Background/Purpose: Osteoarthritis (OA) is the most common chronic joint disorder, characterized by structural cartilage and bone damage that often causes pain and disability. However, the severity of joint damage does not always correlate with the level of pain. The limited effectiveness of current treatments may stem from difficulties in identifying patients at high risk of rapid structural progression. The Rheumatology and Health Research Group (GIR-S) defines progression as moving from a Kellgren and Lawrence (KL) grade 0–1 to ≥3, or from KL 2 to KL 4 or total knee replacement (TKR) within 48 months.Our group has identified molecular biomarkers that, together with clinical and imaging data, may enhance the application of machine learning (ML) to detect patterns associated with OA subgroups. Advanced statistical tools enable us to extract multivariate signatures to differentiate patient phenotypes more precisely. Our objective is to develop a predictive model for the rapid structural progression of osteoarthritis by integrating various machine learning techniques and using clinical, genomic, proteomic, and epigenetic data to predict structural progression over time in well-characterized prospective patient cohort, Osteoarthritis Initiative (OAI).

Methods: Rapid structural progression was defined according the GIR criteria, patients with an increase in the Kellgren-Lawrence (KL) score from ≤1 to ≥3, or from KL = 2 to KL = 4, or joint replacement over a follow-up period of 48 months. We used clinical, imaging, molecular biomarkers (including 20 proteins, 8 nuclear SNPs, and 3 mitochondrial haplogroups). The selected features were subsequently filtered using Recursive Feature Elimination (RFE) for dimensionality reduction. Using this data, we generated six machine learning models, dividing the data into 75% for training and 25% for testing. Figure 1.

Results: After applying Recursive Feature Elimination (RFE), the dataset included 2812 individuals and 20 variables, with 240 rapid progressors. We then applied various machine learning algorithms and compared their performance across training and testing subsets in both datasets. Among the models tested, the GLMnet model demonstrated the best performance with AUC of 0.67 (Figure 2a and 2b).

Conclusion: While promising, its full potential requires validation with additional cohorts to ensure robustness. This tool could prove invaluable for the early identification of patients at risk of rapid structural progression of OA, enabling timely interventions that could significantly improve both patient outcomes and the effectiveness of early treatment strategies. Funding: This work was funded by Pfizer inc & Elli Lilly and Company through the 3rd Global Awards for Advancin Chronic Pain Reserah, ADVANCE (Grand ID#64122119)

Figure 1. Flowchart of the analysis.

Figure 2. Variable importance for GLMnet dataset.

Figure 2b. ROC comparison for test in the dataset.

« Back to ACR Convergence 2025

ACR Meeting Abstracts - https://acrabstracts.org/abstract/machine-learning-based-model-to-predict-rapid-structural-progression-in-knee-osteoarthritis/