B1003
Title: Accounting for verification bias in prevalence estimation using verbal autopsies
Authors: Zehang Li - University of California, Santa Cruz (United States) [presenting]
Abstract: Monitoring data describing the cause of death is an essential component for understanding the burden of disease and evaluating public health interventions, especially during public health emergencies when new diseases emerge. Verbal autopsy (VA) is a well-established method to gather information about deaths outside of hospitals in many low- and middle-income countries. VAs collect symptoms and covariates of a deceased person through a questionnaire conducted to caregivers or people who are familiar with the death. We propose a novel Bayesian hierarchical model framework for estimating the fraction of deaths due to the emerging disease using VA data with reference causes only available for a potentially biased sample of deaths. We use a latent class model to capture the distribution of symptoms given causes that accounts for symptom dependence in a parsimonious way. We discuss several potential sources of bias due to the informative data selection process of cause-of-death verification and adapt our framework to account for the non-ignorable verification mechanism. We demonstrate the performance of our model using both simulation and a mortality surveillance dataset that includes suspected COVID-19-related deaths in Brazil in 2021.
