A0899
Title: Semiparametric Bayesian two-stage meta-analysis between ambient temperature and daily confirmed cases of COVID-19
Authors: Dongu Han - Korea University (Korea, South) [presenting]
Kiljae Lee - Ohio state University (United States)
Yeonseung Chung - Korea Advanced Institute of Science and Technology (Korea, South)
Genya Kobayashi - Meiji University (Japan)
Taeryon Choi - Korea University (Korea, South)
Abstract: Environmental epidemiological studies often use a two-stage meta-analysis to explore the short-term link between environmental exposure and health outcomes across various locations. Initially, location-specific exposure-response relationships are estimated using a generalized linear model with splines and lag structures. Then, these location-specific associations are combined in a second stage, alongside location-level predictors, to identify factors contributing to location differences. While traditional methods use frequentist frameworks, our study introduces a Bayesian approach, improving both stages' models. The first stage employs a Bayesian distributed lag nonlinear model accommodating interactive nonlinearities and decaying lag effects. The second stage utilizes a nonparametric Bayesian kernel mixture multivariate meta-regression, explaining association parameters with meta-predictors and addressing violations of linearity, normality, and homoscedasticity assumptions. Markov Chain Monte Carlo and variational Bayes algorithms are developed for estimation. To validate, these methods are applied to study the short-term relationship between ambient temperature and COVID-19 incidence in the United States. Results show superior accuracy of the first-stage model with small sample sizes and decaying lag effects, while the second-stage model captures distributional structures and nonlinear relationships effectively, outperforming conventional methods.
