Statistical justification of expansion cohorts in phase 1 cancer trials

Abstract

Background: Phase I cancer trials increasingly incorporate dose-expansion cohorts (DECs), reflecting a growing demand to acquire more information about investigational drugs. Protocols commonly fail to provide a sample-size justification or analysis plan for the DEC. In this study, we develop a statistical framework for the design of DECs.

Methods: We assume the maximum tolerated dose (MTD) for the investigational drug has been identified in the dose-escalation stage of the trial. We use the 80% lower confidence bound and the 90% upper confidence bound for the response and toxicity rates, respectively, as decision thresholds for the dose-expansion stage. We calculate the operating characteristics with reference to prespecified minimum effective response rates and maximum safe DLT rates.

Results: We apply our framework to specify a system of DEC plans. The design comprises three components: 1) the number of subjects enrolled at the MTD, 2) the minimum number of responses necessary to indicate provisional drug efficacy, and 3) the maximum number of dose-limiting toxicities (DLTs) permitted to indicate drug safety. We demonstrate our method in an application to a cancer immunotherapy trial.

Conclusions: Our simple and practical tool enables creation of DEC designs that appropriately address the safety and efficacy objectives of the trial.

Keywords: design; dose expansion; dose-finding; efficacy; safety.

© 2018 American Cancer Society.

Figures

Figure 1

80% lower confidence bounds (LCBs, green) for response and 90% upper confidence bounds (UCB, blue) for toxicity for sample sizes A) 15, B) 20, C) 30, and D) 40 subjects; DEC sizes are A) 9, B) 14, C) 24, and D) 34 subjects, respectively. DLT= dose-limiting toxicity.

Figure 1

80% lower confidence bounds (LCBs, green) for response and 90% upper confidence bounds (UCB, blue) for toxicity for sample sizes A) 15, B) 20, C) 30, and D) 40 subjects; DEC sizes are A) 9, B) 14, C) 24, and D) 34 subjects, respectively. DLT= dose-limiting toxicity.

Figure 1

80% lower confidence bounds (LCBs, green) for response and 90% upper confidence bounds (UCB, blue) for toxicity for sample sizes A) 15, B) 20, C) 30, and D) 40 subjects; DEC sizes are A) 9, B) 14, C) 24, and D) 34 subjects, respectively. DLT= dose-limiting toxicity.

Figure 1

80% lower confidence bounds (LCBs, green) for response and 90% upper confidence bounds (UCB, blue) for toxicity for sample sizes A) 15, B) 20, C) 30, and D) 40 subjects; DEC sizes are A) 9, B) 14, C) 24, and D) 34 subjects, respectively. DLT= dose-limiting toxicity.