Background/Purpose: MRI imaging provides insights of tissue-specific lesions of osteoarthritis (OA) and has the advantage of identifying earlier pathological changes that are not evident on radiographs. While MRI lesions often co-occur, the co-occurrence patterns and their relation to the risk of incident OA have not been well evaluated. We identified distinct subgroups of knees using a latent class model according to patterns of pathological changes on MRI and examined their relation to incident radiographic knee OA (incROA).

Methods: The MOST Study recruited 3,026 subjects with or at risk for knee OA. We obtained baseline knee MRI and knee radiographs at each visit. MRIs were scored using the Whole Organ Magnetic Resonance Score (WORMS). Tibiofemoral (TF) incROA was defined by a new occurrence of KL≥2 on PA view radiograph by Month 84, and patellofemoral (PF) incROA was defined by a new occurrence of PF OA on lateral view. For specific lesions on MRI in the TF joint, i.e., cartilage morphology, meniscal tear, meniscal extrusion, bone marrow lesion, synovitis, and effusion, we used the worst WORMS score among all sub-regions to represent the severity of that lesion in the knee. We performed latent class modeling (SAS: Proc LCA) to identify subgroups of knees by patterns of MRI lesions. Each knee was assigned to a specific subgroup according to its highest membership probability. We then examined the relation of subgroups of MRI lesions to the risk of TF incROA after adjusting for age, sex, race, clinic site, history of knee injury and surgery using logistic regression model. We took the same approach to identify subgroups of MRI lesions in the PF joint and assess their relation to PF incROA.

Results: Among 579 knees without TF OA (mean age: 60.1 years, BMI 29.2 kg/m², 59% women), we identified 4 subgroups based on baseline MRI lesions with average posterior probability 0.82: mostly normal (Group 1, 48.0%); predominantly cartilage lesion (Group 2, 24.9%); predominantly meniscal lesions (Group 3, 13.8%); and combined cartilage and meniscal lesions (Group 4, 13.3%) (Table). In Group 3, meniscal tear was more prevalent than meniscal extrusion. Bone marrow lesion, synovitis, and effusion were common in Groups 2 and 4. The risk of TF incROA was 12.2%, 22.9%, 31.3% and 44.2%, for Group 1, 2, 3 and 4, respectively. The corresponding odds ratios (ORs) of TF incROA and 95% CI were 1.0, 2.5 (1.4, 4.4), 5.4 (2.8, 10.5), and 8.2 (4.3, 15.7). Similarly, 4 subgroups of MRI lesions were identified in PF joints among 660 knees without PF OA. The ORs of PF incROA for each subgroup were 1.0, 4.3, 8.9, and 17.0, respectively.

Conclusion: The latent class analysis allowed insights of the patterns of MRI lesions. Among the four subgroups of MRI lesions we identified, the co-occurrence of cartilage and meniscal lesion markedly increased the risk of incident ROA, and the meniscal lesion subgroup posed a higher risk than the cartilage lesion subgroup.