Predicting Achilles tendon and patellofemoral joint forces during running with consumer-grade wearable sensor data

Estimating internal forces at common injury locations may identify injury-prone runners but requires in-lab gait analysis. Consumer-grade wearable-sensors track key gait metrics, which may enable prediction of internal forces during real-world running. This study aimed to use gait metrics from consumer-grade wearable-sensors to predict Achilles tendon (AT) and patellofemoral joint (PFJ) forces during running. Fifty-five runners completed a 38-minute run at 70-125% of preferred speed on an instrumented treadmill while equipped with a consumer-grade foot-pod and chest strap. Motion capture data drove a scaled musculoskeletal model to estimate stance phase criterion AT force and PFJ contact force. Gradient-boosted regression trees were used to predict these forces from synchronously recorded wearable-sensor data. Several predictive models were compared against a baseline speed-only model using only sensor-based running speed to evaluate the added performance of incorporating additional sensor-based metrics. Compared with the speed-only model, the model incorporating sensor-based gait metrics (e.g., speed, stance time, vertical oscillation, leg stiffness) reduced mean absolute error (MAE) by 0.069 body weights (BW) for AT force (p = 0.003) and 0.039 BW for PFJ force (p = 0.036). Replacing these sensor-based metrics with motion capture-based counterparts further decreased MAE by 0.053 BW at the AT and slightly increased MAE by 0.005 BW at the PFJ. The model including a comprehensive set of motion capture-based gait metrics achieved the lowest MAE at both locations, further reducing MAE by 0.040 BW for AT and 0.110 BW for PFJ, suggesting improvements in sensor-based internal force estimates requires additional sensors to capture more gait variables.