Using machine learning to examine predictors of treatment goal change among individuals seeking treatment for alcohol use disorder

Abstract

Introduction: The goals of individuals seeking treatment for alcohol use disorder (AUD) are typically quantified as abstinent or nonabstinent (e.g., moderate drinking) goals. However, treatment goals can vary over time and be influenced by life circumstances. This study aims to identify predictors of treatment goal change and direction of change from baseline to six-month follow-up among individuals seeking treatment for AUD.

Methods: This study is a secondary analysis of data from the Relapse Replication and Extension Project. The study included participants who completed assessments at baseline and six-month follow-up in the analysis (n = 441). We used decision trees to examine 111 potential predictors of treatment goal change. The study cross-validated results using random forests. The team examined changes in goal between baseline and follow-up (Decision Tree 1) and quantified them as being toward or away from a complete abstinence goal (Decision Tree 2).

Results: Nearly 50 % of the sample changed their treatment goal from baseline to 6 months, and 68.7 % changed from a nonabstinence goal toward a complete abstinence goal. The study identified seven unique predictors of goal change. The most common predictors of changing a treatment goal were number of recent treatment sessions prior to enrolling in the study, other substance use, negative affect, anxiety, social support, and baseline drinks per drinking day. Participants with a greater number of recent treatment sessions and who sought social support were most likely to change their goal. Additionally, individuals with more substance use tended to change away from complete abstinence goals. Cross-validation supported baseline drinks per drinking day, social support, baseline maximum blood alcohol concentration (BAC), lifetime tobacco use, baseline average BAC, lifetime cocaine use, Inventory of Drinking Situations total score, and Situational Confidence Questionnaire average score as important predictors.

Conclusions: This study identified seven unique predictors of treatment goal change while in AUD treatment. Prior treatment, drinking to cope, and social support were most associated with goal changes. This information can inform providers who seek to understand factors associated with treatment goal selection and changes in goals during treatment.

Keywords: Alcohol use disorder; Machine learning; Random forest; Recursive partitioning; Treatment goal; Treatment goal change.