Tests and Sample Size Determination for Noninferiority Trial of Two Survival Functions Using a Proportional Odds Survival Model

• Activity Number: 171
• Type: Contributed
• Date/Time: Monday, August 4, 2014 : 10:30 AM to 12:20 PM
• Sponsor: Biometrics Section
• Author(s): Elvis Martinez*+ and Debajyoti Sinha and Wenting Wang and Stuart R. Lipsitz and Richard J. Chappell and Juliana Cobre
• Companies: Florida State University and Florida State University and MD Anderson Cancer Center and Harvard Medical School and University of Wisconsin-Madison and University of São Paulo
• Keywords: Clinical importance ; Non-inferiority ; Critical region ; Proportional Odds Survival Model ; Sample Size Determination

Abstract

In bio-pharmaceutical industry, there is substantial interest in developing a new treatment that may have other crucial advantages such as easier administration, lower cost, and better tolerability than an existing treatment of proven efficacy. Although these advantages of a new treatment can be self-evident (e.g., oral versus intravenous administration), an essential step for approving new treatment is a non-inferiority trial to prove that the new treatment is not less efficacious than the existing treatment in terms of the main survival outcome. We propose a non-inferiority test for two survival curves using an alternative to Cox's Proportional Hazards (PH) model based test (e.g., Wellek, 1993). We introduce a practical formulation of the non-inferiority hypothesis as well as the corresponding test based on Bennett's (1981) Proportional Odds Survival (POS) model. Unlike the commonly used procedure of Wellek (1993), our test achieves a type I error rate lower than the desired nominal rate when two treatment arms satisfy either the POS model or Cox's (1972) PH model. We present a practical method for determining sample sizes to guarantee a desired power between two survival curves.