This talk is about the analysis and the issue of bias in the estimate of
treatment effects in multi-arm multi-stage randomised clinical trials designs.
I'm Babak Oskooei, I'm a statistician
at the MRC Clinical Trials Unit at UCL in London.
The outline of my talk is, first,
I'm going to give you some introduction to clinical trials in general,
and then I'm going to move on to just give a brief overview of
multi-arm multi-stage randomised trials which is one type of adaptive designs.
I'm going to go through the basic principles and give you some details,
and then I'll move on to the issue of estimation of treatment effect.
I'm going to go through that relevant issues at the definition
of bias in a two-arm multi-stage setting,
that is TAMS setting.
Then for this presentation,
I use real trial data to show how the stopping rules that we have
in the multi-arm multi-stage setting which is
a generalised version of group sequential designs,
might affect their estimates of treatment effect in those trials.
I'm going to re-design and re-analyse those four trials, mainly cancer trials.
Then I'm going to explore the proposed methods to reduce
the potential bias we might have because of these stopping rules in our design,
and then finally, conclusions.
Usually, treatments or treatment regimens
go through different phases in the course of their development.
In this talk, I'm focusing on mainly Phase II/Phase III setting, where
several such treatments or treatment combinations or
those regimens might be available for testing in those late-phase trials.
We know that in those settings,
especially in cancer trials,
those trials can be quite large and lengthy trials.
The aim is to minimise patient exposure to toxic or ineffective treatments,
and also reduce the cost.
They can cost sometimes millions of dollars or pounds,
and they can take up to 5-15 years depending on the disease area to do.
Another aim would be to reduce costs incurred to ineffective treatments.
For that reason, we need a design that allows faster and more effective evaluation of
several treatments against the gold standard of
control group as we have in this setting at the same time.