Machine learning models for temporally precise lapse prediction in alcohol use disorder.

Area of Science:

• Computational psychiatry
• Behavioral addiction research
• Machine learning in healthcare

Background:

• Alcohol use disorder (AUD) recovery is challenging, with high relapse rates.
• Predicting lapse risk is crucial for timely and personalized interventions.
• Ecological Momentary Assessment (EMA) offers real-time data on risk factors.

Purpose of the Study:

• To develop and validate machine learning (ML) models for predicting alcohol use lapse risk.
• To assess the predictive accuracy of ML models at different temporal granularities (week, day, hour).
• To identify key risk factors influencing short-term and long-term lapse probabilities.

Main Methods:

• Developed three ML models using EMA data from 151 individuals in early AUD recovery.
• Features included raw scores and changes in theoretically implicated risk factors.
• Employed grouped, nested cross-validation for model selection and performance evaluation.

Main Results:

• Models achieved high predictive performance, with AUCs of 0.89 (week), 0.90 (day), and 0.93 (hour).
• Past use and self-efficacy were consistently important predictors across all timeframes.
• Time-varying factors like craving and stress were more critical for next-hour lapse prediction.

Conclusions:

• ML models can accurately predict imminent alcohol use lapse risk.
• Temporal precision of prediction influences the importance of different risk factors.
• This work supports the development of smart sensing systems for proactive addiction support.