Evaluating Heterogeneity in the Effect of Reduced Nicotine Content Cigarettes

• Activity Number: 358 - Contributed Poster Presentations: Section on Statistics in Epidemiology
• Type: Contributed
• Date/Time: Tuesday, July 30, 2019 : 10:30 AM to 12:20 PM
• Sponsor: Section on Statistics in Epidemiology
• Author(s): Chuyu Deng*
• Companies: University of Minnesota-Div of Biostatistics
• Keywords: Machine Learning; Causal Analysis; Cigarettes; Clinical Trial; Virtual Twins

Abstract

On average, participants randomized to reduced nicotine content (RNC) cigarettes exhibit reduced cigarette use, dependence, and biomarkers of exposure compared to participants randomized to a control condition. However, understanding the potential public health impact of mandated nicotine reduction requires a full characterization of the treatment effect, including any treatment effect heterogeneity. We analyzed data from a randomized trial comparing 20 weeks of 15.5 mg/g vs. 0.4 mg/g SPECTRUM among participants with biochemically verified adherence. Our analysis used the Virtual Twins (VT) algorithm. Individual treatment effects (TEs) for cigarettes per day (CPD) were estimated using the LASSO and regression trees were used to identify sub-groups with differential TEs. The mean TE was 8.97 CPD with a SD of 4.28. The optimal regression tree for predicting individual TEs was defined by age, baseline CPD and PANAS (Positive). Our results suggest that there is heterogeneity in the effect of RNC cigarettes on CPD but that the vast majority of smokers benefit from the intervention. We also summarize our experience implementing the VT algorithm in an applied setting.