Abstract:
|
In recent pharmaceutical drug development, adaptive clinical trials become more and more appealing due to ethical considerations, and the ability to accommodate uncertainty while conducting the trial. Several methods have been proposed to optimize a certain study design within a class of candidates, but finding an optimal hypothesis testing strategy for a given design remains open and challenging, mainly due to the complex likelihood function. We propose a novel application of the Deep Neural Networks (DNN) to construct the test statistics as well as the critical value with controlled type I error rates and enhanced power. Simulation studies are performed to demonstrate that our proposed method essentially establishes the underlying known Uniformly Most Powerful (UMP) unbiased test in several scenarios, and provides an alternative solution to the Behrens-Fisher problem with non-inferior performance compared with the popular Welch approximate t-test. As a case study, we apply the proposed method to a sample size reassessment confirmatory adaptive study MUSEC (MUltiple Sclerosis and Extract of Cannabis), demonstrating the proposed method outperforms the existing alternatives.
|