Chest radiography of acute paediatric lower respiratory infections: Experience versus interobserver variation

Yoel Levinsky, Francis B. Mimouni*, Drora Fisher, Matityahu Ehrlichman

*Corresponding author for this work

Research output: Contribution to journalArticlepeer-review


Aim To determine radiological and clinical chest radiographs (CRs) interpretation agreement in children with acute respiratory disease (ARD) versus clinical experience in multiple observers. Methods Chest radiographs obtained in 70 consecutive children at the emergency department in 2010-2011 for ARD were reviewed. They were interpreted by 1-10 paediatric residents, three board-certified paediatricians (BCPs), three paediatric pulmonologists and one paediatric radiologist. Chest radiographs were analysed for presence of 10 radiological features and five diagnoses. A short clinical and laboratory context was given. Each child was given a clinical decision. Statistical analysis was by Fleiss' kappa for multiple observers. Results Kappas by selected major diagnostic features and by observer experience were expressed relative to diagnosis by paediatric radiologist. Best agreements were for pleural effusion and pneumonia and worst for normal X-ray, hyperinflation and atelectasis. Years of experience were influential. Antibiotics for pneumonia diagnosed by radiologist would not have been prescribed in 23% of cases by residents, 25% by BCPs and 15% by pulmonologists. Conclusion In ARD in children, there is little interobserver agreement, especially among residents, which may impact on major clinical decision. There is a need to systematically train physicians in CRs reading.

Original languageEnglish
Pages (from-to)e310-e314
JournalActa Paediatrica, International Journal of Paediatrics
Issue number7
StatePublished - 2013
Externally publishedYes


  • Emergency medicine
  • Interobserver variability
  • Pneumonia
  • Radiography
  • Thoracic


Dive into the research topics of 'Chest radiography of acute paediatric lower respiratory infections: Experience versus interobserver variation'. Together they form a unique fingerprint.

Cite this