Title: Statistical methods for analysing discordance trials of biomarkers or
other diagnostic tests

Co-supervisors: Professor Sandra Eldridge&  Dr Richard Hooper

Diagnostic tests are usually studied in terms of their accuracy – how often
they correctly diagnose disease – but in practice the test result will be
used to make decisions about the management of patients, and a clinical
trial may therefore be the most appropriate way to compare two diagnostic
tests. There is a rapidly growing interest in the use of tests to guide
treatment choices rather than specifically to diagnose: biological tests of
this kind are called biomarkers, and are seen as a key element of an
approach to healthcare known as stratified medicine, in which a treatment is
matched to a patient in order to optimise its effect.  It has been suggested
that an efficient approach to trialling is to give both tests to all
participants, and to randomise and follow up those with discordant results.
We are interested in putting these ideas into a more formal framework.

The regulatory authorities who oversee new treatments require that any new
treatment tool – including a biomarker – must be validated using a large
Phase III clinical trial before it can be accepted as the standard of care.
Although there have so far been relatively few such trials comparing the use
of a biomarker with a non-stratified control, and almost none comparing two
biomarkers, this sort of application is likely to blossom as stratified
medicine takes off, and more and more biomarkers begin to be investigated
for their use in treatment decision-making. Our theoretical work has already
allowed us to put a figure on the huge efficiency saving of a discordance
trial design over a conventional trial for comparing two biomarkers or other
diagnostic tests. We found a four-fold reduction in the number of trial
participants required using figures from one published protocol as an
example (with a further reduction in the number whom it was necessary to
follow up). There are clear ethical and financial imperatives to adopting
such an efficient design in trials, but the details of the statistical
methods that should be used are still unclear, creating an obstacle to the
development of trials of this kind. Much more work is needed to show how to
extend the standard generalised linear modelling approach used for binary,
continuous and survival outcomes in conventional trials to the context of a
discordance trial. This would provide the focus of the PhD project.

For an informal discussion about this project, please contact Dr Richard
Hooper
Tel: 020 7882 7324  Email:r.l.hooper@qmul.ac.uk

This is one of a number of studentships being advertised together by Barts&
The London School of Medicine&  Dentistry, for which students will be
selected by competition. The closing date for applications is 20 February,
and interviews will be held on 14 March. To apply, go to
http://www.smd.qmul.ac.uk/graduatestudies/index.html, click on the button
marked "PhD studentships currently available", and follow the instructions.