Analysis of Self-Report and Biochemically Verified Tobacco Abstinence Outcomes with Missing Data: A Two-Stage Imputation Approach

• Activity Number: 428
• Type: Contributed
• Date/Time: Tuesday, August 11, 2015 : 2:00 PM to 3:50 PM
• Sponsor: Biometrics Section
• Author(s): Xianghua Luo* and Yiwen Zhang and Janet L. Thomas
• Companies: University of Minnesota and University of Minnesota and University of Minnesota
• Keywords: binary outcome ; imputation ; missing data ; tobacco abstinence

Abstract

Missing data are common in tobacco cessation studies. In this paper, we focus on a tobacco cessation study with missing binary cessation outcomes. In this study, a two-by-two factorial design was adopted to examine the effects of extended vs. standard Quit and Win contests and counseling vs. no counseling. Self-report tobacco abstinence through an online survey and urine-verified tobacco abstinence were collected at follow-up time points. Missing in cessation outcomes could occur either (1) when the survey was not responded or (2) when the urine sample was not provided by those who self-reported abstinence and were invited for urine verification. To deal with this type of two-stage missing process, we developed a two-stage imputation procedure, which allow missing status depend on cessation outcomes with different levels of dependence. Two special cases of the proposed methods include (1) the widely adopted "missing = smoking" imputation method which corresponds to the situation of a perfect correlation between smoking and missing and (2) the imputation methods based on a missing at random assumption which corresponds to a zero smoking-missing correlation within each treatment.