Development of a Magnetic Resonance Imaging–Based Definition of Knee Osteoarthritis: Data From the Multicenter Osteoarthritis Study

Objective Although magnetic resonance imaging (MRI) is the imaging modality of choice for research, there is no widely accepted MRI definition of knee osteoarthritis (OA). We undertook this study to test the performance of different MRI definitions of OA. Methods We studied Multicenter Osteoarthritis Study participants with knee symptoms using posteroanterior and lateral knee radiographs and MRIs. Radiographic OA was defined as Kellgren/Lawrence grade ≥2 in the tibiofemoral (TF) and/or patellofemoral (PF) joint. Symptomatic OA was defined using a validated questionnaire. MRI findings of cartilage damage, osteophytes, bone marrow lesions (BMLs), and synovitis were scored using the Whole‐Organ MRI Score system. We compared definitions using combinations of MRI features to the validation criteria of prevalent radiographic OA and symptomatic OA. All combinations included cartilage damage score ≥2 (0–6 scale) and osteophyte score ≥2 (0–6 scale); addition of BMLs and synovitis score was also tested. We also evaluated a Delphi panel definition that defined OA differently for the PF and TF joints. For each definition, we calculated sensitivity, specificity, and the area under the curve (AUC). Results We included 1,185 knees from 1,185 participants (mean age 66 years, 62% female, 89% White). Among the 1,185 knees, 482 knees had radiographic OA, and 524 knees had symptomatic OA. The MRI definitions with a cartilage score ≥2 and osteophyte score ≥2 and definitions which added BMLs or synovitis score ≥1 had the highest sensitivities (95.2% and 94.5%, respectively) for prevalent radiographic OA (AUCs 0.67 and 0.69, respectively), and also had the highest sensitivities for symptomatic OA. The Delphi panel definition had similar performance but was more complex to apply. Conclusion An MRI OA definition requiring cartilage damage and a small osteophyte with or without BMLs or synovitis had the best performance and was simplest for identifying radiographic OA and symptomatic OA.