A0550
Title: Semiparametric Bayesian two-stage meta-analysis for association between ambient temperature and new cases of COVID-19
Authors: Dongu Han - Korea University (Korea, South)
Kiljae Lee - Ohio state University (United States)
Yeonseung Chung - Korea Advanced Institute of Science and Technology (Korea, South)
Taeryon Choi - Korea University (Korea, South)
Genya Kobayashi - Meiji University (Japan)
Dongu Han - Korea University (Korea, South) [presenting]
Abstract: Two-stage meta-analysis has been a popular tool to investigate a short-term association between environmental exposure and a healthy response; in the first stage, a generalized linear model with distributed lag structure is typically fitted for each location and in the second stage, the location-specific association parameters estimated in the first stage are pooled to generate a combined estimate. A novel Bayesian approach for a two-stage meta-analysis is proposed, and an efficient MCMC algorithm and fast variational Bayes algorithms are developed as an alternative for each stage of the proposed model. Precisely, for the first stage, a new Bayesian distributed lag nonlinear model, which accommodates complex nonlinearities among time, covariate and lag, are proposed, and the model is estimated by utilizing non-conjugate variational message passing and importance sampling. A robust matrix-variate Dirichlet process mixture multivariate meta-regression is proposed for the second stage model, and a fast online variational Bayes approach is developed to estimate the model. The proposed methods are illustrated by applying them to study a short-term association between ambient temperature and new cases of COVID-19 in the United States.
