Estimating the effect of very low nicotine content cigarettes on abstinence by menthol status using novel statistical methods: A reanalysis of a randomized trial in an experimental marketplace with alternative nicotine systems

Abstract

Abstract Introduction Very low nicotine content (VLNC) cigarettes have been proposed to reduce cigarette consumption. However, previous studies were not adequately powered to analyze the effects of VLNC cigarettes on abstinence, especially within subpopulations (e.g., people who smoke (PWS) menthol cigarettes). Here, we developed and applied two statistical approaches to address these power limitations in analyses stratified by menthol status. Methods Using data from a recent randomized trial of VLNC cigarettes available within an experimental marketplace with alternative nicotine delivery systems (ANDS), we estimated the treatment effect on abstinence stratified by menthol status. Two novel methods leveraged secondary endpoints (total nicotine equivalents, NNAL) to improve efficiency: the multiple endpoint multisource exchangeability model (ME-MEM) and the super learner structural equation model (SL-SEM). Results Among 206 PWS menthol cigarettes, VLNC cigarettes were estimated to increase the probability of abstinence by 11.3 percentage points (95% credible interval: 4.7–19.6) using the ME-MEM and 18.4 percentage points using the SL-SEM (95% confidence interval: 7.7–27.0), increasing precision by 46% and 12%, respectively. For 232 PWS non-menthol cigarettes, the ME-MEM estimated an increase of 10.2 percentage points (95% credible interval: 2.3–17.3) and the SL-SEM estimated an increase of 8.1 percentage points (95% confidence interval: -1.0–13.2), increasing precision by 35% and 99%, respectively. Conclusions In the presence of ANDS, randomization to VLNC cigarettes increases the probability of abstinence for PWS menthol cigarettes and may increase the probability of abstinence for PWS non-menthol cigarettes. Methods that leverage secondary endpoints may improve precision when conventional methods are underpowered.