MRI cartilage thickness and KL grades show different associations with osteoarthritis pain: a study using accurate 3D MRI data from 9592 normal and osteoarthritic knees in the OAI

Purpose: Understanding the tissue associations of osteoarthritis (OA) pain is fundamental to the study of OA. Automated, highly accurate measurement of knee cartilage is now feasible, and this allows the study of large datasets of MR images, such as the OAI, where previous studies have described pain using radiographs or relatively small MRI studies. Our aim was to evaluate the association of accurate MRI cartilage thickness, and KL grade with pain, incorporating a group of factors previously described as being associated with pain using the entire Osteoarthritis Initiative (OAI) dataset. Methods: All 9592 baseline knee images from the OAI were automatically segmented using a method based on active appearance models, a form of statistical shape model. This method produces highly accurate segmentations (mean accuracy = 0.05 mm, 95th percentile = 0.45 mm) with a measurement repeatability error (SDD) of 0.12 mm. As most knees had healthy levels of cartilage (defined as within the 95% confidence limit of cartilage thickness from healthy knees), they were grouped together and used as the reference in the model. The remaining knees were divided into equally sized tertiles: mild, moderate and severe cartilage loss. KL grades were provided from a central reading of the OAI dataset at baseline. A mixed-effect linear model was used to evaluate factors associated with the WOMAC pain score (primary outcome) at baseline in all knees. Measurements were set as (level 1) unit and participants (level 2) unit. Models including either KL grade or cartilage tertile were assessed separately for femoral and tibial cartilage. Tertiles were calculated independently for each gender, as males and females have a systematic difference in cartilage thickness. Baseline complexity and model selection was evaluated using the Akaike information criterion (AIC), Bayesian information criterion (BIC) and likelihood-ratio test. Adjusted r2 was calculated for the full models, and for models without KL grades or cartilage tertiles. Results: There were 4796 participants, age 61.2, 58.5% women. Tables 1 and 2 show coefficients and effect sizes for models built with either KL grade, or cartilage tertile, along with a range of covariates. The largest coefficients within the model were for knees with KL grade 3–4, or severe cartilage loss. KL grade 4 coefficients were larger than those of severe cartilage loss. Effect sizes were very small for all covariates, including KL grade and cartilage thickness. African-Americans, and NSAID use were associated with higher pain scores, and previous surgery and smoking had small effects. Age, index knee, alignment, height, weight, physical activity and blood pressure had negligible effects. In all cases, models containing KL grade had a higher adjusted r2 than those containing cartilage thickness. For example, in the male femur models adjusted r2 was 12% for the full model including cartilage tertiles, 10% for the model without cartilage tertiles. For the models with KL grade, the corresponding values were 17% and 10%. Conclusions: This is the largest study of the associations of accurate MRI cartilage thickness and KL grade with pain. Nothing in these models explained a high proportion of people's pain. Radiographic (KL) grade explains more of the pain (almost twice as much) than MRI cartilage thickness, suggesting that KL grading measures, at least in part, additional constructs. People with more pain were more likely to be on NSAIDs and be African-American; previous surgery and smoking had slightly higher levels of pain. Age, height, weight and alignment had negligible effects on pain. A limitation of this study is that it was performed at the knee level and not patient level.View Large Image Figure ViewerDownload Hi-res image Download (PPT)View Large Image Figure ViewerDownload Hi-res image Download (PPT)