Use of chest computed tomography in the staging of pediatric Wilms' tumor: interobserver variability and prognostic significance.

Abstract

To determine the specificity and prognostic significance of computed tomography (CT) of the chest in pediatric Wilms' tumor. Patients treated for newly diagnosed Wilms' tumor at St Jude Children's Research Hospital between December 1978 and July 1995 were included in the study if an initial chest radiograph and CT were available and if pulmonary involvement (determined by chest radiographs) was absent. For the 202 patients studied, radiographs and CT scans were reviewed blindly and independently by three experienced radiologists for the presence of pulmonary nodules. Outcome variables consisted of intraobserver variability (in a subsample of 40 cases) and concordance between ratings on radiographs and CT scans (both by McNemar's test), interrater variability (by logistic regression), and the cumulative incidence of pulmonary relapse for patients with and without positive CT scans, by reviewer. As expected, ratings of pulmonary involvement on radiographs were discordant with CT ratings. There was marked variability among reviewers in CT ratings (P = .0001). Of 202 CT scans, 78 were read as positive by at least one reviewer, 41 were rated positive by only one reviewer, 18 by two reviewers, and 19 by all three. Intrarater variability on repeat reviews was not significant. Patients with nodules identified on CT had a significantly higher pulmonary relapse rate when analyzed separately by reviewer. However, for the 14 patients who had pulmonary relapse, CT scans were rated positive by all three reviewers in only five cases and as negative by all three in another five cases. The variability in interpretation of chest CT scans in patients with Wilms' tumor limits the predictive utility of these studies. Optimal, standardized techniques and central review are essential if chest CT is to be used for staging in cooperative studies.