American Journal of Emergency Medicine (2012) 30, 1884-1894

Original Contribution

Patterns and factors associated with intensive use of ED services: implications for allocating resources

Jennifer Prah Ruger PhD ^a,?, Lawrence M. Lewis MD ^b, Christopher J. Richter MD ^b

^aYale School of Public Health, Health Policy and Administration, New Haven, CT 06520

^bWashington University School of Medicine, St Louis, MO

Received 10 January 2012; revised 29 February 2012; accepted 1 April 2012

Abstract

Aim: This study aims to better understand the patterns and factors associated with the use of emergency department (ED) services on high-volume and intensive (defined by high volume and high-patient severity) days to improve resource allocation and reduce ED overcrowding.

Methods: This study created a new index of “intensive use” based on the volume and severity of illness and a 3-part categorization (normal volume, high volume, intensive use) to measure Stress in the ED environment. This retrospective, cross-sectional study collected data from hospital clinical and financial records of all patients seen in 2001 at an urban academic hospital ED.

Results: Multiple logistic regression models identified factors associated with high volume and intensive use. Factors associated with intensive days included being in a motor vehicle crash; having a gun or Stab wound; arriving during the months of January, April, May, or August; and arriving during the days of Monday, Tuesday, or Wednesday. Factors associated with high-volume days included falling from 0 to 10 ft; being in a motor vehicle crash; arriving during the months of January, April, May, or August; and arriving during the days of Monday, Tuesday, or Wednesday.

Conclusion: These findings offer inputs for reallocating resources and altering staffing models to more efficiently provide high-quality ED services and prevent overcrowding.

(C) 2012

Introduction

Emergency department (ED) overcrowding is an interna- tional problem that affects both rural and urban areas [1,2]. In one study in the United States, 91% of ED directors reported overcrowding in the ED as a problem, with nearly 40% reporting overcrowding on a daily basis. Overcrowding was defined as patients in hallways, full occupancy of ED beds, full waiting rooms, and long waiting times to receive

* Corresponding author.

E-mail addresses: [email protected] (J.P. Ruger), [email protected] (L.M. Lewis), [email protected] (C.J. Richter).

treatment [3]. Another study created a model to quantify overcrowding in academic institutions and found that overcrowding occurred in academic medical centers an average of 35% of the time [4].

In recent years, studies have documented the increasing problem of ED overcrowding in various communities [3,5-8] and have shown that overcrowding negatively affects patient health [2,9-11]. Crowding was found to compromise 2 domains of quality of care: safety and timeliness [9]. A literature review of studies addressing the effects of crowding on clinical outcomes found that ED crowding was associated with negative clinical outcomes such as in-hospital mortality as well as negative effects on clinically important Processes of care such as longer wait

0735-6757/$ – see front matter (C) 2012 http://dx.doi.org/10.1016/j.ajem.2012.04.001

Acuity level rating”>times to treatment in patients with time-sensitive condi- tions. time delays in access to treatment can increase the risk of poorer outcomes in time-sensitive conditions, including pain and anxiety [10,11].

There is a growing consensus in the literature that ED crowding is caused by system-wide factors, in particular the ability of hospitals to handle patient ftows through the ED [2,10-14]. A study of strategies to reduce ED crowding in Rochester, New York, found that strategies such as Ambulance diversion had little effect on ED overcrowding, whereas strategies that addressed external factors such as ftexibility of inpatient resources and the use of transition teams to care for patients awaiting beds were more successful [5]. Thus, appropriate resource allocation should be a priority when addressing overcrowding in EDs. As one of the most expensive and adjustable costs, staffing should be allocated to reduce waiting times and control ED overcrowding by increasing staffing levels during the busiest times of the day, week, or year. This has been attempted in the pediatric ED setting with some success using on-call teams during the viral seasons [15].

Finding a way to measure ED crowding in a valid and reliable way is a first step to addressing the problem [16]. Several studies have explored the existence of patterns of increased ED use (higher patient volume) using complex Mathematical modeling, calendar, and weather data to better predict when overcrowding will occur [17-21]. One study used intensity of service use as a part of a statistical formula to calculate physician staffing hours. A multifactorial formula that included intensity of services performed predicted ED staffing needs better than patient volume [22]. A recent study looking at patient acuity measures and overcrowding at a children’s hospital in Arkansas found a strong correlation between overcrowding measures (length of stay and left- without-being-seen) and number of patients admitted, but a poor correlation between overcrowding measures and total ED census [12]. This suggests that volume alone may not be the only factor contributing to overcrowding.

This study aims to better understand the patterns and factors associated with the use of ED services on high- volume and intensive (high volume and severity) days to improve ED capacity and resource allocation and reduce ED overcrowding. We created a new index of intensive use based on volume (in excess of average use) and severity of illness (higher levels of acuity) and a 3-part categorization (normal volume, high volume, and intensive use) to measure stress in the ED environment. We attempted to link various independent factors to our index and categories of ED stress.

Materials and methods

A retrospective, cross-sectional study was conducted at a large, 1200-bed level 1 trauma center at an urban academic hospital ED in St Louis, Missouri. Hospital clinical and admissions data and financial records were obtained for

every patient visiting the ED between January 1, 2001, and December 31, 2001. Combining clinical and admissions data with financial data provided a comprehensive set of ED Patient variables. Hospital visits with incomplete acuity level data (n = 2500) or financial data (n = 5152) were excluded from analysis. Of the 80 209 total hospital visits, 71 287 (89%) visits had complete data on all variables studied and were included in analyses. Although there did not appear to be systematic differences in data collection, missing data were significantly more likely to be from patients who were younger, were male, and left the ED against medical advice (AMA) and were less likely to be from patients who were acutely ill or injured. This study was approved by the Barnes- Jewish Hospital and the Human Studies Committee of Washington University in St Louis, Missouri.

Dependent variable

The primary outcome measures for this study were the volume and triage acuity level of ED visits.

Acuity level rating

When patients arrive at the ED, trained Triage nurses assign patients a triage acuity level using a 5-level Canadian ED Triage and Acuity Scale [23]: (1) acuity level A (resuscitation): threatening conditions requiring immediate evaluation and management by an emergency physician; (2) acuity level B (emergent): potential threatening conditions requiring evaluation by a physician within 15 minutes (3) acuity level C (urgent): conditions that cause significant discomfort and/or could potentially become serious, requir- ing evaluation by a physician within 30 minutes; (4) acuity level D semi-urgent conditions that might deteriorate or benefit from intervention or reassurance within 60 minutes but do not require rapid evaluation by a physician; and (5) acuity level E non-urgent acute but nonurgent conditions that may be seen in a delayed fashion and could be referred for evaluation to other areas of the health care system. Acuity level data were analyzed in bivariate association with patient Sociodemographic and clinical characteristics.

High-volume rating

The volume of visits was obtained from the hospital’s billing and administrative records. A dummy variable was created for “high volume” days in which the total ED census was greater than 110% of the average Daily census. The average number of visits was 220/d; thus, high-volume days were those for which the average daily census was more than 240 visits/d.

Intensive-use rating

A second dummy variable was created for “intensive” days by combining high-volume (>= 240 visits/d) days with days that had an increased absolute number of significant acuity (acuity levels A and B). Because acuity A and B visits

payment method“>constitute the most emergent presenting conditions, combin- ing acuity with volume provides a rough measure of an intense ED environment whereby both the visit volume and severity are high.

These 2 dummy variables (high-volume days and intensive days) were included as dependent variables in separate multiple logistic regression models.

Independent variables

Demographic characteristics

Age and sex were obtained from hospital records. Age was continuous, ranging from 17 to 90 years (mean, 43 years; SD, 19 years), and sex was dichotomous. For multiple logistic regression analysis, dichotomous variables were defined for each of 5 age groups (<= 25, 26-35, 36-40, 41-64, and >= 65 years). Age and sex were independent factors in multiple logistic regression analyses.

Table 1 Patient characteristics and primary concern by intensity of use

Payment method

Payment method was obtained from billing and admin- istrative records routinely collected by the hospital and included the following payor classifications: Preferred Provider Organization (PPO), health maintenance Organi- zation (HMO), Commercial (indemnity), Worker’s Com- pensation, Medicaid Risk (managed care), Medicaid Traditional, Medicare Risk (managed care), Medicare Traditional, Self-pay, and Other. These categories were further collapsed into dichotomous variables used as control variables in multiple logistic regression analysis.

Primary concern

Primary concerns as reported by patients arriving at the ED were categorized by triage nurses into 1 of roughly 300 categories. Medical concerns were classified by the most troubling symptom (eg, shortness of breath, weakness, and fever) and by location (eg, chest pain or abdominal pain).

Normal			High volume (volume only)	Intensive (volume and severity)
No. (%)	64 394	(79.5)	16 642 (20.5)	6057 (7.6)
Sex, male	25 701	(39.9)	6708 (40.3)	2649 (43.7) ??
Age (y)
<= 25	13 916	(21.6) ?	3449 (20.7) ?	755 (12.5) ??
26-35	12 933	(20.1)	3334 (20.0)	893 (14.7) ??
36-40	6426	(9.9)	1690 (10.2)	568 (9.4)
41-64	20 647	(32.1)	5457 (32.8)	2340 (38.6) ??
>= 65	10 472	(16.3)	2712 (16.3)	1501 (24.8) ??
Payor status
Medicaid Traditional	7730	(12.9)	2028 (13.4)	894 (15.4) ??
Medicaid Risk	8458	(14.1)	2058 (13.6)	452 (7.8) ??
Medicare Traditional	12 144	(20.3) ?	3249 (21.4) ?	1679 (29.0) ??
Medicare Risk	1421	(2.4)	386 (2.5)	219 (3.8) ??
PPO	10 777	(18.0)	2677 (17.6)	1000 (17.3)
HMO	6046	(10.1)	1534 (10.1)	612 (10.6)
Commercial	1460	(2.4)	363 (2.4)	123 (2.1)
Worker’s Compensation	1070	(1.8) ?	223 (1.5) ?	68 (1.2) ?
Self-pay	9840	(16.4) ?	2388 (15.7) ?	664 (11.5) ??
Length of ED stay (min), mean (SD)	303.6	(206.2)	310.7 (209.9) ??	358.5 (216.0) ??
Primary concern
Pain (other)	5873	(9.1) ??	2170 (13.7) ??	276 (4.6) ??
Abdominal pain	6805	(10.6)	1657 (10.5)	453 (7.5) ??
Chest pain	4787	(7.4) ?	1252 (7.9) ?	991 (16.4) ??
Shortness of breath	3497	(5.4) ??	999 (6.3) ??	727 (12.0) ??
Motor vehicle crash	2761	(4.3)	680 (4.3)	280 (4.6)
Back pain	2022	(3.1) ?	553 (3.5) ?	60 (0.9) ??
Fall 0-10 ft	1312	(2.0) ??	491 (3.1) ??	166 (2.7) ?
Headache	1887	(2.9)	494 (3.1)	126 (2.1) ??
Sore throat	1541	(2.4)	420 (2.7)	18 (0.3) ??
Cough	1159	(1.8) ??	352 (2.2) ??	51 (0.8) ??
Weakness	1395	(2.2)	330 (2.1)	190 (3.1) ??
Sickle cell	337	(0.5)	76 (0.5)	54 (0.9) ??
* Statistically significant at the P b.05 level as compared with all other groups combined. ?? Statistically significant at the P b.001 level as compared with all other groups combined.

Traumatic concerns were categorized by injury cause (eg, motor vehicle crash, fall, and gunshot wound). The 12 most frequent primary concerns (other pain, abdominal pain, chest pain, dyspnea, wheezing or shortness of breath, motor vehicle crash, gunshot or stab wound, back pain, a fall under 10 ft, headache, sore throat, cough, weakness or dizziness, and sickle cell) were analyzed in bivariate associations with use intensity (Table 1). Dummy variables for the 12 most frequent primary concerns and for all other concerns combined (reference group) were included in multiple logistic regression models as independent factors.

Temporal variables

Patient volume and intensity of services were compared by time of day (7 AM to 3 PM, 3 PM until 11 PM, and 11 PM until 7 AM), day of the week, and month of the year to determine if there were consistent trends in ED resource use.

Diagnostic-Related Group severity index

The Diagnostic-Related Group (DRG) severity index, an ordinal scale ranging from 0 to 4 (4 being most severe), ranks illness severity based on principal diagno- sis, comorbidities, and procedures. Financial coders used information from medical records to retrospectively assign patients to DRG severity categories. Severity data were included as control variables in multiple logistic regres- sion analyses.

Analyses

Univariate analysis was used to assess overall sample characteristics. Bivariate analysis examined unadjusted re- lationships between high volume and intensive use and other factors including patient demographics (sex and age), DRG severity level, primary concern, payment method, and disposition status. ?2 Tests were used to analyze dichotomous and categorical variables. t Tests were used to analyze bivariate correlations between continuous variables. Analysis of variance among multiple groups was used to assess differences in unadjusted mean values between each pair of means for each group, and P values were used for each group- wise comparison. We were able to perform multiple comparisons by applying Bonferroni corrections to signifi- cance levels.

statistical models were constructed to estimate associa-

tions between use intensity and age, sex, primary concern, and payment method controlling for DRG severity level. Separate multiple stepwise logistic regression models estimated adjusted odds ratios (ORs) for high volume and intensive use, controlling for a number of relevant and potentially confounding factors. The covariates chosen for statistical modeling were significant at the P <=.05 level in bivariate analysis, although final models were controlled for all factors studied. We used Stata statistical software [24] for all analyses.

Results

Patient characteristics

In 2001, 16,642 (20.5%) of all ED visits (or 75 days/year) were classified as high volume, 6,057 (7.6%) (or 28 days/ year) of which were classified as intensive (high-volume day and triaged as either acuity level A or B) (Table 1). On days classified as normal and high volume, about 60% of patients were female. On days classified as intensive, statistically significantly fewer patients (56%) were female (Table 1).

The mean age of patients was 43 years, but the mean age of patients presenting on days of intensive use was 49.5 years, roughly 7 years older than the average age for patients presenting on normal-volume days (P b 0.001). Patients presenting on normal- and high-volume days did not have statistically significant differences in age. On days classified as intensive, the percentage of patients 25 years or younger was nearly half as much as that of normal-volume days (12.5% vs 21.6%, P b.05), whereas the percentage of patients 65 years or older was almost double that of normal-volume days (24.8% vs 16.3%, P b.001).

As might be expected, the length of stay increased with use intensity; average length of stay ranged from nearly 6 hours (mean, 358.5 minutes; P b.001) for intensive days to just more than 5 hours for high-volume (mean, 310.7 minutes; P b.001) and normal-volume (mean, 303.6 minutes) days.

Payor status by use intensity

Payor status was related to use intensity in some payor classifications (Table 1). Intensive-use days had a higher percentage of traditional Medicare patients compared with normal-volume days (29.0% vs 20.3%, P b.001) but a lower percentage of Medicaid Risk patients compared with normal- volume days (7.8% vs 14.1%, P b.001). These associations did not hold true, however, when controlling for confound- ing factors. The reduction in association between Medicare and use intensity, for example, may have been primarily explained by the fact that age was significantly related to condition severity.

Primary concern by use intensity

Table 1 shows how primary concerns that potentially signified a serious medical condition (eg, chest pain) differ by use intensity. Other pain accounted for 13.7% of visits during high-volume days, 9.1% of visits during normal- volume days, and 4.6% of visits during intensive days (P b.001). Several primary concerns were more prevalent as ED volume and intensity increased. Chest pain accounted for 16.4% of visits during intensive days, 7.9% of visits during high-volume days, and 7.4% of visits during normal- volume days (P b.001). In addition, shortness of breath accounted for 12.0% of visits during intensive days, 6.3%

Fig. 1 Normal volume, high volume, and intensive use of ED by day of the week.

of visits during high-volume days, and 5.4% of visits during normal-volume days (P b.001). Other primary concerns showed the reverse pattern. Cough was associated with

Table 2 Use intensity and monthly and daily status

0.8% of visits during intensive days, 2.2% of visits during high-volume days, and 1.8% during normal-volume days. Somewhat similar ratios were seen for chief concerns of fall

Normal		High volume (volume only)		Intensive (volume and severity)
No. (%)	64 394 (79.5)	16 642	(20.5)	6057 (7.6)
Time of day
Weekday 8 AM to 5 PM	22 453 (34.9) ??	7561	(47.8) ??	2786 (46.0) ??
Weekday 5 PM to Midnight	17 026 (26.4) ??	5443	(34.4) ??	2112 (34.9) ??
Weekend 5 PM Friday to 8 AM Monday	24 915 (38.7) ??	2811	(17.8) ??	1159 (19.1) ??
ED shifts
7 AM to 3 PM	27 042 (41.9) ?	6851	(43.3) ?	2509 (41.4)
3 PM to 11 PM	25 722 (39.9)	6302	(39.9)	2402 (39.7)
11 PM to 7 AM	11 630 (18.1) ??	2662	(16.8) ??	1146 (18.9) ?
Month of the year
January	4134 (6.4) ??	3012	(19.1) ??	1272 (21.0) ??
February	5148 (7.9) ??	1038	(6.6) ??	454 (7.5)
March	5522 (8.6) ??	1511	(9.6) ??	472 (7.8) ?
April	4431 (6.9) ??	2517	(15.9) ??	1005 (16.6) ??
May	5074 (7.9) ??	2012	(12.7) ??	770 (12.7) ??
June	5598 (8.7) ??	1251	(7.9) ??	431 (7.1) ??
July	5459 (8.5) ??	1508	(9.5) ??	566 (9.3)
August	5183 (8.1) ??	1723	(10.9) ??	626 (10.3) ??
September	5406 (8.4) ??	1001	(6.3) ??	373 (6.2) ??
October	6479 (10.1) ??	1	(0.01) ??	0 (0.0) ??
November	5833 (9.1) ??	241	(1.5) ??	88 (1.5) ??
December	6127 (9.5) ??	0	(0.0) ??	0 (0.0) ??
* Statistically significant at the P b.05 level as compared with all other groups combined. ?? Statistically significant at the P b.001 level as compared with all other groups combined.

Table 3 Factors associated with use intensity

Variable	High volume (volume Adjusted OR ?	only) ?? 95% CI	P	Intensive (volume and severity) ??
				Adjusted OR ?	95% CI	P
Demographics
Male	1.01	0.96-1.07	NS	1.08	1.01-1.16	b.05
Age<= 25 y	1.00	–	–	1.00	–	–
Age 26-35 y	1.06	0.98-1.14	NS	1.13	1.01-1.27	b.05
Age 36-40 y	1.04	0.95-1.14	NS	1.32	1.16-1.51	b.001
Age 41-64 y	1.05	0.97-1.13	NS	1.43	1.28-1.59	b.001
Age >= 65 y	1.01	0.91-1.13	NS	1.53	1.32-1.77	b.001
DRG severity index
Level 0	1.00	–	–	1.00	–	–
Level 1	0.88	0.73-1.06	NS	0.63	0.49-0.79	b.001
Level 2	0.84	0.69-1.02	NS	0.74	0.59-0.94	NS
Level 3	0.88	0.71-1.08	NS	0.76	0.59-0.97	b.05
Level 4	0.81	0.62-1.05	NS	0.67	0.49-0.89	b.05
Primary concern
Other concerns	1.00	–	–	1.00	–	–
Shortness of breath	1.01	0.91-1.14	NS	1.34	1.19-1.51	b.001
Weakness	0.95	0.81-1.11	NS	0.85	0.71-1.02	NS
Abdominal pain	0.97	0.88-1.06	NS	0.70	0.62-0.79	b.001
Back pain	0.97	0.83-1.13	NS	0.33	0.25-0.45	b.001
Fall 0-10 ft	1.27	1.10-1.46	b.001	1.01	0.83-1.22	NS
Headache	1.12	0.95-1.32	NS	0.64	0.50-0.81	b.001
Motor vehicle crash	1.19	1.04-1.36	b.05	1.41	1.18-1.68	b.001
Chest pain	0.93	0.83-1.04	NS	1.57	1.39-1.76	b.001
Cough	1.14	0.96-1.36	NS	0.36	0.26-0.49	b.001
Sore throat	1.09	0.92-1.31	NS	0.17	0.11-0.28	b.001
Gun or stab wound	1.16	0.84-1.60	NS	2.71	2.01-3.65	b.001
Other pain	1.05	0.97-1.13	NS	0.36	0.31-0.42	b.001
Sickle cell crises	0.99	0.71-1.40	NS	1.59	1.11-2.27	b.05
Payment method
Other payor	1.00	–	–	1.00	–	–
Medicaid Traditional	0.99	0.82-1.18	NS	0.98	0.74-1.28	NS
Medicaid Risk	0.95	0.79-1.15	NS	0.78	0.59-1.03	NS
Medicare Traditional	0.99	0.82-1.19	NS	0.92	0.70-1.21	NS
Medicare Risk	0.96	0.76-1.22	NS	0.90	0.66-1.24	NS
PPO	0.94	0.79-1.13	NS	0.99	0.76-1.29	NS
HMO	0.96	0.79-1.16	NS	1.05	0.80-1.39	NS
Commercial (Indemnity)	0.92	0.73-1.15	NS	0.89	0.64-1.24	NS
Self-pay	0.97	0.81-1.17	NS	0.85	0.65-1.12	NS
Worker’s Compensation	0.89	0.69-1.14	NS	1.11	0.77-1.61	NS
Day and time of the week
Weekdays 8 AM-5 PM	1.00	–	–	1.00	–	–
Weekdays 5 PM-12 PM	0.98	0.93-1.03	NS	1.13	1.05-1.21	b.05
Weekend 5 PM Friday-8 AM	0.94	0.85-1.03	NS	0.99	0.88-1.12	NS
Day of the week
Sunday	4.21	3.64-4.87	b.001	3.9	3.19-4.9	b.001
Monday	92.38	78.77-108.3	b.001	41.6	33.3-51.9	b.001

Tuesday	13.95	11.80-16.48	b.001	11.9	9.43-15.10	b.001
Wednesday	10.51	8.88-12.43	b.001	7.77	6.11-9.87	b.001
Thursday	2.68	2.24-3.20	b.001	2.65	2.10-3.42	b.001
Friday	4.87	4.16-5.71	b.001	5.35	4.25-6.73	b.001

(continued on next page)

	Table 3 (continued)
	Variable	High volume (volume only) ?? Adjusted OR ? 95% CI		P	Intensive (volume and severity) ??
					Adjusted OR ?	95% CI	P
	Month of the year
	January	34.79	29.73-40.69	b.001	23.8	18.87-30.13	b.001
	February	5.88	4.99-6.93	b.001	6.32	4.94-8.10	b.001
	March	11.52	9.79-13.56	b.001	7.10	5.56-9.08	b.001
	April	24.43	20.87-28.61	b.001	16.2	12.8-20.5	b.001
	May	16.47	14.06-19.29	b.001	12.11	9.55-15.36	b.001
	June	7.99	6.79-9.39	b.001	5.99	4.68-7.67	b.001
	July	8.79	7.49-10.33	b.001	7.15	5.61-9.11	b.001
	August	14.33	12.22-16.79	b.001	9.45	7.44-12.02	b.001
	September	5.94	5.03-7.01	b.001	5.16	4.01-6.62	b.001
	October	Zero observation	–	–	Zero observation	–	–
	November	1.00 –		–	1.00	–	–
	December	Zero observation	–	–	Zero observation	–	–
	Abbreviation: NS, not significant. * Each variable in the model was adjusted for other factors in the table and for acuity level, primary medical diagnosis, and arrival method as compared with normal-volume days. ?? Compared with normal volume.

0 to 10 ft (2.7% of visits for intensive days, 3.1% of visits during high-volume days, and. 2.0% of normal-volume days; P b .001) and back pain (0.9% of visits during intensive days, 3.5% of visits during high-volume days, and 3.1% of normal-volume days; P b .001).

Use intensity and weekly, monthly, and daily status

On high-volume days, the largest percentage of patients presented on Mondays (45.0%), followed by Tuesdays

(18.8%) and Wednesdays (15.5%); the smallest percentage of patients presented on Saturdays (1.7%) and Thursdays (4.7%). On intensive days (high volume and most severe), the largest percentage of patients presented on Mondays (45.2%), followed by Tuesdays (19.2%) and Wednesdays (13.3%) (Fig. 1). During high-volume days, the largest percentage of patients presented on weekdays between 8 AM and 5 PM (47.8%), whereas fewer patients presented on the weekends (17.8%) (Table 2). During high-volume days, the highest percentage of patients presented during the 7 AM to 3 PM shift (43.3%), with fewer patients presenting during the 11 PM to 7 AM shift (16.8%). Table 2 also shows seasonal

Fig. 2 Differences in mode of arrival and discharge status (disposition) among normal volume, high volume, and intensive use.

differences in use, with the greatest use at the beginning of the year in January (19.1% of visits on high-volume days and 21.0% of visits on intensive days, P b.001), peaking again in the spring months of April (15.9% of visits on high-volume days and 16.6% of visits on intensive days, P b.001) and May (12.7% of visits on high-volume days and 12.7% of visits on intensive days, P b.001). October (0.01% on high- volume days and 0.0% on intensive days, P b.001), November (1.5% on high-volume days and 1.5% on intensive days, P b.001), and December (0.0% on high- volume days and 0.0% on intensive days, P b.001) had the smallest percentage of ED visits. By contrast, the percentage of visits presenting on normal-volume days remained fairly constant across most months–ranging from 6.4% in January to 10.1% in October (Table 2). Controlling for all other factors, these relationships held true in multiple logistic regression analyses (Table 3).

Use intensity, mode of arrival, and discharge status

Fig. 2 shows differences in mode of arrival and discharge status by normal-volume days, high-volume days, and intensive days. Although more patients arrived by car than any other arrival method, the percentage arriving by car on high-volume days was not different from that on normal- volume days. The percentage of patients arriving by foot (walk-in) and county emergency medical services (EMS) was similar on normal-volume and high-volume days, as well. A greater percentage of patients arrived by bus (6.2% vs 5.5%, P b.05) on high-volume days as compared with normal-volume days. A slightly lower percentage of patients arrived by city EMS (16.1% vs 17.1%, P b.05) and helicopter (0.2% vs 0.4%, P b.05) on high-volume days as compared with normal-volume days. The reduced percent- age of arrivals by city EMS and helicopter on high-volume days as compared with normal-volume days may be explained by ambulance diversion status. As expected, the percentage of patients arriving by city EMS, county EMS, and helicopter on intensive days was greater than that on normal- and high-volume days, and the percentage of patients arriving by car, walk-in, or bus was less than that on normal- and high-volume days (Fig. 2).

In terms of discharge status, there was little or no difference in in-hospital or intensive care unit/operation room admission between high-volume and normal-volume days. On intensive days, the percentage of hospital admissions (43.9%) and intensive care unit/operation room admissions (4.6%) was nearly double the rate of admissions on both high-volume days (22.5% and 2.1%, respectively) and normal-volume days (23.6 % [P b.05] and 2.1 % [P b.001]) (Fig. 2). These relationships disappeared, howev- er, in multiple logistic regression analyses. On high-volume days, there was a slightly greater percentage of patients who eloped or left AMA (10.2%, P b.001) as compared with

normal-volume days (7.3%, P b.001) and intensive days (4.4%, P b.001).

Multivariate analyses: independent factors associated with use intensity

Table 3 shows results of 2 multiple logistic analyses of high-volume days and intensive days (reference, normal- volume days) after adjusting for all other variables studied. Patients in a motor vehicle crash were 1.2 times more likely to present to the ED (confidence interval [CI], 1.0-1.4), and patients having fallen 0 to 10 ft were 1.3 times more likely to present to the ED (CI, 1.1-1.5) on high-volume days (Table 3). Patients were more likely to arrive on a high- volume day during the months of January (OR, 34.8; CI, 29.7-40.7), April (OR, 24.4; CI, 20.9-28.6), May (OR, 16.5;

CI, 14.1-19.3), and August (OR, 14.3; CI, 12.2-16.8) as

compared with those arriving in November (reference group) (Table 3). In addition, patients were the most likely to arrive on a high-volume day on Monday (OR, 92.4, CI, 78.8- 108.3), Tuesday (OR, 14.0; CI, 11.8-16.5), and Wednesday (OR, 10.5; CI, 8.9-12.4) as compared with patients arriving on Saturday (reference group). Patients arriving by helicopter were half as likely to arrive on high-volume days as compared with patients arriving by foot (reference group) (OR, 0.6; CI, 0.3-0.9).

Patients 65 years or older were 1.5 times more likely to present (OR, 1.5; CI, 1.3-1.8) on intensive-use days (Table 3). In terms of primary concern, patients experienc- ing shortness of breath (OR, 1.3; CI, 1.2-1.5), a motor vehicle crash (OR, 1.4; CI, 1.2-1.7), chest pain (OR, 1.6; CI, 1.4-1.8), or a gun or stab wound (OR, 2.7; CI, 2.0-3.7) were more likely to present on intensive-use days. Patients with a primary concern of sore throat and cough were 5.9 and 2.8 times less likely to present to the ED on intensive- use days (Table 3). Patients arriving during the week from 5 PM to 12 PM were 1.1 times more likely to arrive on an intensive-use day (OR, 1.1; CI, 1.1-1.2). Patients arriving during the months of January (OR, 23.8, CI, 18.9-30.1), April (OR, 16.2; CI, 12.8-20.5), and May (OR, 12.1; CI,

9.6-15.4) were the most likely to arrive on an intensive-use day as compared with those arriving in November (reference group) (Table 3). In addition, patients presenting to the ED on Monday (OR, 41.6; CI, 33.3-51.9), Tuesday (OR, 11.9; CI, 9.4-15.1), and Wednesday (OR, 7.8; CI, 6.1-

9.9) were the most likely to arrive on an intensive-use day as compared with patients arriving on Saturday (reference group). Patients arriving by city (CI, 1.2-2.0) and county (CI, 1.2-2.0) EMS were each 1.5 times more likely to arrive on an intensive-use day as compared with patients arriving by foot (reference group). In terms of the DRG severity index, patients presenting on intensive days were 1.5 (CI, 1.1-2.0) times less likely to be rated level 4 (most severe) and 1.6 (CI, 1.3-2.0) times less likely to be rated level 1 compared with the reference group (patients rated level 0 [least severe]).

Discussion

This study examined the patterns of ED use at an urban, academic hospital with respect to time of day, day of the week, month of the year, chief concern, payor status, severity of illness, and triage classification as a mechanism to better understand how to allocate resources to prevent inefficiency and overcrowding. We defined new measures of ED overcrowding–high volume and intensive use–that take into account ED volume and severity of patient illness.

Investigators have identified 3 distinct components of the health care system that contribute to overcrowding: input, throughput, and output [25,26].This analysis attempts to understand input as a means of providing adequate resources to achieve rapid throughput–measured by length of stay in the ED [26]. A study by Chan and colleagues [27] analyzed the impact certain resource use factors had on throughput times. Although inpatient admissions, main ED census, pediatric volume, and ambulance arrivals significantly affectED throughput times, Correlation coefficients suggest that these factors only explain a small portion of throughput time. Understanding how different factors affect ED use is important for resource allocation because the literature has shown a significant increase in ED use, admission rates, and acuity over time [28].

To appropriately allocate resources in the ED, it is crucial to grasp not only daily census but also perceived severity of illness. To examine the effects of both volume and severity on ED overcrowding, we stratified our data by volume (normal volume vs high volume) and then, again, by patient acuity (triage categories A and B vs triage categories C-E). By categorizing in this way, we were able to show that, because the profile and percentage of high-acuity (triage categories A and B) patients do not differ between normal- volume and high-volume days, the actual number of patients who, based on triage acuity, would require a greater amount of resource is significantly higher on high-volume days. Furthermore, length of stay is significantly longer on high- volume and intensive days. Thus, high volume and severity of illness are vital to any discussion of resource allocation because both factors stress the health care system by increasing the demand for resources.

This study suggests that intensive days (days with higher patient volume and more High-acuity patients) are a good marker for an increased demand for resources. We found that patients presenting on intensive days were statistically more likely to have primary concerns of shortness of breath, chest pain, motor vehicle crash, and gun or stab wound and were statistically less likely to have primary concerns of cough and sore throat. Chest pain and shortness of breath, in particular, are both indications of more serious medical conditions that will require a higher level of resource use. motor vehicle crashes and gun or Stab wounds are substantial mechanisms of injury that will likely require higher levels of resources. Conversely, cough and sore throat, 2 conditions that are more benign and require less resources to treat per patient

presentation, are less common on intensive days. This is consistent with other studies that have found strong correlations between ED crowding measures and rate of emergent patients (triage category B), with emergent patients having an increased risk of being admitted and requiring longer evaluation in the ED [12].

Findings from previous studies generally concur with this study’s results. Our data found that Mondays had the greatest number of high-volume days during the year by far (45% of high-volume days were Mondays), followed in succession by each day of the week except that Friday was higher than Thursday (Table 3). Holleman et al [20] and Diehl et al [17] similarly found that Mondays were the busiest days of the week and that Saturdays were the slowest. A study in Israel identified Sunday (first day of the work week in Israel) as the busiest day of the week [29]. In terms of ED use throughout the year, we found that January, April, May, and August were the busiest months and October, November, and December were the slowest. Holleman et al [20] found winter months to be busiest, whereas Diehl et al [17] found summer months to be the busiest. Emergency department volume by time of day would seem to have significant implications for staffing models. Tandberg and Qualls [30] showed that ED use is at its highest between 8 AM and midnight, and Morris and colleagues [31] found that trauma visits peaked between 4 PM and midnight. Our research confirms this trend, with a statistically significant drop in census from 11 PM to 7 AM on both normal- and high-volume days (Table 2).

Several studies have investigated the demand for ED health care service during other predictable events. One study looked at the association between poison exposure calls during lunar events [32], another investigated the rate of dog bites during dates surrounding a full moon [33], and a third studied the rate of ED visits during the Super Bowl [34].

One indication of the amount of overcrowding and inefficiency on high-volume days is the significant increase in those who left AMA or eloped. These data suggest that increasing staffing levels during periods of highest ED volume might ease system pressure and help reduce crowding. A review of access block and ED overcrowding reports reductions in ED length of stay after increasing staff capacity [11]. Another study compared crowding measures during viral epidemic seasons before and after the Viral Epidemic Supplemental Attending and Staff (VESAS) plan was implemented [15]. The VESAS plan included a team of medical personnel that could provide care during busier periods as needed. After implementation, the left-without- being-seen rate was reduced by 37%, and the average wait time was reduced by 15 minutes. Other studies suggest that using resources differently–such as using scribes to help with documentation, use of mental health nurses to provide specialist support, or using social workers to help with discharge–may help improve patient ftow and decrease ED crowding [2].

This study has several limitations. First, the study population was obtained from a single Urban academic ED in the Midwest serviced largely by 1 EMS system. Our findings are specific to a large, academic level I trauma center and thus may not be generalizable to other types of EDs. Second, our data set did not provide information on Patient race or income, precluding our ability to draw conclusions about independent relationships between race or income and intensive use. However, using payment type as a proxy for economic status may provide some idea of overall socioeconomic status. Third, our data set did not provide information on whether patients had been referred to the ED by a physician. Knowing whether patients received a physician referral would have helped us understand whether patients had a primary source of health care and if they had exhausted out-of-hospital sources of care. Fourth, our sample was reduced by roughly 11% due to missing data on acuity level, payment method, and mode of arrival. The data suggest that patients without complete data are more likely to be younger, to be male, to have eloped or left AMA, and to be less acutely ill and injured than patients with complete data. This may have slightly biased our conclusions; however, because of the small reduction in sample size, imputation was not pursued. Finally, the Canadian Triage and Acuity Scale and the DRG severity index were used as proxies for the potential or actual seriousness (severity) of a patient’s medical condition. Although these scores work well when assessing either extreme of the spectrum, prognostic value in the middle categories–which constitute a large percentage of the overall patient population–is lost. To adjust for this, we used the top 2 acuity levels (A and B) as indicators of the patient severity on intensive days.

Conclusions

Efforts to control ED overcrowding involve steps at the input, throughput, and output stages of the ED system of care. This study created a new index of intensive use based on volume and severity of illness and measured stress in the ED environment using a 3-part categorization (normal volume, high volume, and intensive use). Intensive use was newly defined in relation to average patient volume and severity per day and related to monthly, weekly, and daily status. Patients arriving during intensive days displayed distinct patterns of presenting concerns. Length of stay was significantly greater on high-volume and intensive days. These findings offer input for reallocating resources and altering staffing models to more efficiently provide high- quality ED services.

These analyses raise some important issues. Using these data, it is possible to imagine a staffing model that increases physician, nursing, radiologic, and laboratory coverage early in the week and in the evening shifts when ED volume is highest and decreasing somewhat the

coverage toward the end of the week and in the early morning hours when ED volume is lowest. Using a model similar to the VESAS model [15], on-call staff could be used during the highest-volume months of January, April, May, and August. Would this low cost reallocation of resources improve length of stay and decrease patients who leave AMA or leave without being seen? If staffing models are changed, will it only lengthen the throughput times on the low-volume days and not improve the high-volume days because of fundamental problems with output or disposition? Future studies may benefit from using a measurement of “bed hours” [25] instead of input.

Acknowledgments

Jennifer Prah Ruger would like to acknowledge that the study was funded, in part, by the Washington University Center for Health Policy and to clarify that this study did not receive NIH funding.

References

1. Institute of Medicine (IOM). Future of emergency care: hospital-based emergency care at the breaking point. Committee on the future of emergency care in the United States health system. Washington, DC: National Academies Press; 2007.
2. Morris ZS, Boyle A, Beniuk K, Robinson S. Emergency department crowding: towards an agenda for evidence-based intervention. Emerg Med J 2011 [Epub ahead of print].
3. Derlet R, Richards J, Kravitz R. Frequent overcrowding in U.S. emergency departments. Acad Emerg Med 2001;8(2):151-5.
4. Weiss SJ, Derlet R, Arndahl J, et al. Estimating the degree of emergency department overcrowding in academic medical centers: results of the National ED Overcrowding Study (NEDOCS). Acad Emerg Med 2004;11(1):38-50.
5. Schneider S, Zwemer F, Doniger A, Dick R, Czapranski T, Davis E. Rochester, New York: a decade of emergency department over- crowding. Acad Emerg Med 2001;8(11):1044-50.
6. Reeder TJ, Garrison HG. When the safety net is unsafe: real-time assessment of the overcrowded emergency department. Acad Emerg Med 2001;8(11):1070-4.
7. Derlet RW, Richards JR. Overcrowding in the nation’s emergency departments: complex causes and disturbing effects. Ann Emerg Med 2000;35(1):63-8.
8. Velianoff GD. Overcrowding and diversion in the emergency department: the healthcare safety net unravels. Nurs Clin North Am 2002;37(1):59-66.
9. Bernstein SL, Aronsky D, Duseja R, et al. The effect of emergency department crowding on clinically oriented outcomes. Acad Emerg Med 2009;16(1):1-10.
10. Moskop JC, Sklar DP, Geiderman JM, Schears RM, Bookman KJ. Emergency department crowding, part 1–concepts, causes, and moral consequences. Ann Emerg Med 2009;53(5):605-11.
11. Forero R, McCarthy S, Hillman K. Access block and emergency department overcrowding. Crit Care 2011;15(2):216.
12. Graham J, Aitken ME, Shirm S. Correlation of measures of patient acuity with measures of crowding in a pediatric emergency department. Pediatr Emerg Care 2011;27(8):706-9.
13. Richardson LD, Asplin BR, Lowe RA. Emergency department crowding as a health policy issue: past development, future directions. Ann Emerg Med 2002;40:388-93.
14. Trzeciak S, Rivers EP. Emergency department overcrowding in the United States: an emerging threat to patient safety and public health. Emerg Med J 2003;20:402-5.
15. Shaw KN, Lavelle JM. VESAS: a solution to seasonal fluctuations in emergency department census. Ann Emerg Med 1998;32(6):698-702.
16. McCarthy ML, Aronsky D, Jones ID, Miner JR, Band RA, et al. The emergency department occupancy rate: a simple measure of emergency department crowding? Ann Emerg Med 2008;51(1): 15-24.
17. Diehl AK, Morris MD, Mannis SA. Use of calendar and weather data to predict walk-in attendance. South Med J 1981;74(6):709-12.
18. Christoffel KK. Effect of season and weather on pediatric emergency department use. Am J Emerg Med 1985;3(4):327-30.
19. Leo PJ, Bove J. ED census and day of the week. Am J Emerg Med 1996;14(1):114-5.
20. Holleman DR, Bowling RL, Gathy C. Predicting daily visits to a walk- in clinic and emergency department using calendar and weather data. J Gen Intern Med 1996;11(4):237-9.
21. Vassilacopoulos G. Allocating doctors to shifts in an accident and emergency department. J Opl Res Soc 1985;36(6):517-23.
22. Graff LG, Radford MJ. Formula for emergency physician staffing. Am J Emerg Med 1990;8(3):194-9.
23. Beveridge R, Clarke B, Janes L, et al. Implementation guidelines for the Canadian Emergency Department Triage & Acuity Scale (CTAS). Canadian Association of Emergency Physicians, 16 Dec 1998. http:// www.triagefirst.com/downloads/libfiles/CTASImplementation.pdf
24. Stata Statistical Software, Stata Corporation, College Station, Tex.
25. Solberg LI, Asplin BR, Weinick RM, Magid DJ. Emergency department crowding: consensus development of potential measures. Ann Emerg Med 2003;42(6):824-34.
26. Asplin BR, Magid DJ, Rhodes KV, Solberg LI, Lurie N, Camargo CA. A conceptual model of emergency department crowding. Ann Emerg Med 2003;42(2):173-80.
27. Chan L, Reilly KM, Salluzzo RF. Variables that affect Patient throughput times in an academic emergency department. Am J Med Qual 1997;12(4):183-6.
28. Meggs WJ, Czaplijski T, Benson N. Trends in emergency department utilization, 1988-1997. Acad Emerg Med 1999;6(10):1030-5.
29. Rotstein Z, Wilf-Miron R, Lavi B, Shahar A, Gabbay U, Noy S. The dynamics of patient visits to a public hospital ED: a statistical model. Am J Emerg Med 1997;15(6):596-9.
30. Tandberg D, Qualls C. Time series forecasts of emergency department patient volume, length of stay, and acuity. Ann Emerg Med 1994; 23(2):299-306.
31. Morris RW, Leikin JB, Eckenrode P, Boston D. The effects of time of trauma patient presentation on emergency department utilization. Prog Clin Biol Res 1990;341A:201-11.
32. Oderda GM, Klein-Schwartz W. Lunar cycle and poison center calls. J Toxicol Clin Toxicol 1983;20(5):487-95.
33. Chapman S, Morrell S. Barking mad? Another lunatic hypothesis bites the dust. BMJ 2000;321:1561-3.
34. Reich NT, Moscati R, Jehle D, Ciotoli M. The impact of a major televised sporting event on emergency department census. J Emerg Med 1994;12(1):15-7.