1999 Jun 25 (Fri), 10:30 AM
陳 達 教授
University of Maryland School of Medicine U.S.A
The slippage test, considered by Mosteller, Paulson, Karlin and Truax,
only deals with the situation where one treatment is better than all
others. The usual selection procedures, developed by Bechhofer and Gupta,
only select the best treatment or a random subset containing the best
treatment. These methods are not appropriate for phase III clinical
trials where more precision is needed.
We pool all the hypotheses considered in the above two approaches into a
family of hypotheses. The statistical procedures are developed to
differentiate between many hypotheses in this big family. This solution
is more precise and suitable for application to clinical trials. The
interpretation of our new procedure is straight-forward and more relevant
than the multiple-comparison procedures of Duncan, Newman, and Keuls.
In comparison of clinical interventions, in addition to the efficacy main
endpoint, consideration should also be given to toxicity and cost.
Consideration of these other factors leads to prior preference ranking of
treatments, and results in one-sided, two-sided, or equivalence design in
We extend our procedures to cover the situation when the prior
preferences are not equal in multiple-treatment trials. Our procedures
include Dunnett's many-one test as a sub-procedure.
These statistical procedures are developed under the assumption of
continuous normal data. They can be used for clinical trials with binary
response and time-to-event endpoints.