Statistical Analysis

Power analyses indicated that the target enrollment of 2250 smokers would provide 80% power to detect a 3% or larger difference in quit attempts, assuming an α = .05. We examined baseline differences between trial arms using χ² tests for categorical variables and independent-samples t tests for continuous variables. We examined differential attrition using logistic regression models.

Intent-to-treat analyses of trial outcomes included all participants randomized.²⁵ Analysis of the primary trial outcome of quit attempts used logistic regression to examine the association with trial arm, controlling for any variables that differed at baseline between arms or for which we found differential attrition. To examine if warning effects differed by participant characteristics, we added participant characteristics and their interaction with trial arm to a separate logistic regression model for each characteristic. To understand whether any effects of warnings emerged over time, exploratory analyses examined differences in quit attempts during the trial by each follow-up visit. For continuous secondary outcomes, we used independent samples t tests, examining whether the outcomes differed by trial arm. Analyses of continuous outcomes using nonparametric tests yielded an identical pattern of statistical significance. Analyses of secondary outcomes used the last observation available. We did not plan or conduct interim analyses. Analyses used SAS version 9.4 (Cary, NC). We set critical α = .05 and used 2-tailed statistical tests.