Osteoarthritis progression pattern based on patient specific characteristics using machine learning.

Abstract

This study aimed to identify key factors associated with different progression patterns of early osteoarthritis (OA) by developing a predictive model using patient-specific characteristics. From 2003 to 2017, data from 833 knees were analyzed. Demographic factors included age, body mass index, bone mineral density (BMD), metabolic diseases, and other comorbidities. Radiographic factors included joint space narrowing (JSN) and osteophyte formation grades. Three classification models were developed using logistic regression and a light gradient boosting machine: unicompartmental/tricompartmental OA, tricompartmental JSN-dominant OA, and tricompartmental osteophyte-dominant OA. Feature importance was evaluated using SHapley Additive exPlanations feature explanations. Patients with osteoporosis were likely to progress to tricompartmental OA with JSN, while those with a high BMD were likely to progress to unicompartmental OA. Metabolic disease-related OA was associated with tricompartmental OA involving large osteophytes. Identifying OA progression patterns and patient information may enable more effective personalized treatment and prevention strategies in the future.