REF-994 · Düking et al., 2021 · Journal of Science and Medicine in Sport, Vol. · Paper · PubMed

Monitoring and adapting endurance training on the basis of heart rate variability monitored by wearable technologies: A systematic review with meta-analysis

A-fairly strong evidenceMeasured on a mixed groupWritten before October 2026, being re-checked

A systematic review and meta-analysis asking whether endurance training steered by heart rate variability (HRV) leads to better outcomes than predefined training. The search covered PubMed and Web of Science in March of 2020 and found 8 studies (198 participants) with 9 interventions that took varied approaches.

Most HRV-guided interventions scheduled fewer moderate- and/or high-intensity sessions than the predefined plans.

In a fixed-effects meta-analysis HRV guidance had a significant positive effect, g = 0.296 (95% CI from 0.031 to 0.562; p = 0.028), on submaximal physiological parameters, which the authors call medium-sized; the effects on performance (g = 0.079, p = 0.597) and VO2peak (g = 0.171, p = 0.130) were small and fell short of statistical significance.

For performance, HRV-guided training was linked to fewer non-responders, and to more positive responders.

Participants

8 studies, 198 participants, 9 interventions. Sports, sex, age and training status are not given in the abstract.

Protocol

Sources: PubMed and Web of Science, searched systematically in March of 2020 with keywords related to endurance, the autonomic nervous system and training.
Comparison: HRV-guided vs predefined training (between groups), pooled as Hedges' g with 95% confidence intervals; fixed-effects model.
Outcomes: submaximal physiological parameters, performance, VO2peak; responder counts for performance.
Interventions: "a variety of approaches"; the HRV decision rules are not described in the abstract.

Key Results

HRV-guided training was favoured for submaximal physiological parameters, with g = 0.296 (95% CI from 0.031 to 0.562), p = 0.028.
Performance: g = 0.079, 95% CI -0.050 to 0.393, p = 0.597 — small and not statistically significant.
VO2peak: g = 0.171, 95% CI -0.213 to 0.371, p = 0.130 — small and not statistically significant.
On performance, HRV-guided training was linked to fewer non-responders, and to more positive responders (counts not given in the abstract).
Most HRV-guided interventions prescribed fewer moderate- and/or high-intensity sessions than the predefined plans did.

Practical Application

"Is training steered by morning HRV better than following the plan as written?" In a meta-analysis of 8 studies (198 participants; between-group comparison with predefined training), HRV-guided training had a significant effect on submaximal physiological markers (g = 0.296, 95% CI 0.031 to 0.562) but only small, non-significant effects on performance (g = 0.079) and VO2peak (g = 0.171); it was associated with fewer performance non-responders. Letting a response marker steer the load helps at the margins and trims the bad outcomes; it is not shown to make the group faster.
Read the mechanism with care: most HRV-guided arms did FEWER moderate- and/or high-intensity sessions. Similar or slightly better outcomes from fewer hard sessions is itself useful, but it means the benefit may come from holding hard days back, not from HRV as such.
The pooled outcome that moved is the submaximal one. Inference, not tested here: submaximal markers may be the more sensitive place to look for an effect of this kind of adjustment.
Not from this paper; the base's own practice: which HRV rule to use and when the athlete's own report overrides it — see SYNTH-30.

Limitations

Only 8 studies and 198 participants; a fixed-effects model across "a variety of approaches".
The confidence intervals for performance and VO2peak are quoted as printed in the abstract; they are not symmetric about the point estimates.
Most HRV-guided arms did fewer moderate/high-intensity sessions, so HRV guidance is confounded with training content.
Populations, training status, intervention length and the HRV rules are not given in the abstract.
How responders and non-responders were defined is not given in the abstract; responder counts from small single blocks are noisy.
Search ended March 2020; later trials are not included.
Written from the abstract and record metadata only (full text not read).
How this entry was checked and changed
Why this level

meta-analysis of only 8 small studies (198 participants) using a variety of approaches; one of three pooled outcomes is significant