Title: A Bayesian Measurement Error Model for the Effect of Cigarette Filter Ventilation on Biomarker Exposures of Smokers
Presented by Can Zhang
Masters Candidate in Biostatistics
Plan B Adviser: Dr. Xianghua Luo
Abstract: Cigarette Filter Ventilation (FV) is a design feature that uses small holes to mix air and smoke, to reduce the inhalation of harmful substances. To assess potential health effects of ventilated cigarettes, researchers usually used a simple approach by averaging repeated laboratory measurements of FV for each type (e.g., brand or sub-brand) of cigarette and then use the average FV levels to correlate with outcomes such as biomarkers of exposure (BOE) of smokers. However, this simple approach overlooks possible measurement errors in FV, which may cause the estimation of the effects of FV on outcomes to be biased, as well known in the classical measurement error model literature. In this study, we develop a Bayesian measurement error model (BMEM) to take into account measurement errors in the repeatedly measured FV levels properly. The proposed Bayesian model also has the flexibility to allow the FV level (with a range of 0 to 100%) to follow distributions which are not Gaussian, a typical assumption required by frequentist measurement error models. The data motivating this research were merged from a smoker dataset which is from the baseline of 839 smokers randomized to a multi-center trial and a laboratory dataset of triply measured FV of 114 sub-brands, linked through cigarette brand/sub-brand ID. We conducted extensive simulations to compare the performance of the proposed BMEM and the simple averaging method, which showed that while the performance of the simple method can be improved by increasing the number of repeated measures within-brand, our proposed method outperformed the simple method, with less bias and more precise variance estimates for the parameters of interest in all studied scenarios. For the data application, we found that the FV level of cigarettes was negatively associated with total NNAL, a tobacco specific carcinogen measured from smokers’; urine, though not statistically significant.
