Event Details

Predicting emergency admissions to hospital specialties using real-time data from Electronic Health Records: A Machine Learning application in operational use.

Hospitals in the UK are currently at capacity for much of the time. Bed planners have the difficult job of ensuring enough emergency beds are available within each clinical speciality, to avoid admission delays and outlying patients. In previous work [1] we reported on an analytical pipeline that applied Machine Learning to real-time health record data to predict aggregate numbers of emergency admissions in the next 8 hours. That application has been in operational use by bed planners at UCLH for over a year to help them manage patient flow into the hospital.

Since then, I have been working as an embedded researcher alongside the bed planners at UCLH. Building on a richer knowledge of their role, I will describe the work of bed planners and their use of data products, and explain how we re-formulated the modelling problem and the technical implementation to make our application better meshed with what they need to know. My talk will cover both the academic elements of the work (the problem framing and formulation, and evaluation of model performance), and the practical aspects of engagement with bed managers and their wider system to achieve an application supported by UCLH as part of their business-as-usual suite of data products.

[1] King, Z., Farrington, J., Utley, M. et al. Machine learning for real-time aggregated prediction of hospital admission for emergency patients. npj Digit. Med. 5, 104 (2022). https://doi.org/10.1038/s41746-022-00649-y