Prediction Rules Identify Which Young Adults Have Higher Rates of Heavy Episodic Drinking After Exposure to 12-Week...

Area of Science:

• Behavioral Science
• Public Health
• Digital Health

Background:

• Text message interventions show promise in reducing heavy episodic drinking (HED) among young adults with hazardous drinking histories.
• Identifying baseline predictors is crucial for personalizing interventions and optimizing treatment strategies.
• Prediction Rule Ensembles (PREs) offer a method for creating interpretable rules to identify subgroups that respond differently to interventions.

Purpose of the Study:

• To identify baseline participant characteristics that predict heavy episodic drinking (HED) at three-month follow-up in young adults receiving an alcohol text message intervention.
• To develop interpretable decision-making tools for clinical use and intervention development.

Main Methods:

• A secondary analysis of data from 916 young adults in an alcohol text message randomized clinical trial was conducted.
• A Prediction Rule Ensemble (PRE) with 10-fold cross-validation was employed to predict HED at three months.
• The PRE model incorporated 21 baseline variables, including sociodemographics, alcohol consumption patterns, impulsivity, readiness to change, and perceived peer influences.

Main Results:

• The PRE model generated 12 interpretable rules that predicted three-month HED, explaining 23% of the variance (R² = 0.23).
• The most significant predictors of HED were baseline alcohol consumption, negative urgency scores, and perceived peer drinking.
• The model demonstrated high classification accuracy, with only 0.2% of cases unclassified.

Conclusions:

• Interpretable rules derived from baseline characteristics can predict intervention response for heavy episodic drinking (HED).
• Interventions should consider addressing negative urgency and peer influences to improve outcomes for young adults at risk for HED.
• These findings support personalized approaches in digital health interventions for alcohol use.