Statistical issues in the use of genetic variants to evaluate drug
effects
Supervisors: John Whittaker and Liam Smeeth
The average current cost of taking a new drug from discovery phase
to clinical use is estimated to be £500 million and is projected
to increase because of the high failure rate; currently only one in
every 104 new molecular entities enters clinical practice. The problem
of developing new drugs that provide incremental benefits over existing
drugs is particularly sever in the case of cardiovascular disease.
Initially, drug targets are discovered and then validated (investigated
for their relevance to a disease) through experimental findings in cells
and tissues (that may not translate to the whole animal) or from experiments
in animal models (that may not translate to human beings). Non-experimental
observational studies in humans provide another route to target discovery
and have yielded plausible drug targets (e.g. > 100 potential targets
for coronary heart disease such as circulating biomarkers of inflammation,
coagulation, and oxidant stress as well as multiple lipid sub-fractions).
However, causal relevance is often difficult to establish in observational
studies because of confounding (where a target marks a causal factor
without being causal itself) and reverse causation (where even a subclinical
disease process alters the target rather than vice versa).
An alternative source of information about the validity of drug targets
comes from genetics. Genes are not subject to reverse causation (because
they are fixed at birth) and are less vulnerable to confounding than
other risk factors. Utilising genetic effects in this way has been termed
Mendelian randomisation, but the approach has been little used in drug
development at present. Recent advances in molecular biology mean
that very extensive genetic information can now be obtained relatively
cheaply, so this is an excellent time to exploit the potential of this
approach.
The focus of this PhD project can either be epidemiological or statistical
depending on the skills and interests of the applicant.
The work will be undertaken in collaboration with colleagues at University
College London and GlaxoSmithKline. We would expect the applicants to
already have an MSc in an appropriate quantitative area e.g. statistics,
or epidemiology. Previous knowledge of genetics or drug development
is not required; training will be provided as needed.
|