Objective: To test the predictive accuracy and reporting reproducibility of digital chest radiographs under low-resource conditions.
Methods: One hundred thirty four tachypneic children who presented to two Indian hospitals were enrolled. Based on review of 16 variables recorded in the Emergency Room (ER) by a senior pediatrician, children were given one of the four clinical diagnoses: pneumonia, wheezy disease, mixed and non-respiratory. Every child also had a digital CXR. It was interpreted by ER physician, pediatrician and two independent radiologists. All used the same standardized interpretation system (one or more of: normal, minor patches, major patches, hyperinflation, lobar change, pleural effusion).
Results: The 10% of CXRs showing pleural effusions reliably predicted pneumonia and disease severity. For all other CXR findings, the correlation between CXR interpretation and clinical diagnosis was moderate to poor. Apart from pleural effusions, inter-observer agreements between interpretations made by ER physician, pediatrician and radiologist were also poor (kappa <0.4).
Conclusions: With the exception of pleural effusions, CXR findings, interpreted by a radiologist, had moderate to poor power to predict respiratory diagnosis or disease severity defined by a pediatrician. Value of CXRs was further reduced by poor inter-observer agreement. When investigating tachypneic children under low-resource conditions, CXRs should be used with a clear understanding of their limitations.