Accounting for Informative Dropout in Longitudinal Clinical Trials of Smoking Cessation: Application of Inverse Probability Weighting and Sensitivity Analysis

• Activity Number: 260
• Type: Contributed
• Date/Time: Tuesday, August 13, 2002 : 2:00 PM to 3:50 PM
• Sponsor: Biometrics Section*
• Author(s): Christina Korkontzelou*+ and Joseph Hogan
• Affiliation(s): Brown University and Brown University
• Keywords: Informative dropout ; semiparametric regression ; nonignorable missing data ; behavioral medicine ; repeated measures ; selection bias

Abstract

Missing data caused by dropout occur widely in smoking cessation trials. To assess the potential biases and losses in efficiency it is important to know the missing data mechanism and whether the missingness probability depends on the outcomes of the study. The application comes from the Commit to Quit Study (CTQ), a clinical trial which enrolled 281 subjects and randomized them on either the exercise condition (n=134) or behavioral therapy (control) (n=147). The primary outcome variable is longitudinal cessation status, assessed weekly for twelve weeks of follow up. During the study, 31% on exercise and 35% on control dropped out. The objective is to draw inferences about treatment effects while adjusting for dropout under several assumptions about the missing data mechanism (MCAR, MAR, MNAR), and to assess whether treatment effects are sensitive to non-identifiable selection biases. We use inverse probability weighted estimating equations (Robins et. al.,1995 and Rotnitzky et. al.,1998) and associated sensitivity analysis procedures. In our analysis of the CTQ data, estimates of cessation rates depend on missing data assumptions, but treatment effects remain relatively constant.