American Statistical Association
New York City
Metropolitan Area Chapter

Memorial Sloan-Kettering Cancer Center
Biostatistics Seminar


Elizabeth Garrett-Mayer, Ph.D.
Department of Biostatistics, Bioinformatics and Epidemiology
Medical University of South Carolina

LIKELIHOOD-BASED DESIGNS FOR EARLY STOPPING
IN PHASE II CANCER CLINICAL TRIALS

In single arm phase II trials (in cancer research in particular) where clinical response is the outcome, Simonís two-stage design has become most commonly used due to its simplicity, its ability to allow early stopping for futility, and preservation of type I and type II errors. When time-to-event outcomes, such as progression free survival, are of interest in single arm phase II trials, designs incorporating early stopping rules have proven to be challenging. Bayesian adaptive designs (BAD) have, however, been proposed in these settings to allow additional possibilities for early stopping when the evidence becomes strong enough to declare the alternative hypothesis unlikely. Critics of the BAD find fault with choices of priors and the sometimes complex posterior distributions that these analyses require. As an alternative, likelihood-based approaches which allow frequent looks at the data and are similar to the BAD whereby there is not the accumulation of type I errors can be used. The likelihood ratio can be used to evaluate evidence for two competing hypotheses, and the universal bound (Birnbaum, 1962) provides a limit for the probability of misleading evidence in favor of the incorrect hypothesis. In the case of clinical response as the outcome, the likelihood ratio for the binomial likelihood has a very simple form and can be easily computed. For time to event outcomes, the choice or parametric (or semi-parametric model) should be made with care. Commonly used trial scenarios are simulated to demonstrate the increased efficiency of likelihood-based designs, in terms of frequency of early stopping under the null hypothesis and smaller sample sizes, as compared to similar Simon two-stage designs. Simulations for time-to-event outcomes are also considered and include considerations of model choice.


Date: Wednesday, May 14, 2008
Time: 4:00 P.M. - 5:00 P.M.
Location: Memorial Sloan-Kettering Cancer Center
Department of Epidemiology and Biostatistics
307 East 63rd Street
(between First and Second Avenues)
3rd Floor Conference Room
New York, New York
Note: To gain access to the building, please follow the directions by the telephone in the foyer.

RESERVATIONS ARE NOT REQUIRED


Home Page | Chapter News | Chapter Officers | Chapter Events
Other Metro Area Events | ASA National Home Page | Links To Other Websites
NYC ASA Chapter Constitution | NYC ASA Chapter By-Laws

Copyright © 2008 by New York City Metropolitan Area Chapter of the ASA
Designed and maintained by Cynthia Scherer
Send questions or comments to nycasa@mindspring.com