A method for therapeutic dose selection in a phase II clinical trial using contrast statistics

This paper proposes a statistical method for determining the therapeutic dose of a test drug in a confirmatory clinical trial based on a phase II clinical trial using 3 or 4 doses of the drug. This method assumes the primary variable has a normal distribution with a common variance, that a test-drug effect is seen when the population means show a response pattern indicating a monotonic increase with dose, and that there is a prior distribution for the population means. Under the proposed method, multiple contrast statistics are determined, such as contrast statistics for linear increase and plateau, and a response pattern is selected based on the maximum contrast statistic. The posterior probability that the selected response pattern is the true one is evaluated, and if this exceeds the cut-off value a therapeutic dose is selected based on the estimated response pattern. To select the appropriate cut-off value, a simulation study was conducted using a loss function for which the loss due to overestimation is greater than the loss due to underestimation. It was found that, as a rule, the appropriate cut-off value to reduce the expected loss for various response patterns is 0.75 for a 3-group trial and 0.70 for a 4-group trial. Using these cut-off values, the proposed method was applied to a previous clinical trial of a leukotriene receptor antagonist in patients with bronchial asthma. The method enabled the selection of what are considered appropriate response patterns and a therapeutic dose. Thus, the proposed method appears reasonable.