An Analysis of the Time Aggregation Influence on Patients Forecasting in Emergency Services

Abstract

The COVID-19 pandemic has underlined that Emergency Department (ED) overcrowding is a critical factor in care services. Getting an approximation of the number of patients attending the department can assist in service resources planning and prevent overcrowding. In this manuscript we present the forecasting results for the admissions, inpatients and discharges series in ED by using different time aggregations (eight hours, twelve hours, one day and the service workers official shifts) and classical time series algorithms. Moreover, series forecasting is performed in two terms: long (four months ahead) and short (seven days ahead). The results show that time aggregations strongly influence the forecast quality, decreasing the effectiveness for one-day aggregations. In addition, best metrics are not obtained in the same aggregation, so there is no best aggregation for all cases. Therefore, it is essential to analyse the ED-related problem faced for the time aggregation selection.