C2DREAM #4

NIH RePORTER link

Abstract

African Americans have higher morbidity and mortality than do White Americans. One contributing factor to this health disparity is tobacco use and access to and utilization of smoking cessation treatment. Just in time adaptive interventions use predictive models to identify high risk situations for smoking and alert the user to use coping skills such as medication or behavioral strategies. They are typically delivered via smartphone technology. Recent advances in artificial intelligence technology using large language models (e.g., ChatGPT) provide an opportunity to create lifelike and flexible conversational agents to support health behavior change in those with low healthcare access. However, there are racial disparities in automated speech recognition that could limit utility among African Americans. The overarching goal of this proposed new project for the Center for Cardiovascular Disease Reduction and Equity Promotion Across Minnesota is to develop a Just In Time Adaptive Intervention for trigger management for African Americans who smoke.

Specific Aims

In support of this goal, we have two specific Aims. (A1) Develop a novel and racially unbiased Personal Assistant for Smoking Cessation with Artificial Intelligence and Large Language Models (PASCAL) for managing smoking triggers. (A2) Pilot test PASCAL to determine its feasibility and acceptability; and collect preliminary efficacy data. In support of these aims we will conduct 3 studies. First, we will conduct focus groups (N=20-24) to determine the acceptability of a conversational agent for smoking cessation among African Americans who smoke cigarettes. Next, we will conduct a Wizard of Oz study where the smoking cessation app will be trained using conversations between African American people who smoke (N=20) and a tobacco treatment specialist impersonating a conversational agent. Following the development of the Just In Time Adaptive Intervention app, we will conduct a third study to pilot test the App among African Americans who smoke (N=100). All participants will be given a two-week supply of nicotine lozenges and self-help cessation materials. Participants will track their smoking for 1 week pre-quit to identify smoking triggers. At the end of the week, participants will be randomly assigned to quit smoking with the support of PASCAL or to quit on their own. Participants will be followed for 8 weeks post quit. The co-primary outcomes are feasibility of study procedures and acceptability of PASCAL. We will also collect data on lozenge use, behavioral skills use, and CO-verified point prevalence abstinence at 2 and 8 weeks post quit. This proposal is significant as it is one of the first to test the use of large language models to generate content for conversational agents for health behavior change. In addition, it centers African Americans, a priority population for cardiovascular disease prevention in the development of a new health behavior change intervention.