected, none significant, in a network meta-analysis of 24 crossover studies (302 athletes, 765 observations) — and across studies the cost of a slow start grew with the share of the trial spent slow (about 0.03 SMD per percentage point, a between-study meta-regression): slow phases of half the trial or more went with slower mean times, most clearly slow pacing over 73% of a 1500-m run, while in the trials the authors class as long distance a slow start did not differ from self-pacing (SMD −0.21, −0.78 to 0.36)

Main caveat: The authors': no cycling study was outdoors, and outdoor cycling time-trial performance differs from laboratory performance.
fairly strong evidenceMeasured on recreational to highly trained athletesRead in full

Ramos-Campo and colleagues pooled crossover experiments in healthy endurance athletes in which the opening of a set-distance time trial was imposed — faster than, equal to or slower than the athlete's own self-paced intensity — and compared with self-selected pacing or with another imposed start.

Twenty-four studies with 302 participants entered a frequentist random-effects network; the trials ran from 400 m to 20 km in cycling, running, swimming, rowing, skiing, skating and the run of a triathlon. Across the network no imposed start differed from self-selected pacing, and peak oxygen uptake, mean oxygen uptake and end-test lactate did not differ either.

The slow-start comparison was heterogeneous, and a between-study meta-regression on the length of the slow phase removed that heterogeneity: the longer the slow phase lasted as a share of the trial, the worse the result against self-pacing.

Slow phases of half the trial or more went with slower mean times; the only individual study with a significant deficit held slow pacing, continued after the start, over 73% of an outdoor 1500-m run, and the running, outdoor and slow-after-the-start moderators rest on it and an 800-m run by the same group.

In the trials the authors class as long distance (cycling over 20 km, 20 km in the heat and 16.1 km with slow phases of 12-25% of the distance, and over 10 km with a slow phase of 50%), a slow start did not differ from self-pacing.

The authors conclude that self-pacing is a reliable default, that excessively prolonged slow starts should be avoided, and that the laboratory setting limits transfer to competition.

Participants

24 studies, 302 participants (249 men, 53 women), 765 observations; 270 classed as competitive and 32 as recreational; training status "trained" in 22 samples and "active" in 2.
Mean age 16-42 years and VO2peak 47-72 mL/kg/min across studies (several studies report neither).
By exercise mode: cycling in 13 studies (one of them also speed skating), running in 5 (one the run of a triathlon), swimming in 3 (one by triathletes), cross-country skiing in 2, rowing in 1.
Sample sizes 7 to 32; half the studies had ten or fewer participants.

Protocol

Search and selection: five databases from inception to 15 February 2025; 5493 records, 2628 screened after duplicates, 383 full texts assessed, 24 included. Protocols without a fixed distance were excluded.
Inclusion: peer-reviewed controlled experiments (within- or between-subject) in healthy endurance athletes of any level; at least one manipulated pacing strategy against self-selected or another strategy; time-trial completion over a set distance; tasks lasting more than 1.5 min and less than 4 h. The included trials ranged from 400 m to 20 km (the longest: 20-km and 16.1-km cycling, 10-km running).
Strategies: the start was coded as self-selected, fast, even or slow — initial intensity above, equal to or below the athlete's self-paced intensity (a fast start is defined against the familiarisation trial mean). Where a study had two intensities of one strategy, the one further from the mean ("high") and the one closer ("low") were analysed separately.
How slow and how long the slow starts were (supplement, Table S3): slow phases covered 10% to 73% of the trial at 0.85 to 0.98 of the reference intensity.In the 10- to 20-km trials: 12% of a 20-km cycling trial at 0.85, 13% of a 20-km cycling trial in the heat at 0.90, 25% of a 16.1-km cycling trial at 0.95, 10% of a 10-km triathlon run at 0.90 and 0.95 (a study with no self-selected arm, so outside the slow-versus-self-selected analyses), and 50% of a 10-km cycling trial at 0.98, coded as slow after the start too.In the 800-m and 1500-m running studies: 50% at 0.98 and 73% at 0.90 and 0.95, with the slow pacing continued after the start.
Setting: every cycling study was indoors; outdoor trials were three running studies and two skiing studies. Eight studies had no familiarisation trial; three used a pacer; the minimum time between trials was 0 to 10 days.
Analysis: standardised mean differences (Hedges' g; under 0.20 read as no effect); network meta-analysis (netmeta, DerSimonian-Laird); heterogeneity by I², global inconsistency by a design-by-treatment model, local by separating direct from indirect evidence; pairwise meta-analyses with meta-regression on study, participant, intervention and outcome covariates.
Funding and interests: no funding; no conflicts of interest directly relevant declared.

Key Results

Network estimates against self-selected pacing (positive favours self-selected):

Imposed start"High" analysis, SMD (95% CI), p"Low" analysis, SMD (95% CI), p
Fast0.02 (−0.22 to 0.26), 0.87−0.02 (−0.23 to 0.19), 0.83
Even0.03 (−0.30 to 0.35), 0.87−0.03 (−0.32 to 0.25), 0.82
Slow0.19 (−0.05 to 0.44), 0.120.11 (−0.10 to 0.32), 0.31
Heterogeneity: I² 22.7% (95% CI 0.0-48.6) in the high analysis, 0.0% (0.0-37.0) in the low. Global inconsistency in the high analysis (Q = 25.84, p = 0.02), not in the low (Q = 10.97, p = 0.61); local inconsistency for self-selected vs fast (p = 0.04) and fast vs slow (p < 0.01).
Pairwise against self-selected: fast SMD −0.13 (−0.38 to 0.13), p = 0.33, I² 0.0%; even 0.02 (−0.35 to 0.40), p = 0.90, I² 0.0%; slow 0.24 (−0.15 to 0.63), p = 0.23, I² 59.5%. No sign of small-study or publication bias in these comparisons.
In the individual studies, one showed a significant difference between self-selected and slow starts, favouring self-selected (the 1500-m running study).

What moved the slow-start result (pairwise slow vs self-selected, high analysis; positive favours self-selected):

ModeratorResult
Duration of the slow phase, % of the trialSMD rose by 0.026 per percentage point (95% CI 0.015 to 0.038), p < 0.0001, I² 0.0%; the figure's line is y = 0.03x − 0.83
Intensity of the slow startnot significant (intensities spanned only 0.85-0.98 and several studies gave none): −0.002 (−0.115 to 0.111), p = 0.97, I² 68.4%
Event type (the authors' classes)middle distance 0.49 (0.06 to 0.91), p = 0.02; long distance −0.21 (−0.78 to 0.36), p = 0.47; difference between classes p = 0.055
Sportrunning 0.88 (0.39 to 1.38), p = 0.0005; cycling −0.13 (−0.57 to 0.31); swimming −0.21 (−1.07 to 0.65); rowing 0.24 (−0.63 to 1.10); difference p = 0.02
Pacing after the startslow 0.88 (0.41 to 1.34), p = 0.0002; self-selected −0.13 (−0.49 to 0.24); fast 0.19 (−0.70 to 1.07); difference p = 0.004
Locationoutdoor 0.87 (0.43 to 1.31), p = 0.0001; indoor −0.08 (−0.41 to 0.24); difference p = 0.0006
Event distance and event durationmeta-regressions in the supplement only: the slow-start disadvantage was smaller in longer trials (p = 0.04 and p = 0.03); the text does not list them among the significant covariates
The fitted line crosses zero near 28% of the trial by the figure's rounded equation (0.83 / 0.03) and near 32% with the table's slope (0.83 / 0.026), our arithmetic; below that the line's estimate is at or under zero. This is a between-study fit with a confidence band around it, not a tested threshold below which a slow start is shown to do no harm.
The authors note that the running, outdoor and slow-after-the-start levels all come from the same two studies (800 m and 1500 m), so those three moderators cannot be separated; start duration removed the heterogeneity (I² 0%) and did not interact with those studies (p = 0.71), and they name it the key modifier.
Secondary outcomes: no significant differences between strategies in VO2peak (4 studies, 82 observations), mean VO2 (4 studies, 119 observations) or end-test lactate (11 studies, 287 observations).

Practical Application

In an individual time trial of about 2 to 45 min, letting the athlete pace by feel did as well as any imposed start: fast, even and slow starts did not differ from self-selected pacing (SMD 0.02, 0.03 and 0.19, none significant). What went with slower times was a slow start held too long: across studies the deficit grew with the share of the trial spent slow; it was largest with slow pacing over 73% of a 1500-m run (mean times about 3-7% slower, the only individual study with a significant difference), while slow phases of half an 800-m run, a 10-km cycling trial and a 2-km row left mean times about 1-1.5% slower, none significant on its own (our arithmetic from the supplement).
In the longer trials a short, mild conservative opening did not differ from self-pacing. The long-distance slow-versus-self-selected class holds four cycling trials: 20 km, 20 km in the heat and 16.1 km with slow phases of 12-25% of the distance at 0.85-0.95 of the reference intensity, and 10 km with a 50% slow phase at 0.98; in that class a slow start did not differ from self-pacing (SMD −0.21, −0.78 to 0.36; not significant, and the interval does not rule out a small deficit).The intensity of the slow start showed no significant association with the result over its narrow range; its length did, across studies.
How [SYNTH-10] uses this: the guide's opening for mass-start events longer than about 2 h — the first fifth at about 90-95% of the planned average — is the base's own practice and is not tested here: no trial was longer than 20 km, none was a mass start, and no cycling trial was outdoors.It resembles the slow starts of the 16.1- and 20-km cycling trials here (12-25% of the distance at 0.85-0.95), which did not differ from self-pacing, and from the between-study dose-response the base sets its own bound: do not stretch the conservative opening beyond about a quarter of the event. Both are transfers from trials of 20 km or less to events of several hours.
Against [SYNTH-62]: even power on flat, still-air courses is not contradicted (imposed even pacing did not differ from self-pacing, SMD 0.03, −0.30 to 0.35, in trials of 20 km or less), and it was not shown to improve on a self-paced effort either.
Inference, not tested here: for an athlete who already paces well by feel, a rigid opening target adds little; for one who habitually goes out far too hard in a group, the field data in SYNTH-10 remain the reason to cap the first part of the event.

Limitations

The authors': no cycling study was outdoors, and outdoor cycling time-trial performance differs from laboratory performance; half the studies had ten or fewer participants; men dominate (53 of 302 participants were women); the self-selected pacing profiles were not consistently reported; the links between performance and the physiological outcomes could not be analysed.
The moderators that single out harmful slow starts (running, outdoors, slow pacing after the start, a slow phase of half the race or more) come largely from two studies of 800 m and 1500 m, both rated at high risk of bias, and are comparisons between studies.
Risk of bias: 4 of 24 studies high (three running studies and one swimming study), the other 20 "some concerns"; none low.
The longest trial is 20 km, and apart from the sprint-triathlon study (whose listed times of about 66-68 min appear to cover the whole race, with only the 750-m swim paced) every trial lasted under about 45 min: nothing here tests a pacing plan for a multi-hour event, a mass start, or terrain and wind.
The supplement's table of intensities does not state its reference in the table itself (the Methods define fast starts against the familiarisation trial mean), and three studies give no duration or intensity for their slow start.
2026-11-10: rewritten from the full text and its supplementary tables (the entry of the same day was written from the abstract); level B+ -> A-.
How this entry was checked and changed
Why this level

regraded 2026-11-10 from B+ after the full text was read: a network meta-analysis of crossover experiments with the risk of bias read (RoB 2: 4 of 24 studies at high risk, the other 20 with some concerns); held at A- because half the studies had ten or fewer participants and the main analysis showed inconsistency between direct and indirect evidence