Kerry Costello, PhD: Predicting medial knee cartilage worsening

Poster presenter: Kerry Costello, PhD, Boston University

Poster title: Using Machine Learning to Predict Medial Knee Cartilage Worsening over 2 Years Using Gait and Physical Activity: The MOST Study

Poster Session A
8:30 – 10:30 a.m. ET Saturday, Nov. 6
All ACR Convergence 2021 poster presentations are available on demand to registered meeting participants through March 11, 2022.

What is your poster about?
We used an ensemble machine learning algorithm to predict cartilage worsening over two years in people with and without knee osteoarthritis from gait and physical activity measures taken at baseline.

Why did you decide to investigate this topic?
Mechanical loading on the joint is one of only a few modifiable risk factors for knee osteoarthritis. A better understanding of the way people walk and how their activity is related to cartilage worsening could lead to interventions to optimize both gait and physical activity for patient-specific OA management. However, correlation between various measures of gait and physical activity make traditional statistical analysis challenging. Machine learning gave us a way to look at a number of gait and physical activity measures together to start to understand how they can predict who is at risk and to identify potential intervention targets. The Multicenter Osteoarthritis Study provided the large sample size that we needed to do this work.

What are you working on next related to this research?
We are taking this a step further using deep learning to remove bias introduced by pre-selecting specific variables. Instead, we will use raw gait and physical activity data and let the algorithms decide what is important in the data.

What excites you most about your work?
We can collect a tremendous amount of information about gait and physical activity with today’s technologies, but traditionally only a very small part of these data has been used to examine how they relate to OA outcomes and to look for potential intervention targets. Our approach takes a more holistic view of the loading environment than the joint experiences by incorporating more information about gait and physical activity into our models. We hope this will lead to more effective intervention strategies for knee OA.