Reproducibility of statistical tests in pharmaceutical products development
EPSRC-CASE PhD project at Durham University in collaboration with AstraZeneca (Cambridge)
Prof. Frank Coolen, Department of Mathematical Sciences, Durham University (frank.coolen@durham.ac.uk)
Project description
Throughout the development process of new medicines, many statistical tests are used
to support decisions. Statistical analysis of the reproducibility
of such test results is important: would a repeat of the experiment lead to the same
test result and decision? The student will investigate reproducibility of
several tests at AstraZeneca, which will require further development of statistical
methods.
Statistical inference about reproducibility of tests is a non-trivial topic about
which there has been considerable confusion in the literature. Recently,
it has started to receive increasing attention, but it remains a topic for which
classical statistics approaches are not well suited. Prof. Frank Coolen has
recently, in collaboration with PhD student Sulafah Bin Himd, developed nonparametric
predictive inference (NPI) for reproducibility of tests. NPI is a
frequentist statistical method with explicit focus on prediction, so considering events
in terms of future observations. NPI is based on relatively few
modelling assumptions, enabled by the use of imprecise probabilities to quantify
uncertainty. It seems logical to consider reproducibility of tests as a
predictive problem. The use of NPI to derive lower and upper probabilities of reproducibility,
that is the event that a future repeat of the test would
lead to the same overall test conclusion (typically rejecting a null-hypothesis or not),
provides an attractive alternative to classical statistical methods.
In this research project, the student will first study the initial results on NPI for
reproducibility, in which some very basic statistical tests were
considered. The student will also gain an insight into practical tests during the
product development processes at AstraZeneca, and a choice of these
tests will be made to focus on first for the development of NPI methods to investigate
the test reproducibility. It is expected that the student can
study reproducibility of a variety of tests during this project, leading to valuable
insights at AstraZeneca and to novel statistical methodology supported
by algorithms in the statistical software R to implement the new methods.
The following are examples of possible test scenarios at AstraZeneca that could be
considered during this project:
- Assessment of growth inhibition of tumour cell xenografts implanted into rodent
species to assess compound potency.
Tumour cells are implanted into rodents and allowed to develop. Rodents will be treated with
(increasing) concentrations of
potential drug compounds. Growth of the tumour (either change in tumour size or final/start size)
is monitored, potentially
alongside a number of biochemical measurements describing the within tumour. The analyses
typically involve ANOVA and comparison
of multiple groups against a control. The goal of a reproducibility study is to understand
if repeating the study under the same
conditions would lead to the same results in terms of significant differences in growth
inhibition or other biomarkers.
- Experiments similar to the above description, but analysis undertaken by fitting a
dose response curve to inhibition data
to derive an IC50 estimate. This can be used to rank potential compounds according to
efficacy, or to compare them, i.e. are
IC50 values from two different compounds significantly different from each other? Would
repeating such an experiment lead to
a similar IC50 value, or with similar confidence intervals?
- It is also of interest to consider how inference on reproducibility might influence
experimental design, and to compare the
NPI methods for reproducibility of tests to alternative methods, for example a priori
(or posteriori) power calculations based on
the variability within the data. The methods can also be compared to alternative
approaches for investigating reproducibility, e.g.
Bayesian or bootstrap-based methods.
Being able to address such issues in terms of a reproducibility probability would be
extremely beneficial for AstraZeneca in terms of having confidence
in the results taken forward for investment decisions.
Supervision Team
The student will be supervised by Dr. Claus Bendtsen (Head of Quantitative Biology) and
Alan Sharpe MSc (Senior Statistician) at AstraZeneca, Cambridge, and by
Prof. Frank Coolen (Department of Mathematical Sciences) and Dr. Tahani Coolen-Maturi
(Durham University Business School) at Durham University.
Practicalities
This is a studentship in line with general UK research council (EPSRC) funding rules,
for a period of 3.5 years, starting 1 October 2017. This also defines
eligibility of candidates according to the EPSRC rules.
The student will be based at Durham University and benefit from the usual training for
PhD students in Statistics, including
attendance at APTS training weeks. The student will benefit from regular stays
at AstraZeneca in Cambridge to ensure
excellent communication with Dr Claus Bendtsen and Alan Sharpe and their colleagues
at the Quantitative Biology group, and to gain good awareness of the
practical relevance of the research work. Exact planning of visits to AstraZeneca will
depend on progress of the
research, it is expected that the student will spend at least two weeks per
half year at AstraZeneca.
Stipend
The stipend consists of the standard EPSRC PhD studentship stipend, enhanced by an additional
GBP 2,500 per year.
Further Information and Applying
Information about Nonparametric Predictive Inference (NPI) is available: NPI webpage.
For an initial idea about the NPI approach to reproducibility of tests, please have a look
at the PhD thesis by Sulafah Bin Himd:
PhD thesis Bin Himd.
Information about AstraZeneca is available here: AstraZeneca,
and about the Durham University Mathematical Sciences Department here: Maths Department.
If you are interested in this opportunity, or want further information, please contact
Prof. Frank Coolen (frank.coolen@durham.ac.uk) by Friday 24 March.
Frank Coolen