BACKGROUND
From 2006/2007 to 2017/2018, there was a 26% increase in emergency department (ED) attendances and 32% increase in total admissions in the National Health Service in England (NHS). Growing demand puts severe strain on hospitals, resulting in bed, nursing, clinical and equipment shortages. Nevertheless, scheduling issues can still result in significant under-utilization of beds. It is imperative to optimize the allocation of existing healthcare resources, including hospital beds. More accurate and reliable long-term hospital bed occupancy rate prediction would help managers plan ahead for their population’s hospital requirements, ultimately resulting in greater efficiencies and better patient care.
OBJECTIVE
This study aimed to compare widely used automated time series forecasting techniques to predict short-term daily non-elective bed occupancy at all trusts in the NHS.
METHODS
Bed occupancy models that accounted for patterns in occupancy were created for each trust in the NHS. Daily non-elective midnight trust occupancy data from April 2011 to March 2017 for 121 NHS trusts were utilized to generate these models. Forecasts were generated using the three most widely used automated forecasting techniques: Exponential Smoothing (ES); Seasonal Autoregressive Integrated Moving Average (SARIMA); Trigonometric, Box-Cox transform, ARMA errors, Trend and Seasonal components (TBATS). The NHS Modernization Agency’s recommended forecasting method prior to 2020, was also replicated.
A comparative analysis of forecast accuracy was conducted by comparing forecasted daily non-elective occupancy with actual non-elective occupancy in the out-of-sample dataset for each week forecasted. Percentage root mean squared error (RMSE) was reported.
RESULTS
The accuracy of the models varied based on the season during which occupancy was forecasted. For the summer season, percent RMSE values for each model remained relatively stable across six forecasted weeks. However, only the TBATS model (median error 2.45% for six weeks) outperformed the NHS Modernization Agency’s recommended method (median error 2.63% for six weeks).
In contrast, during the winter season, percent RMSE values increased as we forecasted further into the future. ES generated the most accurate forecasts (median error 4.91% over four weeks), but all models outperformed the NHS Modernization Agency’s recommended method prior to 2020 (median 8.5% error over four weeks).
CONCLUSIONS
It is possible to create automated models, similar to those recently published by the NHS, that can be used at a hospital level for a large, national healthcare system in order to predict non-elective bed admissions and thus schedule elective procedures.
CLINICALTRIAL
N/A
Temporal Changes in Central-Line–Associated Bloodstream Infection Time Between Events, 2017–2018 Versus 2015–2016