Application of Random Forest for Identifying Key Demographic, Engagement, and Psychosocial Predictors of Smoking Cessation

Introduction: The use of tobacco continues to be a problem across the United States, and most quit attempts are unsuccessful. Determining which demographic, engagement, and psychosocial variables best predict smoking cessation outcomes can help future smoking cessation programs and public health strategies focus on factors that will lead to greater success in smoking cessation attempts. Methods: This study used data from an NIH-funded smoking cessation clinical trial. Random forest was used to determine which variables were the most important in predicting smoking cessation outcomes at week 24 by considering psychosocial variables only, engagement and psychosocial variables, and demographics and psychosocial variables. After random forest, a principal component analysis was performed due to high correlation among the selected important variables. Logistic regression was then performed for the two major principal component variables for the psychosocial model and the demographics and psychosocial model at week 24 and earlier weeks (week 12 and week 8) to evaluate their ability in predicting the biochemically confirmed smoking abstinent status at week 24. Results: The most important predictors for smoking cessation outcomes include the contrast between self-efficacy and dependence motives, the difference between self-efficacy and cravings, and income. Higher self-efficacy, lower cravings, and higher income were associated with increased probability of being abstinent at the final time-point. Conclusion: Future research and smoking cessation programs should focus on increasing individuals’ confidence in their ability to decrease smoking, while also working to decrease cravings and other dependence motives.