Accuracy Of Commercially Available Heart Rate Monitors Using Consumer Technology Association Standards

Abstract

New Consumer Technology Association (CTA) testing guidelines have been released for validation of heart rate (HR) sensing wearable devices. However, few investigations have implemented these standards. PURPOSE: To evaluate the accuracy and reliability of commercially available HR monitors using CTA guidelines. METHODS: Participants (n = 22) simultaneously wore a Polar chest strap, Apple Watch 4, and Garmin Forerunner 735XT during sitting, activities of daily living, walking, jogging, running, and cycling, totaling 57 minutes of monitored activity. The chest strap served as the criterion measure for HR comparisons. HR data were obtained every second from the chest strap, while Garmin and Apple Watch data were only available in varied increments. Accuracy was assessed for each device within each activity via mean bias, mean absolute percent error (MAPE), and intraclass correlation coefficient (ICC). A device was considered accurate using the CTA standard of MAPE <10% and a common acceptance of ICC > 0.90. RESULTS: Apple Watch had a mean bias within ±1 bpm, and mean absolute percent error (MAPE) <3% and an ICC of >0.96 in all six conditions. Apple Watch was equivalent to the Polar chest strap for all six testing conditions (p < 0.001). Garmin tended to underestimate HR in all conditions, except cycling. MAPE was >10% during sedentary, lifestyle, walk-jog, and running while ICC was between 0.44 and 0.81 in all conditions. Garmin tested as not equivalent to the Polar chest strap for the sedentary condition (p = 1.00), but equivalent for the remaining five conditions (p = 0.006 for lifestyle, p < 0.001 for walking, dynamic walk-jog, running, and cycling). DISCUSSION: While employing CTA testing guidelines, Apple Watch displayed acceptable HR agreement in all six conditions. However, Garmin only met acceptable MAPE accuracy guidelines while walking and cycling. Garmin’s higher MAPE and lower ICC may in part be a result of fewer data points available in the extracted file.