Bayesian semi-parametric ZIP models with space-time interactions: an application to cancer registry data

Area of Science:

• Epidemiology
• Biostatistics
• Spatial Analysis

Background:

• Analysis of lymphoid leukemia incidence data from 1988-2002 in Haut-Rhin, France.
• Standard Poisson models showed poor fit due to a high frequency of zero counts in the incidence data.
• Need for models accommodating zero-inflated count data and complex space-time interactions.

Purpose of the Study:

• To model spatio-temporal variations in lymphoid leukemia incidence.
• To address challenges posed by zero-inflated count data and space-time interactions.
• To provide a foundation for further epidemiological research on leukemia.

Main Methods:

• Utilized a flexible zero-inflated Poisson model for semi-parametric regression.
• Incorporated space-time interactions using a varying coefficient model extension.
• Employed Bayesian inference with Markov chain Monte Carlo (MCMC) methods via BayesX software.

Main Results:

• Successfully modeled spatio-temporal variations in lymphoid leukemia incidence.
• The zero-inflated model provided a better fit than standard Poisson models for the observed data.
• Identified geographical patterns and temporal evolution of the disease.

Conclusions:

• The developed zero-inflated Poisson model effectively handles excess zeros and space-time dynamics in disease incidence data.
• This approach offers a robust framework for analyzing epidemiological data with similar characteristics.
• The study provides a baseline for future investigations into lymphoid leukemia patterns and risk factors.