Tree-informed Bayesian multi-source domain adaptation: cross-population probabilistic cause-of-death assignment using verbal autopsy

• DOI: 10.1093/biostatistics/kxae005
• Publication Date: 2024-02-24T11:44:10Z
• Authors: Zhenke Wu; Zehang R Li; Irena Chen; Mengbing Li
• Keywords: Verbal autopsy; Tree (set theory)

Determining causes of deaths (CODs) occurred outside civil registration and vital statistics systems is challenging. A technique called verbal autopsy (VA) widely adopted to gather information on in practice. VA consists interviewing relatives a deceased person about symptoms the period leading death, often resulting multivariate binary responses. While statistical methods have been devised for estimating cause-specific mortality fractions (CSMFs) study population, continued expansion new populations (or "domains") necessitates approaches that recognize between-domain differences while capitalizing potential similarities. In this article, we propose such domain-adaptive method integrates external similarity encoded by prespecified rooted weighted tree. Given cause, use latent class models characterize conditional distributions responses may vary domain. We specify logistic stick-breaking Gaussian diffusion process prior along tree mixing weights with node-specific spike-and-slab priors pool between domains data-driven way. The posterior inference conducted via scalable variational Bayes algorithm. Simulation studies show domain adaptation enabled proposed improves CSMF estimation individual COD assignment. also illustrate evaluate using validation dataset. article concludes discussion limitations future directions.