A Simulation Study Approach to Assess the Accuracy of the Weibull Shape Parameter Based on Historical Studies for Time-to-Event Outcomes

• Activity Number: 53 - New Developments in Survival Analysis
• Type: Contributed
• Date/Time: Sunday, August 8, 2021 : 3:30 PM to 5:20 PM
• Sponsor: Biometrics Section
• Author(s): Palash Sharma* and Prabhakar Chalise and Nadeesha Thewarapperuma and Milind A. Phadnis
• Companies: University of Kansas Medical Center and University of Kansas Medical Center and University of Kansas Medical Center and University of Kansas Medical Center
• Keywords: Survival; KM Curve; Weibull Distribution

Abstract

In case of time-to-event endpoint, recent advances in literature propose using the Weibull distribution due to its flexibility in modeling the hazard function. Phadnis (2019) proposed an exact parametric test statistics based sample size calculation method and compared it with optimal log rank test which also required an assumption of the Weibull distribution. However, these methods requires an estimate of the Weibull shape parameter from the prior study or historical data. The aim of this paper is to assess how accurate is the estimated shape parameter when it is obtained from the published results of the median survival time and/or corresponding interquartile range. Furthermore, we also assess for a given sample size of the historical study and censoring rate, how many survival quantiles we need from the published results. Our simulation results show that the estimate of the shape parameter is quite accurate when the prior study is large enough (>= 50) with the censoring rate of less than 20%. If the prior study has a smaller sample size or high censoring rate, we need more information form the KM curves to improve the reliability and accuracy of the estimated shape parameter.