949 - Parametric effect size estimates from sipuleucel-T randomized trials

Date 28 September 2012
Event ESMO Congress 2012
Session Publication Only
Topics Prostate Cancer
Cancer Immunology and Immunotherapy
Presenter Brent Blumenstein
Authors B.A. Blumenstein
  • Trial Architecture Consulting, 20008-3462 - Washington, DC/US

Abstract

Introduction

Median overall survival (OS) difference is frequently used as a measure of effect size because it is easily understood. Other effect size estimates can more fully represent the OS distribution, and are less subject to local variations. Analysis of the sipuleucel-T D9901 and IMPACT trials using non-parametric (median) or semi-parametric (hazard ratio [HR]) methods provide 4.5- and 4.1-month estimated median OS differences, respectively, with HRs of 0.586 and 0.752 (not stratified or adjusted). The OS curves from these trials exhibit delayed separation, suggesting a delayed treatment effect consistent with that shown for other immunotherapies. In this research, parametric statistical models are used to account for delayed effect and obtain alternative effect size estimates from D9901 and IMPACT.

Methods

Parametric models based on the Weibull distribution and a modification of the Weibull distribution that allows for delay of effect were applied. These models require specification of the general shape of the hazard function.

Results

The table shows the original OS and HR estimates in comparison to those obtained through parametric modeling. D9901 estimates from the Weibull model suggested a larger median OS difference vs the original estimates. The OS difference and the HR estimate using the delayed effect Weibull model were both indicative of greater sipuleucel-T effect. For IMPACT, the OS difference and HR estimates from the Weibull model indicated greater effect, but to a lesser degree than in D9901. The delayed effect modified Weibull model for IMPACT did not support a delayed effect.

Discussion

This statistical modeling illustrates alternative methods of obtaining effect size estimates that may be particularly applicable to clinical trials of cancer immunotherapies. For the sipuleucel-T trials, these models suggest a greater sipuleucel-T effect than obtained from non-parametric or semi-parametric methods.

Study Estimate Type Median Difference (mos) Hazard Ratio
D9901 Non-parametric/semi-parametric Weibull Modified Weibull 4.5 10.3 6.6 0.586 0.585 0.486*
IMPACT Non-parametric/semi-parametric Weibull Modified Weibull 4.1 4.7 NS 0.752 0.749 NS

NS, delayed effect not supported *Applicable after delayed efffect

Disclosure

All authors have declared no conflicts of interest.