Forecasting Severe Acute Patient Harm Resulting from Emergency Department Delays
Date: September 23 2026 14:45
Event: NHS SPHERE-PPL Forecasting Contest
Organizer: NHS Bristol, North Somerset and South Gloucestershire ICB
Context
In this presentation, I shared our joint 1st-place winning solution for the NHS SPHERE-PPL Forecasting Contest (1–5 Day Horizon Category), developed by our team at the Data Lab for Social Good (Cardiff University).
We tackled the challenge of forecasting daily Estimated Avoidable Deaths (EAD)—a critical outcome metric measuring acute patient mortality risk caused by prolonged emergency department (ED) delays awaiting inpatient beds. Standard linear and single-regime models fail because hospital operations experience extreme regime shifts and severe heteroskedasticity explosion in error variance during operational crisis states.
To address this, we developed a Markov-Switching AutoRegressive Regression (MS-ARR) model. By coupling regime-switching filters and the Forward-Backward algorithm, the framework captures persistent crisis lock-in, tracks daily regime escalation risks, and delivers actionable reliable forecasts for NHS trust capacity planning and clinical bed management.