Wrist optical heart rate is close to a chest strap during steady running but less reliable while cycling and in sprints and short efforts, where an electrode strap remains the better reference
Interpretation
Consistent direction across a meta-analysis and several validation studies. Zhang 2020 (44 studies, 15 brands): mean wrist error -0.51 bpm on the treadmill but -4.55 bpm cycling and -7.26 bpm in resistance training. Gillinov 2017 (50 adults, 54% women): chest strap rc = 0.996 vs ECG; wrist devices 0.67-0.92; only three of five optical devices acceptable cycling. Pasadyn 2019 (50 athletic adults, 68% men): strap rc = 0.98, wrist 0.89-0.96 running. Muggeridge 2021 (20 adults, 9 women): optical sensors less valid in sprints; wrist band limits of agreement -46 to 33 bpm. Gilgen-Ammann 2019 (10 adults): strap detected 99.6% of beats. Graded moderate, not strong: samples are mostly healthy non-athletes, stages were short, and every device tested (2015-2019) has since been replaced, so current-model accuracy is unknown.
Practical implication
Use wrist heart rate for steady, easy running; use an electrode chest strap for cycling, intervals, zone tests and beat-to-beat data, or check your own watch against a strap on typical sessions. Device studies date quickly, and no study tested trained triathletes over long sessions. A wearable is not a medical device: wrist sensors underestimated heart rate in atrial fibrillation (Al-Kaisey 2020, 32 patients), so symptoms or rhythm alerts need a doctor.
Sources
8 sources, independently verified.
Against ECG across treadmill, bike and elliptical at rest to vigorous intensity: electrode chest strap rc = 0.996; best wrist device rc = 0.92, others 0.83 and 0.81, a forearm optical band 0.75 and one wrist device 0.67. Most devices rc 0.88-0.93 on the treadmill; only three of five optical devices acceptable (rc > 0.80) cycling; none accurate on the elliptical with arm levers. Devices of the time; healthy adults, not athletes; short stages.
Treadmill running at 4-9 mph (2 min per stage) against ECG: chest strap rc = 0.98; best wrist device rc = 0.96; three other wrist devices rc = 0.89. Running only; 2017-2018 devices.
Mean difference versus criterion: -0.01 bpm at rest, -0.40 bpm in sleep, -0.51 bpm during treadmill walking and running, -4.55 bpm cycling and -7.26 bpm in resistance training, with error rising about 3 bpm per 10 bpm increase in heart rate during resistance training. Acceptable validity for most common activities; mostly non-athlete samples and older devices.
Against an electrode chest strap: an upper-arm optical sensor had bias -1 bpm (limits of agreement -20 to 19, r = 0.95); a wrist band underestimated by 7 bpm (limits -46 to 33, r = 0.8). Both were less valid during maximal bike and treadmill sprints than steady exercise; the wrist band was particularly poor cycling. One model of each device type.
No statistically significant difference in optical heart rate accuracy across skin tones; significant differences between devices and activity types, with absolute error during activity on average about 30% higher than at rest; devices differed in responding to changes in activity. Rest, breathing, walking and typing, not hard exercise.
Consumer chest strap correctly detected 99.6% of RR intervals overall and 99.4% during high-intensity activities, versus 94.6% and 89.8% for a Holter ECG; correlation r = 0.997. Very small sample.
Smartwatches agreed with Holter in sinus rhythm (bias < 1 beat) but underestimated heart rate in atrial fibrillation (bias -9 beats overall), worsening at higher rates: -28 at 100-120, -48 at 120-140 and -69 above 140 bpm. Patients, not athletes.
Best-practice validation framework for PPG heart rate wearables across six domains (target population, criterion measure, index measure, testing conditions, data processing, statistics), with checklists; calls for testing against a proper criterion across realistic activities and intensities.