Abstract
Introduction: Despite substantial health benefits, smoking cessation attempts have high relapse rates. Neuroimaging measures can sometimes predict individual differences in substance use phenotypes – including relapse – better than behavioral metrics alone. No study to date has compared the relative prediction ability of changes in psychological processes across prolonged abstinence with corresponding changes in brain activity. Methods: Here, in a longitudinal design, measurements were made 1 day prior to smoking cessation, and at 1 and 4 weeks post-cessation (total n = 120). Next, we tested the relative role of changes in psychosocial variables versus task-based functional brain measures predicting time to nicotine relapse up to 12 months. Abstinence was bio-verified 4–5 times during the first month. Data were analyzed with a novel machine-learning approach to predict relapse. Results: Results showed that increased electrophysiological brain activity during inhibitory control predicted longer time to relapse (c-index = 0.56). However, reward-related brain activity was not predictive (c-index = 0.45). Psychological variables, notably an increase during abstinence in psychological flexibility when experiencing negative smoking-related sensations, predicted longer time to relapse (c-index = 0.63). A model combining psychosocial and brain data was predictive (c-index = 0.68). Using a best-practice approach, we demonstrated generalizability of the combined model on a previously unseen holdout validation dataset (c-index = 0.59 vs. 0.42 for a null model). Conclusion: These results show that changes during abstinence – increased smoking-specific psychological flexibility and increased inhibitory control brain function – are important in predicting time to relapse from smoking cessation. In the future, monitoring and augmenting changes in these variables could help improve the chances of successful nicotine smoking abstinence.