Evaluation of Time-to-Event Data in the Presence of Informative Censorship

• Activity Number: 359
• Type: Contributed
• Date/Time: Tuesday, August 2, 2011 : 10:30 AM to 12:20 PM
• Sponsor: Biometrics Section
• Author(s): Susan Y. Zhou*+ and Guoxing Soon
• Companies: U.S. Food and Drug Administration/CDER and U.S. Food and Drug Administration/CDER
• Keywords: Kaplan Meier ; Hodge Lehmann Estimator ; informative censorship ; imputation ; simulation

Abstract

When analyzing time-to-event data for the estimation of treatment effect, we noticed significant discrepancies using different approaches such as Kaplan Meier (KM) and Hodge Lehmann Estimator (HLE). In the clinical trial setting it may not be as evident that the censoring is non-informative. Hence the use of the KM may not be suitable. To resolve this issue, several imputation scenarios including worst-case scenario were conducted to complete the missing data and the HLEs were obtained and compared. A simulation study was carried out to further investigate the effects of different patterns of informative censorship associated with the misuse of the KM. Several imputation approaches were performed on missing data prior to the estimation of treatment differences. In the present work, the median of all paired treatment difference placebo vs. treatment was obtained for the HLE, and the difference in median time between placebo arm and treatment arm was used for the KM approach. We will summarize the performances of different statistical approaches for selected missing patterns and discuss alternative analysis approaches for time-to-event data with informative censoring.