Machine learning to predict incident radiographic knee osteoarthritis over 8 Years using combined MR imaging features, demographics, and clinical factors: data from the Osteoarthritis Initiative

• Published in: Osteoarthritis and cartilage 2022-02, Vol.30 (2), p.270-279
• Main Authors: Joseph, G.B., McCulloch, C.E., Nevitt, M.C., Link, T.M., Sohn, J.H.
• Format: Article
• Language: English
• Subjects: Aged; Cartilage imaging; Female; Humans; Knee Joint - diagnostic imaging; Machine Learning; Magnetic Resonance Imaging; Male; Middle Aged; MRI; Osteoarthritis; Osteoarthritis, Knee - diagnosis; Osteoarthritis, Knee - diagnostic imaging; Predictive Value of Tests; Retrospective Studies; Time Factors; XGboost
• ISSN: 1063-4584; 1522-9653
• DOI: 10.1016/j.joca.2021.11.007

To develop a machine learning-based prediction model for incident radiographic osteoarthritis (OA) of the knee over 8 years using MRI-based cartilage biochemical composition and knee joint structure, demographics, and clinical predictors including muscle strength and symptoms. Individuals (n = 1,044) with baseline Kellgren Lawrence (KL) grade 0–1 in the right knee from the Osteoarthritis Initiative database were analyzed. 3T MRI at baseline was used to quantify knee cartilage T2, and Whole-Organ Magnetic Resonance Imaging Scores (WORMS) were obtained for cartilage, meniscus, and bone marrow. The outcome was set as true if a subject developed KL grade 2–4 OA in the right knee over 8 years (n = 183) and false if the subject remained at KL 0–1 over 8 years (n = 861). We developed and compared three models: Model 1: 112 predictors based on OA risk factors; Model 2: top ten predictors based on feature importance score from Model 1 and clinical relevance; Model 3: Model 2 without the imaging predictors. We compared the models using the area under the ROC curve derived from hold-out data. The 10-predictor model (Model 2, that includes cartilage and meniscus WORMS scores and cartilage T2) had a slightly lower AUC (0.772) compared to the model with 112 predictors (Model 1: AUC = 0.792, p = 0.739); and had a significantly higher AUC compared to the model without MR imaging predictors (Model 3, AUC = 0.669, p = 0.011). A 10-predictor model including MRI parameters coupled with demographics, symptoms, muscle, and physical activity scores provides good prediction of incident radiographic OA over 8 years.