REF-1238 · Rothschild et al., 2025 · European Journal of Applied Physiology · Paper · PubMed · Free full text
Does heart-rate drift tell me my threshold has slipped?
In ~23 trained cyclists with repeat tests, a bigger heart-rate rise at a fixed power went with a bigger fall in VT1 power (r = −0.76) over 1-5 h. Lab steps; no cut-off tested.
Rothschild et al. asked whether the drift of heart rate, breathing frequency or ventilation at a fixed power tells how far the moderate-to-heavy boundary (the first ventilatory threshold, VT1) has fallen during a long ride.
They pooled four of their laboratory's studies, in which 51 trained cyclists had VT1 power measured fresh and after about 2.5 h of riding at 90% of VT1 (in one study every hour until task failure, up to 6 h), and at each test also recorded heart rate, breathing frequency and ventilation at the same fixed power.
In 48 riders with usable data (85 change scores), a larger rise in heart rate at the fixed power went with a larger loss of VT1 power (repeated-measures r = −0.76, 95% CI −0.87 to −0.58), a larger rise in breathing frequency more weakly (−0.40) and ventilation not significantly (−0.25, P = 0.136).
A model of baseline VT1, VO2peak, breathing-frequency drift and heart-rate drift × time predicted the fatigued VT1 with a mean absolute error of 7.2 W in-sample and 8.3 W under bootstrap resampling; a simpler model using only baseline VT1 and heart-rate drift was nearly as good (8.7 W).
Across the cohort, heart rate at VT1 after 150-180 min was higher than fresh (147 vs 142 beats/min) while VT1 power had fallen.
Our reading: the first evidence that the heart-rate drift the base tracks moves with an actual loss of threshold power, within riders — but measured as a controlled fixed-power step, in riders the base already cites for the threshold slide (REF-179, REF-425, REF-181, REF-143), and with no cut-off.
Protocol
Design: secondary, exploratory analysis of four published studies from the Auckland University of Technology laboratory; ethics approvals 21/253, 22/163, 23/37, 23/197; not registered; no specific sample-size estimate.
Who (n = 51 unique trained cyclists; Table 1): Study A 13 men and 1 woman, 34 ± 10 y, VO2peak 59.9 ± 6.8 mL/kg/min, VT1 217 ± 42 W, 9 ± 3 h/week; Study B 11 and 2, 57.4 ± 5.2, 211 ± 40 W, 12 ± 2 h/week; Study C 10 and 2, 59.3 ± 5.3, 229 ± 37 W, 14 ± 5 h/week; Study D 10 and 2, 41 ± 8 y, 52.3 ± 5.2, 184 ± 36 W, 10 ± 3 h/week.
Source studies (identified from the reference list and protocol): A = Stevenson 2022 (REF-179; overnight fast, water only, 2 h at 90% VT1); B = Hamilton 2024 (REF-425; breakfast with about 2 g/kg carbohydrate, electrolyte drink, 2.5 h at 90% VT1); C = Dudley-Rode 2024 (REF-181; about 1 g/kg carbohydrate breakfast, then 60 g/h glucose or a sweetened placebo for the first 120 min); D = Gallo 2024 (REF-143; about 2 g/kg breakfast, VT1 measured every hour to task failure, 234 ± 66 min, maximum 6 h).
VT1 measurement: a five-step test (A-C: from 50 W below to 50 W above the fresh VT1 power in 25-W, 4-min steps) or a six-step test (D: 70-120% of fresh VT1 power), with VT1 from the oxygen uptake vs ventilatory-equivalent breakpoint, converted to power by regression.
Decoupling: the difference in heart rate, breathing frequency and ventilation in the last minute of the step at the fresh VT1 power (third step in A-C, fourth step in D) between the fresh test and each fatigued test. Expressed in beats/min, breaths/min and L/min, not as a percentage.
Data: 85 change scores from 48 participants (A 13, B 11, C 23, D 38); one point (VT1 power +29 W after 150 min) removed as an outlier.
Analysis: repeated-measures correlation; generalised estimating equations with candidate predictors chosen by a genetic search on AICc (up to five predictors), then 2000 bootstrap resamples to pick the model with the lowest mean absolute error. R 4.3.1.
Funding and interests: open-access funding through CAUL; Crossref lists Auckland University of Technology as funder; no competing interests declared.
Key Results
| Measure | Result |
|---|---|
| Heart-rate drift vs change in VT1 power | rrm = −0.76 (95% CI −0.87 to −0.58), P < 0.001 |
| Breathing-frequency drift vs change in VT1 power | rrm = −0.40 (−0.64 to −0.09), P = 0.013 |
| Ventilation drift vs change in VT1 power | rrm = −0.25 (−0.52 to 0.08), P = 0.136 |
| Best model: baseline VT1 + breathing-frequency drift + heart-rate drift × time + VO2peak | MAE 7.2 W, R² 0.95 in-sample; 8.3 W (8.3-8.4), R² 0.93 bootstrap |
| Baseline VT1 + heart-rate drift + VO2peak | MAE 7.6 W in-sample, 8.4 W bootstrap; heart-rate coefficient −0.91 (−1.60 to −0.37) |
| Baseline VT1 + heart-rate drift | MAE 8.3 W in-sample, 8.7 W bootstrap; heart-rate coefficient −0.75 (−1.60 to −0.38), baseline VT1 coefficient 0.96 (0.89 to 1.00) |
| Heart rate at VT1, fresh vs after 150/180 min | 142 ± 11 vs 147 ± 10 beats/min (N = 58), P = 0.0002 |
Data points by time: 48 at minute 0 and 12, 12, 47, 9, 3 and 2 at 60, 120, 150, 180, 240 and 300 min.
Range of the slide (introduction, from their earlier studies): after 2.5 h of moderate riding, VT1 power fell by 1 to 45 W between individuals.
The authors' benchmark does not add up: they place the error within the 3.5% day-to-day coefficient of variation they cite for VT1 power, but 3.5% of the cohort's mean fresh VT1 (about 210 W) is about 7.4 W, and less of a fatigued VT1; the in-sample 7.2 W fits, the resampled 8.3 W (about 4%) does not (our arithmetic).
Numbers that do not add up: the abstract says 51 cyclists; the analysis used 48 (three contributed no usable change score). The heart-rate-at-VT1 comparison after 150/180 min gives N = 58, while the paper lists 47 + 9 = 56 data points at those two times (our arithmetic). The abstract's 7.2 W and R² 0.95 are the in-sample figures; the resampled ones are 8.3 W and 0.93.The correlations' CIs and P values (for example −0.40, P = 0.013; −0.25, P = 0.136) fit about 35 degrees of freedom, not the roughly 84 expected if all 48 riders contributed: riders of studies A and B had one change score each and cannot enter a repeated-measures correlation, so the correlations rest on about 23 riders from studies C and D (our arithmetic).
Mechanisms
Measured: heart rate at a fixed power rose as VT1 power fell, and the two moved together within riders; breathing frequency followed more loosely and ventilation not significantly. Heart rate at the re-measured VT1 itself was higher after 150-180 min than fresh.
Authors' interpretation: heart-rate (or breathing-frequency) decoupling in a controlled session may serve as a practical durability test, and serial measures may track durability adaptations; a fixed heart-rate threshold from a fresh test cannot show proximity to VT1 in real time because heart rate at VT1 moves. The ventilation result surprised them; they suggest noise, or that a harder protocol might show it.
Inference: the high R² comes largely from baseline VT1 being a predictor (riders differ by about 40 W at baseline); the error of about 8 W is the number to read, against individual slides of 1-45 W.
Inference: the paired group means (VT1 power down, heart rate at VT1 up from 142 to 147 beats/min) mean a cap at the fresh VT1 power overshoots the boundary late in a long ride, while a cap at the fresh VT1 heart rate errs on the easy side; individual riders were not shown and may differ.
Inference: because feeding differed between the source studies (fasted with water, electrolyte drink, 60 g/h glucose or placebo) and the relation held in the pool, the heart-rate signal tracked the slide whatever its cause; the paper did not test feeding as a moderator.
Practical Application
Does heart-rate drift tell me my threshold has slipped? In ~23 trained cyclists with repeat tests, a bigger heart-rate rise at a fixed power went with a bigger fall in VT1 power (r = −0.76) over 1-5 h. Lab steps; no cut-off tested.
Decoupling as the tracked metric now has evidence under it, in its controlled form: a rise in heart rate at the same power, within one rider, tracked the loss of threshold power. The closest field version is a repeatable steady block at one fixed power early and late in comparable long rides, fed the same way; a whole-ride power-to-heart-rate ratio over varying terrain and heat was not tested (inference).
The cut-off stays house practice: the study reported the drift in beats/min and gave no threshold; the base's ">10% over 3 h" flag is not tested here (SYNTH-32).
A rough conversion, population-average only: in the simplest model each beat/min of heart-rate rise at the same power went with about 0.75 W less VT1 power (95% CI 0.38-1.60). Do not compute an individual's threshold from it.
Late in a long ride, a fresh power cap overshoots (inference): with VT1 power down and heart rate at VT1 up, a cap at the fresh VT1 power drifts into the heavy domain; bringing the power down is the safer reading, as SYNTH-32 already advises for late-session targets (REF-179, REF-143); on the group means a cap at the fresh VT1 heart rate errs easy, but the authors say a fixed fresh heart-rate threshold cannot show proximity to VT1 in real time.
Breathing data (inference for chest-strap ventilation sensors): breathing-frequency drift tracked the slide weakly and ventilation drift did not; ventilation at VT1 itself stayed stable in REF-180. A ventilation reading may mark where the boundary is, but its drift did not tell how far the boundary had moved.
Limitations
Secondary, exploratory and in-house. Four studies from one laboratory, pooled after the fact; not registered, no sample-size estimate, and the riders are already in the base, so this is not independent confirmation of the threshold slide.
Internal validation only. The model was built and checked on the same 48 riders (bootstrap); the authors say external validation in other cyclists and protocols is needed and estimate at least 22 riders for it.
One kind of ride. Every source protocol held about 90% of VT1 power; intermittent, race-like or hotter rides, and running, were not studied. Feeding was controlled within each study; the authors note a usable model must cope with the varied fuelling of real athletes.
Decoupling as measured is a laboratory step, the last minute of a 4-min step at a fixed power; whole-ride decoupling from training files and any percentage threshold were not tested.
VT1 itself carries error (about 3.5% day to day by the authors' citation), and one implausible point (+29 W) was removed.
Reporting: 51 in the abstract vs 48 analysed; N = 58 for the heart-rate-at-VT1 comparison vs 56 data points listed at 150 and 180 min (our arithmetic); headline error and R² are in-sample. 44 of the 51 riders were men.
How this entry was checked and changed
Why this level
n = 51 recruited across four studies, 48 contributing 85 change scores; correlational and prediction modelling, validated only internally (bootstrap); not registered, no sample-size estimate; all four source studies are from the same laboratory and are already in the base; read from the full text (open access, CC BY 4.0)
Full text (checked 2026-10-26)
Durability of VT1 was assessed via power output at VT1 before and after ~ 2.5-h of initially moderate-intensity cycling in 51 trained cyclists, as part of four studies published elsewhere (full text)
Exercise-induced changes were quantified for 48 participants at 85 timepoints; 48, 12, 12, 47, 9, 3 and 2 data points at minutes 0, 60, 120, 150, 180, 240 and 300 (full text)
Change in VT1 power vs change in HR: rrm = −0.76 [95% CI −0.87, -0.58], P < 0.001; vs FR: rrm = −0.40 [−0.64, −0.09], P = 0.013; vs VE: rrm = −0.25 [−0.52, 0.08], P = 0.136 (full text)
Final model MAE 7.2 W and R2 0.95 in-sample; 8.3 W and R2 0.93 from 2000 bootstrap resamples (full text)
Baseline VT1 + ΔHR model: MAE 8.3 (8.7 from resampling), R2 0.94, ΔHR coefficient −0.75 [−1.60, −0.38], baseline VT1 coefficient 0.96 (full text)
HR at VT1 after 150/180 min was elevated compared to baseline (N = 58, 142 ± 11 vs. 147 ± 10 b.min−1, P = 0.0002) (full text)
All studies involved prolonged exercise (primarily) at 90% of the initial power output at the moderate-to-heavy intensity transition, and pre- and peri-exercise feeding was controlled in each study (full text)
One data point, an increase of 29 W in VT1 power following 150 min of cycling, was removed as an outlier (full text)
Participants: Study A 13/1, B 11/2, C 10/2, D 10/2 men/women; VO2peak 59.9, 57.4, 59.3 and 52.3 mL.kg−1.min−1 (full text)
This study was not registered; no specific sample size estimate was made for this exploratory analysis; the authors declare no competing interests (full text)