Page 135 - Contributed Paper Session (CPS) - Volume 1

CPS1201 M. Iftakhar Alam et al.

Combined criteria for dose optimisation in early
phase clinical trials
1
2
2
M. Iftakhar Alam , D. Stephen Coad , Barbara Bogacka
1 Institute of Statistical Research and Training, University of Dhaka, Dhaka 1000, Bangladesh
2 School of Mathematical Sciences, Queen Mary University of London, London E1 4NS, U.K.

Abstract
The paper aims to investigate whether any bridge is possible between so-
called best intention and D-optimum designs. It introduces combined criteria
for dose optimisation in seamless phase I/II adaptive clinical trials. Each of the
optimality criteria considers efficacy and toxicity as endpoints and is based on
the probability of a successful outcome and on the determinant of the Fisher
information matrix for estimation of the doseresponse parameters. In addition,
one of the criteria incorporates penalties for choosing a toxic or inefficacious
dose. Starting with the lowest dose, the adaptive design selects the dose for
each subsequent cohort that maximises the respective defined criterion. The
methodology is illustrated with a dose-response model that assumes trinomial
responses. Simulation studies show that the method is capable of identifying
the optimal dose accurately without exposing many patients to toxic doses.

Keywords
Adaptive design; Continuation ratio model; D-optimum design; Penalty
function; Phase I/II trial.

1. Introduction
Different methods are being developed to increase the popularity of
seamless phase I/II clinical trials. There are designs proposed by Thall and
Russell (1998), Thall and Cook (2004) and Zhang et al. (2006). These designs
have the intention of allocating the best dose to the cohort of patients based
on the current knowledge and are known as the “best intention designs”. They
may lead to poor learning of the dose-response relationship. In contrast, there
are methods which rely on optimal design criteria for estimation of model
parameters. Heise and Myers (1996) constructed the D-optimal design using
the Gumbel model for bivariate binary data. Fan and Chaloner (2004)
described the D-optimal design for trinomial responses using a continuation
ratio model. Dragalin and Fedorov (2006) considered binary outcomes for
each endpoint and used either Gumbel bivariate binary logistic regression or
the Cox bivariate binary model.
This paper investigates whether any trade-off between the best intention
designs taking care of individual ethics and D-optimal designs focusing more
on collective ethics is possible. The underlying idea is to develop a design that

124 | I S I W S C 2 0 1 9