Predicting treatment recovery and adherence in alcohol behavioral couple therapy using natural language processing

Introduction Alcohol Behavioral Couple Therapy (ABCT) is an efficacious treatment for Alcohol Use Disorder (AUD). Research increasingly focuses on identifying predictors of treatment outcomes to guide the continued refinement of ABCT for diverse AUD populations and relationship typologies, with language spoken by couples emerging as a promising behavioral domain for understanding treatment success. Leveraging natural language processing (NLP) methods and a prospective design, the present study examined a broad range of linguistic features exhibited by couples as predictors and moderators of treatment outcomes.Method Participants represent a subset (n = 84, 42 couples) of individuals enrolled in a larger parent study examining ABCT intervention efficacy. Patients' drinking behaviors were tracked throughout treatment and at 3- and 6-month follow-ups. Using NLP tools, linguistic features of patients' and partners' speech during session 1 of ABCT were extracted, respectively.Results Partners' behaviors moderated patients' drinking trajectories, with patients whose partners more frequently referenced alcohol, β = -0.0015, SE = 0.0006, p = 0.0075, and paired alcohol-related statements with positive sentiment, β = -0.0717, SE = 0.0183, p Conclusions Both AUD patients and their partners' language - particularly how couples discuss alcohol - may shape patients' treatment outcomes in different ways, with partners' language influencing drinking trajectories and patients' language influencing treatment adherence. Findings point to both what is said (alcohol-related language) and who said it (treatment role) as key predictors of ABCT treatment success.