Abstract
OBJECTIVE: To determine reliability among four experienced and calibrated readers in cross-sectional and longitudinal semi-quantitative MRI assessments of knee osteoarthritis (OA) in the Multicenter Osteoarthritis (MOST) study. DESIGN: From all MOST participants with at least one knee with readable 60-month and 84-month paired knee MRIs (1.0 T extremity systems), we selected 10 subjects having a spectrum of baseline disease severity of cartilage, bone marrow lesions, and meniscal damage and a spectrum of longitudinal changes in severity at 24 months follow-up. MRIs were independently assessed using the WORMS grading system by four musculoskeletal radiologists with the chronological sequence known to the readers. Kappa statistics were used to determine agreement between each pair of readers and Kendall's coefficient of concordance to determine average agreement across readers. RESULTS: For most features, cross-sectional reliability was substantial to almost perfect. Regarding longitudinal reliability (detection of longitudinal change), inter-reader reliability as weighted kappa values ranged from 0.62 to 0.78 for cartilage damage, 0.75-0.88 for bone marrow lesions, 0.75-0.92 for meniscal tears, 0.67-0.95 for meniscal extrusion, 0.51-0.77 for bone attrition, 0.43-0.76 for osteophytes, 0.31-0.70 for Hoffa-synovitis, and 0.47-0.85 for effusion-synovitis. Kendall's coefficient ranged from 0.65 to 0.98. CONCLUSION: High levels of cross-sectional reliability and moderate to high longitudinal reliability was achieved using four experienced readers in semiquantitative MRI-assessment of most knee OA features.