Multi-level zero-inflated poisson regression modelling of correlated count data with excess zeros

Area of Science:

• Biostatistics
• Statistical Modeling
• Biomedical Data Analysis

Background:

• Count data with excess zeros are prevalent in biomedical research.
• Standard zero-inflated Poisson (ZIP) models are inadequate for data with simultaneous zero-inflation and lack of independence.
• Hierarchical or longitudinal study designs often lead to correlated count data.

Purpose of the Study:

• To present a multi-level ZIP regression model with random effects.
• To account for both excess zeros and inherent correlation in count data.
• To generalize the model for complex correlation structures.

Main Methods:

• Developed a class of multi-level ZIP regression models with random effects.
• Utilized an expectation-maximization algorithm for model fitting.
• Estimated variance components via residual maximum likelihood (REML) estimating equations.
• Presented a score test for zero-inflation.

Main Results:

• The multi-level ZIP model effectively handles excess zeros and correlated observations.
• The approach was generalized to accommodate more complex correlation structures.
• Demonstrated the model's utility in analyzing longitudinal infant feeding study data.

Conclusions:

• The multi-level ZIP regression model provides a robust framework for analyzing complex biomedical count data.
• This method is particularly useful for longitudinal or hierarchical studies with excess zeros and correlated outcomes.
• The developed statistical tools enhance the analysis of zero-inflated, correlated count data in various biomedical applications.