Sankhya: The Indian Journal of Statistics

2001, Volume 63, Series B, Pt. 1, pp. 27--42

ROBUST ADAPTIVE DESIGNS IN CLINICAL TRAILS FOR CONTINUOUS RESPONSES

By

ATANU BISWAS

and

AYANENDRANATH BASU *Indian Statistical Institute,
Kolkata, India.*

SUMMARY. We consider the clinical trials scenario where patients enter the trial sequentially, and the experimenter has to adaptively select the better of the two competing treatments for future applications. This is particularly important since the subjects are human patients. However in almost all such techniques currently available, the responses are assumed to be dichotomous (success or failure). If the responses are continuous, one may consider dichotomizing them by setting an appropriate threshold value. But this may not have any real basis except its mathematical tractability. In this paper we introduce an adaptive design which uses all the information from the previous continuous responses for the next allocation without dichotomizing the responses. The method may be suitably modified to handle problems of lack of robustness (which may occur when the actual continuous measurements are used), as well as some other possible contingencies arising in such situations, such as delayed response.

*AMS (1991) subject classification.* Primary 62L05; secondary 62F35.

*Key words and phrases. *Adaptive designs; clinical trials;
delayed response; robustness.