App-Based Mindfulness Training Predicts Reductions in Smoking Behavior by Engaging Reinforcement Learning Mechanisms: A Preliminary Naturalistic Single-Arm Study

Abstract

Mindfulness training (MT) has been shown to influence smoking behavior, yet the involvement of reinforcement learning processes as underlying mechanisms remains unclear. This naturalistic, single-arm study aimed to examine slope trajectories of smoking behavior across uses of our app-based MT craving tool for smoking cessation, and whether this relationship would be mediated by the attenuating impact of MT on expected reward values of smoking. Our craving tool embedded in our MT app-based smoking cessation program was used by 108 participants upon the experience of cigarette cravings in real-world contexts. Each use of the tool involved mindful awareness to the experience of cigarette craving, a decision as to whether the participant wanted to smoke or ride out their craving with a mindfulness exercise, and paying mindful attention to the choice behavior and its outcome (contentment levels felt from engaging in the behavior). Expected reward values were computed using contentment levels experienced from the choice behavior as the reward signal in a Rescorla–Wagner reinforcement learning model. Multi-level mediation analysis revealed a significant decreasing trajectory of smoking frequency across MT craving tool uses and that this relationship was mediated by the negative relationship between MT and expected reward values (all ps p p = 0.357 after controlling for the mediator). Results indicate that the use of our app-based MT craving tool is associated with negative slope trajectories of smoking behavior across uses, mediated by reward learning mechanisms. This single-arm naturalistic study provides preliminary support for further RCT studies examining the involvement of reward learning mechanisms underlying app-based mindfulness training for smoking cessation.

Keywords

• digital therapeutics
• tobacco smoking
• mindfulness training
• reinforcement learning
• computational modeling
• reward value