Analysis of multivariate longitudinal substance use outcomes using multivariate mixed cumulative logit model

Area of Science:

• Statistics
• Biostatistics
• Longitudinal Data Analysis

Background:

• Longitudinal studies frequently assess multiple substance use (e.g., cigarettes, alcohol, marijuana).
• Research often focuses on understanding the associations between different substances.
• Existing methods may not fully capture the complexity of multivariate ordinal substance use data over time.

Purpose of the Study:

• To propose a multivariate longitudinal modeling approach for jointly analyzing ordinal multivariate substance use data.
• To extend binary random slope logistic regression to multi-category ordinal outcomes.
• To relax the proportional odds assumption by allowing differential covariate effects.

Main Methods:

• Developed a multivariate mixed cumulative logit model.
• Extended binary random slope logistic regression for multi-category ordinal outcomes.
• Analyzed data from a P01 study with 1263 participants across 8 measurement waves over 7 years.

Main Results:

• Identified significant differences in substance use time trends between males and females.
• Males showed steeper trends for cigarette and marijuana use compared to females.
• Age effects varied across cumulative logits, violating the proportional odds assumption for all three substances.

Conclusions:

• The multivariate mixed cumulative logit model provides flexibility for analyzing inter-substance associations.
• This approach is particularly useful when the proportional odds assumption is violated.
• The model allows for a more nuanced understanding of longitudinal substance use patterns.