Please see below the funded PhD opportunity at University of
Manchester, supervised by Dr Niels Peek and I. Interested candidates
are welcome to contact me for informal discussion
[log in to unmask] For further details and to apply,
see https://www.findaphd.com/search/ProjectDetails.aspx?PJID=82132 and
for information on the scheme more generally, see
https://www.bmh.manchester.ac.uk/study/research/funded-programmes/gmchc-studentships/
Best wishes
Matt
==
Connected Health Cities is a government-funded programme that aims to
create "learning health systems" across North England by harnessing
data that is routinely captured by health services, applying advanced
data analytics methods, and feeding the results back to clinicians,
patients, public health professionals, and other stakeholders in the
health service; in other to deliver better outcomes for patients and
communities. https://www.connectedhealthcities.org/
A core element of learning health system is the analysis of healthcare
processes and outcomes that are measured downstream to inform decision
making that takes place earlier along the same care pathway. In the GM
Connected Health City, an example is the Stroke Mimics project which
aims to improve decision making by the ambulance staff in the
assessment of patients when there is a suspicion stroke. The ambulance
staff currently use a simple algorithm, called the FAST test, to
determine which patients should be taken directly to a stroke centre.
However a recent study in London demonstrated that 37% of FAST+
patients identified by paramedics subsequently had a non-stroke
diagnosis (they were "stroke mimics"). The GM-CHC Stroke Mimics
project hypothesizes that systematic learning and improvement of this
process can be achieved by analysing linked data from the ambulance
service and the stroke unit – in particular, the diagnosis that is
made at the unit.
Analytically, the key question is whether we can predict the outcome
of interest (in this case, the diagnosis) using data that are
available at the earlier time point (in this case, at the time of the
paramedic assessment). If so, the predictive model/algorithm can be
made available to the relevant decision makers (in this case,
ambulance personnel) to improve their decision making process. Thus,
predictive methods are key to the success of this approach. However,
there are important methodological challenges to enable the use
predictive methods in the context of a learning health system:
• Data are collected in routine clinical practice and can therefore
contain recording errors and missing values, including missing
outcomes. This leads to significant risks of bias (and thus,
prediction errors) if they are not handled properly.
• For some patients much more data is available than for others – a
situation that traditional predictive methods (e.g. based on logistic
regression) cannot deal with. For instance for some patients we may
have access to a rich, longitudinal health record from primary care,
whereas for others hardly any information is available at all.
• As the health system changes over time, the model will need
updating to reflect those changes. Some parts of the model must then
be discarded while other parts can be retained. If decision makers
follow the model’s predictions, this will further change the pathway
and complicate updating the model. In the Stroke Mimics project, we
might increasingly lose sight of potential stroke mimics. This is an
improvement of the service, but it diminishes the opportunities for
learning.
This PhD project investigates predictive methods for learning health
systems. Building on existing methods from statistical modelling and
machine learning, the project will develop new methods to address the
above challenges. The resulting methods will be applied in the Stroke
Mimics project and other relevant projects within the GM CHC.
Student background:required We require a student with a minimum of an
upper second class bachelor’s degree in mathematics, statistics or a
closely related discipline, with an interest in developing new
statistical and machine learning methods that can be applied to
challenging data streams. Training will be available in learning
health systems and dealing with routinely collected data.
Funding Notes
Studentships are for 3 years and provide stipend and fees for UK/EU
applicants only – anticipated start date April or September 2017.
Students must first discuss the project with the named academic
supervisors and obtain permission to apply.
Students should have a minimum of a 2:1 in their first degree in an
appropriate discipline, preferably holding (or be completing) a
Masters qualification in an appropriate field.
Applications should be submitted online. On the application form,
select PhD Health Informatics. Applications should include the
following documents – CV, supporting statement – outlining your
research experience and interests and two academic references.
You may leave the list at any time by sending the command
SIGNOFF allstat
to [log in to unmask], leaving the subject line blank.
|