BiMM tree: A decision tree method for modeling clustered and longitudinal binary outcomes

Area of Science:

• Biostatistics
• Clinical Research Methodology
• Data Science

Background:

• Clustered binary outcomes are common in longitudinal clinical studies.
• Generalized linear mixed models (GLMMs) face challenges with complex data structures like multi-way interactions and unknown nonlinear predictors.
• Existing methods may not be optimal for all clustered binary outcome scenarios.

Purpose of the Study:

• To introduce and evaluate the Binary Mixed Model (BiMM) tree, a novel data-driven method for analyzing clustered binary outcomes.
• To provide an alternative to GLMMs that can handle complex predictor relationships.
• To assess the performance of BiMM tree against standard methods.

Main Methods:

• Developed the Binary Mixed Model (BiMM) tree, integrating decision tree algorithms with generalized linear mixed models.
• Conducted simulation studies to compare BiMM tree accuracy with established statistical methods.
• Applied the BiMM tree method to a real-world dataset from the Acute Liver Failure Study Group.

Main Results:

• BiMM tree demonstrated slightly higher or similar accuracy compared to standard methods in simulation studies.
• The method effectively handles complex data structures, including multi-way interactions and nonlinear predictors.
• Successful application to the Acute Liver Failure Study Group dataset validates its practical utility.

Conclusions:

• The Binary Mixed Model (BiMM) tree is a viable and accurate data-driven alternative for analyzing clustered binary outcomes in clinical research.
• This method offers advantages in scenarios with complex predictor relationships where traditional GLMMs may struggle.
• BiMM tree provides a robust framework for uncovering patterns in complex clinical data.