David Marples

Research Topic: Breaking the Wave - Predicting surges in unplanned hospital attendance before they happen

Supervisors: Nishant Ravikumar, Kieran Zucker, Andrew Webster, Stephen Bush

About David: My first degree was in medicine, graduating from Oxford in 1990, and going on to do my house jobs (equivalent of foundation years!). In the meantime I had also done a D.Phil. in renal physiology, and after my house jobs I returned to research in the same area, first in Oxford and then in Denmark. In 1996 I took up a research fellowship at Leeds, progressing to a lectureship and then senior lectureship in physiology and anatomy. In 2019 I changed career course, getting a degree apprenticeship with PwC to study computer science, from which I graduated with first class honours in 2023, before starting the CDT programme in October 2023.

Project description: My project is looking into whether we can predict unplanned hospital admissions.

The first part looks at the general public going to A&E, and is based on the idea that the number of people who turn up depends on things like the day of the week, what the weather is like, and if there are any special events on. The idea is that we WON’T look at any medical data of individuals, but just look at things affecting the population as a whole. The goal is to help the hospital plan what resources they will need on a given day – whether they need to get more staff into A&E, or whether they can release them to help with planned operations, for example. This part of the study is nice, because all the data needed for the predictions is in the public domain. On the other hand, there’s quite a big random element in how many people go to A&E on any given day, and often spikes will be due to a totally unpredictable event, such as a big pile-up on the motorway. The challenge is to get a useful prediction, despite these variations.

The second part of the project looks at existing patients with particular conditions, and we’re starting with people with cancer. In this case, we know quite a lot about them, in terms of their main disease, and also what other medical problems they have, and we know they will often get treated with chemotherapy drugs that can have severe side-effects, so there is a high chance of unplanned admissions. This part of the project faces the common problems of data confidentiality - but because we are looking at more detailed information about each case, which means we are likely to be able to make a more accurate prediction of their chances of being admitted as an emergency. One problem with this kind of modelling  is that of correlation vs causation: many previous studies have looked at individual patients, and tried to predict which ones will become ill, to try to prevent those admissions. However, factors that are associated with emergency admissions may not be causing the admission, so changing them won’t help. That’s less important for our project, because we are only trying to forecast the overall number of admissions, rather than trying to intervene in the care of the individual patients.