REF-1496 · Olivares-Llorente et al., 2026 · BMC Sports Science, Medicine and Rehabilitation · Paper

Do slower PRO athletes pace their runs differently?

Slightly: in 8,422 faster-half PRO athletes the running shape was similar across levels (η²p ≤ .054), with more fade at km 8; level showed more at stations. Shares, not what to train.

Main caveat: Fixed order, part-whole. Wall balls are always last and the sled push always second, so a station cannot be separated from its place.
moderate evidenceMeasured on trained to elite athletes

Olivares-Llorente and colleagues took official HYROX PRO results from 149 events (2018-2025) and kept each athlete's best race. Of 28,619 race observations (20,316 men, 8,303 women) that left 12,070 men and 4,597 women; only those faster than their sex's mean were classified, leaving 6,081 men and 2,341 women.

Four z-score groups (G2 fastest to G5, just faster than the mean) were compared, with the three fastest of each sex (G1) kept as description only. Each segment was expressed as a share of total running time or total station time, with Roxzone removed.

Running had a similar shape in all groups: kilometres 3, 4 and 7 did not differ, and the differences at kilometres 1, 2 and 8 were small (η²p ≤ .054); slower athletes spent a smaller share on the opening kilometres and a larger share on kilometre 8.

Station time differed more: the SkiErg and row shrank as a share in slower groups (η²p .470 to .505, Hedges' g up to 4.15), which the authors read as a mirror of time lost at the other stations, not as more effort on the ergometers.

Sex differences were trivial in running (η²p about .005) and present at stations (η²p .060 to .098), which the authors say seem to follow the load scaling of the PRO division. Our reading: a useful table of mean PRO splits and a clear description, in a sample with repeated athletes properly handled. A share of time is a share.

The authors' own discussion says the same data fit both better station execution and more aerobic capacity, and the design cannot separate a station from its place in the race.

Protocol

Design: observational, retrospective, secondary analysis of public results; ethics approval and consent stated as not required.
Sample: n = 8422 PRO athletes (6,081 men, 2,341 women; the total is our sum), results collected on 6 November 2025 from the official results database; 149 events: 5 in 2018, 12 in 2019, 4 in 2020, 14 in 2021, 23 in 2022, 39 in 2023, 50 in 2024, 2 in 2025.
Repeated athletes: handled. From 28,619 race observations (20,316 men, 8,303 women) duplicate participations were removed, keeping each athlete's best overall time: 12,070 men and 4,597 women. Unique athletes, one race each.
Who was kept: only athletes faster than the sex-specific mean (z < 0). The slower half of unique PRO athletes was dropped from the classification and from all inference.
Groups: z-score bands of overall time within sex: G2 −2 to < −1.5 (250 men, 58 women), G3 −1.5 to < −1.0 (1,331, 470), G4 −1.0 to < −0.5 (2,294, 955), G5 −0.5 to 0 (2,203, 855). G1 = the three fastest of each sex, descriptive only, excluded from inference.
Measures: chip times of eight runs and eight stations. Relative time = segment time / (overall time − Roxzone time) × 100; also as a share of total running and of total station time. Roxzone excluded.
Analysis: Welch's ANOVA per segment with Benjamini-Hochberg correction and Games-Howell post hoc tests; sex × segment mixed-design ANOVA in each group with Greenhouse-Geisser correction; SPM1D for sex differences across the race; η²p and Hedges' g. JASP 0.19.3, Python 3.12.
Funding and interests: funded by the Universidad Alfonso X el Sabio Foundation; no competing interests declared. Processed data on request. No registration reported.

Key Results

Table 1 of the paper, reproduced in full (PRO division, each athlete's best race; running and station times exclude Roxzone; n in brackets). Not percentiles. The G1 columns do not add up and are not a reference (see below).

Mean ± SD (mm:ss)G1 men (3)G1 women (3)G2 men (250)G2 women (58)G3 men (1,331)G3 women (470)G4 men (2,294)G4 women (955)G5 men (2,203)G5 women (855)
Overall time53:58 ± 0:3359:32 ± 1:1960:29 ± 1:5264:15 ± 1:4267:45 ± 2:1271:51 ± 2:1575:17 ± 2:1879:16 ± 2:2082:54 ± 2:1787:01 ± 2:25
Run 13:17 ± 0:293:30 ± 0:193:21 ± 0:353:39 ± 0:333:38 ± 0:414:01 ± 0:383:59 ± 0:424:18 ± 0:434:16 ± 0:474:35 ± 0:50
Run 23:26 ± 0:103:52 ± 0:063:36 ± 0:153:53 ± 0:153:55 ± 0:184:12 ± 0:184:13 ± 0:204:33 ± 0:204:33 ± 0:214:56 ± 0:22
Run 33:32 ± 0:094:02 ± 0:093:49 ± 0:174:05 ± 0:184:11 ± 0:204:28 ± 0:224:35 ± 0:244:51 ± 0:235:00 ± 0:255:18 ± 0:26
Run 43:34 ± 0:044:02 ± 0:063:51 ± 0:184:04 ± 0:174:13 ± 0:204:28 ± 0:214:36 ± 0:234:52 ± 0:245:02 ± 0:255:19 ± 0:25
Run 53:32 ± 0:044:06 ± 0:133:54 ± 0:174:07 ± 0:174:17 ± 0:204:32 ± 0:214:41 ± 0:234:57 ± 0:235:07 ± 0:255:25 ± 0:26
Run 63:36 ± 0:124:05 ± 0:133:51 ± 0:184:05 ± 0:174:13 ± 0:204:29 ± 0:214:36 ± 0:234:53 ± 0:225:01 ± 0:245:21 ± 0:25
Run 73:30 ± 0:053:58 ± 0:063:54 ± 0:184:07 ± 0:174:16 ± 0:214:31 ± 0:214:39 ± 0:234:56 ± 0:235:05 ± 0:255:25 ± 0:26
Run 83:53 ± 0:064:02 ± 0:094:06 ± 0:224:20 ± 0:234:31 ± 0:294:44 ± 0:285:02 ± 0:345:13 ± 0:335:37 ± 0:385:50 ± 0:38
1000 m SkiErg3:48 ± 0:114:20 ± 0:103:53 ± 0:084:24 ± 0:114:00 ± 0:094:32 ± 0:114:07 ± 0:114:40 ± 0:124:13 ± 0:124:47 ± 0:14
50 m sled push2:15 ± 0:042:37 ± 0:202:35 ± 0:232:38 ± 0:272:55 ± 0:333:02 ± 0:333:16 ± 0:423:26 ± 0:413:34 ± 0:463:48 ± 0:46
50 m sled pull2:59 ± 0:163:42 ± 0:143:52 ± 0:304:01 ± 0:274:39 ± 0:464:53 ± 0:485:29 ± 1:005:35 ± 0:546:18 ± 1:086:16 ± 1:04
80 m burpee broad jump2:37 ± 0:203:27 ± 0:372:51 ± 0:233:21 ± 0:293:21 ± 0:333:56 ± 0:343:56 ± 0:414:28 ± 0:444:27 ± 0:475:02 ± 0:50
1000 m row3:55 ± 0:024:24 ± 0:164:03 ± 0:084:32 ± 0:094:15 ± 0:104:42 ± 0:114:24 ± 0:114:53 ± 0:124:33 ± 0:125:02 ± 0:14
200 m farmers carry1:26 ± 0:021:37 ± 0:071:35 ± 0:101:46 ± 0:121:47 ± 0:162:04 ± 0:192:02 ± 0:212:21 ± 0:242:17 ± 0:262:39 ± 0:26
100 m sandbag lunges3:05 ± 0:313:40 ± 0:193:21 ± 0:203:28 ± 0:224:00 ± 0:304:00 ± 0:284:32 ± 0:394:34 ± 0:345:06 ± 0:445:04 ± 0:38
Wall balls3:37 ± 0:093:25 ± 0:164:25 ± 0:384:05 ± 0:345:12 ± 0:524:47 ± 0:446:01 ± 1:075:31 ± 0:536:43 ± 1:116:13 ± 0:59

Our arithmetic from that table, not printed in the paper: running was 50.2% of overall time in G2 men and 47.9% in G5 men (50.3% and 48.4% in women); stations 44.0% to 44.9% (men), 44.0% to 44.7% (women); the remainder (Roxzone) was 3:32 to 6:02 in men and 3:40 to 6:01 in women.

Between G2 and G5 men the overall gap of 22:25 was 9:19 in the runs, 10:36 at stations and 2:30 in the remainder; in women 22:46, 9:49, 10:36 and 2:21. The largest single gaps in men were sled pull 2:26, wall balls 2:18, lunges 1:45 and burpee broad jumps 1:36, against SkiErg 0:20 and row 0:30. These are gaps inside a total, not what would close them.

Between-group effect sizes (η²p) for the share of running or station time:

SegmentMenWomen
Km 1.002.007
Km 2.054.021
Km 8.035.028
Km 3, 4, 7not significantnot significant
Km 5, 6significant, p = .035 and .034, no η²p printednot significant
SkiErg.505.489
Sled push< .001 (p = .358, not significant).004
Sled pull.071.033
Burpee broad jump.040.019
Row.486.470
Farmers carry.007.018
Sandbag lunges.023.020
Wall balls.021.043
Kilometre 2 and 8 contrasts: at km 2 all six group pairs differed in men (g 0.23 to 0.80) and five of six in women (g 0.19 to 0.60); at km 8 G5 differed from G2, G3 and G4 (men g −0.50, −0.47, −0.26; women g −0.37, −0.44, −0.27).
Largest station effects: SkiErg and row, G2 vs G5: g 4.15 and 3.97 in women, 3.97 and 3.64 in men. Sled pull: g −0.84 (women), −0.96 (men).
Sex × segment, running: significant in G3 to G5 (η²p .006, .005, .005), not in G2 (p = .446).
Sex × segment, stations: significant in all groups (G2 to G5 η²p .098, .085, .068, .060). Women held a larger share at the SkiErg, burpee broad jumps, row and farmers carry; men at sandbag lunges and wall balls in every group.
A row that does not add up: in G1 women the means leave 0:43 for Roxzone across the race (58:49 of 59:32, our arithmetic), against 3:40 in G2 women; G1 men leave 1:56 against 3:32. G1 is three athletes per sex and descriptive, and a mean Roxzone of 0:43 is not credible against 3:40 to 6:01 in the other women's groups (our reading), so the columns may carry a data problem; do not use any G1 value as a reference.

Mechanisms

Measured: the shares of time and absolute splits above. No physiology, load, age or body mass.
Authors' interpretation: the ergometer effect largely mirrors time lost by slower athletes at the other stations, because shares within a block sum to 100%; it does not mean faster athletes put more effort into the ergometers.The sex pattern seems to follow the scaling of PRO loads: distance-matched stations (SkiErg, row, burpee broad jumps) take women a larger share, the stations whose load falls most for women take men a larger share (lunges and wall balls in every group, sled pull only in G4 and G5).
Authors' interpretation: the greater cost of kilometre 8 in slower athletes may reflect pacing, residual fatigue from the lunges, and lower durability. Not measured.
Inference: the run-to-run steps in the table fit the pattern in REF-1498 (a preprint): from run 2 to run 3, after the sled push, means rise 13 to 27 s in men and 12 to 22 s in women; from run 5 to run 6, after the row, they fall 3 to 6 s in men and 2 to 4 s in women; from run 7 to run 8, after the lunges, they rise 12 to 32 s in men and 13 to 25 s in women (our arithmetic). Station and race position cannot be told apart here either.

Practical Application

Do slower PRO athletes pace their runs differently? Slightly: in 8,422 faster-half PRO athletes the running shape was similar across levels (η²p ≤ .054), with more fade at km 8; level showed more at stations. Shares, not what to train.
Placing a PRO split: use the table to see where a PRO athlete sits against the mean of a band of faster-than-average best races. It is not a percentile, covers only the faster half of unique athletes, and pools 2018-2025. For percentiles use REF-901; for Open, this table does not apply.
What to train is not in this paper. The authors state both readings: better execution at loaded and bodyweight stations, or more aerobic and running capacity so athletes arrive at stations less fatigued, and they did not test either. Not from this paper; the base's own practice for the weekly split is in SYNTH-71.
Pacing: the data describe what finishers did, not what works. Run 8 is the slowest run in every G2 to G5 column, and in slower groups by more (run 8 minus the mean of runs 2 to 7: +17 s in G2 men, +39 s in G5 men, +16 s and +33 s in women; our arithmetic; run 8 may differ in length by venue). No pacing was tested; REF-1508 is the nearest test, of a run opening in triathlon; REF-1521 tests ride intensity.
Women: nearly the same running shape as men in this PRO sample (sex × segment η²p about .005, not significant in G2); SYNTH-71 lists women's pacing as a gap. The sex split at stations reads as load scaling, so it does not by itself say women or men should train any station more.

Limitations

Fixed order, part-whole. Wall balls are always last and the sled push always second, so a station cannot be separated from its place; shares within a block sum to 100%, so one shrinking forces others to grow.
Faster half only. Athletes slower than their sex's mean were dropped, and G1 is three athletes per sex: by the authors' own statement the findings cannot be extended to the very fastest or to slower PRO finishers. Open, Doubles and Relay are absent.
Best race per athlete. Keeping each athlete's best day handles repeats but selects good days; the table is not what a typical race of that athlete looks like.
Pooled years and venues. 2018-2025 are pooled and course layout, sled-track surface and equipment calibration vary, by the authors' statement; the first and last runs may differ in length by venue, which clouds kilometres 1 and 8. Run 1's SD is about twice run 2's in every G2 to G5 column (our reading of the table).
No explanatory data. No age, body mass, physiology or training; finishers only.
Unedited manuscript. Read from the Article in Press PDF; the final version may change. Figures 1 to 4 were not read for values; the numbers above come from Table 1 and the text.
Interests. Funded by a foundation of the authors' university; none declared.
How this entry was checked and changed
Why this level

large observational sample with repeated athletes removed (best race per athlete), read in full (open access, CC BY-NC-ND 4.0). Associations in finishers only, within a fixed station order, in shares that sum to 100%; no training or physiological data

Full text (checked 2026-10-30)
6,081 men and 2,341 women in the final sample, from 149 events held between 2018 and 2025 (full text)
Initial database 28,619 race observations (20,316 male, 8,303 female); after keeping each athlete's best time, 12,070 men and 4,597 women; only athletes with z < 0 retained (full text)
Running, men: km 2 F = 133.50, η²p = .054; km 8 F = 73.14, η²p = .035; km 1 η²p = .002 (full text)
Stations, men: SkiErg η²p = .505, RowErg η²p = .486; women: SkiErg η²p = .489, RowErg η²p = .470 (full text)
SkiErg and RowErg, Group 2 vs 5: g = 4.15 and 3.97 in women, 3.97 and 3.64 in men (full text)
Sex × segment for stations: Group 2 η²p = .098, Group 3 .085, Group 4 .068, Group 5 .060; for running, Groups 3 to 5 η²p = .006, .005, .005, Group 2 p = .446 (full text)
Funded by the Universidad Alfonso X el Sabio Foundation; no competing interests declared (full text)