An optimal two-stage phase II design utilizing complete and partial response information separately.

Abstract

Phase II clinical trials in oncology are performed to evaluate the therapeutic efficacy of a new treatment regimen. A common measure of efficacy for these trials is the proportion of patients who obtain a response measured by tumor shrinkage. It is standard practice to classify this response into the following categories: (1) complete response (CR); (2) partial response (PR); (3) stable disease; and (4) progression of disease. Tumor response is then treated as a binary variable whereby patients who achieve either a CR or a PR are considered responders and all others nonresponders. A two-stage design that allows for early termination of the trial if the treatment shows little efficacy such as Gehan or Simon gives equal weight to a CR and a PR. However, a CR, defined as complete disappearance of the tumor, is more likely than a PR to signal an important antitumor effect and result in a survival advantage. We argue that CRs and PRs should be considered separately, and hence we propose a two-stage design with a multilevel endpoint (i.e., CR, PR, and nonresponders). This design is an extension of Simon's optimal two-stage design and is based on a trinomial model. For most scenarios the proposed design results in an improvement in expected sample size compared to Simon's optimal design. Design optimization was performed by a direct search based on enumerating exact trinomial probabilities. Sample size tables are provided for parameter sets commonly used in the oncologic setting. Software is available by contacting the authors.