The Discordance Between Pain and Imaging in Knee Osteoarthritis

Introduction: Clinicians use imaging studies to help gauge the degree to which structural factors within the knee account for patients' pain and symptoms. We aimed to determine the degree to which commonly used structural features predict a patient's knee pain and symptoms. Methods: Using Osteoarthritis Initiative data, a 10-year study of 4,796 patients with knee osteoarthritis (KOA), participants' KOA was characterized by radiographs and MRI scans of the knee. Salient features were quantified with two established grading systems: (1) individual radiographic features (IRFs) and (2) MRI Osteoarthritis Knee Scores (MOAKS) from MRI scans. We paired participants' IRFs (24,256 readings) and MOAKS (2,851 readings) with side-specific Knee Injury and Osteoarthritis Outcome Scores (KOOS). We trained generalized linear models to predict KOOS from features measured in IRF and MOAKS. We repeated the analysis on four subsets of the cohort. The models' predictive performance was evaluated using root mean square errors and coefficient of determination (R 2 ). Results: Neither radiographic features used to determine IRF grades nor MOAKS were predictive of patient pain or symptoms. MOAKS's performance was slightly more predictive of KOOS than IRF's. IRF's prediction of KOOS achieved a maximum R 2 of 0.15 and 0.28 for MOAKS, indicating a low level of accuracy in predicting the target variable. Discussion: Commonly used structural features from radiographs and MRI scans cannot predict KOA pain and symptoms—even when imaging features are codified by established grading systems like IRF or MOAKS. The predictive performance of these models is even worse as symptom severity worsens. Level of Evidence: IV