Pneumococcal pneumonia prevalence among adults with severe acute respiratory illness in Thailand - Comparison of Bayesian latent class modeling and conventional analysis

Abstract

Background: Determining the etiology of pneumonia is essential to guide public health interventions. Diagnostic test results, including from polymerase chain reaction (PCR) assays of upper respiratory tract specimens, have been used to estimate prevalence of pneumococcal pneumonia. However limitations in test sensitivity and specificity and the specimen types available make establishing a definitive diagnosis challenging. Prevalence estimates for pneumococcal pneumonia could be biased in the absence of a true gold standard reference test for detecting Streptococcus pneumoniae. Methods: We conducted a case control study to identify etiologies of community acquired pneumonia (CAP) from April 2014 through August 2015 in Thailand. We estimated the prevalence of pneumococcal pneumonia among adults hospitalized for CAP using Bayesian latent class models (BLCMs) incorporating results of real-time polymerase chain reaction (qPCR) testing of upper respiratory tract specimens and a urine antigen test (UAT) from cases and controls. We compared the prevalence estimate to conventional analyses using only UAT as a reference test. Results: The estimated prevalence of pneumococcal pneumonia was 8% (95% CI: 5-11%) by conventional analyses. By BLCM, we estimated the prevalence to be 10% (95% CrI: 7-16%) using binary qPCR and UAT results, and 11% (95% CrI: 7-17%) using binary UAT results and qPCR cycle threshold (Ct) values. Conclusions: BLCM suggests a > 25% higher prevalence of pneumococcal pneumonia than estimated by a conventional approach assuming UAT as a gold standard reference test. Higher quantities of pneumococcal DNA in the upper respiratory tract were associated with pneumococcal pneumonia in adults but the addition of a second specific pneumococcal test was required to accurately estimate disease status and prevalence. By incorporating the inherent uncertainty of diagnostic tests, BLCM can obtain more reliable estimates of disease status and improve understanding of underlying etiology.