Monitoring data on causes of death is an important part of understanding the burden of diseases and effects of public health interventions. Verbal Autopsy (VA) is a well-established method for gathering information about deaths outside of hospitals by conducting an interview to 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. We propose to develop and compare a series of latent variable models for modeling VA data for mortality surveillance. First, when a new disease emerges, little information is available about the relationship or the dynamics between symptoms and the new cause of death. This research proposes a Bayesian hierarchical model framework that can be used to estimate the fraction of deaths due to the emerging disease using VA data collected with partially verified cause of death. We use a latent class model to capture the distribution of symptoms and their dependence in a parsimonious way. We discuss potential sources of bias that may occur due to the cause-of-death verification process and adapt our framework to account for the verification mechanism. We also develop structured priors to improve prevalence estimation for sub-populations. We demonstrate the performance of our model using a mortality surveillance dataset that includes suspected COVID-19 related deaths in Brazil in 2021. Next, we extend the estimation of sub-population prevalence from binary to multi-class cause-of-death scenarios. Advanced tensor decomposition techniques, the r-group independent PARAFACs model and the c-Tucker model, are introduced to model the conditional distribution of symptoms given causes of death in order to capture the distinct patterns for multiple groups of indicators and improve model interpretability and flexibility. Lastly, we propose to conduct a comprehensive analysis studying the influence of causal structure among the collected variables on the predictive performance of various models. In particular, we focus on evaluating the robustness of cause-of-death assignment algorithms under data shift and model misspecification.

Event Host: Yu Zhu, Ph.D. Student, Statistical Science PhD

Advisor: Zehang (Richard) Li

Join us in person or on Zoom: https://ucsc.zoom.us/j/91768465417?pwd=N2puT2hGbGRqVWc2NGJ2V0MxK0JQdz09

Passcode: 985290