Abstract:
|
In clinical research and development, interim Go/No-Go decision making is critical for minimizing risk of exposing patients to possible ineffective therapies and making better decisions. For such decision making, statistical methods based on predictive probabilities have been actively advocated. Although those methods have been well studied for univariate data, in many clinical trial settings where more complicated data structure exist, computational burdens solutions often limit their feasibility.
For longitudinal data, we derive the closed-form solutions for predictive probabilities (e.g., Bayesian predictive probability, Bayesian-frequentist hybrid approach for predictive probability, and the conditional power), which takes into account all available longitudinal information. We show that our methods utilizing all information is more informative than using completers only in longitudinal data case. In addition, based on the obtained closed-form solutions, we study their distributions and provide guidelines on the selection of cut-off values for those tools. We demonstrate the benefit of our methods by numerical studies.
|