Hidden Markov models for zero-inflated Poisson counts with an application to substance use

Area of Science:

• Addiction research
• Statistical modeling
• Neuroscience

Background:

• Substance abuse research often uses stress or cues to study cocaine dependence.
• The impact of study-induced stress on drug-seeking behavior remains unclear.
• Existing models may not adequately capture the complexities of cocaine use patterns.

Purpose of the Study:

• To develop and validate a statistical model for analyzing cocaine abuse frequency.
• To investigate the relationship between participation in stress and cue-reactivity studies and subsequent drug use.
• To assess the utility of a novel Hidden Markov Model for longitudinal count data in addiction research.

Main Methods:

• A 2-state Hidden Markov Model (HMM) was proposed to model weekly cocaine abuse counts.
• A zero-inflated Poisson distribution was incorporated to handle the high frequency of zero counts.
• Bayesian methods were used for model fitting, with the conditional predictive ordinate statistic for model comparison.

Main Results:

• The proposed zero-inflated Poisson Hidden Markov Model effectively captured longitudinal cocaine use patterns.
• The model demonstrated superior performance compared to other models for analyzing count data with many zeros.
• This statistical approach provides a robust framework for analyzing changes in substance abuse.

Conclusions:

• The developed Hidden Markov Model offers a powerful tool for understanding cocaine abuse dynamics.
• The findings highlight the importance of considering zero-inflation and state-dependency in addiction research.
• This methodology can inform future studies on the effects of experimental paradigms on substance use behavior.