A0537
Title: A Bayesian hierarchical model for mortality surveillance using partially verified verbal autopsy data
Authors: Zehang Li - University of California, Santa Cruz (United States) [presenting]
Abstract: Monitoring data on causes of death is an integral part of understanding the burden of diseases and evaluating public health interventions. Verbal autopsy (VA) is a well-established method for gathering information about deaths outside of hospitals by interviewing family members or caregivers of a deceased person. Data from VA can be used to infer causes of death based on the collected symptoms and covariates. However, little information about the relationship or dynamics between symptoms and the new cause of death is available when a new disease emerges. A Bayesian hierarchical model framework is proposed that can be used to estimate the fraction of deaths due to the emerging disease using the VA data stream collected with partially verified cause-of-death. A latent class model is used to capture the distribution of symptoms and their dependence parsimoniously. Several potential sources of bias are discussed that may occur in the data selection process of cause-of-death verification, and our framework is adapted to account for the verification mechanism. Also, structured priors are developed to improve prevalence estimation for sub-populations. Our model's performance is demonstrated using simulation and a mortality surveillance dataset that includes suspected COVID-19-related deaths in Brazil.
