ASTRAZENECA PhD STUDENTSHIP
AT THE UNIVERISTY OF LEICESTER 

Applications are invited for a three-year full-time PhD studentship in
the Centre for Biostatistics and Genetic Epidemiology at the University
of Leicester starting 1 October 2006. The project is entitled "Modelling
phenotypic variation in large medical databases using latent variables"
and will look at theoretical and practical problems that arise when
latent variable and latent class models are used with large clinical
databases. The initial application will be to the study of patterns of
disease in a database of patients with difficult to treat asthma. A
brief description of the project is given at the end of this email.

This studentship is open to EU citizens. 

Applicants should be interested in working on medical applications of
statistics and have at least a 2(i) Honours degree or a recognised
postgraduate qualification containing a substantial element of
statistics. 

The Centre for Biostatistics and Genetic Epidemiology is part of the
Department of Health Sciences within the University's Medical School.
For more information about the Department, see our website at
www.hs.le.ac.uk 

The studentship is funded by AstraZeneca and the PhD student will have
the opportunity to visit AstraZeneca for a short period each year,
although they will be based at the University of Leicester. 

For written details on the studentship and the department please contact
Joy Fox at: [log in to unmask] . 

If you have specific questions concerning the studentship or the
project, please email the main PhD supervisor, Professor John Thompson
at [log in to unmask]

An application form can be downloaded from the University of Leicester
Graduate School at www.le.ac.uk/gradschool . Alternatively, email
[log in to unmask], phone +44 (0)116 252 3206 or write to:

Mrs J Fox
Centre for Biostatistics and Genetic Epidemiology
Department of Health Sciences
University of Leicester
22-28 Princess Rd West
Leicester LE1 6TP

 The closing date for applications is Friday 11th August 2006.



Brief Project Description: Modelling phenotypic variation in large
medical databases using latent variables

The Leicester difficult asthma cohort was collected by clinicians in
Leicester and contains nearly 400 people recruited through a difficult
asthma clinic. There is extensive baseline information on these patients
and they are now being followed longitudinally with collection of data
on lung function and airway inflammation.

The investigators have commented on the degree of phenotypic variation
in their patients, that is the large variation in the way that the
disease manifests, and they have questioned whether this is due to the
range of disease severity or to the presence of subgroups of patients
with diseases that follows different natural histories. This will be
investigated using latent variable models.

Latent variable methods assume that underlying a condition there are
unmeasured quantities, such as disease severity or diagnostic grouping,
that influence the measured data. This approach is the basis for factor
analysis, mixture modelling and a whole range of other statistical
techniques.  The technique is commonly applied to classification
problems in medicine particularly in Psychiatry were disease
classification is a major problem. However, these studies are almost
exclusively based on the analysis of relatively small numbers of
variables. There have been very few applications to very large databases
similar to the difficult asthma database.

The fitting of latent variable models is not a trivial exercise and if
the models are to be flexible enough to cope with binary, categorical
and continuous measures, missing data, covariates and longitudinal
measurements then fitting is complex. Recently the statistical
literature has included a number of articles that have looked at the use
of Markov chain Monte Carlo (MCMC) methods for the fitting of latent
variable models. MCMC algorithms are extremely flexible. 

The clinical investigators are keen to work with the PhD student to help
them develop meaningful models.

This project will investigate the use of latent variable analysis with
large medical databases and as such will be of interest well beyond the
specific asthma application. The methods developed will be applied to
other locally collected clinical databases.