Structural biomarkers predict onset of knee pain: data from the osteoarthritis initiative

Abstract

Purpose: Joint pain is the most significant symptom of OA. However, morphologic changes in bones and cartilage may precede onset of pain. The purpose of this study was to evaluate in a case-control setting which morphological changes predict the development of different subcategories of knee pain at 12, 24 or 36 months in the future. Methods: Osteoarthritis initiative (OAI) data sets were mined to find subjects whose right knee was free of joint pain at baseline but had a WOMAC pain score of 5 or greater at the 36 month observation. 61 Subjects from the OAI Incidence Cohort met the inclusion criteria. A control cohort was created by matching age, BMI, and gender of each case subject to an incident cohort subject whose 36 month WOMAC pain scores were zero. The 122 subjects’ 3D DESS WE MRI series [1] were analyzed for cartilage thickness and for bone-cartilage interface curvature and signal contrast of the entire femur and tibia (Qmetrics, Rochester, NY). Thickness measurements were normalized using a standard reference atlas. Receiver operating characteristic curves (ROC) and the area under the curve (AUC) of the ROC were used to assess the potential of the quantitative measurements as predictors of the 12, 24 and 36-month WOMAC pain scores for following sub-categories: sitting, walking, lying in bed, climbing stairs and standing pain. Each non-zero score was considered a positive response. The quantitative measurements that best predicted future pain were used to build a logistic regression model of the observed pain. The strength of model was then characterized by ROC analysis. Results: The age, gender and BMI of the case subjects matched to control subjects with no statistical differences between the groups. The ROC analysis identified several morphological variables (Standard deviation of cartilage thickness, and the presence of large thin areas of medial cartilage and in the lateral femur condyle) which weakly predicted future pain (AUC<0.62). The subcategories analysis shows that there is strong preference to predict bed pain at 12 month (AUC<0.73). 24-month pain and 36-month pain were harder to predict based on the baseline observations. Using the pain predicting variables, unconditional logistic models were used to predict the pain. Figure 1a) shows the model ROC analysis of pain in bed at 12 month, and 1b) the 12-month max pain model ROC. Figure 2 shows a plot of the different AUC for all models predicting future pain. AUC at 12-month models were better than 24 or 36-month models. Pain in bed was more predictable and walking pain was harder to predict using the measured variables. Conclusions: Quantitative measurements of bone and cartilage morphology can predict onset of knee pain. The predictive power of those descriptors can be enhanced using a multivariate model of pain with good sensitivity and specificity especially when predicting the onset of pain within 12 months. Furthermore, these models can be viewed as risk factors for developing chronic knee pain; therefore, they may be helpful in the development and test of OA therapies.