Dynamics and Forecasting of Emergency Department Presentations: An Integrative Approach Based on Statistical Analysis and Time-Series Models

Background/Objectives: Emergency Department (ED) attendance is characterized by substantial temporal variability, creating challenges for workforce scheduling, resource allocation and short-term operational planning. This study aimed to characterize the temporal patterns of daily ED presentations and to evaluate the short-term forecasting performance of SARIMA and Prophet models in a Romanian hospital setting. Methods: Daily ED visits recorded at the Craiova County Emergency Clinical Hospital between January 2025 and April 2026 were analyzed (485 consecutive days). The analytical framework combined descriptive statistics, normality assessment, the Kruskal–Wallis test, and post hoc Mann–Whitney U tests with Holm correction and time-series modeling. A SARIMA(1,0,5)(0,0,1)7 model incorporating a Romanian legal-holiday dummy and weekly seasonality and a Prophet model incorporating Romanian holidays and weekly seasonality without yearly seasonality were evaluated using a common rolling-origin out-of-sample framework. Predictive accuracy was assessed using RMSE and MAPE, while differences in forecasting performance were examined using the Diebold–Mariano test with a HAC/Newey–West correction. Results: ED attendance differed significantly across weekdays (H(6) = 52.365, p < 0.001), with the highest mean attendance observed on Mondays (257.6 patients/day). Post hoc analysis confirmed significant differences between several weekday distributions after Holm adjustment, supporting a structured weekly pattern. Time-series analysis demonstrated serial dependence and recurrent weekly seasonality. Both forecasting models showed good short-term predictive performance across the four common validation windows. Over the complete out-of-sample period of 1–30 April 2026, SARIMA achieved an RMSE of 22.59 patients and a MAPE of 7.46%, compared with 23.54 patients and 7.56%, respectively, for Prophet. Although the aggregate results slightly favored SARIMA, the difference in predictive accuracy was not statistically significant (Diebold–Mariano test, p-value = 0.575). Conclusions: Daily ED attendance at SCJU Craiova exhibits systematic weekly variation, serial dependence and calendar-related effects that can be incorporated into short-term forecasting models. SARIMA and Prophet provided comparable out-of-sample predictive accuracy, with neither model demonstrating statistically significant superiority. These findings support the use of short-term forecasting as an operational planning tool for anticipating variations in ED demand and informing adaptive resource allocation.