A Multicenter Reliability Test of a Novel Osteochondritis Dissecans Radiographic Feature Classification System

OBJECTIVES: Approximately 30-50% of skeletally immature patients with stable osteochondritis dissecans (OCD) lesions of the knee fail to heal with non-operative treatment, and about 30% of patients who undergo surgery fail to heal radiographically. Unfortunately it is nearly impossible to predict which patients will heal with non-surgical or surgical treatment. We identified multiple OCD features on standard radiographs that may help to predict healing rates. In this study, we test the inter- and intra- rater reliability of orthopaedic surgeons from multiple institutions on classifying these specific OCD radiographic features. METHODS: Pre-treatment anteroposterior, lateral, and notch radiographs (Figure) of 45 pediatric knees containing OCD lesions of the medial or lateral femoral condyle were reviewed by 7 physician raters at different medical institutions at two time points. Images were viewed over a secure internet portal. Classifications included lesion location (medial/lateral), growth plate maturity (open/closing/closed), visibility of the progeny bone including fragmentation (yes/no), fragment displacement (none/partial/total), boundary (distinct/indistinct), shape (convex/linear/concave), and comparative radiodensity of the center and rim (more/same/less). Condylar width and lesion size were measured on all views. Inter-observer reliability was measured using free-marginal kappa (k(f)) and intraclass correlations. Intra-observer reliability was measured using Cohen’s kappa (k(c)), linear-weighted kappa (k(lw)), and intraclass correlations depending on measurement type. RESULTS: Inter- and intra-observer reliability were excellent for classification of lesion location (k(f)=0.96, k(c)=0.97, respectively) and skeletal maturity (ICC=0.86, k(lw)=0.84, respectively) and for measuring knee and lesion size on all views (ICC=0.92-0.98, ICC=0.84-0.95, respectively). The visibility, fragmentation, and displacement of the progeny bone were classified with substantial reliability over time (k(c)=0.67, k(c)=0.64, k(lw)=0.80, respectively) and moderate reliability between raters (k(f)=0.45, k(f)=0.54, ICC=0.52, respectively). The progeny bone boundary demonstrated substantial reliability between raters (k(f)=0.62) and moderate reliability over time (k(c)=0.55). Fair to moderate inter- and intra-observer reliability was obtained for classifying the shape (ICC=0.33, k(lw)=0.53, respectively) and comparative radiodensity of parent and progeny bone (ICC=0.11-0.52, k(lw)=0.32-0.57, respectively). CONCLUSION: Most of the specific OCD radiographic features tested showed good to excellent reliability. Lesion shape and density had only fair to moderate reliability. The results of the current study support the use of OCD radiographic feature classification in multi-center investigations. Each reliable feature may be correlated with healing in future studies and help to predict OCD outcome at the start of treatment.