Cours intensif par le Professeur Alessandra Giovagnoli, Université de Bologne les 6, 7 et 8 mai 2009. Le thème sera sur « Design for Clinical Trials ».
Le cours sera donné à l'UCL, Institut de statistique, voie du Roman Pays, 20 à Louvain-la-Neuve, selon l'horaire suivant :
Registration is free of charge but mandatory (before May 1st) to edt-stat-actu@uclouvain.be
Note: You are also welcome on May 8 afternoon (14h30 and 16h00) to the two presentations of the seminar series of the Institute of Statistics.
Local organiser : Prof. C Legrand (catherine.legrand@uclouvain.be)
Course summary:
Ethics and statistics in clinical trials: an introduction
Clinical trials are very complex experiments with patients as subjects; this poses a critical
dilemma for investigators, known as the individual-versus-collective ethics, since it is
necessary to minimize potential harm to the patients presently under care and maximize
the experimental information. Some fundamental principles of the statistical methodology
employed in planning and analysing experiments of this type are laid out, together with a
sketch of the historical development of the subject, up to simulated clinical trials in the
present time.
Clinical trials for treatment comparison - The ABCD design, sequential experiments,
and Adaptive Designs
Randomized clinical trials, which consist in assigning treatments to subjects randomly, are
widely regarded as the most scientifically sound approach to determining which of two or
more medical treatments is better. However, for the past 15 years the attention has turned
to studying a statistical design of the trial aimed at minimizing the number of subjects
exposed to inferior treatments, increasing the patient’s chances of receiving the best one.
Needless to say, a randomization component in the assignments is always required in
order to mitigate several types of bias. Covariates too come into consideration, which are
usually random but may be known before assigning the treatment.
When the target allocation for comparing 2 treatments is balance, a restricted form of
randomization is used (Efron’s Biased Coin Design). An extension, the so-called ABCD,
which is more flexible, will be presented. However, usually the target allocation derived
from adopting a given criterion depends on the parameters of the statistical model, and as
such is unknown; this problem is known in the statistical literature as “local optimality”. A
possible solution consists in sequential experimentation, so that assignments can be
redressed towards the unknown target. Some aspects of these “adaptive” designs will be
illustrated.
Dose Escalation studies – The up&down design
Traditionally, dose escalation studies are binary response trials where the probability of
positive response (toxicity) is assumed to be an increasing function of the given dose level
and the aim consists in estimating the target dose at which a pre-specified probability of
positive response is associated without any parametric assumptions on the response
curve. Dose escalation studies are aimed at identifying a quantile of interest in Phase I
clinical trials. Classical examples are the median dose, usually denoted by LD50, or the
maximum tolerated dose of phase I clinical trials. Assuming that the set of available
ordered doses is fixed in advance and doses are allocated to either groups or single
patients, the “up-and-down” procedures for binary or continuous responses (U&D)
provide a possible solution for dose-finding problems