Full Scale Randomized Trial of an Innovative Conversational Agent for Smoking Cessation

项目总结/摘要（描述） 吸烟导致每年480，000人过早死亡，占癌症死亡人数的三分之一 我们对吸烟问题进行高影响、高成本效益的人群干预是非常必要的 停止在过去的15年里，移动的电话提供的短信干预，如国家癌症研究所的 SmokefreeTXT一直是解决这一需求的突出技术。然而，非常像所有广泛 戒烟的可用技术（例如，网站），短信干预有适度的退出 率，主要由低参与度驱动。幸运的是，一项新技术提供了一种治疗对话， 解决影响短信和其他当前数字技术的参与问题， 戒烟。机器学习、自然语言处理和云计算的进步现在已经 使创建和广泛传播计算机驱动的会话代理（CA）成为可能 数字教练旨在通过对话与用户建立长期的社交情感联系。 CA是支持性的，同情的，反思性的倾听，提供个性化的回应，并提供目标设定和 根据用户的需要适时提供建议。关于戒烟的核证机关， 知识差距是：（1）其功效，（2）理论机制，（3）成本效益。也 尚未探索的是戒烟的CA的潜在基线调节剂。我们最近开发了一种CA 戒烟，称为“QuitBot”，在日记研究中评估它，然后在一个随机的试点中测试它。 对照试验（N = 306），将其与NCI的SmokefreeTXT进行比较。试验性RCT设计非常可行 三个月随访率为93% QuitBot具有：（a）高参与者参与度和（B）高退出率， 3个月随访-与SmokefreeTXT相比非常有希望。弥补这些知识差距 在我们QuitBot研究的有希望的结果的基础上，该项目将进行一项随机对照研究， QuitBot（n = 760）与SmokefreeTXT（n = 760）的试验，随访12个月，以确定 QuitBot是否：（1）提供比SmokefreeTXT更高的戒烟率，（2）具有戒烟效果 显著介导的治疗联盟过程和参与，（3）是成本效益与。 SmokefreeTXT。此外，本研究还将探讨这些基线因素是否会影响疗效 QuitBot：信任，社会支持和人口统计（例如，性别）。这一创新项目将推动以下领域的发展 研究CA，特别是戒烟和一般健康行为改变- 而不管结果是正还是零。积极的结果可能会产生很大的人口影响 并鼓励对保护性农业的传播和实施进行新的研究， 吸烟者、语言、社区和医疗环境的多个亚群。

PROJECT SUMMARY/ABSTRACT (DESCRIPTION) Cigarette smoking accounts for 480,000 premature deaths and one third of all cancer deaths annually in the US. There is enormous need for high-impact, cost-effective population-level interventions for smoking cessation. For the past 15 years, mobile phone-delivered text messaging interventions such as the NCI’s SmokefreeTXT have been a prominent technology addressing this need. However, very much like all widely available technologies for smoking cessation (e.g., websites), text messaging interventions have modest quit rates, driven largely by low engagement. Fortunately, a new technology provides a therapeutic conversation to address the problem of engagement that impacts text messaging and other current digital technologies for smoking cessation. Advances in machine learning, natural language processing, and cloud computing are now making it possible to create and widely disseminate conversational agents (CAs), which are computer-powered digital coaches designed to form long-term social-emotional connections with users through conversations. CAs are supportive, empathic, reflectively listen, provide personalized responses, and offer goal setting and advice appropriately timed to the needs of the user. Regarding CAs for smoking cessation, the major knowledge gaps are: (1) their efficacy, (2) theoretical mechanisms, and (3) the cost-effectiveness. Also unexplored are the potential baseline moderators of CAs for smoking cessation. We recently developed a CA for smoking cessation, called “QuitBot,” evaluated it in a diary study, and then tested it in a pilot randomized controlled trial (N = 306), comparing it with the NCI’s SmokefreeTXT. The pilot RCT design was very feasible with 93% three-month follow-up. QuitBot had: (a) high participant engagement and (b) high quit rates at the three-month follow-up—very promising in comparison with SmokefreeTXT. Addressing these knowledge gaps and building on the promising results of our QuitBot research, the project will conduct a randomized controlled trial of QuitBot (n = 760) versus SmokefreeTXT (n = 760) with 12-month follow-up in order to determine whether QuitBot: (1) provides higher quit rates than SmokefreeTXT, (2) has smoking cessation outcomes significantly mediated by therapeutic alliance processes and engagement, and (3) is cost-effective vs. SmokefreeTXT. In addition, this study will explore whether these baseline factors moderate the effectiveness of QuitBot: trust, social support, and demographics (e.g., sex). This innovative project will advance the fields of research on CAs both for smoking cessation in particular and for health behavior change in general— regardless of whether the results are positive or null. Positive results could have high population-level impact and stimulate new lines of research into CA dissemination and implementation, and the adaptation of CAs for multiple subpopulations of smokers, languages, and community and medical settings.