Individuals with knee osteoarthritis (OA) exhibit altered walking patterns that cause repetitive, abnormal forces on the knee joint, leading to disease progression. Existing interventions to reduce knee loading during walking have not resulted in meaningful change in knee OA symptoms or joint structure. A key limitation of existing research has been the use of simplified metrics to describe walking patterns and not accounting for walking amount and intensity (i.e. physical activity). The goal of this research is to comprehensively characterize walking patterns and activity in people with and without knee OA and to assess the associations of walking with 2-year change in knee OA outcomes. Machine learning approaches will be used to analyze ground reaction force (GRF) data and accelerometer-derived physical activity metrics in an existing, large, well- characterized cohort (n=2575) from the Multicenter Osteoarthritis Study (MOST). Machine learning approaches that use selected features and those that are agnostic and utilize all available information from time-varying GRFs will be used in combination with physical activity metrics to classify symptomatic and structural change. The results from machine learning approaches will be compared to those of common statistical approaches. This research will allow for characterization of the complex relationships between walking patterns and activity, providing novel insights into OA disease processes. Further, this research will provide valuable training in applying machine learning approaches to biomechanics data and may inform patient-specific strategies to optimize walking patterns and physical activity for personalized knee OA management. The principal investigator will leverage prior training in biomedical engineering applied to OA research to further advance her skills in traditional and agnostic machine learning approaches for analyses of biomechanics and physical activity data. The sponsor and co-sponsor at Boston University (BU) will provide mentorship in clinical aspects of OA, implementation of the proposed studies in a large cohort, grantsmanship, and career development. The team will work with a collaborator in computational biomedicine at BU with expertise in machine learning to achieve the scientific and training goals of this project. In addition to hands-on training, the principal investigator will enroll in didactic coursework and workshops at BU related to machine learning and computer programming. Other key aspects of training include participation in research and networking opportunities at BU and other local Institutions, as well as national and international meetings. The sponsors and the Institution provide an environment where the PI will work and learn as a part of a diverse and interdisciplinary team of OA researchers across rehabilitation, rheumatology, epidemiology, computational methods, and imaging specialties. This postdoctoral training environment will prepare the principal investigator for a career as an independent researcher with expertise in machine learning approaches as applied to biomechanics for the study of musculoskeletal diseases such as osteoarthritis.
This research will use machine learning to understand how walking patterns and physical activity in people with knee osteoarthritis are related to pain and disease worsening. The results may identify ways to improve walking patterns and physical activity to maintain knee joint health in people with knee osteoarthritis.
