Chatbot for Online Support Groups to Treat Tobacco Addiction

We will conduct a pilot RCT to test the efficacy of an intelligent chatbot to aid small, private, quit-smoking peer support groups. Participants will be randomized to an intervention arm (chatbot-enhanced support group), or a control arm (support group only). In the intervention arm (N=60), each support group will be connected to an intelligent chatbot running on a secure local server as a trained LLM (large language model). The intelligent chatbot will function as an additional member of the GroupMe support group, but a member that only responds if no human does so. In the control arm (N=60), the support groups will be connected to an automated message-posting bot running on our secure local server. This automated message-posting bot will lack the response capabilities of the intelligent chatbot. But both the intelligent chatbot and the automated message-posting bot will post a pre-written daily discussion topic to encourage participants to discuss issues known to facilitate tobacco cessation or group bonding.

We will conduct a pilot RCT to test the efficacy of an intelligent chatbot to aid small, private, quit-smoking peer support groups. For this pilot RCT, we will use a 2-armed parallel group trial design with an active treatment concurrent control. Participants will be randomized to the intervention arm (chatbot-enhanced support group), or the control arm (support group only). All groups will run on GroupMe, a free group chat platform. We will recruit cohorts of 30 smokers, in 4 waves (N=120). After recruitment is complete, our statistician will randomly assign 15 participants to the intervention arm and 15 participants to the control arm. We will place 15 people in each support group arm to ensure a critical mass even with the expected participant attrition. After randomization, each participant will stay in their assigned arm for study duration. Prior to study start, each participant will receive 8 weeks of combination NRT, including nicotine patch and gum or lozenges (their choice), consistent with clinical practice guidelines for tobacco cessation. In the intervention arm (N=60), each support group will be connected to our intelligent chatbot running on our secure local server as a trained LLM (large language model). This chatbot will monitor all posts in the group and seek to comprehend these posts using the training it has been provided. If a group member makes a post and no one responds with about 10 seconds, the chatbot will respond using one of its 25 response libraries created from knowledge bases, which contain over 1k responses in total. In effect, the intelligent chatbot will function as an additional member of the GroupMe support group, but a member that only responds if no human does so. Furthermore, every day at 5 PM Pacific, 8 PM Eastern, the chatbot will post a pre-written daily discussion topic, to encourage participants to discuss issues known to facilitate tobacco cessation or group bonding. In the control arm (N=60), the support groups will be connected to our original automated message-posting bot running on our secure local server. This automated message-posting bot will lack the response capabilities of the intelligent chatbot; it will not respond to posts if no human group member does but, instead, remain silent. However, it will post the same daily discussion topic, and at the same time of day, as the intelligent chatbot. Each night, we will download each group's past 24-hour posts. These downloads will show each post verbatim along with the datetime stamp and poster username (including posts by the chatbot and automated message-posting bot). Using these downloaded posts, we will measure our primary outcome: the total number of posts made by each human participant, excluding the chatbot or bot. In addition, we will measure participants' bio-confirmed smoking abstinence and their NRT usage at 1-month and at 3-month intervention end. Trained research staff who are blind to study arm will contact participants via email, text and phone. They will ask participants the following 4 questions: "Over the past 7 days, how many cigarettes have you smoked? Over the past 7 days, how many times have you vaped or use e-cigarettes with nicotine? Over the past 7 days, how many times have you used tobacco products other than cigarettes, e.g., cigars, pipe, snuff, chew, snus, or hookah? Over the past 7 days, how many times have you used Nicotine Replacement Therapy or NRT, e.g., patches, gum or lozenges?" At 3-month intervention end, we will seek to bioconfirm self-reported smoking abstinence. If a participant reports no use of tobacco and no NRT use over the past 7 days, or refuses to answer, we will mail out a cotinine saliva test and set up a video (e.g., zoom or FaceTime) appointment at a convenient time for them to take it. Each participant's saliva test will be observed by trained research staff blind to study arm. If a participant recently used NRT, the saliva test will pick up that NRT, and so the test will not be a valid indicator of their smoking status. Therefore, we will not give a saliva test current NRT users and instead record them as NRT users. We will incentivize participants to take our saliva test with gift cards.