Abstract
AirPods Pro 3 incorporate an in-ear optical heart-rate sensor, but independent validation during exercise is limited. We evaluated agreement between AirPods Pro 3 heart rate and an ECG-derived chest-strap reference (Polar H10) during graded treadmill exercise in a controlled laboratory setting. Forty adults (mean age 23·8 years; 37·5% female) completed a protocol comprising rest and progressive exercise stages targeting ~40-85% of age-predicted maximal heart rate, including rapid workload transitions. Heart-rate time series from both devices were synchronised by timestamp and aggregated into non-overlapping 5-s epochs. Agreement was assessed using a repeated-measures Bland-Altman approach implemented via a linear mixed-effects model with participant-level random effects; absolute error metrics were calculated at the participant level and summarised overall and by intensity category. Across 16,735 paired epochs, mean bias was -0·03 beats·min ⁻ ¹ (AirPods Pro 3 minus Polar H10; 95% CI -0·22 to 0·17), indicating negligible systematic error. The total standard deviation of differences was 5·23 beats·min ⁻ ¹, yielding 95% limits of agreement from -10·27 to 10·22 beats·min ⁻ ¹, with greater dispersion at higher heart rates. Overall mean absolute error was 2·08 beats·min ⁻ ¹ and mean absolute percentage error was 2·02%, with mean absolute error ranging from 1·31 to 2·4 beats·min ⁻ ¹ across intensity categories. AirPods Pro 3 therefore provided heart-rate estimates closely aligned with a validated chest-worn reference during graded treadmill exercise in healthy adults, with minimal bias and low average error but wider epoch-to-epoch variability at higher intensities.