Bayesian Poisson Regression and Tensor Train Decomposition Model for Learning Mortality Pattern Changes during COVID-19 Pandemic

TL;DR

This paper introduces a Bayesian Poisson tensor train model to analyze changes in cause-specific mortality during COVID-19, revealing how pandemic measures affected various death causes in Italy.

Contribution

It combines Poisson regression with tensor train decomposition in a Bayesian framework to analyze high-dimensional mortality data, providing new insights into pandemic impacts.

Findings

Identifies differential effects of interventions on mortality causes

Reveals relationships between COVID-19 and other death causes

Discovers latent demographic and temporal patterns

Abstract

COVID-19 has led to excess deaths around the world, however it remains unclear how the mortality of other causes of death has changed during the pandemic. Aiming at understanding the wider impact of COVID-19 on other death causes, we study Italian data set that consists of monthly mortality counts of different causes from January 2015 to December 2020. Due to the high dimensional nature of the data, we develop a model which combines conventional Poisson regression with tensor train decomposition to explore the lower dimensional residual structure of the data. We take a Bayesian approach, impose priors on model parameters. Posterior inference is performed using an efficient Metropolis-Hastings within Gibbs algorithm. The validity of our approach is tested in simulation studies. Our method not only identifies differential effects of interventions on cause specific mortality rates through the Poisson regression component, but also offers informative interpretations of the relationship between COVID-19 and other causes of death as well as latent classes that underline demographic characteristics, temporal patterns and causes of death respectively.