We will test Adapt2Quit, an innovative Machine-Learning, Adaptive Motivational Messaging System. Adapt2Quit uses complex, machine-learning algorithms to adaptively select the best messages for a smoker, based upon multiple attributes, including: 1) the smoker?s profile; 2) the smoker?s explicit feedback over time to the system; and 3) data from thousands of prior smokers? profiles and their feedback patterns. Adapt2Quit?s type of machine- learning is called a recommender system. Outside healthcare, companies (like Amazon) use recommender systems to continuously learn from user feedback (e.g.: liked product, products purchased) to improve, thus enhancing personal relevance and customer engagement. Engagement is a huge challenge for digital health. In the field of computer-tailored health messaging, Adapt2Quit is the first to use machine-learning to continuously adapt to feedback and select new personalized messages to send to smokers. To evaluate the impact of the recommender system, Adapt2Quit will be compared with a robust, active control, a simple but effective messaging system. In our pilot experiment, Adapt2Quit outperformed the control, especially among socio- economically disadvantaged (SED) smokers. SED smokers are harder to engage in interventions. Thus, Adapt2Quit?s increased engagement will be of particular importance for targeting SED smokers. In addition to the potential impact of the Adapt2Quit messages in inducing and engaging smokers in cessation, our goal is to increase use of the state Quitline. We will recruit 700 SED smokers at two sites. All smokers will complete a baseline interview and receive a paper brochure with information about the state?s Quitline. Smokers will then be randomized to: Adapt2Quit or the standard messaging. As the system is designed to enhance engagement, and through engagement lead to positive actions, Aim 1 will focus on engagement [Hypothesis (H1a) Among Adapt2Quit smokers, those with higher engagement levels (completed more ratings) will have greater scores on the perceived competence scale (PCS)].
Aim 2 compares (Adapt2Quit and control) behavior change processes including perceived competence for smoking cessation and cessation supporting actions (calling a Quitline) [H2a: Adapt2Quit smokers will have greater scores on the PCS than control smokers; H2b: Adapt2Quit smokers will adopt more cessation supporting actions (Quitline, NRT) than control smokers].
Aim 3 will assess effectiveness of the system [H3a: (primary outcome) Adapt2Quit smokers will have greater smoking cessation rates (6-month point prevalence biochemically verified) than control smokers; H3b: (secondary outcome) Adapt2Quit smokers will have lower time to first quit attempt than control smokers; H3c: (mediation analysis) Measured internal and external processes will mediate the effect of Adapt2Quit on smoking cessation]. To accomplish the above aims, we have brought together a multidisciplinary team with relevant expertise, and a strong track record of collaboration.
We propose testing Adapt2Quit ? an innovative motivational texting ?recommender system.? Adapt2Quit enhances tailored motivational messaging systems using machine-learning algorithms to learn from, and adapt to, user feedback (prior and daily message ratings), thereby increasing message personal relevance. Our study will test Adapt2Quit motivational messaging texting with socioeconomically disadvantaged (SED) smokers.
