←Back

611 – Missing Data Imputations

An Extended Kaplan-Meier Estimator for Time-to-Event Data with Informative Censoring

at Astellas Pharma, Inc.

In many clinical trials, an important endpoint is rate of treatment success which can be analyzed by crude rate or time-to-event approaches. In trials where patients may discontinue treatment or die due to poor response to treatment, the use of standard survival methods such as the Kaplan-Meier estimator and log-rank test may yield severely biased results. This is because the independence of censoring (or noninformative censoring) assumption is often violated. The analysis of time-to-success must therefore account for informative censoring. We propose a two-step method that is capable of handling multiple censoring mechanisms including informative censoring in a trial. Step one is to utilize observed data to assign a unique likelihood index to each censored patient that quantitatively measures the censored patient's likelihood of achieving success relative to the remaining patients. In Step two, we propose an extended Kaplan-Meier (EKM) estimator that adjusts number of patients "at risk" based on the likelihood index of patient censored at each time point. Unlike standard KM estimator, the EKM is flexible in accounting for various censoring mechanisms and does not ignore other observed data of censored patients. We illustrate the application of EKM using a case study. While developed for time-to-success, the EKM is applicable to general time-to-event endpoints.