A two-part mixed-effects pattern-mixture model to handle zero-inflation and incompleteness in a longitudinal setting.

Area of Science:

• Biostatistics
• Longitudinal Data Analysis
• Statistical Modeling

Background:

• Longitudinal count data often exhibit excess zeros and heterogeneity.
• Missing data due to dropouts is a common challenge in longitudinal studies.
• Existing models may not adequately address both heterogeneity and non-ignorable missingness.

Purpose of the Study:

• To propose a flexible statistical framework for analyzing longitudinal count data with excess zeros.
• To address individual-level heterogeneity using random effects.
• To handle non-ignorable missing values using a pattern-mixture approach.

Main Methods:

• Development of a finite mixture of hurdle models.
• Incorporation of random effects with discrete distributions for heterogeneity.
• Specification of a pattern-mixture model for non-ignorable missing data.
• Consideration of overdispersed counts and association between model parts.

Main Results:

• The proposed model effectively handles heterogeneity and non-ignorable dropouts.
• The method allows for overdispersed counts and associations within the model.
• Simulation studies demonstrate the effectiveness of the proposed approach.
• Application to skin cancer data illustrates practical utility.

Conclusions:

• The finite mixture of hurdle models provides a robust approach for longitudinal count data analysis.
• This framework successfully integrates solutions for heterogeneity and non-ignorable missingness.
• The method is applicable to real-world epidemiological and clinical studies.