Explainable machine learning for risk prediction in later life care

The Advanced Care Research Centre (ACRC) is a new, multi-disciplinary, £20M research centre at the University of Edinburgh. The ACRC will lead society’s response to the grand challenge of an ageing population that is growing in size, longevity and needs through the pursuit of research intended to deliver “high‐quality data‐driven, personalised and affordable care to support the independence, dignity and quality‐of‐life of people living in their own homes and in supported care environments”.

This project sits within the ACRC Academy , a dedicated Centre for Doctoral Training, co-located with the ACRC, whose students will deliver key aspects of the ACRC research agenda through a new doctoral-level research and training programme that will also equip them for careers across a wide range of pioneering and influential leadership roles in the public, private and third sectors.

The PhD with Integrated Study in Advanced Careis a novel, structured, thematic, cohort-based, programme of 48 months duration. Each PhD research project within the Academy has been devised by a supervisory team comprising academic staff from at least two of the three colleges within the University of Edinburgh. Each annual cohort of around twelve will include students with disciplinary backgrounds spanning from engineering and data science to humanities, social science, business and commerce, social work, medicine and related health and care professions. This unique level of diversity is a key attribute of our programme.

Project

Aim

The project aims to develop explainable machine learning algorithms to predict the risk of harms from common healthcare interventions in older people approaching the end of life. 

Objectives

• Develop explainable machine learning algorithms to predict the risk of benefits and harms in older patients where there is clinical uncertainty (e.g. using medications to prevent clots causing stroke that also increase the risk of bleeding). 
• Use the explainable models to provide insights to clinicians of the relative contribution of risk factors, rather than just estimating risk. 
• Design and validate these novel models in routine healthcare datasets (SAIL Databank and DataLoch) to test effectiveness in ‘real world’ older populations. 

Description

Risk prediction in healthcare is important, particularly when there is uncertainty that treatment benefits outweigh risks of harm. This is more often true in older populations due to complexity from multiple health issues. Machine learning may improve prediction of outcomes, but many such models are ‘black-box’ and lack explainability to help clinicians make better decisions [1]. This project will develop explainable machine learning models to predict and explain the benefits and risks from treatments. We will rely on recent advances in machine learning surrounding explainability, e.g., gradient- or attention-based approaches, to investigate how changes in risks relate to input factors.

Research Group:

https://vios.science

Eligibility 

We are specifically looking for applicants who will view their cutting-edge PhD research project in the context of the overall vision of the ACRC, who are keen to contribute to tackling a societal grand challenge and who can add unique value to – and derive great benefit from – training in a cohort comprising colleagues with a very diverse range of disciplines and backgrounds. We advise prospective candidates to engage in dialogue with the named project supervisor and/or the Director of the Academy prior to submitting an application. 

Recruitment  

We are running a rolling recruitment process. We will assess applications on a monthly basis, and will continue to do so until our places are filled. The next deadline is 31 March, which will then be moved to 31 April, then 31 May, if places are remaining. 

You must read How to apply prior to application

Please Apply here