link.springer.com
US adults’ complete switching away from cigarettes by menthol- and tobacco-flavored ENDS and by menthol cigarette preference: testing robustness to missing data - Internal and Emergency Medicine
Use of menthol-flavored (vs. tobacco-flavored) electronic nicotine delivery systems (ENDS) may be associated with higher rates of complete switching among adults who smoke (AWS). This paper evaluates the association over 2 years, while considering robustness across different missing-data approaches. 22,905 US AWS and purchased JUUL ENDS were enrolled in a longitudinal cohort study and completed up to 10 follow-ups. To assess potential bias due to missingness, analyses: (a) compared participant characteristics by levels of missingness and (b) assessed whether tobacco or menthol JUUL was associated with missingness. The association of menthol (vs. tobacco) JUUL with switching (no past 30-day smoking) was evaluated using four missing-data treatments: (1) non-missing data-only; (2) imputing smoking for missing (‘missing-as-smoking’); imputing outcome from participant characteristics and previous switching (3) one time, and (4) Multiple Imputations. Survey completion was minimally associated with participant characteristics (each explaining < 0.4% of variance) or flavor (< 2% between-flavor difference). In observed data, menthol was significantly associated with a higher switch rate than tobacco (adjusted risk ratio = 1.12 [95% CI = 1.09–1.16], model-based average switching probability: 51.6% vs. 45.9%). Effects were similar across all approaches, including Missing-as-Smoking (1.15 [1.11–1.18]; 44.1% vs. 38.4%), Single Imputation (1.12 [1.09–1.16]; 51.3% vs. 45.7%), and Multiple Imputation (1.11 [1.08–1.15]; 51.9% vs. 46.6%). The added benefit of menthol- (vs. tobacco-flavored) JUUL was concentrated among adults smoking non-mentholated cigarettes. AWS using menthol JUUL, especially those smoking non-menthol cigarettes, were more likely to switch completely than those using tobacco JUUL. Consistent results across several approaches suggested minimal bias due to missing data. Graphical Abstract
