A Feasibility and Acceptability Study of a Large Language Model-based Chatbot for Brief Alcohol Intervention Among Emerging Adults

Overview

American emerging adults (EAs; aged 18-29 years) have the highest rates of alcohol use disorder (AUD) and the lowest levels of treatment engagement of any age group. Innovative, scalable, and cost-effective strategies are needed to expand early detection and intervention for EAs engaged in patterns of drinking associated with AUD. Because digital technology use is frequent among EAs, digital interventions may be a particularly suitable way to reach this population. Prior studies of digital alcohol interventions demonstrate modest but consistent reductions in alcohol use, but these tools are often limited by a lack of interactivity and personalization. Large language model (LLM)-based chatbots, such as ChatGPT, may address these limitations by enabling personalized, adaptive, and human-like engagement. These features have the potential to increase uptake and engagement with screening and brief interventions among EAs. This study will develop, validate, and conduct an open trial of an LLM-based chatbot-delivered brief intervention designed to reduce alcohol use and problems among EAs, with the primary goal of establishing preliminary feasibility and acceptability.

This feasibility and acceptability study will develop, validate, and conduct a Phase I single-arm open trial of a large language model (LLM)-based chatbot-delivered brief intervention designed to reduce alcohol use and problems among EAs. To develop the augmented LLM, the investigators will use instruction fine-tuning to enhance conversational abilities within the context of brief interventions based on high-fidelity recordings of sessions from prior clinical trials and simulated patient-provider interactions. A retrieval augmented generation system will be developed to ensure the model delivers accurate information. The augmented LLM will be incorporated into a chatbot interface delivered through a user-friendly web application. To validate the chatbot's capability for delivering brief alcohol interventions, patient actors (clinical or counseling psychology PhD students) will be assigned clinical vignettes depicting diverse EAs with patterns of drinking associated with alcohol use disorder. Patient actors will engage in two randomly ordered online text-based brief intervention sessions for each vignette (one with the chatbot and one with a human clinician). Blinded transcripts from sessions will be reviewed by experts to assess treatment fidelity. To maximize and test initial feasibility and acceptability of the intervention, the investigators will conduct semi-structured interviews with 20 EAs who report hazardous drinking, followed by an open trial with another 20 EAs.