Information-based Group Sequential Tests With Lagged or Censored Data
No Thumbnail Available
Files
Date
2000-06-23
Authors
Journal Title
Series/Report No.
Journal ISSN
Volume Title
Publisher
Abstract
Conventionally, values of nuisance parameters given in a statistical design are often erroneous, thus may result in overpowering or underpowering a test using traditional sample size calculations. In this thesis, we propose to use Fisher Information data monitoring in group sequential studies to not only allow an early stopping in a clinical trial but also maintain the desired power of the test for all values of nuisance parameters. Simulation studies for the simple case of comparing two response rates are used to demonstrate that a test of a single parameter of interest with a specified alternative achieves the desired power in information-based monitoring regardless of the value of the nuisance parameters, provided that this parameter of interest can be estimated efficiently. The emphasis in this part is to show how information-based monitoring can be implemented in practice and to demonstrate the accuracy of the corresponding operating characteristics in some simulation studies.When there is lag time in reporting, standard statistical techniques often lead to biased inferences on interim data. A maximum lag estimator ensures complete information by using data before a lag time period. The estimator is unbiased but less powerful. We propose an inverse probability weighted estimator which accounts for censoring and is consistent and asymptotically normal in estimating mean of dichotomous variables. The joint distribution of test statistics at different times have the covariance structure of a sequential process with independent increments. This allows the use of information-based monitoring. Simulation study shows that our estimator preserves the type I and type II errors, and reduces the number of participants required in a trial. Future approach in finding an efficient estimator is also suggested in chapter 3.
Description
Keywords
Citation
Degree
PhD
Discipline
Statistics
