Skip to main content
. 2022 Sep 7;9(9):e37354. doi: 10.2196/37354

Table 1.

Summary of Apple Watch validation studies (N=19).

Study Study focus Outcome
Binsch et al [21], 2016 Resilience and workload monitoring

  • PPGa reliable in the at-rest condition; wide-ranging outcomes during movement

  • Apple Watch showed the most variance in steps and distances compared with ground truth measurements, followed by the comparison, Fitbit Surge and Microsoft Band

  • Such variances are surmised to be because of differences in data resolution and access and underlying algorithms using accelerometer and GPS data for step count estimation
Shcherbina et al [22], 2017 HRb and EEc

  • Lowest error in HR and EE for cycling; highest error for walking

  • Apple Watch achieved the lowest overall error in HR and EE of the tested devices (Basis Peak, Fitbit Surge, Microsoft Band, Mio Alpha 2, PulseOn, and Samsung Gear S2)
Dooley et al [23], 2017 HR and EE

  • Apple Watch HR mean absolute percentage error was between 1.14% and 6.70%, not significantly different during baseline and vigorous-intensity treadmill exercise; lower HR in light- or moderate-intensity treadmill exercise and recovery

  • EE mean absolute percentage error was between 14.07% and 210.84%, measuring higher EE in all states compared with the criterion measure (Parvo Medics TrueOne 2400), with greater errors for higher BMI and the male population

  • HR and EE results were mostly better than other tested devices (Fitbit Charge HR and Garmin Forerunner 225)
Wang et al [24], 2017 HR

  • Apple Watch had 95% differences between −27 bpmd and +29 bpm; concordance correlation coefficient was 0.91; accuracy diminished with exercise.
Hernando et al [25], 2018 HRVe

  • Apple Watch RR interval data were found to contain gaps lasting 6.5 seconds per gap, averaging 5 gaps per recording, not correlated with stress or relaxation case

  • The cause is surmised to be because of failure to detect reliable pulses from PPG data

  • Temporal HRV indices were not significantly affected, but frequency-based LFf and HFg power showed a significant decrease

  • Apple Watch was able to successfully detect mild mental stress
Abt et al [26], 2018 Moderate-intensity exercise

  • Apple Watch threshold for moderate-intensity exercise was lower than the defined criterion of 40% to 59% VO2Rh, leading to overestimation of moderate-intensity exercise minutes
Abt et al [27], 2018 Maximal HR

  • Apple Watch had good to very good criterion validity for measuring maximal HR with no substantial under- or overestimation

  • Moderate and small errors were found for simultaneous recording from left versus right watches
Roomkham et al [28], 2019 Sleep monitoring

  • Apple Watch raw acceleration data were used to compute ENMOi for classification

  • Apple Watch had high accuracy (97.3%) and sensitivity (99.1%) in detecting sleep and adequate specificity (75.8%) in detecting wakefulness
Perez et al [29], 2019 AFj

  • Apple Watch irregular rhythm notification was triggered on 0.52% of 419,297 participants

  • Of those who returned an ECGk patch, 84% of subsequent notifications were confirmed to be AF

  • A total of 34% of ECG patches returned identified AF in part because of the transient nature, suggesting that Apple Watch may be useful for ongoing monitoring
Nuss et al [30], 2019 EE

  • Apple Watch overestimated EE in women and underestimated EE in men

  • Pooled relative error was 24.3%, 18.6% for men, and 19.9% for women

  • Neither device showed accurate results compared with EE measured with a MetCart
Thomson et al [31], 2019 HR

  • ECG correlation was strongest for very light intensity with a >0.90 concordance correlation coefficient

  • Most relative error rates were <5% with a maximum of 5.73%

  • Apple Watch was more accurate in recording HR than the Fitbit Charge HR 2
Nelson and Allen [32], 2019 HR and passive monitoring

  • Apple Watch 3 was generally accurate across a 24-hour period compared with ECG; the mean difference was −1.8 bpm, the mean absolute error was 5.86%, and the mean agreement was 95%

  • Apple Watch was more accurate than Fitbit Charge 2
Falter et al [33], 2019 HR and EE in patients with cardiovascular disease

  • Apple Watch showed good correlation without systematic error comparing Apple Watch PPG HR with ECG ground truth

  • Apple Watch showed a systematic overestimation of EE compared with indirect calorimetry

  • Apple Watch HR accuracy was clinically acceptable
Düking et al [34], 2020 HR and EE

  • Apple Watch 4 showed the highest validity in measuring HR, followed by Polar Vantage V, Garmin Fenix 5, and Fitbit Versa

  • The coefficient of variation for HR was 0.9% to 4.3% and, for EE, it was 13.5% to 27.1%
Espinosa et al [35], 2020 Step counting and HR

  • The walking error was 2.6%; jogging error was 5.1%

  • HR limit of agreement was −2.2 to 1.8 bpm for walking and −3.5 to 4.3 bpm for jogging

  • Apple Watch displayed a high level of agreement and was highly accurate
Seshadri et al [36], 2020 HR in patients with AF

  • Patients with AF showed a correlation coefficient of 0.7 between Apple Watch 4 and telemetry

  • Apple Watch 4 HR was more accurate for patients in the AF condition than for those not in the AF condition

  • Caution suggested in Apple Watch HR monitoring in patients with arrhythmia
Seshadri et al [37], 2020 AF

  • Apple Watch 4 notification correctly identified AF in 34 of 90 instances (41% sensitivity), with no false positives and 31% inconclusive

  • The agreement between Apple Watch 4 and telemetry was 61%

  • Apple Watch–exported ECG PDF files showed AF in 84 of 90 instances (96% sensitivity), no false positives, and 2 failures to generate PDFs

  • Agreement between Apple Watch 4 ECG PDFs and telemetry was 98.9%

  • Further validation is required because of the high inconclusive result rate
Glasheen et al [38], 2021 Wheelchair use

  • Apple Watch 1 only showed good agreement on higher-rate fixed-frequency tasks, with significant overestimation at low frequency

  • Arm ergometry showed good agreement across all cadences

  • Overground tasks showed poor agreement, with significant differences found
Huynh et al [39], 2021 HR in patients with obstructive sleep apnea and AF

  • Apple Watch 1 variability increased as the magnitude of the HR measurement increased

  • The Lin concordance correlation coefficient was 0.88, suggesting acceptable agreement between Apple Watch 1 and telemetry

aPPG: photoplethysmography.

bHR: heart rate.

cEE: energy expenditure.

dbpm: beats per minute.

eHRV: heart rate variability.

fLF: low-frequency.

gHF: high-frequency.

hVO2R: reserve oxygen consumption.

iENMO: Euclidean norm minus one.

jAF: atrial fibrillation.

kECG: electrocardiogram.