Novel radiographic feature classification of knee osteochondritis dissecans: a multicenter reliability study

Osteochondritis dissecans (OCD) is a vexing condition for patients, parents, and physicians because of the frequent slow healing and nonhealing that leads to prolonged treatment. Several features on plain radiographs have been identified as predictors of healing, but the reliability of their measurement has not been established.

To determine the inter- and intrarater reliability of several radiographic features used in the diagnosis, treatment, and prognosis of OCD femoral condyle lesions.

Cohort study (Diagnosis); Level of evidence, 3.

Pretreatment anteroposterior, lateral, and notch radiographs of 45 knees containing OCD lesions of the medial or lateral femoral condyle were reviewed in blinded fashion by 7 orthopaedic physician raters from different institutions over a secure web portal at 2 time points over a month apart. Classification variables included lesion location, growth plate maturity, parent bone radiodensity, progeny bone fragmentation, progeny bone displacement, progeny bone contour, lesion boundary, and radiodensity of the lesion center and rim. Condylar width and lesion size were measured on all views. Interrater reliability was assessed using free-marginal kappa and intraclass correlations. Intrarater reliability was assessed using the Cohen kappa, linear-weighted kappa, and intraclass correlations based on measurement type.

Raters had excellent reliability for differentiating medial and lateral lesions and growth plate maturity and for measuring condylar width and lesion size. In the subset of knees with visible bone in the lesion, the fragmentation, displacement, boundary, central radiodensity, and contour (concave/nonconcave) of the lesion bone were classified with moderate to substantial reliability. The radiodensity of the lesion rim and surrounding epiphyseal bone were classified with poor to fair reliability.

Many diagnostic features of femoral condyle OCD lesions can be reliably classified on plain radiographs, supporting their future testing in multifactorial classification systems and multicenter research to develop prognostic algorithms. Other radiographic features should be excluded, however, because of poor reliability.