MRI of knee osteonecrosis in children with leukemia and lymphoma: Part 1, observer agreement

The aim of this study was to determine whether a new MRI-based staging system for osteonecrosis of the knee in pediatric patients could be used with an acceptable level of intra- and interobserver agreement.

We conducted a retrospective analysis of MRI studies of the knee performed in a single institution between April 1994 and July 2003. Knee osteonecrosis was identified in 168 children with a primary diagnosis of hematologic malignancy. This substantial number prompted us to design a staging system for use with pediatric patients. To assess interobserver reliability of two primary observers in using the system, they reviewed and interpreted the same 36 imaging studies of randomly chosen patients. For the assessment of intraobserver reproducibility, each observer rereviewed 16 studies. A senior observer coded potential causes of disagreement between the primary observers.

Interobserver agreement was substantial: the kappa value was 0.66 (95% confidence interval [CI], 0.58-0.75) in locations where the observers had to record only the presence or absence of a lesion, and the weighted kappa value was 0.65 (95% CI, 0.59-0.72) in locations where they had to classify the extent of involvement. The presence of marrow edema, punctate foci of altered signal, and mottled marrow changes was associated with a higher level of disagreement between the primary observers.

Our proposed classification system, developed specifically for use with MRI, was used with substantial intra- and interobserver agreement. We think its use can contribute to a standardized approach to the interpretation of MRI findings in pediatric osteonecrosis of the knee.