Observer variation in detecting the radiologic features associated with bronchiolitis

Chest radiographs are commonly obtained to assess children for bronchiolitis, both to corroborate the diagnosis and to exclude other diagnostic possibilities. Their utility in this setting has not previously been examined. Using a blinded, randomized study design, we examined the interobserver and intraobserver variation in the detection of the radiologic features of bronchiolitis from the chest radiograph using "weighted kappa" statistics. This observer variation was compared with that found by other authors for other diagnoses. We also determined the reported presence of these radiologic features in radiographs from patients with bronchiolitis as compared with normal controls. Our study showed acceptable interobserver (kappa = 0.40-0.66) and intraobserver agreement (kappa = 0.50-0.78) on the radiologic features of bronchiolitis relative to other diagnoses. We demonstrated a higher reported presence of these accepted radiologic features in patients with bronchiolitis as compared to controls. Although kappa statistics are widely used in studies of observer variation, "weighted kappa" has received little attention in the radiologic literature. This statistical analysis allows observers to equivocate on the presence or absence of a feature and therefore allows the format of observer variation studies to simulate more closely the normal clinical setting.