Patient visits to the Emergency Department of an Acute Hospital; dataset to accompany the patientflow repository

These datasets have been provided by University College London Hospitals (UCLH) NHS Trust, to support PatientFlow, a Github repository containing code and explanatory notebooks for predicting short-term hospital bed capacity using real-time data, which can be found here: https://github.com/zmek/patientflow. 

The PatientFlow repo is designed to support hospital bed management through predictive modelling. The repository shows methods for forecasting short-term bed capacity, a crucial aspect of hospital operations that impacts patient care and resource allocation.

UCLH have agreed to release the data to demonstrate to researchers, analysts, hospitals managers and others how this predictive modelling can be done, and to inform and educate anyone who wishes to adopt a similar approach.

The inpatient_arrivals dataset contains arrival times of all patients who visited the UCLH Emergency Department (ED) and/or the Same Day Emergency Care (SDEC) unit, over the period of the data, and were later admitted to a ward. It includes

the specialty they were admitted to (grouped into one of four categories: medical, surgical, haematology/oncology, paediatric)

their sex M = male; F = female

whether adult or child (boolean)

The ed_visits datasets contains snapshots of visits by all patients who visited the ED and/or SDEC over the period of the data, including those who were later admitted to a ward and those who were discharged. It includes various columns, including

visit_number: to identify unique visits

snapshot_date: the date on which the snapshot of the visit as taken was taken

prediction_time: the time of day at which the snapshot of the visit as taken

is_admitted: whether the patient was admitted at the end of the visit

elapsed_los: elapsed length of stay in seconds

demographic information

locations visited in the ED or SDEC

observations

lab orders

lab results

consult requests (referals to inpatient specialty), as a sequence of requests

specialty of admission

Dates have been shifted into the future by a whole number of weeks. Detailed data dictionaries are available here. Notebooks explaining the data structure and explaining its rationale are available in this folder.

We are grateful to University College London Hospitals (UCLH) NHS Trust and its Caldicott Guardian for agreeing to release an anomymised dataset based on real patient data, and for the scrunity of our anonymisation process undertaken by the Information Governance team at UCLH.