Advisory Committee Chair
Leslie A McClure
Advisory Committee Members
Gary Cutter
Stephen Glasser
David T Redden
Jenifer Voeks
Document Type
Dissertation
Date of Award
2009
Degree Name by School
Doctor of Philosophy (PhD) School of Public Health
Abstract
Three reasons to review accumulating data in clinical trials include: ethical issues, financial concerns and administrative concerns. Interim analysis is a good way to monitor accumulating data in clinical trials. Interim analysis allows for the possibility that a study may be terminated early; that is, if the currently observed data convincingly favor the null or the alternative hypothesis then the study ends early. In addition, many clinical trials are conducted to compare a treatment group to a standard group on multiple endpoints. Combining interim analyses with multiple endpoints allows for more information to be provided from the trial than either testing a single endpoint alone or testing multiple endpoints without interim looks. Stochastic curtailment procedures are frequently used in interim monitoring of clinical trials. These methods allow for the possibility of early termination of a study at any interim time point if a significant result (or null result) is highly likely given the current data. One such stochastic curtailment method is conditional power. Conditional power gives the probability of rejecting the null hypothesis at the end of the study given the current data and some fixed parameter of interest. One criticism of conditional power is that it is computed under values of the parameter of interest which may not be supported by the current data. The predictive power method, a mixed Bayesian-frequentist method, is another stochastic curtailment procedure for which this is no longer a concern. The predictive power method extends conditional power to assume that the parameter of interest is random. The conditional power function is averaged with respect to the posterior distribution of the parameter of interest given the observed data. We present an extension of the univariate predictive power method to allow for two correlated endpoints. We assess the properties of the extended predictive power under a variety of conditions, and compare the performance of the extended predictive power method to other commonly used methods for assessing data at interim time points. Finally, we illustrate our extended predictive power method by applying it to data from the Inhaled Nitric Oxide Study.
Recommended Citation
Hamilton, Kiya R., "Design of the Predictive Power Method With Two Endpoints" (2009). All ETDs from UAB. 1851.
https://digitalcommons.library.uab.edu/etd-collection/1851