Emergency Department Contributors to Ambulance Diversion: A ...

EMERGENCY MEDICAL SERVICES/ORIGINAL RESEARCH

Emergency Department Contributors to Ambulance Diversion: A Quantitative Analysis

Michael J. Schull, MD, MSc Kate Lazier, BA Marian Vermeulen, BNSc, MSc Shawn Mawhinney, MD Laurie J. Morrison, MD, MSc From the Clinical Epidemiology Unit and the Department of Emergency Services, Sunnybrook and Women’s College Health Sciences Centre, Toronto, Ontario, Canada (Schull, Morrison); the Prehospital Research Group, Toronto Emergency Medical Services, Toronto, Ontario, Canada (Schull, Vermeulen, Morrison); the Division of Emergency Medicine, Department of Medicine (Schull, Mawhinney, Morrison), and the Department of Health Administration (Schull, Vermeulen, Morrison), University of Toronto, Toronto, Ontario, Canada; and the Faculty of Medicine, Dalhousie University, Halifax, Nova Scotia, Canada (Lazier).

Copyright © 2003 by the American College of Emergency Physicians.

See editorial, p. 477.

Study objective: We determine the relationship between physician, nursing, and patient factors on emergency department use of ambulance diversion. Methods: Data were collected at 1 ED in Toronto, Ontario, Canada, on the duration of ambulance diversion during consecutive 8-hour intervals from January to December 1999 (intervals=1,095). By using time series methods, the association between ambulance diversion and nurse hours, physician on duty, and boarded patients was determined. Covariates included patient volume, assessment time, and boarding time. Results: A total of 37,999 patients were treated in the ED over the study period (2% major trauma, 16% ambulance arrivals, and 22% admitted). Nurse hours per interval averaged 60. A mean of 3.2 admitted patients were boarded in the ED each interval. For admitted patients, the time from registration to admission order and from admission order to ED departure averaged 5.2 and 3.5 hours, respectively. There was no ambulance diversion during 170 (15.5%) intervals, whereas 17 (1.5%) intervals were continuously on diversion. In time series analyses, ambulance diversion increased with the number of admitted patients boarded in the ED (6.2 minutes per patient; 95% confidence interval [CI] 2.6 to 9.8 minutes), the number admitted per interval (4.6 minutes per patient; 95% CI 0.1 to 9.1 minutes), assessment time (9.9 minutes per hour; 95% CI 3.3 to 16.5 minutes), and boarding time (11.3 minutes per hour; 95% CI 5.6 to 17.0 minutes). Thirteen of 15 emergency physicians were not associated with ambulance diversion, 1 was associated with reduced use (–36.3 minutes; 95% CI –65.2 to –7.5 minutes), and 1 was associated with increased use (47.6 minutes; 95% CI 4.5 to 90.6 minutes). ED nurse hours were not associated with diversion. Ambulance-delivered patient volume was associated with diversion (5.2 minutes per patient; 95% CI 2.7 to 7.8 minutes), but walk-in patients and patients with major trauma were not. Conclusion: Admitted patients in the ED are important determinants of ambulance diversion, whereas nurse hours and most emergency physicians are not. Reducing the volume of walk-in patients is unlikely to lessen the use of diversion. [Ann Emerg Med. 2003;41:467-476.]

0196-0644/2003/$30.00 + 0 doi:10.1067/mem.2003.23

APRIL 2003

41:4

ANNALS OF EMERGENCY MEDICINE

4 6 7

EMERGENCY DEPARTMENT CONTRIBUTORS TO AMBULANCE DIVERSION Schull et al

INTRODUCTION

Providing acutely ill patients with rapid access to emergency care is the prime mission of emergency departments.1 In recent years, worsening ED crowding across North America1-12 has cast doubt on the capacity of some systems to consistently fulfill this mandate.1,10,12 The problems of crowding and resulting ambulance diversion have gained wide attention from practitioners, policymakers, and patients, yet there has been little systematic investigation into its causes,12 despite at least 3 anecdotal reports that link crowding to patient fatalities.13-15 Multiple factors have been reported to lead to ED crowding and ambulance diversion. These have included increasing numbers of elderly and acutely ill patients,16 lack of access to primary care, influenza outbreaks,17 staffing shortages,10,12 investigation and consultation delays, inadequate physical space, and delayed transfer of admitted patients to ward beds.1-5,7,11,12,18 All these studies were limited by the possibility that important determinants were overlooked and because their analyses could not establish the relative importance of the selected factors. In a previous study,19 we developed a model of crowding that divided determinants into 4 domains (community, patient, ED, and hospital factors) and have used ambulance diversion as a proxy measure of crowding.19-21 We have based the present study on this theoretic model with the aim of developing a more systematic understanding of the causes of ambulance diversion. We sought to quantify the relative importance of 3 determinants from our model, all of which might be directly affected by ED personnel. The specific factors were variations in the number of staffed ED nurse hours (which might affect the efficiency of patient management), the number of admitted patients boarded in the ED (which might affect the physical capacity to accept new patients), and emergency physician on duty (whose individual practice characteristics might affect the speed of patient assessments and resource demands). Our hypothesis was that ambulance diversion would be associated with nurse staffing and admitted patients

4 6 8

boarded in the ED but not with the emergency physician on duty. M AT E R I A L S A N D M E T H O D S

The study was conducted at the Sunnybrook site of the Sunnybrook and Women’s College Health Sciences Centre, a 1,200-bed tertiary-care hospital in Toronto, Ontario, Canada, from January 1, 1999, to December 31, 1999. This hospital serves as the sole burn unit and 1 of 2 adult trauma centers in the city of 2.5 million people. The ED is functionally divided into a “minor” section, with 6 stretcher cubicles for patients with low-severity complaints, such as minor trauma and ear-nose-throat, upper respiratory tract, and psychiatric complaints, and a “major” section, with 17 stretcher cubicles (6 with dedicated monitors) for patients with high-severity complaints or for patients requiring resuscitation, as well as a dedicated trauma room with 2 more stretchers. In situations of crowding, additional stretchers on the major section were placed in hallways, and boarded patients could be held in either section. Nurses were moved between sections depending on demand. The trauma room was reserved exclusively for multisystem trauma patients; an on-call trauma team provided physician staffing, but nurses came from those present in the ED. Fifteen staff emergency physicians worked during our study period, all with certification in emergency medicine. A single staff emergency physician was present in the department, except from 12 PM to 8 PM, when a second staff physician was present. All staff physician shifts were 8 hours in length. When patient volumes were heavy, physicians extended their shifts on an ad hoc basis, but additional emergency physicians were called in only under exceptional conditions. Zero to 4 residents (mostly junior off-service staff) staffed the ED each day; there were no nurse clinicians or physicians’ assistants. Admission decisions were made and orders were written by the admitting service. All EDs in the city were served by a single out-ofhospital care provider, Toronto Emergency Medical Services, which operated a centralized dispatch system and an ambulance diversion system. EDs continuously


41:4

APRIL 2003


signaled their ambulance diversion condition to Toronto Emergency Medical Services through the Internet. Three conditions were possible: normal (no diversions); redirect consideration (all diverted except critically ill patients); and critical care bypass (all diverted). Conditions other than normal automatically expired after a standard interval: critical care bypass was downgraded to redirect consideration after 30 minutes, which was then downgraded to normal after 120 minutes. The condition could be renewed, upgraded, or downgraded at any time. Paramedics could override a hospital’s status if their patient required specialized care available only at certain hospitals (eg, trauma, obstetrics, neonates and pediatrics, burns, stroke). The ambulance diversion condition of the hospital was determined by the attending emergency physician on the major section and the ED charge nurse on the basis of the degree of ED crowding in accordance with standard criteria. The decision was guided by these criteria and by individual physician judgment. The criteria were established for all hospitals by the Ontario Ministry of Health, and the use of ambulance diversion by EDs is monitored by the government.10 The present study’s outcome was the total duration (in minutes) of ambulance diversion during each 8hour interval. The 3 main independent variables studied were the number of admitted patients boarded in the ED (ie, held in the ED pending an available ward bed) at the beginning of each interval, the number of ED nurse hours worked per interval, and the emergency physician on duty in each interval. The following covariates were measured for each interval: (1) volume of walk-in patients, (2) ambulancedelivered patients, and (3) trauma patients; (4) nursing workload; (5) number of patients admitted through the ED; (6) time of day (day, evening, or night); (7) day of week; (8) mean assessment time (time from patient registration to admission order for admitted patients); (9) mean boarding time (time from admission order to departure from the ED for admitted patients); and (10) number of inpatient acute care beds occupied by patients awaiting placement in chronic care facilities in the community.

APRIL 2003

41:4


In all analyses, total patient volume was the sum of walk-in and ambulance-delivered patients. Because all trauma patients arrive by ambulance, they are included in the ambulance-delivered patient count. We included trauma patients as a separate variable given the heavy burden they represent compared with that of most other ambulance-delivered patients.22 In multivariate analyses, the ambulance-delivered patient volume was lagged by 1 interval (ie, the number of ambulances that arrived in the previous interval was used) to correct for the fact that the duration of ambulance diversion in a given interval directly influences the number of ambulance arrivals in that interval. However, we did not lag the trauma volume because these patients are not affected by ambulance diversion. Data regarding the minute-by-minute real-time ambulance diversion status for the ED over the study period were obtained from Toronto Emergency Medical Services, and total minutes of ambulance diversion during each 8-hour interval was calculated. Data on nurse staffing were obtained from the nursing records used for payroll purposes. For each interval, nurse hours were calculated as the number of nurses working multiplied by the number of hours worked by each nurse. Because nurses in the Sunnybrook ED work 14 different shift types, nurse hours from these various shifts were reallocated to the 8-hour intervals used in this study. The number of admitted patients boarded in the ED was obtained from a census undertaken daily at 8 AM, 4 PM, and 12 AM. Information on the emergency physician on duty during each 8-hour interval was obtained from physician staffing records used for billing purposes. Although more than 1 physician was sometimes on duty, we selected the emergency physician designated to manage ambulance diversion for that interval. For purposes of confidentiality, nurse’s names were not collected in the data set, and physicians were identified by code. This study was approved by the hospital research ethics board. All covariate data were obtained from prospectively collected hospital administrative data. The nursing workload measure is used for administrative purposes and is calculated by assigning each ED patient a score

4 6 9


based on presenting problem and intensity and duration of nursing care required. A total nursing workload score is generated for each day. Because only daily total workload scores were available, the workload in each 8hour interval was calculated by prorating the total daily score by the number of patients seen in the ED during the interval. The hospital reported the number of inpatients awaiting chronic-care placement and the assessment and boarding times for admitted ED patients on a daily basis; we assumed the same values for each of the 3 intervals in a given day. Double data entry was carried out independently by 2 clerks, both blinded to the study hypothesis. These 2 data sets were then merged, and the discrepancies were resolved by the principal investigator (MJS) with reference to the original data. The observations in this study represented consecutive 8-hour intervals, which are therefore likely to be autocorrelated (ie, the amount of ambulance diversion in 1 interval is likely to be associated with that of the previous shift or shifts). This violates a core assumption of linear regression analysis, which is that observations are independent of each other. Time series analysis allows for autocorrelation of observations when determining the association between predictor variables and outcome. We used an autoregressive integrated moving average model23 and evaluated its assumptions through standard tests (data stationarity [evaluated by using the Dickey-Fuller test] and seasonality [assessed with the autocorrelation and partial autocorrelation functions]).24 A coefficient (and 95% confidence interval [CI]) was estimated for each predictor variable in univariate and multivariate models, representing the change in minutes of ambulance diversion per unit change in the variable. Model fit was assessed by using the Q statistic to test for residual autocorrelation, the Akaike information criterion, the Schwarz Bayesian information criterion, and adjusted R2 values. SAS software (version 8.0, SAS Institute, Cary, NC) was used for all analyses. R E S U LT S

The ED treated 37,999 patients during the study period, of whom 22% were admitted. A total of 31,959 (84%)

4 7 0

patients were walk-in patients, and 6,040 (16%) arrived by ambulance. A total of 879 (2%) patients had major trauma. Table 1 provides descriptive statistics for continuous variables. On average, walk-in patients outnumbered ambulance-delivered patients by almost 6 to 1 per interval. Among admitted patients, the mean assessment time was almost 50% longer than the mean boarding time (5.2 hours [SD 1.1] versus 3.5 hours [SD 1.9], respectively). At the start of each interval, an average of approximately 3 admitted patients were boarded in the ED, although this increased to as many as 12 patients in some intervals. Patient volumes and nurse staffing varied by time of day. The number of ambulance-delivered and walk-in patients was highest in daytime intervals (mean 6.6 and 40.2, respectively), and both were lowest at night (mean 4.3 and 13.5, respectively). The average number of trauma patients was greatest during the evening interval (mean 1.0). Nurse hours averaged 60 in the daytime, 72 in the evening, and 49 at night but varied most in the daytime (range 36 to 96). Figure 1 shows the distribution of ambulance diversion per interval broken down by time of day. The overall mean, median, and SD were 198, 191, and 147 minutes, respectively. There was no ambulance diversion during 170 (16%) intervals, 1 to 30 minutes of diver-

Table 1.

ED staffing, patient volumes, admission, and hospital indicators per interval (n=1,095). Domain

Variable

Mean (SD)

Range

Staffing

Nurse hours 60 (11.7) Nurse workload, min 2,356 (967) Intervals per physician, No. 73 (36.2) Patient volume Trauma, No. 0.8 (1.0) Walk in, No. 29.2 (13.1) Ambulance, No. 5.5 (2.9) Admitted patients Admissions, No. 7.5 (3.2) No. boarded in ED 3.2 (2.3) Mean assessment time, h 5.2 (1.1) Mean boarding time, h 3.5 (1.9) Hospital Inpatients awaiting placement, 32.7 (6.8) No.* *Patients

32–96 349–5,314 10–134 0–6 1–67 0–15 1–16 0–12 0–8.7 0.8–13 21–49

in acute care beds awaiting chronic-care beds in the community.


41:4

APRIL 2003


sion in 35 (3%) intervals, and continuous diversion in 17 (1.5%) intervals. The box plot insert in Figure 1 shows that ambulance diversion peaked in the evening, with a median of 324 minutes, which also had the widest variation (interquartile range 204 to 406 minutes). The box plot in Figure 2 shows that ambulance diversion was most common on Mondays, with a median of 265 minutes per interval, whereas Saturdays had the least, with a median of 120 minutes. In univariate analyses, admitted patients boarded in the ED and one of the physician variables were associated with ambulance diversion: each admitted patient boarded in the ED per interval was associated with 9 additional minutes of ambulance diversion per interval (95% CI 5 to 12 minutes). Nurse hours and the remaining physician variables were not predictors. Several covariates were associated with ambulance diversion: ambulance-delivered patient volume (6 minutes; 95% CI 3 to 8 minutes), total number of admitted patients (8

No. of intervals

Frequency distribution of crowding duration and box plot of duration by shift (inset).

210 200 190 180 170 160 150 140 130 120 110 100 90 80 70 60 50 40 30 20 10 0

205

500

400

Night shift Day shift Evening shift

Minutes of ambulance diversion

Figure 1.

minutes; 95% CI 4 to 12 minutes), boarding time (15 minutes; 95% CI 9 to 20 minutes), day shift (77 minutes; 95% CI 60 to 93 minutes), evening shift (182 minutes; 95% CI 165 to 199 minutes), and weekend (–50 minutes; 95% CI –68 to –32 minutes). In the multivariate analysis (Table 2), the number of admitted patients boarded in the ED was an important predictor of increased ambulance diversion. For every admitted patient boarded in the ED, there were an additional 6 minutes (95% CI 3 to 10 minutes) of diversion per interval, a 3% increase over the mean. ED nurse hours were not associated with crowding. Thirteen of 15 emergency physicians were not associated with ambulance diversion, and 2 were (1 with a decrease of 36 minutes per interval [95% CI –65 to – 7 minutes] and the other with an increase of 48 minutes per interval [95% CI 5 to 91 minutes]). The volume of ambulance-delivered patients was associated with ambulance diversion; each additional

300

200

100

79

85

79

0 Night

Day

Evening

Shift

69

63

63

60 46

63 55

51

49

48

46 34

30

0–

0

–6

31

0

–9

61

20

–1

91

50

–1

1 12

80

–1

1 15

10

–2

1 18

40

–2

1 21

70

–2

1 24

00

–3

1 27

30

–3

1 30

60

–3

1 33

90

–3

1 36

20

–4

1 39

50

–4

1 42

80

–4

1 45


APRIL 2003

41:4


4 7 1


ambulance-delivered patient added 5 minutes (95% CI 3 to 8 minutes) of diversion per interval. Walk-in and trauma patient volume did not predict ambulance diversion. All covariates relating to admitted patients were independently associated with ambulance diversion. Each additional admitted patient added approximately 5 minutes (95% CI 0.1 to 9 minutes) of diversion per interval, and each additional hour in mean assessment and boarding times added 10 (95% CI 3 to 16 minutes) and 11 (95% CI 6 to 17 minutes) minutes, respectively, per interval. The number of inpatients awaiting placement outside the hospital was not associated with ambulance diversion. The seasonal variable representing the evening time period remained a strong independent predictor of ambulance diversion in the multivariate analysis (135 minutes; 95% CI 98 to 172 minutes), as did weekend

days (–40 minutes; 95% CI –58 to –22 minutes). Nursing workload was not associated with crowding in the univariate analysis. The correlation matrix revealed strong correlations (r≥0.80) between nursing workload and patient volume variables; hence, the workload measure was removed from the multivariate model. We conducted 4 sensitivity analyses. We included the volume of ambulance-delivered patients from the current interval in the model in addition to the lagged ambulance volume, but this did not substantially change the model, and hence, it was not retained. We explored the nursing workload measure in 2 analyses, first by defining it as an average per patient per shift (to render it independent of patient volume) and second by including an interaction term between nurse hours and workload, but neither were associated with the outcome. Finally, we grouped emergency physicians

Figure 2. 500

Median number of minutes of ED crowding per day.


400

300

200

100

0 Sunday

Monday

Tuesday

Wednesday

Thursday

Friday

Saturday

Day

4 7 2


41:4

APRIL 2003


according to the number of shifts worked, but the variable was not associated with ambulance diversion in the final model. The time series data met the assumption of stationarity. Each model included 2 seasonal dummy variables to account for seasonality (ie, evening peaks and nighttime lows) in the dependent variable. The autocorrelation, partial autocorrelation, and inverse autocorrelation functions of the final autoregressive integrated moving average model all showed good fit, and the adjusted R2 value was 0.37.

Table 2.

Multivariate time series model of the association between predictor variables and ambulance diversion. Associated Change in Duration of Diversion (Min/Interval)

Domain

Variable

Physician on duty (No. of intervals worked)*

MD1 (49) MD2 (73) MD3 (117) MD4 (105) MD5 (134) MD6 (70) MD7 (79) MD8 (70) MD9 (76) MD10 (22) MD11 (55) MD12 (71) MD13 (37) MD14 (10) Nurse hours Trauma Walk in Ambulance Admits, No. Boarded in ED, No. Assessment time, h Boarding time, h Inpatients awaiting placement, No.† Day Evening Weekend

Nurse staffing Patient volume, No. Admitted patients

Hospital Shift (vs night) Day of week (vs weekday) *The

–29.2 –10.6 14.3 7.0 13.9 –10.7 12.7 14.4 –36.3 –11.4 –7.0 –5.2 47.6 29.1 –0.4 4.0 1.1 5.2 4.6 6.2 9.9 11.3 –0.3

–65.9 to 7.6 –44.2 to 23.1 –13.3 to 41.8 –21.4 to 35.5 –14.5 to 42.2 –43.4 to 21.9 –22.3 to 47.8 –19.7 to 48.4 –65.2 to –7.5 –49.8 to 27.0 –37.5 to 23.4 –37.2 to 26.7 4.5 to 90.6 –60.1 to 118.3 –1.5 to 0.7 –3.4 to 11.4 –0.2 to 2.4 2.7 to 7.8 0.1 to 9.1 2.6 to 9.8 3.3 to 16.5 5.6 to 17.0 –2.6 to 2.0

27.6 135.4 –39.9

–10.1 to 65.4 98.3 to 172.4 –58.0 to –21.8

reference physician was MD15. in acute care beds awaiting chronic care beds in the community.

†Patients

APRIL 2003

41:4

95% CI


DISCUSSION

Our results demonstrate that admitted patients played a major role in determining the use of ambulance diversion in the ED. The number of patients admitted from the ED, the number boarded because of bed shortages, and delays in their assessment and disposition were all important independent predictors of increased ambulance diversion. In contrast, nurse hours and the large majority of emergency physicians were not associated with the problem in our analysis. The association of patient volume with crowding depended on the mode of arrival of patients; ambulance-delivered patients were highly associated with crowding, whereas walk-in and trauma patients were not. Nursing shortages are thought to be an important cause of ED crowding,3,5,7,10,25 but we did not find an independent association between ED nursing hours and ambulance diversion. These results are similar to those of a British study that found no association between reduced ED nurse staffing and waiting times for ED patients.26 However, there are several reasons why a true association might have been missed in our study. One is reverse causality; that is, during intervals that typically have more crowding and ambulance diversion, such as evening shifts, more nurses are scheduled to work. Without detailed data on severity allowing us to carefully control for variations between intervals, we might fail to identify an association even if one was present. Second, measuring nurse hours alone does not take into account how hard individual nurses are working during those hours. If, during short-staffed intervals, nurses work substantially harder than during fully staffed intervals, we might fail to identify a true association with crowding. Indeed, in our ED, the number of stretchers and cubicles available for patients was the same regardless of nurse staffing levels (ie, when fewer nurses were on duty, each was assigned more patients). The effect of this practice on staff morale, quality of care, and patient outcome is unknown, but in other highseverity environments, such as the ICU, increased nursing workload has been associated with poorer patient outcomes.27,28 Our study was not designed to look

4 7 3


directly at the effect of nursing shortages elsewhere in the hospital, although the effect of these on the ED likely results from prolonged boarding times for admitted patients. Thirteen of 15 emergency physicians were not associated with ambulance diversion when compared with the reference physician, whereas 2 were associated with ambulance diversion. The latter 2 physicians were associated with similar but opposite effects on diversion: one reduced use by 36 minutes per interval (95% CI –65.2 to –7.5), and the other increased it by 48 minutes (95% CI 4.5 to 90.6). This suggests that variations in practice patterns or the implementation of ambulance diversion guidelines did not affect the duration of diversion most of the time, and, when it did, the effect would have had little or no net effect on the total duration of ambulance diversion over the long run. Our results clearly implicate admitted patients and delays in the admission process as important contributors to ambulance diversion in our ED. During our study period, up to 16 patients were admitted in a single interval; this alone would account for a 39% increase in duration of ambulance diversion over the median for those intervals. In situations in which inpatient beds are not readily available for admitted patients or other delays occur in their assessment or disposition, diversions can be expected to be prolonged substantially as well. Previous studies have suggested factors that might influence the duration of time admitted patients spend in the ED. Prolonged assessment times might be the result of delays in moving patients from the waiting room to an assessment area, increased patient illness severity,5,29 or delays in completing investigations or consultations. 5 A commonly cited cause of crowding is the lack of acute care inpatient beds for admitted patients,3,5,9,10,30 which our study confirmed. However, we also found the average boarding time to be an independent predictor of ambulance diversion, implying that reducing either the number of admitted patients or the average time they spend in the ED would help reduce the problem. The effect of patient volume on crowding and ambulance diversion is reported inconsistently in the litera-

4 7 4

ture.5,7,10 In some jurisdictions, increasing volumes caused by population growth or decreased access to primary care7,21,31,32 have been thought to contribute to the problem, whereas in others, crowding has worsened even without an increased volume of patients.16,21 The volume of walk-in patients was not associated with diversion, perhaps because of lower severity, fewer time-consuming investigations, and fewer admissions among these patients. Despite smaller numbers relative to the number of walk-in patients, ambulance-delivered patient volume was associated with ambulance diversion, likely reflecting a higher severity of illness and likelihood of admission.33 Despite the resource-intensive care they require, the number of trauma patients was not independently associated with diversion. We believe this is because these patients are few in number, and at our hospital, there is dedicated physical space in the ED and a designated trauma team that brings extra physician (but not nursing) staff for every case of trauma. In smaller centers without such resources, the effect of trauma patients on ED crowding might be greater. Our results might help administrators and policy makers to prioritize interventions designed to reduce ambulance diversion and ED crowding. For example, efforts to discourage patients with low-severity illness from using the ED or diverting them elsewhere after their arrival might not result in a substantial benefit. In contrast, the presence of 2 fewer admitted patients boarded in the ED per interval or a reduction of the mean assessment or boarding times by 1 hour would each be expected to lead to a reduction in the duration of ambulance diversion of approximately 6%. Achieving this would require a 20% decrease in the assessment time, a 30% decrease in the boarding time, and a 34% reduction in admitted patients boarded in the ED. Because the resources required to achieve such change would need to be balanced with their likely effect, our study provides guidance to decisionmakers, who must consider trade offs when trying to decide where scarce resources should best be directed. As a retrospective observational study relying on administrative data, this study suffers from a number of limitations inherent to its design. First, the study’s


41:4

APRIL 2003


focus on a single hospital system could limit the generalizeability of our results. For example, in other hospitals, nursing shortages might prompt the temporary closure of ED stretchers and therefore contribute to crowding. Second, like several previous studies,19-21 we chose to use ambulance diversion as a proxy measure of ED crowding in the absence of a widely accepted definition. We recognize that ambulance diversion is 1 of several manifestations of the problem of crowding; however, we believe it is highly appropriate because it reflects the compromised ability of an ED to receive acutely ill patients. Providing rapid medical care to acutely ill patients is the prime mission of EDs,1 and this frequently depends on rapid ambulance transport to an accessible ED. Delayed access by ambulance diversions could, therefore, compromise health in the community.34 Nonetheless, the implementation of ambulance diversion might vary within different regions and hospitals and might even vary between individuals within the same hospital if no standard guidelines are followed. In our study setting, such guidelines existed; however, no guideline can be sufficiently precise as to exclude human judgment. Hence, individuals might have varied interpretations of the guidelines in response to similar degrees of crowding. We used a lagged measure of ambulance-delivered patient volume as a result of the relationship between this variable and the outcome. Other proxy measures of crowding might eliminate the need to lag this variable. We believe that lagging the variable was appropriate because ambulance-delivered patients tend to be older, tend to be more acutely ill, tend to spend a longer time in the ED, and are more likely to be admitted. Thus, they are likely to influence conditions in the ED long after their arrival. Furthermore, a sensitivity analysis showed that the results did not change even when current-interval ambulance-delivered patients were included as well. We chose to measure ambulance diversion over 8-hour intervals because this allowed us to best model its seasonality (highest in the evening, lowest at night, medium in the daytime) and because several key variables varied or were measured at this frequency (eg, physician on duty, boarded ED patient

APRIL 2003

41:4


census). However, for some variables, this meant that we had to assume that data collected only on a daily basis were the same for all 3 intervals (eg, alternate-care patients awaiting placement), thus reducing the variation per interval and possibly the association of these variables with ambulance diversion. Finally, there are a number of other potential causes of crowding for which we had no data available, such as patient-specific factors like age, severity of illness, or comorbidity.5,29,35-38 We believe that our results point to 3 main conclusions. First, variations in physician practice and nurse staffing appear to contribute little to ambulance diversion and, hence, crowding in our setting. Second, admitted patients represent a minority of patients seen in the ED but contribute disproportionately to ambulance diversion. Third, reducing the volume of walk-in patients is unlikely to alleviate the need for EDs to use ambulance diversion to alleviate crowding. Therefore, the search for solutions should be focused on increasing the institutional efficiency of the evaluation and disposition of complex patients, as opposed to encouraging ambulatory patients to seek care elsewhere. Author contributions: MJS conceived the study, designed the methodology, obtained funding, and had main responsibility for drafting and revising the manuscript. SM oversaw the data collection. KL assisted with data collection, data analysis, and helped draft the manuscript. MV oversaw the data analysis, managed the data, and ensured quality control. LJM helped develop the study protocol, assisted in the analysis, and manuscript preparation. All authors participated in manuscript editing. MJS takes responsibility for the paper as a whole. Received for publication March 25, 2002. Revisions received August 1, 2002, and September 5, 2002. Accepted for publication September 11, 2002. Presented at the Society for Academic Emergency Medicine annual meeting, St. Louis, MO, May 2002; and the Canadian Association of Emergency Physicians annual meeting, Hamilton, Ontario, Canada, April 2002. Supported by a grant from the Social Science and Humanities Research Council of Canada (No. 416573). Dr. Schull also has a career award from The Canadian Institutes for Health Research and the Peter Lougheed Foundation (No. PLS47850). Address for reprints: Michael J. Schull, MD, MSc, Room G147, 2075 Bayview Avenue, Toronto, Ontario, Canada M4N 3M5; 416-480-3793, fax 416-480-6048; E-mail [email protected].

4 7 5


REFERENCES 1. Henry M. Overcrowding in America’s emergency departments: inpatient wards replace emergency care. Acad Emerg Med. 2001;8:188-189. 2. Thorpe KE. The current hospital crisis in New York City and policy options for resolving it. N Y State J Med. 1990;90:247-252. 3. Gallagher EJ, Lynn SG. The etiology of medical gridlock: causes of emergency department overcrowding in New York City. J Emerg Med. 1990;8:785-790. 4. Andrulis DP, Kellermann A, Hintz EA, et al. Emergency departments and crowding in United States teaching hospitals. Ann Emerg Med. 1991;20:980-986. 5. Derlet RW, Richards JR, Kravitz RL. Frequent overcrowding in US emergency departments. Acad Emerg Med. 2001;8:151-155. 6. Schull MJ, Redelmeier DA, Szalai J, et al. Trends in ambulance diversion for Toronto emergency departments from 1991 to 1999 [abstract]. Acad Emerg Med. 2000;7:520.

27. Tarnow-Mordi WO, Hau C, Warden A, et al. Hospital mortality in relation to staff workload: a 4-year study in an adult intensive-care unit. Lancet. 2000;356:185-189. 28. McCloskey JM. Nurse staffing and patient outcomes. Nurs Outlook. 1998;46:199-200. 29. Pineault R, Roberge D, Lambert J, et al. Les determinants organisationnels de la duree du sejour des patients sur civiere dans les services d’urgence de dix hopitaux au Quebec. Groupe de recherche interdisciplinaire en sante, editor. Montreal, Quebec, Canada: Universite de Montreal, Faculte de Medecine; 1992:R92-09, 1-132. 30. Bagust A, Place M, Posnett JW. Dynamics of bed use in accommodating emergency admissions: stochastic simulation model. BMJ. 1999;319:155-158. 31. Schappert SM. National hospital ambulatory medical care survey: 1992 emergency department summary. Vital Health Stat. 1997;13:1-15. 32. McCaig LF, Burt CW. National Hospital Ambulatory Medical Care Survey: 1999 Emergency Department Summary. Advance Data From Vital and Health Statistics; No. 320. Hyattsville, MD: National Center for Health Statistics; 2001.

Haugh R. ER diverts. Going nowhere fast [news]. Hosp Health Netw. 1999;73:22-24.

33. Lindsay P, Bronskill S, Schull MJ, et al. Clinical utilization and outcomes. In: Brown AD, ed. Hospital Report 2001: Emergency Department Care. Toronto, Ontario, Canada: Hospital Report Research Collaborative (Ontario Hospital Association and Government of Ontario); 2002:29-50.

9. Shih FY, Ma MH, Chen SC, et al. ED overcrowding in Taiwan: facts and strategies. Am J Emerg Med. 1999;17:198-202.

34. Dickinson G. Emergency department overcrowding [editorial]. CMAJ. 1989;140:270271.

10. Ontario Hospital Association, Ministry of Health. OHA Region 3: Emergency Services Working Group. Toronto, Ontario, Canada: Ontario Hospital Association; 1998. Final report, 1-79.

35. Reeder TJ, Tucker JL, Cascio ES, et al. Trends in emergency department utilization: effect of changing demographics [abstract]. Acad Emerg Med. 2001;8:577.

7. Derlet RW, Richards JR. Overcrowding in the nation’s emergency departments: complex causes and disturbing effects. Ann Emerg Med. 2000;35:63-68. 8.

11. Epstein SK, Slate DH. The Massachusetts college of emergency physicians ambulance diversion survey. Acad Emerg Med. 2001;8:526-527. 12. Viccellio P. Emergency department overcrowding: an action plan. Acad Emerg Med. 2001;8:185-187. 13. Thompson J. Coroners and lack of emergency department resources. J Emerg Med. 1999;17:541-542.

36. Baum SA, Rubenstein LZ. Old people in the emergency room: age-related differences in emergency department use and care. J Am Geriatr Soc. 1987;35:398-404. 37. Kendrick S. The pattern of increase in emergency hospital admissions in Scotland. Health Bull (Edinb). 1996;54:169-183. 38. Beland F, Lemay A, Philibert L, et al. Elderly patients use of hospital-based emergency services. Groupe de recherche interdisciplinaire en sante, editor. Montreal, Quebec, Canada: Universite de Montreal, Faculte de Medecine. N89-02, 1-21.

14. Le Coroner en chef, Gouvernement du Quebec. Rapport D’enquete: Deces de Jeannine Lacombe. Montreal, Quebec, Canada, September 1998. Dossier 95356. 15. Regional Coroner Central Region. “Kyle Martin” inquest report #9801205. Toronto, Ontario, Canada: Office of the Chief Coroner; 1998. 16. Chan B, Schull MJ, Schultz S. Atlas of Emergency Department Services in Ontario 1992/1993 to 1999/2000. Toronto, Ontario, Canada: Institute for Clinical Evaluative Sciences. ICES Atlas Report Series; 2001. 17. Schull MJ. Influenza virus as a determinant of emergency department overcrowding. Can J Emerg Med. 2001;3:126-127. 18. Canadian Health Services Research Foundation. Myth: More Money Would Put an End to Emergency Room Crunches. Ottawa, Ontario, Canada: Canadian Health Services Research Foundation; 2000. 19. Schull MJ, Slaughter PM, Redelmeier DA. Emergency department overcrowding: defining the problem and eliminating misconceptions. Can J Emerg Med. 2002;4:76-83. 20. Schull MJ, Morrison LJ, Vermeulen M, et al. Emergency department overcrowding and ambulance transport delays for patients with chest pain. CMAJ. 2003;168:277-286. 21. Schull MJ, Szalai J, Schwartz B, et al. Emergency department overcrowding following systematic hospital restructuring: trends at twenty hospitals over ten years. Acad Emerg Med. 2001;8:1037-1043. 22. Boutros F, Redelmeier DA. Effects of trauma cases on the care of patients who have chest pain in an emergency department. J Trauma. 2000;48:649-653. 23. Pindyck RS, Rubinfeld DL. Econometric Models and Economic Forecasts. New York, NY: McGraw-Hill; 1998. 24. Dickey DA, Fuller WA. Distribution of the estimators for autoregressive time series with a unit root. J Am Stat Assoc. 1979;74:427-431. 25. American College of Emergency Physicians Task Force. Hospital and emergency department overcrowding. Ann Emerg Med. 1990;19:336. 26. Audit Commission. Accident and Emergency. Review of National Findings. London, United Kingdom: The Audit Commission. Acute Hospital Portfolio; 2001. Report 2, 1-16.

4 7 6


41:4

APRIL 2003