The inter-observer variation of chest radiograph reading in acute lower respiratory tract infection among children

Abstract

This study assessed the inter-observer agreement in the interpretation of several radiographic features in the chest radiographs (CXR) of 803 children aged 2–59 months with non-severe acute lower respiratory tract infection (ALRI). Inclusion criteria comprised: report of respiratory complaints, detection of lower respiratory findings, and presence of pulmonary infiltrate on the CXR taken on admission and read by the pediatrician on duty. Data on demographic and clinical findings on admission were collected from children included in a clinical trial on the use of amoxicillin (ClinicalTrials.gov Identifier NCT01200706). CXR was later read by two independent pediatric radiologists blinded to clinical information and pneumonia was finally diagnosed if there was agreement on the presence of pulmonary infiltrate or pleural effusion. The kappa index (κ) of agreement was calculated. The radiologists agreed that 774 (96.4%) and 3 (0.4%) CXR were appropriate or inappropriate for reading, respectively, and that 222 (28.7%) and 459 (59.3%) CXR presented or did not present pneumonia. In intent to treat analysis, that is, considering the 803 enrolled patients, κ for the presence of pneumonia was 0.725 (95% CI: 0.675–0.775). The overall agreement was 78.7% (normal CXR [n = 385, 60.9%], pneumonia [n = 222, 35.1%], other radiological diagnosis [n = 22, 3.5%], inappropriate for reading [n = 3, 0.5%]). The most frequent radiological findings were alveolar infiltrate (33.2%) and consolidation (32.9%) by radiologist 1 and consolidation (28.3%) and alveolar infiltrate (19.3%) by radiologist 2. Concordance for consolidation was 86.7% (k = 0.683, 95%CI: 0.631–0.741). Agreement was good between two pediatric radiologists when diagnosis of pneumonia among children with non-severe ALRI was compared.