(Turing) Can we use machine learning risk prediction models in diverse settings in the healthcare system?

There is considerable potential to use routinely collected electronic health records (EHRs) to estimate the risks of major clinical outcomes (such as heart attack, mortality or hospital admission) which could then be used by clinicians to target treatment. The conventional approach for risk prediction has been to use regression modelling based on known risk factors. There is now major interest in using machine learning techniques which use all information available in the records. Recent research has found that machine learning models can outperform conventional models in predicting risks using EHRs (Steel A PlosOne 2018). However, the application of these approaches to EHRs is challenging due to the clinicians recording and coding information very differently. Recent research in our group has found that the accuracy of risk predictions that are based on conventional models varies considerably between different clinical sites. The objective of this PhD is to assess the generalisability of machine learning models to different clinical sites. This will include review and development of measures of data quality between sites and evaluation of robustness of risk predictions in the presence of heterogeneity in data quality. The performance of machine learning, and conventional regression techniques in presence of these limitations will be evaluated in both simulations and practical examples. This will include the risk prediction of cardiovascular outcomes, mortality or hospital admissions using routinely collected data. The practical outcome of this PhD is to develop and implement methods to measure the generalisability of machine learning with a focus on clinical applicability of these models. The machine learning methods may include random forests and elastic net regressions.

This PhD will provide training in the following priorities: quantitative skills of statistics / machine learning and developing digital excellence as applied to electronic health records). The PhD candidate will work in a multidisciplinary team involving statisticians, data scientists, informaticians clinicians and social scientists. There will be ample training opportunities within the University of Manchester for PhD students (including statistics, machine learning, data science and data management). Also, there may be opportunities for an exchange with international collaborators of the supervisors.

https://www.herc.ac.uk/
https://www.research.manchester.ac.uk/portal/tjeerd.vanstaa.html

The Alan Turing Institute – About the studentship
The Alan Turing Institute and The University of Manchester offer a number of places each year to motivated graduate students to complete a fully funded PhD. The Turing doctoral studentship scheme combines the strengths and expertise of world-class universities with the Turing’s unique position as the UK’s national institute for data science and artificial intelligence, to offer an exceptional PhD programme.

Turing students will have access to a wide range of exceptional benefits:
• Spend time in both a university research environment and at The Alan Turing Institute.
• Access to a range of training, events, seminars, reading groups and workshops delivered by leaders in research, government and industry.
• Opportunities to collaborate on real-world projects for societal impact with current and emerging industry partners.
• Expert support and guidance through all stages of the studentship, delivered by supervisors who are Fellows of the Turing or substantively engaged with the Turing.
• Networking opportunities and brilliant minds researching a range of subjects with opportunities to collaborate and join or start interest groups.
• Opportunities to supercharge your research with access to cutting edge resources.

Find out more at turing.ac.uk/PhD https://www.turing.ac.uk/phd-at-turing

Entry Requirements
Applications are invited from UK/EU nationals only. Applicants must have obtained, or be about to obtain, at least an upper second class honours degree (or equivalent) in a relevant subject.