In team-sport athletes, wearable heart rate, which the sensor measures, came close to a chest strap or ECG in two studies, and a chest strap's breathing rate matched gas analysis in one; energy expenditure, which devices estimate, was under-read against calorimetry by about 10-66% in most studies, worst in multidirectional and collision work; watch VO2max agreed poorly with gas analysis after one run (ICC 0.37) and better after two (ICC 0.54): a systematic review of 11 small validation studies (188 athletes) with no pooled estimate
Main caveat: No pooled estimate; each outcome rests on few, small studies (heart rate 2, VO2max 1, respiratory frequency 1, energy expenditure 8, one of which also reported…
Čokorilo and colleagues gathered studies from 2015-2025 that checked wearables in team-sport athletes against a reference method: electrocardiography or a chest strap for heart rate, indirect calorimetry for energy expenditure, gas analysis for VO2max and respiratory frequency. Eleven studies with 188 athletes (8-26 per study) qualified, two of them added during peer review.
Heart rate from four commercial watches agreed closely with a chest strap in one study (ICC 0.68-0.91) and with ECG in another, although watches under-read at high intensity.
Energy expenditure was the weakest output: armbands, accelerometers, heart-rate-based estimates and GPS or local-position "metabolic power" mostly under-read calorimetry, by about 18-61% in single devices and up to 63-66% in handball, with errors largest in multidirectional, intermittent and collision work.
A smartwatch's VO2max in elite youth footballers agreed poorly after one run and better after a second, with wide limits. A chest strap with a strain gauge matched breath-by-breath respiratory frequency closely in Serie A footballers.
The authors conclude that accuracy depends on the metric and the setting, and that wearable outputs should be treated as supporting data, not definitive measures.
Participants
11 studies, 188 athletes, 8-26 per study, ages about 17-25; all described as trained, competitive or elite.
Samples: 24 elite youth male footballers; 16 female NCAA basketball players; 22 multisport athletes; 10 professional rugby players; 26 field and court athletes; 26 Serie A footballers; 8 competitive athletes (athletics, triathlon, cycling, football); 17 professional female footballers; 16 rugby players; 12 university-standard team-sport players; 11 handball players (6 M / 5 F).
Protocol
Search: PubMed and Scopus, English or Serbian, 2015-2025; 146 records, 70 full texts, 9 included, plus 2 added at revision.
Inclusion: healthy team-sport athletes; a wearable or microtechnology device measuring heart rate, energy expenditure, VO2max or respiratory frequency; a reference method and standard validation statistics.
Quality: QUADAS-2 by one reviewer; per-study ratings in a supplement not available.
Synthesis: narrative, by outcome; no meta-analysis because of heterogeneity in devices, references, protocols and statistics.
Key Results
| Metric (side of the line) | Studies (n) | Reference | Agreement as reported |
|---|---|---|---|
| Heart rate, four commercial watches (measured) | Montalvo 2023 (22) | Polar H10 chest strap | ICC 0.68-0.91 and MAPE 1.8% to 6-8% across the four watches; best: ICC 0.91, r 0.96, MAPE 1.8%, bias −0.77 |
| Heart rate, four watches in a maximal test (measured) | Martín-Escudero 2023 (8) | 12-lead ECG | Watches under-read at high intensity; two of four weakest above about 150 bpm (no summary numbers in the table) |
| Respiratory frequency, chest strap with strain gauge (counted from chest stretch; SYNTH-91 files chest-stretch breathing as estimated) | Di Paco 2024 (26) | Breath-by-breath gas analysis | r 0.970, CCC 0.970, ICC 0.97, bias 0.17, limits −4.58 to +4.92, RMSE 2.42 (units not printed) |
| Energy expenditure, armband (estimated) | Taylor 2018 (16) | Indirect calorimetry | r 0.83-0.84, SEE 14.5-26.7 kcal, under-read about 56.7 kcal in the skills session; error grew with intensity |
| Energy expenditure, four watches over seven activities (estimated) | Montalvo 2023 (22) | Portable gas analysis | Error 20-35%, wide limits |
| Energy expenditure, heart-rate, accelerometer and combined devices over a rugby training day (estimated) | Costello 2022 (10) | Portable gas analysis | Under-read total energy by 20-30% (heart rate only), 30-50% (accelerometer only), 10-20% combined (300-500 kcal/day) |
| Energy expenditure, accelerometer and armband (estimated) | Gastin 2018 (26) | Portable gas analysis | Bias −29.3% and −18.2%; −35% to −61% in sport circuits; RMSE about 140 kJ |
| Energy expenditure, four watches on a treadmill (estimated) | Dasa 2022 (17) | Metabolic cart | R² 0.647-0.956, SEE 0.57-1.52 kcal/min, bias −0.06 to +1.25 and RMSE 0.7-2.3 (units not printed) |
| Energy expenditure, GPS metabolic power (estimated) | Highton 2017 (16); Oxendale 2017 (12) | Spirometry; portable gas analysis | About 45% under in repeated-effort rugby (r 0.63, bias −5.94 kcal/min); 52% under in multidirectional and 34% in linear running |
| Energy expenditure, local position system (estimated) | Fuchs 2022 (11) | Portable gas analysis | 63-66% lower than reference in a handball test; r 0.32, not significant |
| VO2max, smartwatch (estimated) | Düking 2024 (24) | Gas analysis on a treadmill ramp | After one run ICC 0.37, MAPE 5.58%, bias −3.16, limits −15.7 to +9.3; after a second run ICC 0.54, MAPE 1.06%, bias 0.12, limits −8.94 to +9.17 (units not printed); a Yo-Yo IR2 estimate ICC 0.17 |
Overall quality: QUADAS-2 summarised as predominantly moderate to high; no domain results in the text.
Under-reading was the rule for energy expenditure, but not in every study: one watch study gave biases from −0.06 to +1.25 (unit not printed).
Practical Application
In stop-start sport, trust wearable heart rate far more than its calorie number. Heart rate, which the device measures, came within a mean absolute percentage error of 1.8-8% of a chest strap in one study and fell short of ECG mainly at high intensity in another; energy expenditure, which it estimates, was under-read by tens of per cent in most of eight studies, most in multidirectional and contact work.
This extends [SYNTH-91] to team-sport athletes with the same split: heart rate, which the sensor measures, agreed; calories were poor and VO2max mixed. The direction of calorie error belongs to the device and the activity: here devices mostly under-read in stop-start sport, while REF-1254 found two of three watches over-reading in steady walking and running; the VO2max result fits REF-1253's wide individual limits.
A watch VO2max can shift with the number of runs it has seen: in one study agreement improved from the first to the second run, with limits of roughly −9 to +9 units even then. Inference, not tested here: give a new watch several runs before reading its VO2max at all, and still treat it as a range.
Optical heart rate drops off at high intensity: watches under-read near maximal effort in the ECG study; a chest strap is the safer choice for hard efforts (the base's practice, SYNTH-91).
Limitations
No pooled estimate; each outcome rests on few, small studies (heart rate 2, VO2max 1, respiratory frequency 1, energy expenditure 8, one of which also reported heart rate; 8-26 athletes each).
QUADAS-2 was applied by one reviewer and reported only as an overall summary; two studies entered during peer review rather than through the search.
The heart-rate reference in one of the two studies was a chest strap, not ECG; device models, firmware and wear sites are mostly not given in the table.
Units are missing for the VO2max and respiratory-frequency agreement figures, and the VO2max second-run MAPE of 1.06% sits oddly beside limits of about ±9; the figures are reported as printed.
The abstract calls heart-rate validity consistently high, which rests on two studies of 30 athletes in total.
Funding: Provincial Secretariat for Higher Education and Scientific Research of Vojvodina; no conflicts declared.
How this entry was checked and changed
Why this level
a narrative synthesis of 11 small validation studies with no pooled estimate; its QUADAS-2 appraisal was done by one reviewer and is reported only as a one-line summary (per-study ratings in a supplement not read), and heart rate rests on two studies, VO2max and breathing rate on one each, so the claims are no stronger than the validation studies retold (B); no marker step, because device agreement is the subject