Clinical prediction models (CPMs) aim to predict outcomes in patients to inform diagnosis or prognosis in healthcare and also identify potential interventions that might be best suited for the individual [1]. Traditionally the development of such models is siloed on a single site, with training and validation on a single dataset. However, the pandemic has shown that both the demand for services and the respective provision can vary significantly by site which when coupled with changes in security, governance and wider regulatory requirements might introduce barriers that will prevent the implementation of clinical prediction models into practice [2].
There is an opportunity to address this problem as the infrastructure, information governance and data sharing standards now exist to support effective processing at an individual site level and handle significantly larger and disparate data. In particular, federated learning approaches seek to build consensus models across multiple decentralised datasets without actual exchange of individual patient records while ensuring the performance, safety, scalability and fairness of clinical prediction models [3]. Therefore, local prediction models can be re-tuned to improve generalisability and effectiveness even when there are significant differences in service provision and population dynamics [4].
This issue has received great attention with clear evidence that such approaches can bring significant improvements at little cost with respect to model performance. However, the methodological gap is how to best deploy in practice. An implementation framework for federated learning that is tailored to the local regulatory, security and government is needed to help innovators to highlight potential issues and develop solutions. This will include the evaluation of current algorithms used for federated learning with synthetic as well as real-world data with a focus on how the individual site differences in performance might influence the overall strategy and approach. Taken together these are distinct and complementary areas with knowledge gaps that this PhD project will aim to explore.
Aims/Objectives:
1. Review literature on federated learning and usage of multiple models to improve prediction effectiveness at a local (NHS site) and system level
2. Perform a simulation study to illustrate the by-site problem and assess how different sites can impact model effectiveness
3. Assess the impact on performance of model updates at a local and system level taking into account site differences as well as safety, security, governance and data retention
4. Develop strategies as part of an overall framework to identify problems and develop solutions
This project will use simulation to investigate the benefits of federated learning algorithms across multiple criteria including model performance and equity. The project will also develop a framework to implement such decentralised model training approaches in practice as the cost and time to deploy effective clinical prediction models are of increasing importance in healthcare.
Eligibility
Applicants must have obtained or be about to obtain a First or Upper Second class UK honours degree, or the equivalent qualifications gained outside the UK, in a relevant discipline.
Before you Apply
Applicants must make direct contact with preferred supervisors before applying. It is your responsibility to make arrangements to meet with potential supervisors, prior to submitting a formal online application.
How to Apply
To be considered for this project you MUST submit a formal online application form - full details on how to apply can be found on the MRC DTP website https://www.bmh.manchester.ac.uk/study/research/mrc-dtp/
Your application form must be accompanied by a number of supporting documents by the advertised deadlines. Without all the required documents submitted at the time of application, your application will not be processed and we cannot accept responsibility for late or missed deadlines. Incomplete applications will not be considered. If you have any queries regarding making an application please contact our admissions team [Email Address Removed]
Equality, Diversity and Inclusion
Equality, diversity and inclusion is fundamental to the success of The University of Manchester, and is at the heart of all of our activities. The full Equality, diversity and inclusion statement can be found on the website https://www.bmh.manchester.ac.uk/study/research/apply/equality-diversity-inclusion/
Continue with Facebook
