A Bayesian dose-finding trial with adaptive dose expansion to flexibly assess efficacy and safety of an investigational drug.

BACKGROUND: Adaptive dose-ranging trials are more efficient than traditional approaches and may be designed to explicitly address the goals and decisions inherent in learn-phase drug development. We report the design, implementation, and outcome of an innovative Bayesian, response-adaptive, dose-ranging trial of an investigational drug in patients with diabetes, incorporating a dose expansion approach to flexibly address both efficacy and safety.

PURPOSE: The design was developed to assess whether one or more doses of an investigational drug demonstrated superior efficacy to an active control while maintaining an acceptable safety profile.

METHODS: The trial used a two-stage design, in which patients were initially allocated equally to placebo, investigational drug at a low and a medium dose, and an active control. Movement to the second stage was contingent upon evidence of efficacy (measured by change in fasting blood glucose) to add a very low dose of the investigational drug and of safety (measured by weight gain) to add a high dose of the investigational drug. The design incorporated a longitudinal model to maximize use of incomplete data, predictive probabilities to guide the decisions to terminate the trial for futility or move on to Stage 2, and a dose-response model in Stage 2 to borrow information across adjacent doses. Extensive simulations were used to fine tune trial parameters, to define operating characteristics, and to determine the required sample sizes. A data monitoring committee was provided with frequent reports to aid in trial oversight.

RESULTS: In Stage 1, as trial data accrued, the predictive probability that either the low or medium dose of the investigational drug was superior to the active control fell to low values. Stage 1 termination was recommended after 199 subjects were randomized, out of a maximum trial size of 500 subjects, and the final sample size was 218. Thus the trial did not progress to Stage 2.

LIMITATIONS: Because of the relatively narrow dose range to be assessed, and the inability to utilize the highest dose at the beginning of the trial, a fully responsive-adaptive design incorporating dose-response modeling was not considered a viable option. This limited the efficiency gains possible with a full set of adaptive design elements.

CONCLUSIONS: The two-stage dose-expansion design functioned as designed, recommending early termination based on a low probability that the tested doses had efficacy greater than the active control.