Forecasting Daily Arrivals and Peak Occupancy in a Combined Emergency Department

Objective Emergency department (ED) crowding is a global problem associated with negative patient outcomes such as mortality and prolonged length of stay. Forecasting overcrowding would enable pre-emptive strategical maneuvers and is a subject of constant academic interest. However, most studies focus on forecasting arrivals in United States ED setting. We propose a novel and intuitive crowding metric called daily peak occupancy and assess forecasting ED crowding in a Nordic Combined ED using both established and novel predictive algorithms. Methods All episodes of care in Tampere University Hospital ED were acquired from December 1, 2014 to June 19, 2019, amounting to 488 167 individual events. Predictability of two target variables was investigated: total daily arrivals (TDA) and daily peak occupancy (DPO) with forecast horizon of one day. Three models were investigated: Seasonal Autoregressive Moving Average (SARIMA), Facebook’s Prophet algorithm, and General Linear Model (GLM). Calendar variables were used as independent variables. Results SARIMA outperformed other models in predicting both total daily arrivals and daily peak occupancy with mean absolute percentage errors of 6.6% (±5.3) and 12.4% (±10.7) respectively. Next day overcrowding can be predicted using SARIMA with an AUC of 0.74 and accuracy of 79 %. Conclusion Predictive models can be utilized in a Nordic emergency medicine setting with similar or better accuracy as previously reported. Predicting future occupancy is possible but more challenging than predicting arrivals. Predicting peaks in demand remains a significant challenge. Future work should focus on investigating the value of exogenous variables in increasing model sensitivity.