Salon 2
A Latent Class Modeling Approach to Characterize Neonatal Sepsis Etiology (303247)
*Tracy Pondo, U.S. Centers for Disease Control and PreventionStephanie Schrag, U.S. Centers for Disease Control and Prevention
Nong Shang, U.S. Centers for Disease Control and Prevention
Sithembiso Velaphi, Chris Hani Baragwanath Academic Hospital
Matthew Westercamp, U.S. Centers for Disease Control and Prevention
Keywords: biostatistics, latent class methods
The SANISA study, Sepsis Aetiology in Neonates in South Africa, was designed to characterize the etiology of neonatal sepsis using polymerase chain reaction (PCR). Infant cases and controls in the study provided blood and nasopharyngeal/oropharyngeal (NP/OP) samples. Laboratorians tested the samples for the presence of 28 pathogens using TaqMan probe-based assays. The TaqMan array card (TAC) for blood specimens tested for 12 pathogens and the TAC cards for NP/OP specimens tested for 22 pathogens. Six pathogens were tested by PCR on both the NP/OP and the blood TAC cards. Laboratorians also cultured blood and CSF samples from cases. We applied a general Bayesian latent class modeling approach to estimate the proportion of cases that were caused by each pathogen, the true positive rate (TPR) of each test, and the false positive rate (FPR) of each test. A Gibbs sampler was constructed to iteratively update the distribution of the TPR, FPR, and the proportion of cases caused by each pathogen.