PMH26 IDENTIFYING PREDICTIVE ATTRIBUTES OF ADULT SMOKERS WHO CEASE COMBUSTIBLE SMOKING USING THE JUUL ELECTRONIC NICOTINE DELIVERY SYSTEM (ENDS) VIA LOGISTIC REGRESSION AND CART

Abstract

Prior research has shown that Electronic Nicotine Delivery System (ENDS) products such as JUUL may assist adult smokers in switching from combustible cigarettes, however little is known about characteristics of smokers who successfully switch. Using classification and regression tree (CART) methodology, we sought to identify attributes of similar cohorts of adult smokers no longer smoking 90 days after purchasing a JUUL starter kit. Data included 15,116 subjects from two longitudinal surveys of adult smokers who bought a starter kit and completed 90-day follow-up. Thirty-six attribute covariates were defined at time of purchase. The sample was divided into independent training and validation sets, and logistic regression with recursive feature elimination and CART models were used to examine predictors of successful switching. The recursive logistic model identified the following as predictive of switching: smoked more than 100 cigarettes in lifetime, number of days smoked in month prior to purchase, gender, whether the smoker was currently trying to quit, and previous ENDS use. The C-index of the logistic model was 0.75, indicating good discriminatory capability. The model performed similarly well in the validation cohort. The parsimonious CART model identified number of days smoked in month prior to purchase, lifetime years smoking, previous use of ENDS and physical health as key factors. Partial interactions were revealed by CART. The F1 score was 65% for the training sample and 66% for the validation sample, while the misclassification rates for the training and validation samples were ∼35%. The CART model also identified those that might benefit from additional help to switch from combustible cigarettes. This study provides information on attribute predictors of adult smokers who switch successfully from combustible cigarette use, and of those at high risk of continued smoking after initiating ENDS use. Additional research is needed to understand switching patterns among these subgroups.