PhD Studentship: Advancing Joint Modelling Methodology for Clinical Prediction Models

This 3.5 year PhD project is funded by the MADSIM Project, www.madsim.manchester.ac.uk. Funding is for Home students and EU students with settled status. Funding covers home tuition fees and provides a stipend at the UKRI rate (£19,237 for 2024/25). The start date is 1^st October 2024.

This is an interdisciplinary project between the Department of Mathematics (Christiana Charalambous, Timothy Waite) and the Centre for Health Informatics (David Jenkins). Clinical prediction models (CPMs) are algorithms that use information about a patient at a given time point to generate risk estimates for an outcome. These models are widely adopted throughout healthcare and can be used to inform clinical decisions, for example, if an individual should receive an intervention. Traditionally, CPMs have used data from a single time point and often consider a single outcome. However, the adoption of electronic health records and the increase in availability of data provides rich longitudinal (e.g. repeatedly measured biomarkers) and time-to-event (e.g. death or disease progression) data, which are often underutilised. Complex models that use multi-outcome (potentially correlated) data, such as joint models, are beginning to be adopted and evidence from the literature suggests this could improve predictive accuracy and in turn patient outcomes.

Objectives and outcomes of the project:

1. Review the existing literature on joint modelling for clinical prediction.
2. Undertake methodological development for the formulation and validation of joint models, testing the developed methods throughout a range of scenarios in simulated and real-world health data.

• Extend time-dependent AUC methodology for joint model validation.
• Perform simulation studies to compare the proposed method to existing approaches for prediction and model validation.
• Extend current minimum sample size methodology for joint models.

3. Assess the impact of different study design approaches to investigate the optimal way to record clinical measurements and monitor patients (through simulation, application to real-world healthcare datasets and methods development).

• Undertake simulation studies to investigate the impact of longitudinal measurement frequency on joint model performance.
• Investigate techniques for Bayesian design of experiments in joint models.

4. Develop guidance for the development, validation and use of joint models in clinical practice.

The combination of simulation and real-world data will allow us to evaluate the methods under a range of scenarios and parameter combinations and assess the real-world impact of the methods. The collaboration with the Centre for Health Informatics will provide access to real-world health data, such as the Greater Manchester care record, UK Biobank and cardiovascular data from the Manchester University NHS Foundation Trust and Wythenshawe Hospital, that the centre regularly utilises. The project will also provide recommendations to determining when to monitor patients and guidance for developing and validating joint models for clinical prediction.

Applicants should have:

• Obtained or working towards a 1st class degree in Mathematics (BSc/MMath) or Distinction level Masters in (Bio)Statistics, Data Science or similar.
• Research experience, e.g. UG/MSc project, research internship or other.
• Background in some or all of the following: longitudinal data analysis, survival analysis, design of experiments, Bayesian statistics.
• Good programming skills in a language such as R or Python
• Good communication skills (oral and written)
• Openness to working across disciplines

Before you apply, please contact Dr. Christiana Charalambous at christiana.charalambous@manchester.ac.uk.

Advert information