Machine learning predicts rate of cartilage loss: data from the osteoarthritis initiative (OAI) and the multicenter osteoarthritis (most) studies. Issue 9 (28th September 2020)

• Record Type: Journal Article
• Authors: Paixão, T.; DiFranco, M.; Goetz, C.; Ljuhar, R.; Meier, P.; Nehrer, S.
• Is Part Of: Orthopaedic journal of sports medicine. Volume 8:Issue 9(2020)Supplement 7
• Publication Date: 2020-09-28
• Subjects: Sports medicine -- Periodicals; Orthopedics -- Periodicals; Arthroscopy -- Periodicals; Arthroplasty -- Periodicals; Knee -- Surgery -- Periodicals; 616.7
• DOI: 10.1177/2325967120S00530
• Languages: English
• ISSNs: 2325-9671

Abstract

Introduction: Loss of cartilage is one of the hallmark radiographic symptoms of osteoarthritis and the direct cause of much of the disability directly related to OA. The rate of cartilage loss can range from a slow deterioration process, lasting decades, to a very rapid deterioration leading to complete loss in as little as 24 months 1. In fact, evidence has been gathering that a subset of OA patients develops an "accelerated" form of knee osteoarthritis 2. Hypotheses: The rate of cartilage loss can vary widely between patients at risk of or suffering from knee osteoarthritis (OA) but its causes remain unknown. We investigate prediction of future joint space width (JSW) loss from single time point quantitative and semi-quantitative radiographic features. Methods: Bilateral knee radiographs acquired at several time points in the context of the MOST study from 2651 patients (1079 female, 1572 male) were collected. Joint space narrowing (JSN), osteophyte and sclerosis OARSI grades, as well as Kellgren-Lawrence (KL) grade and joint space width were obtained from each image using an automated software algorithm. Individuals were classified as fast progressors if the rate of JSW loss, measured via linear regression, was above 10% baseline JSW. Fast progressors were predicted using a logistic regression model trained with KL and OARSI grades at baseline as independent variables. Independent validation was performed on 1900 individuals (1079 female, 821 male) from the Osteoarthritis Initiative (OAI) study. Performance was characterized by the area under the ROC curve (ROC-AUC). Confidence intervals were calculated by bootstrapping. Results: AUCs of 0.84 (0.82; 0.87) were achieved for classifying individual knees as fast progressors on the validation dataset (OAI). KL and sclerosis OARSI grades were the main predictors of rapid cartilage loss. Conclusion: We demonstrate prediction of future rapid cartilage loss from a single plain radiograph with validation on an independent dataset. Sclerosis OARSI grade, but not osteophytes OARSI grade, was a predictor of rapid cartilage loss, suggesting a non-canonical mode of OA progression.