In a 106 km mountain ultra, runners slowed at every gradient until 70-90% of the race and sped up in the final 10%, while heart rate fell and perceived exertion rose
And faster finishers stopped less, not slowed less
Participants
n=15 competitors in a 106 km trail mountain ultramarathon (5870 m total elevation gain and loss); finish time 18.3 ± 3.0 h, calculated distance 105.6 ± 1.8 km.
Protocol
Speed and gradient, heart rate, and RPE split into general (RPEGEN) and knee-extensor fatigue (RPEKE), collected by voice recorder during the race.
Speed at level, negative and positive gradients, HR and both RPEs computed for each 10% section of total event duration (TED).
Key Findings
Positive pacing: speed at all gradients decreased until 70-90% of TED, then increased from 90% to 100% at all gradients.
Heart rate at all gradients fell from 10% to 70%, 80% and 90% of TED (not at 100%).
RPEGEN and RPEKE increased throughout, especially in the last 30% of the event.
What went with a faster average speed: less total time stopped (r = −.772, p = .001) and a greater magnitude of speed loss (r = .540, p = .038); speed variability was not significantly related (r = −.475, p = .073).
Authors' interpretation: consistent with a complex regulatory system modulating motor drive in anticipation of perceived difficulties such as elevation changes.
Practical Application
In 15 runners over a 106 km mountain ultra (5870 m of climbing, 18.3 h on average), speed fell at every gradient until 70-90% of the race and rose in the final 10%, heart rate drifted down while perceived exertion climbed, and faster runners were the ones who stopped less (r = −.772) — they actually lost more speed, not less. Cut stoppage time at aid stations before trying to hold an even pace.
Expect falling heart rate with rising effort late in a mountain ultra. Inference, not tested: a lower HR there is not a sign of spare capacity.
Inference, not tested: the end-spurt and the rising RPE fit anticipatory pacing — the athlete paces to what is left, not to a fixed target.
Not from this paper; the base's own practice: plan and practise power-hiking for the climbs.
Limitations
Observational, one event, n=15; associations with average speed are correlations, not causes.
The abstract does not separate walking from running, so run/walk ratio cannot be read from it.
Run dynamics vary with course profile and runner level.
How this entry was checked and changed
Content correction 2026-09-27
"The slowdown came chiefly from more time walking (less time running)" and "pacing is managing the run/walk ratio" -> not in the abstract, which does not measure walking; removed. Title and lead rewritten to the abstract's findings (positive pacing to 70-90% TED, end-spurt, HR down, RPE up).
Added: faster runners had less stoppage time (r = −.772) and GREATER speed loss (r = .540); n=15, 106 km, 5870 m, 18.3 h.
"Hike the climbs early / power-hiking is decisive" relabelled as the base's own practice.
Full text (checked 2026-10-12)
Participants were 15 male experienced runners (age: 43 ± 10 yr, weight: 74 ± 8 kg) who completed the UTMB, shortened on race day to ~106 km because of adverse weather; nineteen started the study (full text).
Average speed was 5.88 ± 0.9 km ∙ h−1 (range: 4.58–7.58 km ∙ h−1) and average HR 132 ± 10 bpm; finish times ranged 13.8–23.9 h (full text).
Speed loss over the event was -2.91 ± 2.15% on level, -2.61 ± 0.92% on negative and -1.31 ± 0.84% on positive gradients; level speed reached its minimum at 70% of event duration versus 90% on climbs and descents (full text).
HR decreased -34.2 ± 17.2 bpm over the race, but HR did not change across the 6 main climbs (at least 300 m of vertical gain at a 10% average gradient), even after adjusting for climbing intensity (full text).
RPEGEN and RPEKE (Borg CR-10) averaged 4.3 ± 1.1 and 3.8 ± 1.3 and peaked at relatively low group means of 6.3 and 6.7 in the last section; the two scales were highly correlated (r = .980) (full text).
Total time stopped (assumed to be resting, eating, clothing and gear change, toilet) correlated with performance at r = -.772 (95% CI = -1.15, -0.39) (full text).
Running and walking were not separated; gradients were grouped as negative [-100 to -2.5%], level [-2.5 to 2.5%] and positive [2.5 to 100%], and only 9 HR and 6 RPE data sets had at least 80% of data (full text).