Validity of smartwatch-derived estimates of lactate threshold heart rate and pace compared to graded exercise testing

Quantifying exercise intensity through the lactate threshold (LT) is crucial for optimizing athletic training regimes. Traditional methods like maximal lactate steady state and graded exercise testing are valid but invasive and costly. Advances in smartwatch technology offer a non-invasive alternative for monitoring LT, though their measurement protocols and outcomes have been less validated.

This study evaluates the validity of three mainstream smartwatches (Huawei GT Runner®, Garmin Forerunner 265® or 265s®, and Coros Pace3®) in estimating lactate threshold heart rate (LT HR) and pace (LT Pace), comparing these to measurement protocols and results from the modified Dmax method in laboratory standards. One hundred healthy recreational runners underwent indoor graded exercise tests followed by outdoor tests using Huawei (n = 100), Garmin (n = 23), and Coros (n = 17) smartwatches to compare differences in testing protocols and LT HR and LT Pace.

The success rates for a single test were 78% for Huawei®, 65.22% for Garmin®, and 47.06% for Coros®. For LT HR, no significant differences were observed between smartwatch and DmaxMod estimates across all devices (p > 0.05). The Huawei® watch showed MAE = 10.66 bpm, MAPE = 6.32%; Garmin®: MAE = 11.44 bpm, MAPE = 7.15%; Coros®: MAE = 8.93 bpm, MAPE = 5.95%. Corresponding Pearson correlation coefficients ranged from r = 0.13 to 0.67, and R ² values ranged from 0.02 to 0.45. In contrast, LT Pace predictions demonstrated significant overestimation for all devices. Huawei® reported the smallest error (MAE = 1.22 km/h, MAPE = 12.70%, p = 0.01, r = 0.88, R ² = 0.78), followed by Garmin® (MAE = 2.17 km/h, MAPE = 25.78%, p < 0.01, r = 0.73, R ² = 0.53), and Coros® (MAE = 1.93 km/h, MAPE = 22.63%, p = 0.08, r = 0.79, R ² = 0.62). Bland-Altman plots confirmed systematic biases and variable agreement patterns, particularly for LT Pace.

Smartwatches are capable of providing estimates of LT HR and LT Pace in recreational runners, although they tend to overestimate LT Pace and overall accuracy remains to be improved.