Point Estimator In Seamless Phase Ii/iii Clinical Trials For Binary Outcome

Backgrounds &ObjectiveIn this paper, we focus on the point estimation in seamless phase II/III clinical trials for binary outcomes. This study aims to simulate and verify the statistical properties of maximum likelihood estimator, stage 2 estimator, stepwise correction method, uniformly minimum variance unbiased estimation, Lindley’s estimation by Monte Carlo methods. We suggest appropriate point estimation methods according to simulation results. MethodsAssuming several dose treatment groups and a parallel control group in stage1, we consider point estimation of the treatment effect in which stage 1 is used to select the most effective dose group and the selected group continues to stage2 with control group. We proposed the estimators for the treatment difference in two scenarios. Scenario 1: The trial don’t allow early terminationWe propose two methods for this scenario. One is uniformly minimum variance unbiased estimation for binary outcome. Another method is Lindley’s method, based on estimator by converting the binomial outcome into a nearly normal variable using arcsine transformation. Scenario 2: The trial allow early termination for futilityIn this scenario, we proposed an uniformly minimum variance unbiased estimation for binary outcome.We have compared the properties of 5 methods by means of simulations study at different scenarios and parameter sets using SAS 9.2 software.Bias: The simulation study shown that the maximum likelihood estimator(MLE) is always overestimated and very sensitive to some parameters. The stage 2 estimator, as expected, provides unbiased estimates. While Lindley’s estimation and SOC are under-estimated and affected by t and difference between treatments groups. The UMVUE is almost unbiased in most situations, only increasing with b.MSE: In terms of MSE, all the methods perform well except stage 2 estimator. MLE is very sensitive to some parameters. The stage 2 estimator’s MSE is extremely high and sensitive to t. The MSE of Lindley’s estimation and SOC are almost the same and minimum from these methods. The UMVUE is relatively robust in all situation. ConclusionThe UMVUE performs well in terms of bias and MSE. Therefore, we recommend it as the point estimator of the treatment effect in seamless phase II/III clinical trials for binary outcome.