Segmented zero-inflated Poisson mixed effects model with random changepoint

TL;DR

This paper introduces a novel segmented zero-inflated Poisson mixed-effects model with random changepoints, enabling analysis of hospital infection data with variable change points across hospitals, especially relevant during the COVID-19 pandemic.

Contribution

It develops an iterative likelihood-based estimation procedure for segmented ZIP mixed-effects models with random changepoints, addressing variability across hospitals.

Findings

Effective estimation of random changepoints across hospitals.

Simulation studies confirm accuracy under various scenarios.

Practical approach using standard computational tools.

Abstract

The COVID-19 pandemic has had a substantial impact on hospital services, as many institutions have observed a surge in healthcare-associated infections (HAIs) despite heightened adherence to isolation protocols and hand hygiene. According to the World Health Organization (WHO), HAIs are among the leading causes of mortality and morbidity of hospitalized patients. This study aims to examine the effect of the COVID-19 pandemic on the incidence of central venous catheter-related bloodstream infections (CR-BSIs) of hospitals in the city of S\~ao Paulo. Initially we considered segmented zero-inflated Poisson (ZIP) mixed-effects models with known changepoint, which can be estimated applying the standard framework of ZIP mixed-effects models. However, we found that the changepoint could occur at varying times across different hospitals. We present an effective iterative procedure to estimate segmented ZIP mixed-effects models with random changepoints in a likelihood-based framework. The suggested procedure is a practical approach employing conventional computational tools for estimating standard mixed-effects zero-inflated Poisson (ZIP) models. Prior to its implementation to the CR-BSI data, simulation studies were conducted to examine the accuracy of the estimation under various scenarios.