Identification of Distinct Osteoarthritis Subgroups via Cluster Analysis

Osteoarthritis (OA) is a heterogeneous joint disease, with synovial inflammation (synovitis) reflecting one of its dimensions. Our aim was to explore if cluster analysis as a machine learning approach is effective for stratification of OA patients into distinct subgroups based on clinical data, pain assessments, immunostaining of synovial tissue, and synovial fluid analysis, as well discovering the key attributes of each subgroup. We sought to determine if cluster-analysis-based subgroups differ from synovitis-severity-based classification.
Thirty-one OA patients comprised the study. Clinical data acquisition and pain tests were performed for all patients before arthroplasty. Synovium and synovial fluid samples were obtained during surgical intervention. Synovitis severity was evaluated according to Krenn grading system. The expression of NF-κB, TNF-α, and TGF-β was evaluated via immunohistochemistry in synovial tissue and ELISA in synovial fluid. Statistical data analysis
and cluster analysis was conducted using JMP Pro V17.
Four distinct OA subgroups were identified, which did not match with the three-group classification based on synovitis severity. The first subgroup included only males with the shortest symptom duration, severe pain, low-grade synovitis, and higher NF-κB expression in synovium and synovial fluid. The second subgroup consisted of females with diabetes, hyperlipidemia, obesity, and greater serum CRP levels. It coupled with moderate pain, higher synovitis, and elevated TNF-α expression in synovial fluid. The third subgroup showed the longest symptom duration, but the lowest pain, no synovitis and negligible pro-inflammatory marker expression in synovium and synovial fluid. The fourth subgroup showed severe pain, no comorbidities, the highest synovitis grade, and notable expression of pro-inflammatory markers in synovium and synovial fluid.
This study supports the existence of distinct OA subgroups. The mismatch of cluster-analysis-based subgroups with synovitis-severity-based, highlights the power of machine learning approach to capture clinically mean-ingful OA subgroups for uniform stratification of patients for clinical trials and treatment development.