IMPERIAL COLLEGE LONDON: While predicting the risk of emergency admissions is becoming ingrained in healthcare delivery and financing in England, the models that are being used have limited predictive powe

Data-sharing agreement from NHS Digital registers of approved data releases, 2014–2019. Evidence status: source reported.

Organisation

Imperial College London

Source reference

DARS-NIC-39944-D4M0D-v0.3

Jurisdiction

England

Licence

Open Government Licence v3.0, subject to NHS England terms and exclusions

While predicting the risk of emergency admissions is becoming ingrained in healthcare delivery and financing in England, the models that are being used have limited predictive power. Most models are linear regressions, based on simple, static predictor variables. As such, there is significant scope to improve on these methods. The aim of this project is to develop a new method to predict risk in healthcare, going beyond linear regression, by using data science methods. The aim is to explore the different options such as artificial neural networks and to explore which method offers the most pot

Register reference

DARS-NIC-39944-D4M0D-v0.3

Agreement period

not published to no end date published

For commercial purposes (as marked in the register)

Not stated

Release files listed by NHS England

4

Data approved

• Hospital Episode Statistics Admitted Patient Care

Objective, as published

While predicting the risk of emergency admissions is becoming ingrained in healthcare delivery and financing in England, the models that are being used have limited predictive power. Most models are linear regressions, based on simple, static predictor variables. As such, there is significant scope to improve on these methods. The aim of this project is to develop a new method to predict risk in healthcare, going beyond linear regression, by using data science methods. The aim is to explore the different options such as artificial neural networks and to explore which method offers the most potential to revolutionise risk prediction. Simple risk algorithms that require no advanced data analysis will also be explored. Using HES data, the different techniques will be analysed and compared. If successful, the new risk prediction models will be developed into a tool or manual for the NHS to use.

Expected benefits, as published

The preliminary research for this project has identified, through a survey of 150 GPs across England, that currently only 37% of GPs with access to risk stratification think it is actually useful to their practice. Nevertheless, they are financially incentivised to use it though the NHS Enhanced Service specification on Admission Avoidance, which requires GP to identify the top 2% high risk patients on their list, and provide them with special services. There is a need to try and improve the usefulness of risk stratification. This research aims: - to improve the accuracy of risk prediction; and - to provide healthcare professionals with the tools to do the analysis and provide patient-centred, effective care. 1) The research aims to identify new and improved methods to perform risk predic…

Source: https://webarchive.nationalarchives.gov.uk/ukgwa/2019/http://content.digital.nhs.uk/media/23167/Data-Release-Register-September---November-2016/xls/Data_Release_Register__September_-_November_2016.xlsx. Captured 2026-10-30.

Limitations: Agreement as recorded in an NHS Digital register of approved data releases (2014–2019), retrieved from the UK Government Web Archive. That register lists releases made in its period and the agreement's purpose statement as it then stood; it gives no agreement dates, and the agreement may have changed since. Imported only where a paper quotes a reference no current register holds.