Reliability and Accuracy of Cross-sectional Radiographic Assessment of Severe Knee Osteoarthritis: Role of Training and Experience

Objective. To determine the reliability of radiographic assessment of knee osteoarthritis ( OA) by nonclinician readers compared to an experienced radiologist. Methods. The radiologist trained 3 nonclinicians to evaluate radiographic characteristics of knee OA. The radiologist and nonclinicians read preoperative films of 36 patients prior to total knee replacement. Intrareader and interreader reliability were measured using the weighted kappa statistic and intraclass correlation coefficient ( ICC). Scores kappa < 0.20 indicated slight agreement, 0.21-0.40 fair, 0.41-0.60 moderate, 0.61-0.80 substantial, and 0.81-1.0 almost perfect agreement. Results. Intrareader reliability among nonclinicians ( kappa) ranged from 0.40 to 1.0 for individual radiographic features and 0.72 to 1.0 for Kellgren-Lawrence ( KL) grade. ICC ranged from 0.89 to 0.98 for the Osteoarthritis Research Society International ( OARSI) summary score. Interreader agreement among nonclinicians ranged from kappa of 0.45 to 0.94 for individual features, and 0.66 to 0.97 for KL grade. ICC ranged from 0.87 to 0.96 for the OARSI Summary Score. Interreader reliability between nonclinicians and the radiologist ranged from kappa of 0.56 to 0.85 for KL grade. ICC ranged from 0.79 to 0.88 for the OARSI Summary Score. Conclusion. Intrareader and interreader agreement was variable for individual radiograph features but substantial for summary KL grade and OARSI Summary Score. Investigators face tradeoffs between cost and reader experience. These data suggest that in settings where costs are constrained, trained nonclinicians may be suitable readers of radiographic knee OA, particularly if a summary score ( KL grade or OARSI Score) is used to determine radiographic severity.