An unsupervised machine learning analysis of biopsychosocial characteristics and treatment outcome of alcohol use disorder.

Background

Alcohol Use Disorder is a heterogeneous condition where standard severity measures often fail to predict individual treatment responses. Precision medicine requires identifying distinct biopsychosocial profiles to guide targeted interventions.

Objectives

To identify clinically meaningful Alcohol Use Disorder profiles using k-means clustering based on eight baseline biopsychosocial variables and validate their prognostic utility by comparing treatment outcomes.

Methods

A retrospective observational study included 102 patients at a tertiary care center in India. K-means clustering was applied to baseline variables: age, Alcohol use duration, alcohol severity, craving, co-occurring psychiatric conditions, self-efficacy, education, and occupation. Cluster stability was assessed via bootstrap resampling. The primary outcome was 3-month abstinence, corroborated by GGT levels.

Results

Three distinct profiles emerged: (1) Late-Onset (n = 38), with older age and lowest craving, achieving 65.8% abstinence; (2) High-Functioning (n = 34), defined by high socioeconomic status and zero co-occurring disorders; and (3) Severe (n = 30), characterized by early onset, high co-occurring psychiatric conditions (73.3%), and intense craving. Despite similar alcohol severity scores (p = 0.215), the Severe profile group had poorer outcomes (36.7% abstinence; p = 0.004) and a steep return to use trajectory.

Conclusion

Identifying three distinct Alcohol Use Disorder profiles through k-means cluster analysis advances precision medicine in substance use treatment. Return to use risk was driven by craving and co-occurring conditions rather than AUD severity. These findings support a stratified approach, where Severe profile group patients require immediate, intensive "front-loaded" intervention.