Methods for early phase clinical development
Brady, William E.
MetadataShow full item record
We examine three issues in the design and analysis of clinical trials and early drug development. First, we adapt the commonly used single-arm, two-stage, phase II oncology clinical trial, in which the primary outcome is often tumor response. In these designs, if sufficient responses are observed at the end of the first stage, a second stage of accrual is started. At the end of the second stage, the agent is deemed active if sufficient numbers of responses are observed in total. We incorporate the use of increasing disease into an 'early-go' decision to allow the trial to go to the second stage if sufficient responses are observed or if a sufficiently small number of patients with increasing tumor burden are observed. This can potentially shorten the duration of the trial by shortening or eliminating the delay between the first and second stages of accrual. Exact calculations are shown for the operating characteristics of the trial, which are compared to the Simon (1989) design, and the expected trial durations for the designs are compared. Second, in the setting of phase I and II trials and other small-sample situations, we propose exact versions of two tests for monotonic dose-response for binary data (Dosemeci and Benichou (1998) (DB) and Peddada (2001)), which make the tests more appropriate for the case of small samples, e.g., in preclinical studies and early phase clinical trials. The exact versions are based on the E+M method of Lloyd (2008). Exact calculations are used to compare the Type I error and power of the exact versions of the tests versus non-exact versions, and to compare the DB and Peddada tests with an exact version of the Cochran-Armitage test due to Shan et al (2011). Lastly, we propose two new tests for binomial data when we expect an umbrella pattern in response across dose levels: a contrast-based test and a quasi-likelihood ratio test, both of which are exact. We also examine an exact version of the Peddada test for umbrella orders.