Chest radiography for the diagnosis of pneumonia in epidemiological studies
Fancourt, Nicholas S S
MetadataShow full item record
Introduction: Where chest radiographs (CXRs) form a part of the case definition for pneumonia studies, reproducible and standardized interpretations of those images is critical for drawing inferences across studies. We aimed to understand the observer variability, clinical correlations, and etiologic distributions of CXR findings among children meeting a standardized clinical definition of pneumonia, including describing variations in these between epidemiologic contexts. Methods: Pneumonia Etiology Research for Child Health (PERCH) is a standardized, case-control, study of the causes and risk factors of childhood pneumonia with nine sites in seven countries (Bangladesh; The Gambia; Kenya; Mali; South Africa; Thailand; and Zambia). Cases hospitalized with WHO-defined severe and very severe pneumonia, and community controls, provided specimens for laboratory testing. Etiology was estimated using a partial latent class Bayesian analysis. Cases provided a CXR that was interpreted with a standardized method by two members of a 14-person reading panel; discordant interpretations were resolved by two members of a 4-person arbitration panel. CXRs were categorized into one of five mutually exclusive categories: consolidation only, other infiltrate only, both consolidation and other infiltrate, normal, or uninterpretable. Results: 4232 cases and 5118 controls were enrolled in PERCH; 4172 CXRs were interpreted. Agreement was highest for consolidation (with or without other infiltrate), being 78% (Kappa=0.50) for primary readers and 84% (Kappa=0.61) for arbitrators. CXRs were interpretable in 3587 (85%) cases, of which 1652 (46%) were normal (site range 36% to 65%). Cases with abnormal CXRs (consolidation or other infiltrate) were more likely than those with normal CXRs to have hypoxia, crackles, tachypnea, or fever, but CXR-normal cases also had a high proportion of these signs. A finding of CXR-consolidation had a higher case fatality ratio (13.5%) than a finding of other infiltrate (4.7%) or normal (4.9%) CXRs. Consolidation also had a higher proportion of bacterial disease compared to other infiltrate, particularly for S. pneumoniae and H. influenzae. RSV was the predominant pathogen for cases with either consolidation or other infiltrate. Conclusions: Abnormal CXRs, particularly a finding of consolidation, predict cases more likely to have clinical signs of pneumonia, a higher proportion of bacterial disease, and who are at greater risk of death. While CXRs remain an important tool for pneumonia case identification in epidemiological studies, they are challenged by variation in observer interpretation, and clinical and etiologic correlations that differ between epidemiologic settings.