Advancing accelerometer-based mechanical loading assessment: Development and validation of prediction equations and activity classification models

Abstract

Accurate quantification of physical activity-related mechanical loading is crucial for understanding how to optimize bone health through targeted exercise interventions. However, the high cost and lack of portability associated with force plates limit their utility to laboratory settings, highlighting the relevance of developing technological strategies that accurately measure physical activity-related mechanical loading. This thesis aimed to advance accelerometer-based mechanical loading assessment by developing and validating prediction equations for ground reaction forces and loading rates during walking, running, and jumping, and activity classification models to differentiate between these activities. To accomplish that, a series of cross-sectional studies explored the relationship between the signal obtained by laboratory force plates during the performance of several distinct physical activities and the signal obtained by accelerometers placed on different anatomical locations. The first two studies demonstrated that peak ground reaction force could be accurately predicted using accelerometer data from the ankle, lower back, and hip, with mean absolute percent errors of around 7% for walking and running and around 14% for jumping. However, peak loading rate prediction equations yielded lower accuracy, with mean absolute percent errors ranging from 17.7% to 24.7%. The third study, an external validation of existing prediction equations, revealed substantial variability in accuracy, with peak loading rate equations showing the highest errors. The fourth study developed machine learning models for classifying walking, running, and jumping activities, achieving high accuracy (>90% of agreement) across all accelerometer placements. In conclusion, this thesis advances the field of accelerometer-based mechanical loading assessment by providing validated prediction equations for ground reaction forces and loading rates during walking, running, and jumping, as well as accurate activity classification models. Future research should focus on refining these models, expanding their application, and exploring their potential for informing personalized interventions to optimize bone health.