�Forecasting A&E Attendances to inform staff planning and decision-making

Presenter – Yu Qiao 24/06/2025

Image source: https://www.bbc.co.uk/news/uk-england-merseyside-58348981

Introduction

Who?

• Yu Qiao
• Clinical Data Scientist
• Liverpool University Hospital Foundation NHS Trust

_______________________________________________________________________

​

Coding experience:

• Prior coding knowledge from University and Scientist Training Programme, continuous self-directed learning.
• Connect between problems and solutions
• Joined HSMA to expand on technical skills and modelling capacities.

NEWS

A&E Demand,

Winter Pressure,

Patient Safety, �Limited resource

But… �what is it like?

In A&E as a patient

Patient care, safety and experience

Inform incoming patient attending

A&E to manage A&E operation and staffing

Data Exploration

Limitation of current tools

Model Choice

Explainable, multiple trends, fit for purpose

Multi Seasonal-Trend decomposition using LOESS

Model Choice

Attendees

Monthly Trend

24-hour trend

Weekly Trend

Left over variations

Breakdown into

9% difference on average between predicted and actual patient attendance.

​

Out of 100 attendees, 91 to 109 are predicted to attend on that day.

​

Great predictive accuracy

​

Plan and it is progressing

Murphy's Law

A storm

All came crashing down.

We could not deploy the model.

New Plan

Integration of prediction tools and align with Organisational plans

and existing software

Deployed to Power BI Dashboard

Impact

• Currently in use
• Helps A&E management and staff who look after 500+ patients a day x 10 months x 30days = ~150,000

​

Impact

How?

• Resource planning (staffing) both day to day and hour to hour (every 6 weeks)
• Forecasting admissions against discharges and triangulating with ED attendances to escalate potential pinch points ahead of time (daily)
• Map out impact of any QI projects to support benefits realisation
• Enhanced decision making
• Streamline operational planning
• Adds additional mitigation to risk management

Patient care, safety and experience

HSMA

Personal impact:

• Expanded skills on modelling & deployment options

Wider organisational

• Improve overall capacity of open-source tools use (Tools that can be seen, used and adapted)
• Set up of Infrastructure to support R/Python deployment and use for colleagues
• Improved capacity to support wider organisation and resolve challenges
• Alternative automation tools

• Publication of the code to GitHub
• Future projects related to Predictions and Natural Language Processing

Next steps