Comparator and Outcome are genuinely clean — studies use ECG and measure exactly the marketed construct. The failure is scope: 'accurate' is unqualified.
Evidence located
10
studies located
8
independent
2
field setting (one n=1)
60
largest exercise n
Findings
Where it holds. A meta-analysis of 44 articles and 738 effect sizes found mean differences of −0.01 bpm at rest and −0.51 bpm on a treadmill. Median relative error across seven devices was 1.8–5.5%.
High intensity. At 8–9 mph treadmill, no wrist device reached concordance ≥0.70 — despite 0.89–0.96 at 4–7 mph.
Resistance and arm-driven work. Pooled bias −7.26 bpm in resistance training, −4.55 bpm cycling; error grows about +3 bpm per 10 bpm rise. All five optical devices fell below 0.80 concordance on an elliptical with arm levers.
Skin tone × intensity. Equivalent at rest (~2.8 bpm), but 11.7–11.8 bpm error in medium and dark skin above 60% HRR versus ~4 bpm in light skin. The wider review literature is genuinely split — 4 of 10 studies found significant degradation, 4 found none — and only 3 studies included Fitzpatrick 6.
Device spread is large. Concordance ranged 0.67 to 0.92 within a single protocol. 'Wrist PPG' is not one technology. Moving the same manufacturer's sensor to the upper arm cut MAPE from 6.82% to 1.35%.
What would change this verdict
Independent free-living validation, n≥200, skin-tone-stratified with objective measurement, spanning running, cycling, HIIT and resistance work against ambulatory ECG — showing 95% limits of agreement within ±5 bpm in every activity × intensity × skin-tone cell, replicated on two current devices. No study of that design exists; the largest field study of exercise HR located had n=32.
Practitioner read
Trust it for resting heart rate and steady running — it will be within a few beats of a chest strap. It gets unreliable exactly when you are working hard: during sprints, heavy lifting, or anything where the wrists are gripping and moving it can read 10–20 beats low, and that gap is larger for darker skin and much larger in atrial fibrillation. For zone training, intervals, or any clinical decision, use a chest strap or move the sensor to the upper arm.
Could not be verified
A 2026 living review's figures were taken from the publisher abstract page, not full text; funding unknown.
One frequently-cited skin-tone null result could not be retrieved in full and is formally contested in the same journal.
Two MDPI-hosted papers returned HTTP 403 and are excluded rather than cited.
Listed rather than dropped. Counts are floors, not censuses.
Sources
Zhang Y, et al. (2020). J Sports Sci 38(17):2021–2034 — meta-analysis, 44 articles ↗
Gillinov S, et al. (2017). Med Sci Sports Exerc 49(8):1697–1703 ↗
Disagree with this grade? Every card carries “what would change this verdict” so
the specification is explicit. Send evidence and it will be logged and the card
re-versioned in public — including when it moves a grade in a manufacturer's favour.
Submit evidence →
