a b s t r a c t

Background: Since providing timely care is the primary concern of emergency departments (EDs), long waiting times increase patient dissatisfaction and adverse outcomes. Especially in overcrowded ED environments, emer- gency care quality can be significantly improved by developing predictive models of patients’ waiting and treatment times to use in ED operations planning.

Methods: Retrospective data on 37,711 patients arriving at the ED of a large urban hospital were examined. Or- dinal logistic regression models were proposed to identify factors causing increased waiting and treatment times and classify patients with longer waiting and treatment times.

Results: According to the proposed ordinal logistic regression model for waiting time prediction, age, Arrival mode, and ICD-10 encoded diagnoses are all significant predictors. The model had 52.247% accuracy. The model for treatment time showed that in addition to age, arrival mode, and diagnosis, triage level was also a significant predictor. The model had 66.365% accuracy. The model coefficients had negative signs in the corre- sponding models, indicating that waiting times are negatively related to treatment times.

Conclusion: By predicting patients’ waiting and treatment times, ED workloads can be assessed instantly. This enables ED personnel to be scheduled to better manage demand supply deficiencies, increase patient satisfaction by informing patients and relatives about expected waiting times, and evaluate performances to improve ED op- erations and emergency care quality.

(C) 2021

1. Introduction

The significant increase in admissions to emergency departments (EDs) has led to ED overcrowding, which is a severe problem for healthcare systems worldwide. Overcrowding in EDs can have serious Negative outcomes for patients and ED staff, such as reduced patient sat- isfaction [1], increased mortality [2], increased Ambulance diversions [3], reduced staff morale and attention [4], and increased waiting and treatment times [5,6]. Of these issues, researchers have paid more atten- tion to waiting and treatment times since EDs are highly sensitive to increased delays and waiting. Therefore, any interventions that can re- duce these improve ED operations significantly.

One main goal of ED operations is minimizing the initial waiting time, defined as the time between a patient’s registration and initial physician evaluation. Various studies have set targets for patients’ initial waiting times in EDs. According to the triage system Emergency Sever- ity Index (ESI) version 4, which categorizes patients based on their

* Corresponding author at: Izmir Bakircay University Cigli Training and Research Hospital, Department of Emergency Medicine, Turkey.

E-mail addresses: [email protected] (M.G. Ataman), [email protected] (G. Sariyer).

priorities from level 1 to 5, treatment patients should be treated within 1, 10, 30, 60, and 120 min respectively [7]. Treatment time, defined as the time from first physician assessment to final Disposition decision, is another quality indicator for ED operations. To reduce or minimize waiting and treatment times, the first essential step is detailed record- ing and continuous reporting [8,9], which has become possible by inte- grating electronic warehouses into ED systems. Using these electronic medical records (EMR), researchers have analyzed the variables causing increased delays and waiting [10-15], while some studies have devel- oped prediction models on ED time-related quality indicators [16-18]. Current studies also use data mining of EMR to analyze, model, and re- duce ED times. This can help to reduce ED overcrowding and improve patient flow [19-22].

EMR systems allow hospitals and other health care organizations to collect a significant amount of data. However, most is raw data that is poorly structured and thus unsuitable for analysis [23]. Health data also has a very complicated structure [24] so the health care industry re- quires the application of data mining techniques to examine and ana- lyze the raw data to extract useful information and knowledge. Based on the knowledge obtained, health organizations can take important decisions cost-effectively while providing complementary clinical solutions.

https://doi.org/10.1016/j.ajem.2021.02.061

0735-6757/(C) 2021

The current study uses data mining techniques to develop models for classifying ED patients regarding their waiting and treatment times. The proposed models identify which characteristics are associ- ated with longer waiting and treatment times in EDs. Grouping patients with similar characteristics and generating predictions based on both waiting and treatment times of each group can significantly improve ED operational planning.

1. Methods

The data for this retrospective study were obtained from the emer- gency department of a tertiary-level public hospital in Izmir, Turkey. As one of the main region’s main urban hospitals, it receives most of the city’s emergency visits. In this hospital, initial assessment is initiated by a triage nurse who enters each patient’s demographic characteristics and vital signs into the database. This ED uses the 3-level ESI system, which categorizes patients as red (urgent), yellow (emergent), or green (non-urgent). Since 2018, this ED has added an additional room for trauma patients. Thus, through registering arriving patients and re- cording their data, the triage nurse assigns them to one of four rooms. Higher-priority patients are generally directed to examination rooms for immediate treatment by physicians. Lower-priority patients must wait in waiting rooms until their turn comes. Patients are examined and treated in one of the four rooms by specialists and/or general prac- titioners, who prescribe further diagnostic measures for the nurses to perform. The physicians who treat the patients assign the diagnosis based on the International Classification of Diseases Version 10 (ICD-10), which is a medical classification system developed by the World Health Organization with codes for diseases, signs, and symp- toms, abnormal findings, complaints, and external causes of injury or diseases [25]. When the required treatment is given, the patient is discharged from ED (to go home, be admitted to a hospital polyclinic or service, or transfer to another hospital).

In 2018, this ED received around 1250 daily arrivals. It was estimated that one month’s data was sufficient for the study’s design and purpose in that the observed variations in patient demographics, and waiting and treatment times are generalizable, and that data mining techniques can be applied. August 2018 was randomly selected for the study data.

The local institutional review board approved the study.

• 1. Study variables

The independent variables were the patients’ demographic charac- teristics (gender and age), and visit characteristics (triage category, arrival mode, and ICD-10 diagnosis). Gender, arrival mode, and diagno- sis were measured on nominal scales while age and triage category were ordinal. For referral diagnosis, first-level ICD-10 was used, which has the form LXX, where L denotes a letter and X denotes a digit from 0 to 9. The respective categories of each variable were defined as follows:

• • • Gender: male, female
    • Age: [0-14, 15-64, 65-84], >=85
    • Triage category: red, yellow, green, trauma
    • Arrival mode: walk-in, by ambulance
    • ICD-10 diagnosis: Certain infectious and parasitic diseases “A00-B99”, Neoplasms “C00-D49”, Diseases of the blood and blood-forming organs, and certain disorders involving the immune mechanism “D50-D89”, Endocrine, nutritional, and metabolic diseases “E00- E89”, Mental, behavioral and neurodevelopMental disorders “F01- F99”, Diseases of the nervous system “G00-G99”, Diseases of the eye and adnexa “H00-H59”, Diseases of the ear and mastoid process “H60-H95”, Diseases of the circulatory system “I00-I99”, Diseases of the respiratory system “J00-J99”, Diseases of the digestive system “K00-K95”, Diseases of the skin and subcutaneous tissue “L00-L99”, Diseases of the musculoskeletal system and connective

tissue “M00-M99”, Diseases of the Genitourinary system “N00-N99”, Pregnancy, childbirth, and the puerperium “O00-O9A”, Certain conditions originating in the perinatal period “P00-P96”, Congenital malformations, deformations, and chromosomal abnormalities “Q00-Q99”, Symptoms, signs, and abnormal clinical and laboratory findings, not elsewhere classified “R00-R99”, Injury, poisoning, and certain other consequences of external causes “S00-T88”, External causes of morbidity “V00-Y99”, Factors influencing health status and contact with health services “Z00-Z99”

Initial waiting time, hereafter waiting time, and treatment time were the two output variables used as quality indicators for ED performance. These variables were measured in minutes on a continuous scale for univariate analysis. For the proposed classification models, waiting and treatment times were converted into ordinal levels with thresholds for the levels of waiting time as 10 and 60 min respectively. Thresholds for the treatment time levels were defined as 10 and 120 min. The cat- egorical definitions of the output variables of the two models were thus defined as follows:

• • • Waiting time: “Patients who wait less than 10 minutes”, “Patients whose waiting time is in the range of 10-60 minutes”, and “Patients who wait more than 60 minutes”
    • Treatment time: “Patients who are treated for up to 10 minutes”, “Pa- tients whose treatment time is in the range of 10-120 minutes”, and “Patients who are treated for longer than 120 minutes”
  1. Statistical analysis and outcome measures

Standard descriptive statistics were used to characterize the study sample with frequency and percentage distributions for the catego- ries of independent and dependent variables. Waiting and treatment times were described with mean and standard deviation for each cat- egory. This analysis was also used to determine the appropriateness of the selected input variables since the significance of mean difference between categories of them was tested. Independent sample t-tests were used to test the significance of the differences between the cat- egories of the binary input variables (gender, arrival mode) while ANOVA was used to test for the other input variables (age, triage cat- egory and ICD-10 codes). P-values less than 0.05 were considered as statistically significant. The relationship between waiting and treat- ment times for each category was tested by Pearson correlation analysis.

Ordinal logistic regression analysis was used to model the ED waiting and treatment times. The model classified patients into pre- determined, ordered categories of the output variable. Since this study aimed to identify the characteristics of patients with shorter and longer waiting and treatment times, the model was appropriate. Additionally, since the output variables of the models, waiting and treatment times, were classified into three different levels which were ordered, this method was chosen. The performances of the proposed classification models are reported based on model accuracies. This metric shows the percentage of correctly classified instances, formulated as follows:

accuracy = Nc ?100%

N_c + N_ic

where N_c shows the number of correct classifications and N_ic represents the number of incorrect classifications. For the proposed waiting time model, the number of correct predictions, N_c, was given by the number of patients whose actual waiting time (<10, 10-59, or >=60 min) fell within the same predicted waiting interval category. A similar calcula- tion was made for treatment time accuracy. In addition to reporting model accuracies, classification tables are presented to show in which categories the models performed better or worse. The statistical analysis was conducted using SPSS statistical software (version 16.0).

1. Results
   1. Univariate analysis

Table 1 presents the frequency and percentage distributions based on the model input variables.

Table 1 shows nearly equal numbers of male and female ED arrivals while almost 10% of arrivals were elderly patients who are at least 65 years old. Almost all arrivals were walk-in patients. Just over half of arrivals were non-urgent, and triaged into green areas accordingly. The most frequent ICD-10 diagnosis group codes were M00-M99 followed by J00-J99.

Table 2 presents the descriptive statistics of the waiting and treat- ment times for each category of the input variables. It also shows the significance of the differences between the categories for both waiting and treatment times.

Table 2 shows that males have significantly longer waiting times but significantly shorter treatment times. Waiting times decrease significantly with increasing age whereas treatment times increase significantly with age. Patients arriving by ambulance wait signifi- cantly less time than walk-in patients but their treatment times are significantly longer. Red room patients (urgent) wait the shortest time, followed by trauma, yellow, and green room patients. Red room patients also have the longest treatment times, followed in order by yellow, trauma, and green room patients. Patients with ICD-10 diagnoses H60-H95, J00-J99, and L00-L99 wait significantly longer than others but have significantly shorter treatment times. Conversely, patients with ICD-10 diagnoses E00-E89, O00-O9A, and P00-P96 wait significantly less time but have significantly longer treatment times. Overall, there are statistically significant differ- ences in waiting and treatment times between the categories of all the input variables (p < 0.05).

Table 3 shows the correlations between waiting and treatment times for each category along with the statistical significance.

Table 3 shows that, for each category of the input variables gen-

der, age, arrival mode and triage category, there is a significant negative correlation between waiting and treatment times. That is, as waiting times decrease, treatment times are expected to in- crease and vice versa. This pattern is similar for the ICD-10 codes, except for A00-B99, C00-D49, H00-H59, O00-O9A, P00-P96 and Z00-Z99.

Table 4 presents the frequency and percentage distributions for the three ordinal categories of waiting and treatment times.

• 1. Ordinal logistic regression (OLR) analysis

The ordinal logistic regression analysis classified patients into one of the predefined output categories based on waiting and treatment times. Table 5 presents the results of the two models.

The OLR analysis showed that age, arrival mode, and ICD-10 diagno- sis all significantly predicted both waiting and treatment times. How- ever, the signs of parameter estimates were opposite in the two proposed models. That is, the parameters for age, arrival mode, and ICD-10 codes had negative signs in the waiting time model whereas they were positive in the treatment time model. More specifically, this shows that increasing age decreases the likelihood of a patient being classified in the longer waiting Time groups but increases the likelihood of being classified in the longer treatment time groups. Likewise, arriv- ing by ambulance significantly decreases the likelihood of being classi- fied in the longer waiting time groups but increases the likelihood of being classified in the longer treatment time groups.

Two variables differed in Predictive power for the two models. First,

while gender was not a significant predictor (p-value = 0.261) in the waiting time model, it was a significant positive predictor in the treat- ment time model (p-value = 0.056, at 90% confidence interval). That is, a female patient is significantly more likely than a male patient to be classified in the longer treatment groups. Second, triage category produced an unexpected result. Triage category had no significant effect on waiting time classification (p-value = 0.153) whereas it was a signif- icant negative predictor in the treatment time model (p-value = 0.000). That is, a patient categorized as red (i.e. urgent) is significantly less likely than a green group patient (i.e. non-urgent) to be classified in the longer treatment time groups. Finally, interpreting the model signs for ICD-10 diagnoses was more complicated because the variable is not ordinal.

Table 6 presents the classification matrixes for the predictive perfor- mance of the OLR models for waiting and treatment time.

Table 6 shows that each model performed significantly better than classifying patients into any of the three output groups ran- domly when the probability of correct classification would be just 33.3%. Additionally, the treatment time classification was more ac- curate (accuracy = 66.365%) than the waiting time classification (accuracy = 52.247%). The waiting time model gave the most accu- rate predictions for patients in the 10-60-min group, with 10,815 of 15,064 (71.794%) patients being correctly classified (see Table 4). The treatment time model gave the most accurate predictions for patients in the <10-min group, with 18,930 of 20,928 (90.453%)

Table 1

Frequency and percentage distributions based on the model inputs

Variable	Categories	Frequency	Percentage	Variable	Categories	Frequency	Percentage
Gender	Male	18,349	48.657	ICD-10 Diagnosis	A00-B99	1232	3.267
	Female	19,362	51.343		C00-D49	23	0.061
Age	[0-14]	6110	16.202		D50-D89	88	0.233
	[15-64]	28,004	74.259		E00-E89	103	0.273
	[65-84]	3153	8.361		F01-F99	427	1.132
	>=85	444	1.177		G00-G99	581	1.541
Arrival mode	Walk in	36,034	95.553		H00-H59	76	0.202
	Ambulance	1677	4.447		H60-H95	487	1.291
Triage levels	Trauma	5401	14.322		I00-I99	642	1.702
	Red	1472	3.903		J00-J99	7240	19.199
	Yellow	11,435	30.323		K00-K95	2143	5.683
	Green	19,403	51.452		L00-L99	971	2.575
					M00-M99	7779	20.628
					N00-N99	1859	4.930
					O00-O9A	44	0.117
					P00-P96	76	0.202
					R00-R99	5520	14.638
					S00-T88	1319	3.498
					V00-Y99	1771	4.696
					Z00-Z99	5330	14.134

Table 2

Descriptive statistics for waiting and treatment times by input variable category and significances of differences

Variable	Categories	Waiting Time (minutes)			Treatment Time (minutes)
		u, ?	p-value		u, ?	p-value
Gender	Male	41.262, 61.859	0.035		61.458, 116.288	0.000
	Female	39.957, 58.334			69.310, 121.500
Age	[0-14]	43.847, 60.009	0.000		56.091, 111.234	0.000
	[15-64]	41.502, 61.529			58.922, 111.025
	[65-84]	28.944, 46.561			125.015, 158.404
	>=85	21.153, 36.133			186.349, 185.567
Arrival mode	Walk in	42.286, 60.826	0.000		58.077, 109.310	0.000
	Ambulance	4.205, 16.729			224.769, 188.327
Triage levels	Trauma	30.125, 61.430	0.000		76.120, 106.475	0.000
	Red	4.658, 13.348			242.763, 189.537
	Yellow	39.705, 53.837			103.679, 128.939
	Green	46.755, 63.846			26.575, 84.162
ICD-10 Encoded Diagnosis	A00-B99	20.986, 43.399	0.000		22.393, 78.753	0.000
	C00-D49	36.253, 64.168			186.348, 191.324
	D50-D89	30.394, 42.247			111.662, 129.759
	E00-E89	21.463, 36.293			133.606, 150.671
	F01-F99	56.845, 69.349			46.033, 95.161
	G00-G99	51.367, 65.976			65.180, 141.593
	H00-H59	32.054, 41.082			48.715, 95.381
	H60-H95	59.180, 67.672			21.292, 55.374
	I00-I99	37.059, 61.368			120.110, 188.661
	J00-J99	58.704, 69.079			32.819, 93.957
	K00-K95	49.051, 59.829			54.247, 114.318
	L00-L99	60.408, 72.229			31.273, 84.227
	M00-M99	39.026, 57.512			58.907, 107.643
	N00-N99	37.650, 51.299			105.345, 137.837
	O00-O9A	12.342, 22.166			147.535, 109.492
	P00-P96	9.532, 15.021			119.710, 109.751
	R00-R99	37.025, 47.460			140.018, 141.356
	S00-T88	37.789, 57.346			54.069, 101.910
	V00-Y99	33.019, 56.959			75.334, 113.448
	Z00-Z99	21.331, 57.368			43.901, 103.143

Table 3

Correlations between waiting and treatment times for each input category and statistical significances

Table 4

Frequency and percentage distributions by waiting and treatment times

	Waiting time				Treatment time
	Categories (in minutes)	Frequency	Percentage		Categories (in minutes)	Frequency	Percentage
	<10	14,391	38.161		<10	20,928	55.496
	[10-60)	15,064	39.946		[10-120)	8904	23.611
	>=60	8256	21.893		>=120	7879	20.893

Variable	Categories	Correlation coefficient	Variable	Categories	Correlation coefficient
Gender	Male	-0.151??	ICD-10 Encoded	A00-B99	-0.016
	Female	-0.173??	Referral Diagnosis	C00-D49	-0.281
Age	[0-14]	-0.149??		D50-D89	-0.253?
	[15-64]	-0.151??		E00-E89	-0.315??
	[65-84]	-0.204??		F01-F99	-0.266??
	>=85	-0.242??		G00-G99	-0.276??
Arrival	Walk in	-0.138??		H00-H59	-0.166

mode Triage

levels

Ambulance -0.189? H60-H95 -0.166??

Trauma -0.182?? I00-I99 -0.252??

Red -0.073? J00-J99 -0.170??

Yellow -0.189?? K00-K95 -0.177??

Green -0.067? L00-L99 -0.122?? M00-M99 -0.159?? N00-N99 -0.175??

O00-O9A -0.041

P00-P96 0.037 R00-R99 -0.222?? S00-T88 -0.227?? V00-Y99 -0.295?? Z00-Z99 0.009

1. Discussion

This study proposed two models to classify patients based on their waiting and treatment times in the emergency department of a metro- politan hospital in Izmir, Turkey. The first noteworthy result concerns the huge volume of ED arrivals as the patient volume for this hospital is considerably greater than for many of other EDs, both in Turkey and other countries [26-29]. This indicates that overcrowding is a major problem for this ED, so models to plan operations more efficiently may lead to considerable improvements in patient care.

Another important result is that more than half of the arrivals are

* corr. significant at 95%

?? corr. sig. at 99%

patients being correctly classified (see Table 4). The two models’ least accurate predictions where for the >=60-min waiting time group (only 760 of 8256-9.205%- patients correctly classified) and the 10-120-min treatment group (only 925 of 8904-10.389%- pa- tients correctly classified).

triaged into green areas for non-urgent patients, which is a higher proportion than reported for other EDs [4,12,30,31]. Conversely, con- siderably fewer patients arrive by ambulance (5%) than in most other EDs studied [4,32,33]. This indicates another important problem in EDs, particularly in Turkey: unnecessary admissions leading to redun- dant use of EDs. Non-urgent arrivals are most frequently from four particularly common diagnosis groups, namely diseases of the respi- ratory system “J00-J99”, musculoskeletal system [22,33,34], symp- toms, signs and abnormal clinical and laboratory findings “R00-R99”,

Table 5

OLR models results

Input variable OLR model for waiting time OLR model for treatment time

Parameter estimate p-value 95% confidence interval Parameter estimate p-value 95% confidence interval

		Lower bound	Upper bound			Lower bound	Upper bound
Gender -0.022	0.261	-0.061	0.016	0.041	0.056	-0.001	0.084
Age -0.116	0.000	-0.154	-0.079	0.151	0.000	0.111	0.190
Arrival mode -3.398	0.000	-3.616	-3.180	1.215	0.000	1.095	1.335
Triage level 0.016	0.153	-0.006	0.037	-0.950	0.000	-0.973	-0.926
ICD-10 diagnosis -0.067	0.000	-0.071	-0.063	0.054	0.000	0.049	0.058

model fitting information: Chi-square = 3740.277; p-value: 0.000 Model fitting information: Chi-square = 10,504.755; p-value: 0.000

Model summary: Cox & Snell R square = 0.194; Nagelkerke R square = 0.207

Model summary: Cox & Snell R square = 0.343; Nagelkerke R square = 0.382

Table 6

Classification matrixes of the OLR models

ambulance, and having higher ICD-10 codes predict shorter waiting times. For treatment time prediction, the coefficients are positive, indi-

cating the opposite interpretation. That is, these factors increase

OLR model for waiting time OLR model for treatment time

treatment time.

Waiting time categories

<10 [10-60) >=60 Treatment time

categories

<10 [10-120) >=120

The prediction accuracies of the waiting and treatment time models were respectively as 52.247% and 66.365%. These performances were

<10 8128 5849 414 <10 18,930 545 1453

[10-60) 3990 10,815 259 [10-120) 5479 925 2500

>=60 2050 5446 760 >=120 2409 298 5172

Accuracy: 52.247% Accuracy: 66.365%

and factors influencing health status and contact with health services

“Z00-Z99” [35].

Regarding the waiting and treatment times, with the analysis re- veals gender differences in that female arrivals have shorter waiting times but longer treatment times than males. This may be related to human psychological factors that need further investigation. Similar to previous studies [12,13,22,30], waiting times decreased significantly with both age, seriousness of triage category, and for patients arriving by an ambulance whereas treatment times significantly increased in these groups.

This study makes a novel contribution to the literature by analyzing the effects of these factors comparatively on both waiting and treatment times. It also analyzed the relationship between waiting and treatment times for each level of the input variables, finding significant negative relationships between waiting and treatment times for the majority of groups. That is decreased waiting times can predict increased treatment times in EDs. Regarding ICD-10 diagnosis, significantly lower waiting and higher treatment times were recorded for endocrine, nutritional, and metabolic diseases “E00-E89”, diseases of the circulatory system “I00-I99”, pregnancy, childbirth, and the puerperium “O00-O9A”, certain conditions originating in the perinatal period “P00-P96”, and symptoms, signs and abnormal clinical and laboratory findings, not elsewhere classified “R00-R99”.

Most of these univariate analysis results were supported by the two

OLR models proposed here. The models showed that age, arrival mode, and ICD-10 codes all significant predicted patients waiting and treat- ment times. More specifically, triage category significantly predicted treatment time but not waiting time. Despite the significant gender dif- ferences in both waiting and treatment times, gender was not a signifi- cant predictor in either model. This finding suggests that the effects of age, arrival mode, and ICD-10 diagnosis suppress the effects of triage category and gender in predicting waiting time while the effect of gender is suppressed by the other four variables in predicting treatment time.

In addition to the significance of the predictors, the signs of the model coefficients were also useful in interpreting the model results. For predicting waiting time, the negative coefficients for gender, age, ar- rival mode, and ICD-10 codes show that being female, older, arriving by

somehow lower compared to existing literature [22] classifying ED pa- tients based on their LOS values and having around 69% accuracy levels. However, it should be noted that while this study presents waiting and treatment time classification for three levels meaning that patients were categorized into three different time intervals, the cited study was designated to classify patients into one of two levels. From this viewpoint, since the performances of the classification models generally decrease when number of categories of dependent variable increase, this study should be seen as improving the prediction performances of the existing literature.

1. Limitations

There are several limitations. First, the data come from only one institution, which limits the ability to generalize the findings to other EDs. Secondly, although the sample size was adequate for the research design, the models could not include monthly or sea- sonal variations, particularly for frequency distributions based on ICD-10 diagnoses. Third, the input variables are all widely used model parameters in the ED literature. However, other parameters could be added to improve the models’ Predictive performance. Fi- nally, since this study used secondary data from the EMR of one hos- pital, which limited the researchers to using only the recorded variables in the models.

1. Conclusion

This study identified the main factors increasing or reducing waiting and treatment times in Eds. Two models were proposed that very accurately predicted patients’ waiting and treatment times based on data from a large urban ED in Turkey. This shows that OLR models can be used to improve prediction performance in ED contexts. Focusing on the factors that increase patient waiting and treatment times may provide insights into practice that can be used to improve the Efficiency of ED operations. This in turn can in- crease patient well-being and satisfaction. More specifically, patient satisfaction could be improved by informing each patient about their expected waiting and treatment times as soon as they register. The proposed models should also be used as a decision support tool to help ED decision makers plan ED operations and manage ED re- sources more effectively, based on the total estimated waiting and treatment times in each ED area. This may also accelerate patient flow so as to reduce ED overcrowding.

Ethics committe approval

Ethics committe approval of this study was received from Izmir Katip Celebi University Non-Interventional Clinical Studies Institutional Review Board at its board meeting 19.04.2017.

Declaration of Competing Interest

The authors declares no conflict of interest.

References

1. Boudreaux ED, Ary RD, Mandry CV, McCabe B. Determinants of patient satisfaction in a large, municipal ED: the role of demographic variables, visit characteristics, and patient perceptions. Am J Emerg Med. 2000;18(4):394-400.
2. Richardson DB. Increase in patient mortality at 10 days associated with emergency department overcrowding. Med J Aust. 2006;184(5):213-6.
3. Fatovich DM, Nagree Y, Sprivulis P. access block causes emergency department overcrowding and ambulance diversion in Perth, Western Australia. Emerg Med J. 2005;22(5):351-4.
4. Graham B, Bond R, Quinn M, Mulvenna M. Using data mining to predict hospital ad- missions from the emergency department. IEEEAccess. 2018;6:10458-69.
5. McCarthy ML, Zeger SL, Ding R, Levin SR, Desmond JS, Lee J, et al. Crowding delays treatment and lengthens emergency department length of stay, even among high- acuity patients. Ann Emerg Med. 2009;54(4):492-503.
6. Sariyer G, Ataman MG, Kiziloglu I. Analyzing main and interaction effects of length of stay determinants in emergency departments. Int J Health Policy. 2020;X(X):1-8. https://doi.org/10.15171/ijhpm.2019.107.
7. Gilboy N, Tanabe P, Travers D. Emergency Severity Index, Version 4 Implementation Handbook. Rockville: Agency for Healthcare Research and Quality; 2005.
8. Wilper AP, Woolhandler S, Lasser KE, McCormick D, Cutrona SL, Bor DH, et al. Waits to see an emergency department physician: US trends and predictors, 1997-2004. Health Aff. 2008;27(Suppl1):84-95.
9. Ludwick DA, Doucette J. Adopting electronic medical records in primary care: les- sons learned from health information systems implementation experience in seven countries. Int J Med Inform. 2009;78(1):22-31.
10. Pathan SA, Bhutta ZA, Moinudheen J, Jenkins D, Silva AD, Sharma Y, et al. Marginal analysis in assessing factors contributing time to physician in the Emergency De- partment using operations data. Qatar Med J. 2016;18:1-17.
11. Mahmoodian F, Eqtesadi R, Ghareghani A. Waiting times in emergency department after using the emergency severity index triage tool. Arch Trauma Res. 2014;3(4): 1-5.
12. Horwitz LI, Green J, Bradley EH. US emergency department performance on wait time and length of visit. Ann Emerg Med. 2010;55(2):133-41.
13. Sariyer G, Ataman MG, Kiziloglu I. Factors affecting length of stay in the emergency department: a research from an operational viewpoint. Int J Healthc Manag. 2018. https://doi.org/10.1080/20479700.2018.1489992.
14. Cheng I, Zwarenstein M, Kiss A, Castren M, Brommels M, Schull M. Factors associated with failure of emergency wait-time targets for high acuity discharges and intensive care unit admissions. Can J Emerg Med. 2018;20(1):112-24.
15. Khanna S, Sier D, Boyle J, Zeitz K. Discharge timeliness and its impact on hospital crowding and emergency Department flow performance. Emerg Med Australas. 2016;28(2):164-70.
16. Sun Y, Teow KL, Heng BH, Ooi CK, Tay SY. Real-time prediction of waiting time in the emergency department, using quantile regression. Ann Emerg Med. 2012;60(3): 299-308.
17. Bruballa E, Wong A, Epelde F, Rexachs D, Luque E. A model to predict length of stay in a hospital emergency department and enable planning for non-critical patients admission. Int J Integr Care. 2016;16(6). https://doi.org/10.5334/ijic.2967.
18. Rathlev NK, Chessare J, Olshaker J, Obendorfer D, Mehta SD, Rothenhaus T, et al. Time series analysis of variables associated with daily mean emergency department length of stay. Ann Emerg Med. 2007;49(3):265-71.
19. D’Etienne JP, Zhou Y, Kan C, Shaikh S, Ho AF, Suley E, et al. Two-step predictive model for early detection of emergency department patients with prolonged stay and its management implications. Am J Emerg Med. 2020. https://doi.org/10.1016/ j.ajem.2020.01.050.
20. Soni V. Predicting Length of Stay of Emergency Department Patients Using Tree- Based machine learning Models; 2021 (Doctoral dissertation, Wayne State University).
21. Rahman MA, Honan B, Glanville T, Hough P, Walker K. Using data mining to predict emergency department length of stay greater than 4 hours: derivation and single- site validation of a decision tree algorithm. Emerg Med Australas. 2019. https:// doi.org/10.1111/1742-6723.13421.
22. Sariyer G, Tasar CO, Cepe GE. Use of data mining techniques to classify length of stay of emergency department patients. BAMS. 2019;15(1):1-15. https://doi.org/10. 1515/bams-2018-0044.
23. Tekieh MH, Raahemi B. Importance of data mining in healthcare: a survey. Proceed- ings of the 2015 IEEE/ACM International Conference on Advances in Social Networks Analysis and Mining; 2015. p. 1057-62.
24. Tomar D, Agarwal S. A survey on data mining approaches for healthcare. Int J Bio-Sci

Bio-Technol. 2013;5(5):241-66.

1. World Health Organization. International Classification of Diseases (ICD) Informa- tion Sheet; 2019 (Retrieved from: 10 October 2019).
2. Yucesan M, Gul M, Celik E. A multi-method patient arrival forecasting outline for hospital emergency departments. Int J Healthc Manag. 2018:1-13.
3. Visintin F, Caprara C, Puggelli F. Experimental design and simulation applied to a paediatric emergency department: a case study. Comput Ind Eng. 2019;128: 755-81.
4. Shah S, Murray J, Mamdani M, Vaillancourt S. Characterizing the impact of snowfall on patient attendance at an urban emergency department in Toronto, Canada. Am J Emerg Med. 2019;37(8):1544-6.
5. Whitt W, Zhang X. Forecasting arrivals and occupancy levels in an emergency de- partment. Oper Res Health Care. 2019 Jun 1;21:1-8.
6. Storm-Versloot MN, Vermeulen H, Lammeren N, Luitse JSK, Goslings JC. Influence of the Manchaster triage system on waiting time, treatment time, length of stay and patient satisfaction; a before and after study. Emerg Med J. 2014;31:13-8.
7. Bird S, Thomson C, Williams KE. Primary contact physiotheraphy services reduce waiting and treatment times for patients presenting with musculoskeletal condi- tions in Australian emergency departments: an observational study. J Physiother. 2016;62:209-14.
8. Golmohammadi D. Predicting hospital admissions to reduce emergency department boarding. Int J Prod Econ. 2016;182:535-44.
9. Meehan E, Williams K, Reid SM, Freed GL, Babl FE, Sewell JR, et al. Comparing emer- gency department presentations among children with cerebral palsy with general childhood presentations: a data linkage study. Dev Med Child Neurol. 2017;59 (11):1188-95.
10. Khatri KL, Tamil LS. Early detection of peak demand days of chronic respiratory dis- eases emergency department visits using artificial neural networks. IEEE J Biomed Health. 2017 Apr 26;22(1):285-90.
11. Sariyer G, Ataman MG. The likelihood of requiring a diagnostic test: classifying emergency department patients with logistic regression. Health Inf Manag J. 2020. https://doi.org/10.1177/1833358320908975.