Integrating Psychosocial Factors into Artificial Intelligence Models for Predicting Addiction Treatment Outcomes: A...

Area of Science:

• Artificial Intelligence in Addiction Research
• Machine Learning Applications in Healthcare
• Predictive Modeling for Treatment Outcomes

Background:

• The rapid integration of AI in addiction research necessitates understanding how psychosocial, behavioral, and social-structural factors are used in predictive models.
• Recovery is influenced by psychological, social, and environmental contexts, making it crucial to assess AI's operationalization of these dimensions for equitable tools.

Purpose of the Study:

• To systematically review how artificial intelligence (AI) and machine learning (ML) models incorporate psychosocial, behavioral, and social-structural predictors for addiction treatment outcomes.
• To assess the predictive accuracy and methodological quality of AI/ML models in addiction research.

Main Methods:

• Systematic review of peer-reviewed studies from January 2020 to October 2025 from PubMed, Scopus, and Web of Science.
• Inclusion of studies applying AI/ML to addiction treatment outcomes with explicit psychosocial, behavioral, or social-structural predictors.
• Independent screening, data extraction, and quality assessment using Joanna Briggs Institute (JBI) and Cochrane Risk of Bias 2 (RoB-2) criteria.

Main Results:

• Fifteen studies utilized diverse data sources (EHR, EMA, NLP/LLM) and identified key predictors like housing instability, psychiatric comorbidity, and neighborhood disadvantage.
• Ecological Momentary Assessment (EMA) and digital phenotyping demonstrated high short-term predictive accuracy, while EHR and claims-based models showed moderate performance.
• Overall methodological quality was moderate, with limitations in external validation, calibration, fairness, transportability, and reproducibility.

Conclusions:

• Psychosocial, behavioral, and social-structural determinants are integral to AI-driven prediction of addiction treatment outcomes.
• Current AI models are preliminary and require external validation and implementation evaluation before guiding clinical decisions.
• Future research should focus on multi-site validation, transparent reporting, fairness assessment, and co-development with stakeholders for equitable and person-centered AI tools.