Bayesian Machine Learning Tools for Alcohol Use Disorder Research: The bpaup R Package

Area of Science:

• Data Science
• Computational Statistics
• Addiction Research

Background:

• Alcohol use disorder (AUD) research struggles with individual variability and temporal drinking patterns.
• Traditional statistical and machine learning methods are inadequate for hierarchical, longitudinal AUD data.

Purpose of the Study:

• To introduce a novel R package with 30 Bayesian machine learning functions tailored for alcohol use research.
• To address challenges in capturing heterogeneity and temporal dynamics in drinking behaviors.

Main Methods:

• Developed an R package with mixed-effects and time-trend extensions for various Bayesian models, including BART.
• Applied the package to longitudinal data from the ABQDrinQ cohort and the COMBINE clinical trial.

Main Results:

• Identified significant associations between concurrent substance use and alcohol consumption.
• Revealed nonlinear age effects on drinking variability, peaking between ages 25-30.
• Achieved high-accuracy daily alcohol consumption predictions (median correlation 0.82).

Conclusions:

• The R package provides essential uncertainty quantification for AUD research and clinical practice.
• The framework offers a balance between model complexity and interpretability for substance use research.
• The methodology is applicable to broader substance use research challenges.