MULTINOMIAL PHASE II CANCER TRIALS INCORPORATING RESPONSE AND EARLY PROGRESSION

• Published in: Journal of biopharmaceutical statistics Vol. 9; no. 2; pp. 351 - 363
• Main Authors: Zee, Benny, Melnychuk, David, Dancey, Janet, Eisenhauer, Elizabeth
• Format: Journal Article
• Language: English
• Published: England Taylor & Francis Group 24.05.1999
• Subjects: Clinical Trials, Phase II as Topic - methods; Clinical Trials, Phase II as Topic - statistics & numerical data; Computer Simulation; Disease Progression; Humans; Mathematical Computing; Missing data; Models, Statistical; Multiple-stage stopping rule; Multivariate Analysis; Multivariate endpoint; Neoplasms - drug therapy
• ISSN: 1054-3406; 1520-5711
• DOI: 10.1081/BIP-100101181

The objective of a phase II clinical trial in oncology is to assess the antitumor activity of a specific treatment regimen. A multiple-testing procedure is commonly used to decide whether the experimental treatment warrants further investigation based on patients' tumor response. There are ethical concerns about exposing patients to a new drug when the response rate is low and a relatively large number of patients have early progressive disease. Ensign et al. [1] proposed a stopping rule that rejects a drug early when there is a long run of early treatment failures. However, this approach may not be sensitive to pick up early progressors mixed with other nonresponders. In this paper, we present a multiple-stage stopping rule for a single-arm trial of an experimental treatment in which both tumor response and early progression are considered simultaneously. We use a multinomial model to accommodate an outcome of discrete multivariate responses in order to improve the efficiency of the stopping rule. The proposed multiple-testing procedure requires that both the numbers of responses and early progressions fall within the boundaries satisfying the stopping criteria in order to stop the study. Simulation is performed to validate these results and to compare them with other commonly used designs. Other applications of this method are also discussed.