The maturation of emergency medicine as a specialty has coincided with dramatic increases in emergency department (ED) visit rates worldwide [1]. Patients who present to EDs often face long wait times for treatment; moreover, those who require admission wait even longer for an inpatient hospital bed [2]. ED timeliness is a critical factor affecting the operational performance of the complex processes required to deliver emergency care. Operational performance ultimately affects the quality of care, patient safety, and operational and financial efficiency [3]. Many interventions to manage ED crowding have been implemented in order to improve the efficiency of ED operation by eliminating waste, reducing variability, and increasing flexibility [4,5]. The problem remains despite these efforts, because ED crowding is not only a problem related with individual EDs, but is rather a reflection of a larger supply and demand mismatch in the healthcare system [2].

In the conceptual model of the ED, previous interventions mainly target throughput and output components. ED supply systems concerning patient care processes and patient disposition are manageable; however, the input component, which includes any condition, event, or system characteristic that contributes to the demand for ED services, is not easily reducible or predictable [2]. Ambulance diversion to other EDs or telephone triage services to reduce ED demand have limited effects on the overall emergency medical system [3]. Furthermore, ED demand at a given time appears impossible to predict, and there is a lack of analytic research. The input component usually predicts the average ED patient volume over a certain period, despite the lack of evidence indicating that the average is adequately representative of the variability of ED patient visits [6-8].

Efficient operation requires individual EDs to estimate how many busy days will occur and can be reduced with additional medical resources as well as how often “on-call” resources will be needed. Thus, hourly ED visit data should be collected and processed to address these questions. Accordingly, processed nationwide ED visit data could be used to construct a road map to develop policies for the entire healthcare system from the perspective of the expanded input-throughput-output conceptual model [9]. Thus, the present study evaluated the daily variability in hourly ED visit volume.

In 2013, the NEDIS annual report stated 4,919,181 patients visited 141 EDs in 16 provinces nationwide (Table 1) [10]. However, as the information transmission rate to the NEDIS was approximately 94%, the available text file contained the arrival times of 4,672,275 patients in 2013 [10].

The bar graphs and box plots of EDs nationwide are shown in Figs. 1 and 2, respectively. The hourly visit pattern in one day showed a bimodal distribution with peaks from 10:00 to 11:00 and 20:00 to 21:00. The period with the lowest visits was 02:00 to 08:00. The box plots (Fig. 2) of the ED visit volume at each time showed upside-skewed interquartile ranges rather than a symmetric distribution in addition to the median volume.

In contrast to the nationwide ED graph, the single ED graphs showed a similar bimodal distribution of hourly visit volume in one day. However, the skewness and kurtosis of box plots of single EDs were smaller than those of nationwide EDs (Figs. 3, 4).

Statistically, the inter-day difference of hourly visit volumes was too large to assume a normal distribution despite the large dataset (P<0.05). Furthermore, both the nationwide and individual ED data were right skewed (upside skewed) and widely dispersed (Tables 2, 3).

The bimodal feature of the average hourly ED patient visit volumes identified in the present study is similar to the findings of previous studies [11,12]. The authors of those studies assumed there was no significant variability in visits over 365 days at a given time of day; therefore, the average was used as the representative value. However, the box plots showed positively skewed and widely dispersed distributions of visit times and this causes the more-frequent “low visit volume days” to decrease the average, consequently masking the less-frequent “high visit volume days.” In the case of a given ED with a visit volume similar to that shown in Fig. 4 (with a volume difference of approximately 1.6 patients per hour between the 1st quartile and average), an ED operator who prepares the ED expecting the average ED visit volume may experience 1.6 more patients per hour in three months (25%) in a given year. Unfortunately, the unbounded positive range also indicates that ED crowding cannot be completely prevented in cases such as disasters. Therefore, “permissive busy ED days” should be determined in order to prevent excessive waste of medical resources. Accordingly, further simulation studies will allow us to predict how much more medical resources might be needed to reduce the number of busy days below a particular percent in a year.

The features of the box plots presented herein may differ from those of other emergency medical systems with different socioeconomic situations and emergency medical service utilization habits. Furthermore, as patient visit volumes can change over time, the annual change should be taken into consideration [1]. Nevertheless, the box plots verify the daily variability of ED visit volumes.