Comparison of Bayesian interim analysis and classical interim analysis in group sequential design.
- Author:
Lingling YUAN
1
;
Zhiying ZHAN
;
Xuhui TAN
Author Information
- Publication Type:Journal Article
- MeSH: Bayes Theorem; Data Interpretation, Statistical; Sample Size
- From: Journal of Southern Medical University 2015;35(11):1638-1642
- CountryChina
- Language:Chinese
-
Abstract:
OBJECTIVETo explore the differences between the Bayesian interim analysis and the classical interim analysis.
METHODSTo compare the means of two independent samples between control and treatment, superior hypothesis test was established. In line with the data requirements for group sequential design, Type Iota error of Bayesian interim analysis based on various prior distributions, Power, Average Sample Size and Average Stage were estimated in the interim analysis.
RESULTSIn the Pocock and O' Brien & Fleming designs, the Type Iota errors in the Bayesian interim analysis based on the skeptical prior distribution and the handicap prior distribution were controlled at around 0.05. When the powers of these two classical designs were both 80%, Bayesian powers of the skeptical prior distribution and the handicap prior distribution were markedly lower. The powers of the non-informative prior distribution and the enthusiastic prior distribution were distinctly higher than 80%.
CONCLUSIONIn the Bayesian interim analysis based on the skeptical prior distribution and the handicap Prior distribution, the Type Iota errors can be well controlled. Bayesian interim analyses using these two prior distributions, compared with the analysis adopting the O' Brien & Fleming method, can markedly increase the possibility of ending the clinical trials ahead of time. The Bayesian interim analyses based on these two distributions do not have practical value for group sequential design of the Pocock method.