A Comparison of Multiple Imputation and Inverse Probability Weighted Estimation for Survival Analysis with Missing Covariates

• Activity Number: 64
• Type: Contributed
• Date/Time: Sunday, August 2, 2009 : 4:00 PM to 5:50 PM
• Sponsor: Biometrics Section
• Author(s): Lihong Qi+ and Ying-Fang Wang* and Yulei He
• Companies: University of California, Davis and University of California, Davis and Harvard Medical School
• Keywords: censoring ; double robustness ; failure time ; multiple imputation ; weighted estimators

Abstract

Missing-covariate problems are common in medical studies with survival outcomes and impose difficult challenges to analysts. In this talk, I will present two approaches, the inverse probability weighted estimation and multiple imputation, and compare their performance using a simulation study. Our results show that when censoring time depends on missing covariates, weighted estimators are still consistent while MI results in larger bias. The bias from the imputation estimates increases when the correlation between the missing and observed covariates increases or when the selection probability decreases. Furthermore, the imputation estimates are inconsistent if the imputation model is misspecified, while the FAWE has the nice property of double robustness. The advantage of the latter approach is especially noteworthy when the missing and observed covariates are highly correlated.