You are on page 1of 16

HEALTH POLICY AND CLINICAL PRACTICE/ORIGINAL RESEARCH

Crowding Delays Treatment and Lengthens Emergency


Department Length of Stay, Even Among High-Acuity Patients
Melissa L. McCarthy, ScD From the Department of Emergency Medicine, Johns Hopkins University School of Medicine,
Scott L. Zeger, PhD Baltimore, MD (McCarthy, Ding, Levin); the Department of Biostatistics, Johns Hopkins University
Bloomberg School of Public Health, Baltimore, MD (Zeger); the Department of Emergency
Ru Ding, MS
Medicine, University of Michigan School of Medicine, Ann Arbor, MI (Desmond); the Department of
Scott R. Levin, PhD Emergency Medicine, George Washington University School of Medicine, Washington, DC (Lee);
Jeffrey S. Desmond, MD and the Department of Biomedical Informatics and the Department of Emergency Medicine,
Jennifer Lee, MD Vanderbilt University Medical Center, Nashville, TN (Aronsky).
Dominik Aronsky, MD, PhD

Study objective: We determine the effect of crowding on emergency department (ED) waiting room, treatment,
and boarding times across multiple sites and acuity groups.

Methods: This was a retrospective cohort study that included ED visit and inpatient medicine occupancy data for
a 1-year period at 4 EDs. We measured crowding at 30-minute intervals throughout each patient’s ED stay. We
estimated the effect of crowding on waiting room time, treatment time, and boarding time separately, using
discrete-time survival analysis with time-dependent crowding measures (ie, number waiting, number being
treated, number boarding, and inpatient medicine occupancy rate), controlling for patient demographic and
clinical characteristics.

Results: Crowding substantially delayed patients’ waiting room and boarding times but not treatment time.
During the day shift, when the number boarding increased from the 50th to the 90th percentile, the adjusted
median waiting room time (range 26 to 70 minutes) increased by 6% to 78% (range 33 to 82 minutes), and the
adjusted median boarding time (range 250 to 626 minutes) increased by 15% to 47% (range 288 to 921
minutes), depending on the site. Crowding delayed the care of high-acuity level 2 patients at all sites. During
crowded periods (ie, 90%), the adjusted median waiting room times of high-acuity level 2 patients were 3% to
35% higher than during normal periods, depending on the site and crowding measure.

Conclusion: Using discrete-time survival analysis, we were able to dynamically measure crowding throughout
each patient’s ED visit and demonstrate its deleterious effect on the timeliness of emergency care, even for
high-acuity patients. [Ann Emerg Med. 2009;54:492-503.]

Provide feedback on this article at the journal’s Web site, www.annemergmed.com.


0196-0644/$-see front matter
Copyright © 2009 by the American College of Emergency Physicians.
doi:10.1016/j.annemergmed.2009.03.006

SEE EDITORIAL, P. 511. treated at EDs in Ontario, Canada, and found that increasing
numbers of low-complexity patients did not significantly
INTRODUCTION lengthen the wait time or ED length of stay of higher-
Background complexity patients.
Length of stay is an important measure of quality of care in
the emergency department (ED).1-3 Several studies have found Importance
that crowding is associated with increased ED length of stay.4,5 Many hospital-based EDs across the country have been
Rathlev et al4 observed that daily ED length of stay was struggling with crowding for more than a decade.7,8 Until
positively associated with the hospital occupancy rate and recently, there were few objective measures of crowding and
number of emergency admissions at their institution. Asaro et modest evidence of the negative effects of crowding on
al5 found that at their ED, crowding factors increased wait time patient care and outcomes. However, advances in crowding
and boarding time but not treatment time. In contrast, Schull et measures during the past 5 years have resulted in a growing
al6 investigated the effect of the number of low-complexity number of studies that quantify the negative consequences of
patients on the length of stay of higher-complexity patients crowding, especially on delays in care for time-sensitive

492 Annals of Emergency Medicine Volume , .  : October 


McCarthy et al Crowding and Length of Stay

medicine occupancy and ED staffing data. The institutional


Editor’s Capsule Summary
review board of each site approved the study by expedited
What is already known on this topic review.
Emergency department (ED) crowding often results
Setting
from limited hospital functional capacity and can
The selection of the 4 study sites was based on previous
affect patient care and outcomes. collaboration and feasibility of aggregating electronic medical
What question this study addressed record data across the different sites.14,15 The 4 study sites are
geographically dispersed; all but 1 is located in an inner city
How does ED crowding alter the time to initiate
with a population of more than 500,000 residents. All of the
care, ED treatment time, and ED boarding time
study EDs are part of tertiary care, academic medical centers
after admission, and are these effects similar across with Level I trauma centers. Three of the 4 study sites have
patient illness acuities? separate pediatric EDs whose visit data were excluded. However,
What this study adds to our knowledge if a pediatric patient was treated in the study ED, the visit was
retained. Table 1 displays the 4 sites by facility, staffing, and
In this 4-ED retrospective analysis of more than
crowding characteristics. The 4 study sites differ in terms of
225,000 visits, increased ED crowding delayed the inpatient medicine bed capacity (range 224 to 461 beds) and
start of care and lengthened boarding time without ED capacity (range 26 to 41 beds in main ED). The annual ED
altering the length of the treatment phase of ED volumes across the study sites range between 50,000 and
care. This pattern was observed in both high- and 62,000. Two of the study EDs have an observation unit and 1
low-acuity patients. has a dedicated laboratory. At 2 study sites, inpatient physicians
are primarily responsible for boarders, whereas at the other 2
How this might change clinical practice
facilities, emergency physicians remain responsible (ED nurses
These data confirm that ED crowding spares few continue to provide care to boarders at all sites).
and that sicker patients are affected, and it confirms The 4 sites also differ substantially by staffing and crowding
the need for creative solutions. factors (Table 1). The difference in the median number of
attending physicians or residents working per hour is 1 across
the sites. However, the difference in the median number of
conditions.7,9-13 One of the major limitations of crowding nurses working per hour is 11 between the ED with the most
studies conducted to date is that they measure crowding in a versus fewest nurses. The median number boarding during a
static way, either at a point in time (eg, patient arrival) or by 30-minute period at one of the study EDs is twice the median
averaging crowding during a specific interval (eg, a shift). number (median⫽10) of another ED (number⫽5). There is
However, many studies have shown that crowding is substantial variation both within and across the 4 study sites in
dynamic and can fluctuate substantially during the course of the hourly ED occupancy rate. The variation in the inpatient
a patient’s ED stay.14-17 medicine occupancy rate within and across the study sites is
smaller but still considerable.
Goals of This Investigation All 235,928 ED visits that occurred at one of the 4 study
The purpose of this study was to quantify the relationship sites during a 1-year period (October 1, 2006, to September
between crowding and ED length of stay at 4 hospital EDs 30, 2007, for 3 sites and October 1, 2005, to September 30,
and to compare the effect across EDs and patient acuity 2006, for 1 site) were eligible for the study. We eliminated
levels. We measured crowding at regular intervals throughout 4% of ED visits across all 4 sites because of registration
each patient’s ED stay and estimated the cumulative effect of errors (N⫽5,986), because patients were sent elsewhere and
crowding on ED waiting room time, treatment time, and not treated in the ED (ie, mainly labor and delivery)
boarding time, using discrete-time survival analysis, which (N⫽2,537), or because patients had multiple missing
allowed us to include time-varying crowding covariates. We (N⫽226) or prolonged ED times (ie, wait time ⬎12 hours,
stratified the analysis by shift, site, and patient acuity and treatment and boarding time ⬎48 hours) (N⫽645). We
determined whether crowding, adjusted for other factors, were left with 226,534 ED visits.
significantly delayed completion of each phase of ED care.
Methods of Measurement
MATERIALS AND METHODS The premise of this study was that when crowding occurred,
Study Design demand exceeded capacity and delays were more likely. During
This investigation was based on a retrospective cohort design the course of a patient’s ED stay, crowding levels fluctuate,
that included all ED visits to one of 4 hospital EDs during a often substantially. Thus, an analysis of crowding must take into
1-year period. Patient visit information was abstracted from the account the time-varying nature of crowding. To capture
information system of each ED and combined with inpatient crowding changes during each patient’s length of stay, we used a

Volume , .  : October  Annals of Emergency Medicine 493


Crowding and Length of Stay McCarthy et al

Table 1. Description of study sites.


Site A Site B Site C Site D
Facility
Annual ED volume 57,691 56,832 50,824 61,187
No. of staffed ED beds (IQR) 46 (46, 53) 43 (43, 48) 43 (41, 45) 36 (26, 36)
Main ED beds 32 27 41 26
Fast track beds 7 5 4 10
Observation beds 14 16 0 0
Level I trauma center Yes Yes Yes Yes
No. of inpatient medicine beds 224 287 461 246
Where majority of laboratory services ED Central Central Central
processed
Point-of-care testing available Yes Yes No Yes
ED physicians responsible for ED boarders Yes No No Yes
Left without being seen rate (%) 3 4 3 3
Staffing
Median no. of physicians per hour (IQR) 3 (1, 4) 3 (2, 3) 3 (2, 4) 2 (1, 2)
Median no. of residents per hour (IQR) 4 (4, 5) 3 (2, 3) 4 (4, 5) 4 (4, 4)
Median no. of nurses per hour (IQR) 15 (14, 17) 21 (18, 22) 14 (12, 15) 10 (8, 11)
Median hourly staffing/bed (IQR) 0.45 (0.42, 0.47) 0.54 (0.49, 0.60) 0.51 (0.45, 0.53) 0.50 (0.47, 0.53)
Median hourly staffing/patient (IQR) 0.52 (0.45, 0.62) 0.56 (0.47, 0.68) 0.58 (0.48, 0.74) 0.43 (0.35, 0.54)
Crowding
Median no. waiting/30 min (IQR) 8 (5, 12) 4 (2, 8) 2 (1, 6) 7 (3, 12)
Median no. being treated/30 min (IQR) 28 (24, 33) 33 (27, 38) 22 (17, 28) 23 (17, 28)
Median no. boarding/30 min (IQR) 5 (3, 7) 7 (4, 11) 10 (5, 16) 6 (4, 9)
Median hourly ED occupancy rate (IQR) (%) 86 (70, 102) 97 (77, 120) 75 (57, 93) 116 (89, 143)
Median hourly inpatient medicine 83 (77, 87) 90 (86, 93) 78 (75, 80) 83 (77, 87)
occupancy rate (IQR)
Median ED length of stay, min (IQR)
Discharged 240 (109, 451) 285 (163, 465) 238 (144, 374) 231 (152, 343)
Admitted 463 (318, 663) 449 (319, 642) 397 (218, 748) 481 (358, 630)

“counting process approach.”18 We divided each patient’s respectively. Thus, the sum of patients during each 30-
length of stay into a sequence of contiguous time periods and minute interval was not limited to integer values, but may be
observed when (ie, which interval) each patient completed a counted as a fraction.
particular phase of ED care. To characterize hospital crowding, we also measured the
We measured 3 phases of ED care: (1) waiting room time, inpatient medicine occupancy rate. Because the majority of ED
defined as time from registration to room placement; (2) patients are admitted to medicine wards, we considered the
treatment time, defined as time from room placement to inpatient medicine units to have a greater effect on ED length of
disposition decision; and (3) boarding time, defined as time stay than the overall hospital census. For 3 of the sites, the
from disposition decision to ED-hospital transfer for patients inpatient medicine occupancy rates were available hourly
admitted. Completion of waiting room time was measured in throughout the study period, so we used the rate closest to the
15-minute intervals (eg, noon to 12:14 PM, 12:15 to 12:29 PM), previous hour (eg, noon) as the crowding value for the intervals
whereas completion of treatment and boarding time was that started on the half hour (eg, 12:30 PM to 12:59 PM). For
measured in 30-minute intervals (eg, noon to 12:29 PM, 12:30 one site, the inpatient medicine occupancy rate was measured
to 12:59 PM). every 6 hours, so we interpolated it linearly between each
We measured multiple crowding factors corresponding to measurement period.19 To validate the linear interpolation
the 3 phases of emergency care. We divided the 1-year study approach we used at this one site, we linearly interpolated the
period into 30-minute intervals and measured crowding hourly measures at the 3 other sites and found a strong
every 30 minutes (starting on the hour and the half-hour) correlation between the interpolated and observed values
throughout the study period. Crowding measures capture the (r2⫽0.77, 0.94, and 0.97).
number of patients within each phase of care at any time The bottom half of Figure 1 depicts how we characterized
during each 30-minute discrete interval. For example, a completion of the different phases of ED care for 6 patients
patient being treated during an entire 30-minute interval is randomly selected from a single day at one of the sites. The
counted as 1 treatment patient. A patient who is treated for first patient on the far left waited 8 minutes before entering
10 minutes in the interval and then boards for the remaining the treatment phase, so the first circle is blackened to
20 minutes is counted as one-third treatment patient and illustrate completion of waiting room time in the first 15-
two-thirds (ie, 20 minutes/30 minutes) boarding patient, minute interval. The first patient completed treatment in

494 Annals of Emergency Medicine Volume , .  : October 


McCarthy et al Crowding and Length of Stay

Figure 1. The top half of the figure displays the values for 3 of the crowding measures at 30-minute intervals for a single,
random day during the study period. This approach allowed us to capture any changes in these crowding factors every 30
minutes throughout the study period. The bottom half of the figure displays how we measured the completion of each
phase of ED care for 6 randomly selected patients during the same day. Completion of waiting room time was measured
every 15 minutes because it is typically short, whereas completion of treatment and boarding time was measured every
30 minutes.

323 minutes, so treatment is depicted as completed by the shift (8 AM to 4 PM, 4 PM to midnight, and midnight to 8 AM).
11th triangle, which is blackened. Finally, the first patient We obtained the physician, resident, and nurse staffing
boarded for 200 minutes, and this is depicted by the seventh schedules from each site and initially included staffing in our
square, which is blackened. analyses. However, this was not necessary after stratifying by
The top half of Figure 1 shows that we also measured shift because staffing did not vary significantly within a shift at
crowding variables throughout each patient’s ED stay. Because each site. We excluded the weekend intervals (ie, from Friday
completion of waiting room time was measured every 15 night midnight until midnight on Sunday) from the main
minutes and the crowding variables were measured every 30 analysis because resources are different on the weekends and
minutes, we used the crowding values from the 30-minute crowding is less common at the study sites. However, Table E1
interval that contained the 2 waiting room intervals (eg, the (available online at http://www.annemergmed.com) includes
number of patients being treated in the interval from midnight the results of the main analysis according to weekend ED visits.
to 12:29 AM was used to reflect crowding in the midnight to
12:14 AM interval, as well as the 12:15 AM to 12:29 AM interval). Data Collection and Processing
For each phase of ED care, we characterized the distribution This study relied primarily on ED visit and hospital census data.
of the probability that a particular phase of care would be For each ED visit, the following data elements were extracted from
completed in each interval, given that it had not been each site’s ED information system: (1) date and time of registration;
completed in previous intervals. To estimate the median time to (2) date and time of room placement; (3) date and time of initial
completion of care, we calculated the probability of completing contact with physician or midlevel provider, if available; (4) date
care during each successive interval, using a survivorship and time of disposition decision; (5) date and time of disposition or
function. Because the survivorship function is a step function, transfer to inpatient ward if admitted; (6) disposition status; (7)
we linearly interpolated the median time to completion of each demographic characteristics (age, sex, insurance status); (8) triage
phase of care.20 At each interval, we also estimated the effect of level; (9) mode of arrival; and (10) chief complaint. All of the sites
crowding on the probability of completion of a particular phase use their ED information system for operational monitoring and
of ED care. In this calculation, the effect of crowding on improvement purposes and have processes in place to ensure data
completion of care accumulates, producing the cumulative accuracy.
effect of crowding on each length of stay outcome. The ED information system at each site is different (2 have
To examine the influence of crowding across intervals with different commercial products and 2 developed their own
similar ED and hospital resources, we stratified the analysis by system). Despite the different systems, the majority of data were

Volume , .  : October  Annals of Emergency Medicine 495


Crowding and Length of Stay McCarthy et al

documented and stored in a similar way, which facilitated insurance status); and (3) clinical characteristics (chief
analysis across the 4 sites. For example, at each site, arrival time complaint, mode of arrival, and acuity level). For the time-
is documented by either a triage nurse or dedicated registration dependent variables (ie, crowding measures), we examined the
clerk immediately after the patient presents to the ED. The time proportion of subjects who completed their care by various
of transition from waiting to a treatment bed is documented by levels of crowding. For the time-invariant variables, we
bed assignment in each ED information system by a nurse or compared the median time to completion for different
technician. If bed assignment was missing, we used time to categories for each variable.
initial contact with emergency physician or midlevel provider. Third, for each site and shift we used discrete-time survival
At one site, we used the disposition time (completed for both analysis to estimate the effect of crowding separately on the
discharged and admitted patients by a physician) to determine probability of completing waiting room time, treatment time,
the time the physician decided to admit the patient, whereas at and boarding time, adjusted for demographic factors, clinical
the other 3 sites, we used an admission decision order data field characteristics, and variation in follow-up time.25 For each
completed by physicians for admitted patients. phase of ED care, we included only crowding factors that would
All 4 sites use the Emergency Severity Index to triage affect that phase of care or a future phase of ED care. For
patients. The Emergency Severity Index is a 5-level triage scale example, in the treatment time model, we included the number
that prioritizes patients according to their severity of illness and of patients being treated, the number of boarders, and the
anticipated number of resources needed. The index inpatient medicine occupancy rate, but we did not include the
differentiates patients in terms of urgency: high-acuity patients number of patients in the waiting room because subjects in the
who should be treated first (Emergency Severity Index levels 1 treatment phase should not be affected by patients who are
and 2) from those less urgent (Emergency Severity Index levels 3 “behind them” in the ED process. Similarly, in the boarding
to 5). Among those less urgent, Emergency Severity Index levels time model, we included the number of boarders and the
3, 4, and 5 are distinguished according to predicted resources inpatient medicine occupancy rate, but we excluded the number
needed to make a disposition.21-23 of patients in the waiting room and the number of patients
Each site’s ED information system has a standard list of chief being treated because those patients are not yet competing with
complaints, with the option of free text. To standardize the the boarders for an inpatient bed.
chief complaints across the 4 sites, we classified each chief To allow for the probability of completion of a phase of care
complaint according to the Reason for Visit Classification changing over time, we included a natural cubic spline of time
System used by the National Hospital Ambulatory Medical with 2 df for the wait time models, 3 df for the treatment time
Care Survey.24 models, and 4 df for the boarding time models. The number of
From the above data, we calculated each patient’s waiting degrees of freedom selected were based on the log likelihood
room time, treatment time, and boarding time. Across the 4 ratio test.26 The discrete-time survival analysis models estimate
sites, 7% of visits were missing a data element needed to the cumulative probability of completing a particular phase of
calculate either the waiting room time or boarding time. For care, and we used these cumulative probability distributions to
these visits, we imputed waiting room time and boarding time linearly interpolate the median predicted waiting room time,
according to acuity, chief complaint, and other timing treatment time, and boarding time for the study sample.
information that was available. Fourth, to evaluate the effect of each crowding variable on
length of stay, we calculated the median completion times of
Primary Data Analysis care from the fitted models under 2 different scenarios: (1)
All analyses were conducted in R, version 2.7 (available at when all crowding measures were at the 50th percentile; and (2)
http://www.R-project.org). The following analyses were when each crowding measure was at the 90th percentile,
conducted separately at each of the EDs. For readers who are holding the other crowding measures at their median values and
interested in our analytical methods, please see Figure E1 the other covariates constant. We did not estimate the effect of
(available online at http://www.annemergmed.com), which all crowding factors at the 90th percentile simultaneously
includes the commands we used and the types of output because this rarely occurred during the 1-year study period. To
generated. illustrate the overall effect of crowding, we calculated the
First, we examined the frequency distribution of ED wait average median waiting room, treatment, and boarding times
time, treatment time, and boarding time. Because all 3 across the 4 sites by different percentiles of each crowding
outcomes were positively skewed, we focused the analysis on the factor, holding the other crowding factors at 50%.
median rather than the mean time. Second, we conducted a
bivariate analysis and examined the relationship between each Sensitivity Analyses
covariate and the median time in each of the 3 phases of ED To examine the validity of our imputation methods, we
care. The following categories of covariates were examined: (1) compared the visits with missing data to those with no missing
crowding measures (number of patients waiting, number of data at each site by patient, clinical, or temporal factors. Because
patients being treated, number of boarders, and inpatient there were no large differences between the 2 groups, we reran
medicine occupancy rate); (2) patient demographics (age, sex, our models without the imputed data and compared the results

496 Annals of Emergency Medicine Volume , .  : October 


McCarthy et al Crowding and Length of Stay

Table 2. Percent distribution of patient and clinical characteristics by site.


Site A Site B Site C Site D
Characteristics Nⴝ57,691 Nⴝ56,832 Nⴝ50,824 Nⴝ61,187
Age,
⬍18 2 4 1 2
18–34 33 33 42 37
35–54 43 34 36 39
ⱖ55 22 29 21 22
Sex
Female 52 55 55 52
Male 48 45 45 48
Mode of arrival
Walk-in 80 80 77 80
Ambulance 20 20 23 20
Insurance status
Medicaid 30 12 27 13
Medicare 7 24 16 5
Self-pay 37 5 17 21
Commercial 26 59 40 61
Disposition
Discharged 77 70 72 81
Admitted 23 30 28 19
Acuity level
1 5 1 3 0
2 20 34 38 13
3 50 51 43 42
4 23 12 15 37
5 2 2 1 8
Chief complaint
Injury 15 14 18 21
Digestive 13 17 15 11
Nervous/eyes/ears 7 8 5 6
Cardiovascular 7 12 9 9
Mental 5 2 4 4
Respiratory 12 9 9 9
Genitourinary 5 5 7 5
Skin 2 2 1 2
Musculoskeletal 18 12 13 14
General symptoms 16 19 19 19
Arrival time
12 midnight–8 AM 17 17 17 17
8 AM–4 PM 47 44 44 48
4 PM–12 midnight 36 39 39 35

to the results we obtained with the models that included the level, the unadjusted median ED length of stay for patients across
imputed data. Because there were no substantial differences, the the 4 sites was similar (Table 1).
results shown include the imputed data (see Table E2 [available Table 3 displays the sites’ unadjusted median waiting room
online at http://www.annemergmed.com] for main results times, treatment times, and boarding times by patient and
without imputed data). clinical characteristics for those visits that occurred during the
week. At all 4 sites, at least 50% of ED patients admitted wait at
RESULTS least 3 hours before transfer to an inpatient bed. The clinical
The patient populations of the 4 study sites differ substantially characteristics are associated with larger differences in ED length
from one another, particularly in terms of insurance status, of stay outcomes compared with the patient characteristics. For
admission rate, and acuity level (Table 2). Site A treats more example, the median waiting room time for patients who walk
patients who are uninsured (37%) compared with the other sites in is more than double the waiting room time of patients who
(range 5% to 21%). Site D admits proportionately fewer patients arrive by ambulance. ED length of stay is longest for acuity level
(19%) than the other EDs (range 23% to 30%). Site D also treats 2 and 3 patients.
fewer highly acute patients (ie, Emergency Severity Index levels 1 Adjusting for differences in patient and clinical
and 2) (13%) compared with the other sites (range 25% to 41%). characteristics, crowding factors consistently delayed waiting
Despite the differences in demographic characteristics and acuity room time and boarding times at all of the sites; however, the

Volume , .  : October  Annals of Emergency Medicine 497


Crowding and Length of Stay McCarthy et al

Table 3. Median ED length of stay in minutes by patient characteristics.*


Waiting Room Time Treatment Time Boarding Time
Site A Site B Site C Site D Site A Site B Site C Site D Site A Site B Site C Site D
43,862 40,897 35,767 45,104 41,575 39,372 34,998 45,104 9,908 12,802 9,746 8,748
Overall 47 19 19 45 155 232 189 152 195 180 207 286
Age, y
0–34 44 26 24 54 120 201 169 145 175 167 148 255
35–54 51 20 19 44 167 252 211 156 200 181 233 280
ⱖ55 44 13 11 30 180 243 195 158 197 185 219 299
Sex
Female 51 21 22 49 166 244 196 161 200 183 228 291
Male 43 17 15 40 142 215 180 142 191 178 195 280
Mode of arrival
Walk-in 55 25 29 62 145 222 186 141 205 183 299 285
Ambulance 23 7 3 8 190 269 203 210 171 175 122 298
Insurance status
Medicaid 52 20 25 49 174 243 187 182 203 185 197 322
Medicare 48 16 14 45 181 244 208 191 205 183 266 316
Self-pay 44 24 19 41 135 216 185 157 190 168 171 256
Commercial 45 20 17 47 155 231 183 146 187 177 187 278
Disposition
Discharged 50 25 25 51 145 240 199 160 N/A N/A N/A N/A
Admitted 39 11 9 20 176 215 163 114 195 180 207 286
Acuity level
1 3 5 0 0 62 99 34 44 75 116 36 194
2 26 10 12 11 225 249 234 162 206 184 219 286
3 94 35 28 49 220 259 200 209 205 179 343 289
4 37 41 28 57 50 95 99 120 200 165 301 276
5 36 33 27 47 35 57 64 81 261 65 191 312
Chief complaint
Injury 24 18 5 39 89 170 134 126 103 157 74 253
Digestive 73 31 28 57 242 291 250 229 210 184 369 292
Nervous/eyes/ears 44 16 17 32 185 239 187 145 195 181 240 289
Cardiovascular 55 10 13 22 195 283 243 175 180 173 198 293
Mental 57 12 33 23 286 392 288 229 272 154 206 237
Respiratory 41 13 17 39 150 189 176 142 205 198 268 312
Genitourinary 64 33 37 64 166 248 206 188 171 177 214 225
Skin 42 32 34 64 65 161 101 108 224 168 394 336
Musculoskeletal 50 37 26 58 102 193 143 133 200 174 279 285
General symptoms 51 20 26 49 170 226 190 151 195 183 343 294
Arrival time
12 midnight–8 AM 39 11 9 15 185 224 201 169 200 211 212 277
8 AM–4 PM 44 17 18 42 150 248 193 146 205 195 223 304
4 PM–12 midnight 55 33 31 69 148 214 180 153 179 152 176 266
N/A discharged patients do not board.
*Patients who arrived on the weekends are excluded from this table.

magnitude of the effect varied by site. The results based on the Crowding affected the waiting room time of the EDs, with
weekday data presented in Table 4 are similar to results the shortest median waiting room times more than that of the
obtained on the weekend data included in Table E1 (available EDs with longer median waiting room times. For example, at
online at http://www.annemergmed.com). site B during the day shift, the adjusted median waiting room
The crowding factor associated with the longest waiting time increased from 26 minutes to 54 minutes (109% increase)
room times was the number of patients in the waiting room when the number of patients in the waiting room was at 90%
(Table 4). For example, during the day shift, an increase in the (11 patients) compared with 50% (4 patients).
number of waiting room patients from 50% to 90% was In the boarding time models, the number of boarders and the
associated with an increase in waiting room times of 44% to inpatient medicine occupancy rate both had independent, negative
158%, depending on the site. The number of patients being effects on boarding time. During the day shift, an increase in the
treated and the number of boarders also significantly increased number of boarders was associated with an increase in boarding
patients’ waiting room time. time of 15% to 47%, depending on the site. Crowding had the

498 Annals of Emergency Medicine Volume , .  : October 


McCarthy et al Crowding and Length of Stay

Table 4. Impact of crowding on adjusted median times in minutes using discrete-time survival analysis models stratified by shift

and site.*
8 AM To 4 PM 4 PM To Midnight Midnight to 8 AM

Site Site Site Site Site Site Site Site Site Site Site Site
Crowding Factors A B C D A B C D A B C D
Waiting room time

Median 70 26 28 68 91 87 55 113 47 15 12 29
No. waiting, 90% 101 54 73 112 115 157 132 161 67 32 18 63
No. being treated, 90% 78 41 46 76 99 108 89 126 57 23 17 37
No. boarding, 90% 74 33 50 82 98 104 96 137 48 20 18 41
Treatment time

Median 258 284 224 201 184 248 193 200 219 258 210 208
No. being treated, 90% 278 289 219 205 195 261 205 213 240 282 215 234
No. boarding, 90% 241 274 214 196 183 252 198 203 220 258 220 219
Inpatient medicine, 90% 264 296 240 205 186 249 195 201 223 258 214 211
Boarding time

Median 250 334 626 334 192 157 288 378 162 125 347 289
No. boarding, 90% 288 410 921 395 208 208 501 420 170 187 967 338
Inpatient medicine, 90% 277 454 896 354 196 159 319 397 169 136 566 314
*Predictions were calculated for patients who were between 35 and 54 years old, were female, had commercial insurance coverage, walked in, and had acuity level 3,
with a chief complaint of general symptoms.

Bolded values are statistically significant (P⬍.05).

All crowding factors estimated at median level.

greatest effect on the boarding times of patients at site C during the was similar when the number of boarders increased from 50%
overnight shift (ie, midnight to 8 AM). The median boarding time to 90% at the different sites (Table 5).
increased from 347 minutes to 967 minutes (179% increase) when
the number of boarders increased from the median (12 boarders) to LIMITATIONS
90% (20 boarders) at that site. The results of this study must be interpreted in the context
In general, crowding factors also had a negative effect on ED of the following limitations. First, this study did not attempt to
treatment times across the sites, but the increases were relatively connect the different phases of care and determine how the
small. For example, during the day shift, the biggest increase completion of care in one phase affects the completion of care in
(8%) in median treatment time (20 minutes) was at site A when another.
the number of patients being treated was high (ie, 37 patients) Second, our models did not include all potential covariates
compared with normal (ie, 29 patients). that affect ED length of stay, such as orders for diagnostic tests,
Figure 2 displays the strong effect of the different crowding treatments, or specialty consultations. Although overall
factors on ED length of stay averaged over all 4 sites during the treatment times were not significantly delayed by crowding
day shift. As observed with the site-specific data, the delay in factors, future research should address how crowding affects
care is most marked for waiting room time and boarding time. different diagnostic and treatment services.
The impact of the number of patients boarding and the Third, the definitions of different ED phases of care and
inpatient medicine occupancy rate is similar for the waiting crowding measures were based on somewhat different
room time and boarding time. operational data, depending on the site. Although this is not
Crowding affected patients differently, depending on their ideal, the benefits of multicenter participation outweighed the
acuity level (Table 5). The highest acuity patients, Emergency weaknesses of variation in operational data and the results were
Severity Index level 1 patients, were not affected by crowding. consistent across the sites.
Crowding had a strong negative effect on high-acuity level 2 Fourth, the inpatient medicine occupancy rate was collected
patients and level 3 patients. For example, the adjusted median every 6 hours at one site rather than hourly at the other sites.
waiting room time for high-acuity level 2 patients increased by However, we developed site-specific models, we linearly
13% to 29%, depending on the site, when the number of interpolated the hourly measures at one site, and the results of
patients in the waiting room increased from 50% to 90%. For inpatient medicine occupancy on ED length of stay were
level 3 patients, the same increase in the number of waiting consistent across the sites. Using the inpatient medicine
room patients was associated with a 24% to 68% increase in occupancy rate rather than the overall hospital occupancy rate
adjusted median waiting room time. The increase in the may limit the generalizability of the study. However, as
adjusted median boarding time of high-acuity level 2 patients previously found by Rathlev et al,4 hospital occupancy can also
(range 12% to 83%) and level 3 patients (range 12% to 72%) be used to measure crowding; the magnitude of the effect may

Volume , .  : October  Annals of Emergency Medicine 499


Crowding and Length of Stay McCarthy et al

Figure 2. The effect of different crowding factors averaged across the 4 sites on the adjusted waiting room time,
treatment time, and boarding time for various levels of crowding. The time patients spend in the waiting room or in the ED
boarding increases sharply as the EDs become increasingly crowded.
Table 5. Impact of crowding on adjusted median times in minutes using discrete-time survival analysis models stratified by

patient acuity and site.*

Acuity Level 1 Acuity Level 2 Acuity Level 3 Acuity Levels 4 and 5
Site Site Site Site Site Site Site Site Site Site Site Site Site Site
Crowding Factors Site A B C A B C D A B C D A B C D
Waiting room time
§
Median 11 8 N/A 21 11 11 21 50 16 12 35 26 20 12 28
No. waiting, 90% 12 8 N/A 24 12 13 27 64 27 15 54 30 26 14 43
No. being treated, 90% 11 8 N/A 22 12 15 23 60 26 19 43 27 28 15 32
No. boarding, 90% 10 8 N/A 21 12 15 27 55 21 22 50 26 23 17 33
Treatment time
§
Median 149 198 115 227 253 270 208 241 268 211 221 47 104 105 102
No. being treated, 90% 147 191 123 234 254 271 223 262 288 219 235 50 111 107 108
No. boarding, 90% 134 233 123 218 252 275 204 238 266 210 223 46 101 105 104
Inpatient medicine, 90% 154 183 113 234 261 290 212 245 271 218 222 47 104 104 103
Boarding time
§
Median 156 178 88 218 278 559 349 232 268 714 375 N/A N/A N/A N/A
No. boarding, 90% 183 206 101 244 440 1025 401 259 413 1228 449 N/A N/A N/A N/A
Inpatient medicine, 90% 141 175 86 229 320 734 384 248 328 937 394 N/A N/A N/A N/A
N/A, At site C waiting room type was not estimated because there was little variation in waiting room time (95% of patients immediately sent to room on arrival);
boarding times not estimated for acuity level 4 and 5 patients because of small number of patients admitted.
*Predictions were calculated for patients who were between 35 and 54 years old, were female, had commercial insurance coverage, walked in, and had a chief com-
plaint of general symptoms during dayshift.

Bolded values are statistically significant (P⬍.05).

Acuity level 1 estimates for site D were not calculated because of inadequate sample size.
§
All crowding variables were estimated at median level.

500 Annals of Emergency Medicine Volume , .  : October 


McCarthy et al Crowding and Length of Stay

be smaller or larger than the result we found, depending on the uninsured or the poor27; however, the results of this study
facility and patient population. and others clearly show that output factors are associated
Fifth, we used the same calendar period for the analysis at all with the biggest delays in emergency care.4,5,28,29
4 sites, but one site’s analysis is based on data provided to the Among the ED patients admitted during weekdays, 58%
data coordinating center 1 year earlier than the other 3 sites. We experienced a boarding time that was longer than their treatment
do not believe the results would have been any different at this time (range across sites 45% to 78%). Patients boarding in the ED
site if the data used were from the same year as the other 3 sites. for prolonged periods negatively affect patient flow in the ED. Each
Sixth, to better understand how hospital operations affect boarder is equivalent to the loss of 1 ED treatment space and, more
ED length of stay, it would have been advantageous to have had important, the ED staff time associated with patient care, family
information on non-ED patients who are competing with ED communication, and documentation requirements for an inpatient.
patients for hospital resources (eg, elective admissions, In this study, during normal periods, the median number of
diagnostic test requests, specialty consultations). boarders in the study EDs ranged from 5 to 13, which represents a
Seventh, a small number of pediatric patients treated at the
reduction in main ED bed capacity of 16% to 33%, depending on
four EDs were left in the analysis, but, given the small
the study site. When the bed capacity of an ED is routinely reduced
proportion, we do not believe it substantially affected the
by boarders, it signifies a critical problem elsewhere in the system.
results.
In this case, a boarding problem most likely reflects a resource
Eighth, we did not conduct any site-specific evaluations of
the accuracy of the study data. However, the sites have been constraint (ie, inadequate inpatient bed capacity) or a policy
using all of the data we relied on in this study for operational constraint (eg, bed management policies, financial incentives to
reporting purposes, so the general accuracy of the study data prioritize elective admissions over emergency ones). The results
should be relatively high. of this study suggest that we must take a system-wide approach
Finally, the study sites were all EDs that are part of tertiary to solve crowding; we cannot limit our analyses and changes
care hospitals with Level I trauma centers; they are not a to the ED.
representative sample of hospital EDs in the United States. Proportionately, crowding affected ED waiting room time
Regardless, we expect that crowding will negatively affect ED the most across the 4 sites. ED waiting room time is the
length of stay at other types of EDs if they experience frequent shortest of the 3 phases of emergency care, so one could
periods of crowding, as did the EDs in this study. misconstrue that decreasing waiting room time is the least
important. However, the ED is the front door of the hospital
DISCUSSION to almost half of all patients admitted (42%).17 It is during
To our knowledge, this is the first study to measure the waiting room time that ED patients and their families
crowding dynamically and to examine its effect on ED length form their first impressions of the facility. A number of
of stay. We found that crowding was associated with studies have found that prolonged wait time is associated
substantial delays in ED length of stay across 4 ED sites. with patient dissatisfaction and an unwillingness to return to
Moreover, crowding prolonged the ED length of stay of the same facility or to recommend it to others.30-33 Thus, it
high-acuity level 2 patients. Output factors, such as the is critical that EDs keep waiting room times to a minimum
number of patients boarding and the inpatient medicine (experts recommend no more than 15 minutes without an
occupancy rate, were associated with large delays in ED care. explanation for the delay) and develop strategies (eg,
Crowding negatively affected ED patients’ waiting room frequent communication about delays, providing
time and boarding time but not treatment time. Some sites
information on how the ED works, offering comforts in the
were more sensitive to crowding than others. The results of
waiting area) to make patients’ waiting room time as short
our study suggest that crowding has a negative effect on the
and pleasant as possible.34-36
timeliness of the delivery of emergency care; however, the
Crowding had little effect on ED treatment time across all of
magnitude of the effect varied by the crowding measure, the
phase of ED care, patient characteristics, and site. the study sites, regardless of the shift. These results suggest that
During patients’ ED stay, the factor most consistently ED providers base their clinical decisions on patient and clinical
associated with delays in ED care was the number of (ie, results of tests) characteristics rather than system factors.
boarders in the ED. Not only did the number of boarders Asaro et al5 also found at their institution that treatment time
significantly increase ED patients’ boarding time but it also was strongly influenced by patient demographic and clinical
substantially delayed their waiting room time. After adjusting characteristics and not crowding factors. Similarly, Forster et
for the number of boarders in the ED, the inpatient al37 reported that ED admission rates did not vary significantly
medicine occupancy rate was also associated with significant by hospital occupancy level. These findings imply that any
delays in boarding time across the 4 sites. There is a substantial reductions in the ED treatment phase will be related
widespread misconception among the public, policymakers, to improvement initiatives or technological advances in the
and the lay press that crowding is primarily caused by large diagnostic, treatment, or discharge process, rather than better
influxes of patient arrivals to the ED, particularly among the management of patient flow.38,39

Volume , .  : October  Annals of Emergency Medicine 501


Crowding and Length of Stay McCarthy et al

Triage did not adequately protect severely ill patients The authors would like to thank Allison Orlina, JD, for her help
from delays in care during periods of crowding. Across all of obtaining the inpatient medicine occupancy rate data for one of the
the sites, the wait time and boarding time of high-acuity study sites.
level 2 patients increased substantially during crowded
periods. These results are consistent with those of other Supervising editor: Donald M. Yealy, MD
studies that have found that crowding causes delays in
Author contributions: All of the authors were involved in the
emergency care for patients with time-sensitive conditions.11-13 study concept and design, drafting of the article, and critical
It may be that EDs need to consider alternative prioritization revision of the article for important intellectual content. The
of patients such as the See-and-Treat approach being objectives, data collection protocol, review of the analysis,
adopted in the United Kingdom that divides patients into and findings were discussed by teleconference calls with all of
those likely to be discharged versus admitted40-42 or a the investigators. MLM, JSD, JL, and DA were responsible for
manufacturing strategy that categorize patients by their acquiring the data from their sites and obtaining institutional
expected treatment rather than reported symptoms or review board approval. RD performed the data analysis under
acuity.43 the supervision of MLM and SLZ. However, all of the authors
had input into the variables considered for the analysis and
There was substantial variation in how crowding affected
how it was conducted. MLM drafted the article, and all
the study EDs. A sample size of 4 facilities is too small to
authors contributed substantially to its revision. MLM takes
draw any conclusions about how structural, procedural, or responsibility for the paper as a whole.
cultural characteristics may contribute to crowding, but the
following qualitative observations are worth noting. First, the Funding and support: By Annals policy, all authors are required
waiting room time during normal periods at 2 of the EDs to disclose any and all commercial, financial, and other
relationships in any way related to the subject of this article
was approximately equivalent to the waiting room time
that might create any potential conflict of interest. The authors
during crowded periods at the other study sites. These long have stated that no such relationships exist. See the
waiting room times during normal periods most likely reflect Manuscript Submission Agreement in this issue for examples
inefficient operations or inadequate capacity. Second, the 2 of specific conflicts covered by this statement.
EDs with the shortest waiting room times during normal
Earn CME Credit: Continuing Medical Education is available for
periods were disproportionately affected by crowding. At
this article at: www.ACEP-EMedHome.com.
these 2 EDs, the number of boarders was associated with
bigger increases in boarding time (23% to 179% across the Publication dates: Received for publication January 14, 2009.
shifts) compared with that of the other EDs (5% to 18% Revision received February 25, 2009. Accepted for publication
across the shifts). Third, the ED with the worst boarding March 3, 2009. Available online May 6, 2009.
problem had the most treatment spaces in the main ED but Address for reprints: Melissa L. McCarthy, ScD, 5801 Smith
lacked an observation unit. Finally, the 2 sites where Avenue, Davis Building Suite 3220, Department of Emergency
crowding affected the boarding times the most relied on Medicine, Baltimore, MD 21209; 410-735-6421, fax 410-735-
inpatient physicians to be responsible for ED boarders. These 6425; E-mail mmccarth@jhmi.edu.
results suggest that individual organizations face different
problems, so each facility should conduct a local assessment
to determine the best solutions to its crowding problem. In REFERENCES
1. Asplin BR. Measuring crowding: time for a paradigm shift. Acad
addition, if EDs in the United States collected a uniform Emerg Med. 2006;13:459-461.
minimum data set with standard data definitions, we could 2. Haradan C, Nolan T, Resar R, et al. Optimizing Patient Flow:
more easily make across-facility comparisons and identify Moving Patients Smoothly Through Acute Care Settings.
best practices. Cambridge, MA: Institute for Healthcare Improvement; 2003. IHI
Innovation Series White Paper.
In summary, we have used a discrete-time survival analysis
3. Joint Commission. The Joint Commission requirements. Available
approach to demonstrate the negative effect that crowding at: http://www.jcrinc.com/Joint-Commission-Requirements.
has on the timely delivery of emergency care, even among Accessed January 2, 2009.
high-acuity patients. As the evidence of the deleterious 4. Rathlev NK, Chessare J, Olshaker J, et al. Time series analysis of
effects of crowding on patient care and outcomes continues variables associated with daily mean emergency department
length of stay. Ann Emerg Med. 2007;49:265-271.
to accumulate, the federal government may need to step in as 5. Asaro PV, Lewis LM, Boxerman SB. The impact of input and
it did in the United Kingdom and set a standard of care that output factors on emergency department throughput. Acad Emerg
requires 98% of all ED patients be treated and discharged, Med. 2007;14:235-242.
admitted, or transferred within 4 hours of presentation. It is 6. Schull MJ, Kiss A, Szalai JP. The effect of low-complexity patients
our belief that without major changes in our health care on emergency department waiting times. Ann Emerg Med. 2007;
49:257-264.
delivery system, such as those undertaken in the United 7. Committee on the Future of Emergency Care in the United States
Kingdom, EDs in the United States will continue to struggle Health System. Hospital-Based Emergency Care: At the Breaking
with crowding and its negative effect on patient care.40 Point. Washington, DC: National Academies Press; 2006.

502 Annals of Emergency Medicine Volume , .  : October 


McCarthy et al Crowding and Length of Stay

8. General Accounting Office. Hospital Emergency Departments: 25. McCullagh P, Nelder JA. Generalized Linear Models. 2nd ed.
Crowded Conditions Vary Among Hospitals and Communities. London, England: Chapman & Hall/CRC; 1989.
Washington, DC: United States General Accounting Office; 2003. 26. Hastie T, Tibshirani RJ, Friedman J. The Elements of Statistical
GAO-03-460 [March 2003]. Learning: Data Mining, Inference and Prediction. New York, NY:
9. Bernstein SL, Aronsky D, Duseja R, et al. The effect of emergency Springer-Verlag; 2001.
department crowding on clinically oriented outcomes. Acad Emerg 27. Weber EJ, Showstack JA, Hunt KA, et al. Are the uninsured
Med. 2008;15:1-10. responsible for the increase in emergency department visits in
10. Hoot NR, Aronsky D. Systematic review of emergency department the United States? Ann Emerg Med. 2008;52:108-115.
crowding: causes, effects and solutions. Ann Emerg Med. 2008; 28. Khare RK, Powell ES, Reinhardt G, et al. Adding more beds to the
52:126-136. emergency department or reducing admitted patient boarding
11. Pines JM, Hollander JE. Emergency department crowding is times: which has a more significant influence on emergency
associated with poor care for patients with severe pain. Ann department congestion? Ann Emerg Med. 2009;53:575-585.
Emerg Med. 2008;51:1-5. 29. Fatovich DM, Nagree Y, Sprivulis P. Access block causes
12. Pines JM, Localio AR, Hollander JE, et al. The impact of emergency department overcrowding and ambulance diversion in
emergency department crowding measures on time to antibiotics Perth, Western Australia. Emerg Med J. 2005;22:351-354.
for patients with community-acquired pneumonia. Ann Emerg 30. Hall MF, Press I. Keys to patient satisfaction in the emergency
Med. 2007;50:510-516. department: results of a multiple facility study. Hosp Health Serv
13. Fee C, Weber EJ, Maak CA, et al. Effect of emergency department Adm. 1996;41:515-532.
crowding on time to antibiotics in patients admitted with 31. Sun BC, Adams J, Orav EJ, et al. Determinants of patient
community-acquired pneumonia. Ann Emerg Med. 2007;50:501- satisfaction and willingness to return with emergency care. Ann
509. Emerg Med. 2000;35:426-434.
14. McCarthy ML, Zeger SL, Ding R, et al. The challenge of predicting
32. Sun BC, Adams JG, Burstin HR. Validating a model of patient
demand for emergency department services. Acad Emerg Med.
satisfaction with emergency care. Ann Emerg Med. 2001;38:527-
2008;15:337-346.
532.
15. McCarthy ML, Aronsky D, Jones ID, et al. The emergency
33. Boudreaux ED, O’Hea EL. Patient satisfaction in the emergency
department occupancy rate: a simple measure of emergency
department: a review of the literature and implications for
department crowding? Ann Emerg Med. 2007;51:15-24.
practice. J Emerg Med. 2004;26:13-26.
16. Flottemesch TJ, Gordon BD, Jones SS. Advanced statistics:
34. Sherman SG, Sherman VC. Total Customer Satisfaction: A
developing a formal model of emergency department census and
Comprehensive Approach for Health Care Providers. San
defining operational efficiency. Acad Emerg Med. 2007;14:799-
Francisco, CA: Jossey-Bass; 1999.
809.
35. Fottler MD, Ford RC. Managing patient waits in hospital
17. National Center for Health Statistics. National Hospital Discharge
emergency departments. Health Care Manag. 2002;21:46-61.
Survey 2006 Hyattsville, MD: National Center for Health
Statistics; 2008. 36. Rondeau KV. Managing the clinic wait: an important quality of
18. Klein JP, Moeschberger ML. Survival Analysis: Techniques for care challenge. J Nurs Care Qual. 1998;13:11-20.
Censored and Truncated Data. 2nd ed. New York, NY: Springer- 37. Forster AJ, Stiell IG, Wells GA, et al. The effect of hospital
Verlag; 2009. occupancy on emergency department length of stay and patient
19. Twisk J, de Vente W. Attrition in longitudinal studies: how to deal disposition. Acad Emerg Med. 2003;10:127-133.
with missing data. J Clin Epidemiol. 2002;55:329-337. 38. Gardner RL, Sarkar U, Maselli JH, et al. Factors associated with
20. Singer JD, Willet JB. Applied Longitudinal Data Analysis: Modeling longer ED lengths of stay. Am J Emerg Med. 2007;25:643-650.
Change and Event Occurrence. New York, NY: Oxford University 39. Storrow AB, Zhou C, Gaddis G, et al. Decreasing lab turnaround
Press; 2003. time improves emergency department throughput and decreases
21. Eitel DR, Travers DA, Rosenau AM, et al. The Emergency Severity emergency medical services diversion: a simulation model. Acad
Index triage algorithm version 2 is reliable and valid. Acad Emerg Emerg Med. 2008;15:1130-1135.
Med. 2003;10:1070-1080. 40. Alberti G. Emergency care ten years on: reforming emergency care
22. Wuerz RC, Milne LW, Eitel DR, et al. Reliability and validity of a [Department of Health]. Accessed December 30, 2008.
new five-level triage instrument. Acad Emerg Med. 2000;7:236- 41. Leaman AM. See and treat: a management driven method of
242. achieving targets or a tool for better patient care? one size does
23. Wuerz RC, Travers D, Gilboy N, et al. Implementation and not fit all. Emerg Med J. 2003;20:118.
refinement of the Emergency Severity Index. Acad Emerg Med. 42. Kinsman L, Champion R, Lee G, et al. Assessing the impact of
2001;8:170-176. streaming in a regional emergency department. Emerg Med
24. Schneider D, Appleton L, McLemore T. A reason for visit Australas. 2008;20:221-227.
classification for ambulatory care. Vital Health Stat 2. 1979;78:1- 43. Walley P. Designing the accident and emergency system: lessons
63. from manufacturing. Emerg Med J. 2003;20:126-130.

Volume , .  : October  Annals of Emergency Medicine 503


Table E1. Impact of crowding on adjusted median times in minutes using discrete-time survival analysis models stratified by shift

and site for weekend visits only.*
8 AM To 4 PM 4 PM To Midnight Midnight to 8 AM

Site Site Site Site Site Site Site Site Site Site Site Site
Crowding Factors A B C D A B C D A B C D
Waiting room time

Median 70 26 28 68 91 87 55 113 47 15 12 29
No. waiting, 90% 101 54 73 112 115 157 132 161 67 32 18 63
Pts. being treated, 90% 78 41 46 76 99 108 89 126 57 23 17 37
No. boarding, 90% 74 33 50 82 98 104 96 137 48 20 18 41
Treatment time

Median 258 284 224 201 184 248 193 200 219 258 210 208
Pts. being treated, 90% 278 289 219 205 195 261 205 213 240 282 215 234
No. boarding, 90% 241 274 214 196 183 252 198 203 220 258 220 219
Inpatient medicine, 90% 264 296 240 205 186 249 195 201 223 258 214 211
Boarding time

Median 250 334 626 334 192 157 288 378 162 125 347 289
No. boarding, 90% 288 410 921 395 208 208 501 420 170 187 967 338
Inpatient medicine, 90% 277 454 896 354 196 159 319 397 169 136 566 314
*Predictions were calculated for patients who visited the EDs on the weekends and were between 35 and 54 years old, were female, had commercial insurance cover-
age, walked in, and had acuity level 3, with a chief complaint of general symptoms.

Bolded values are statistically significant (P⬍.05).

All crowding factors estimated at median level.

Table E2. Impact of crowding on adjusted median times in minutes using discrete-time survival analysis models stratified by shift

and site with unimputed data only.*
8 AM To 4 PM 4 PM To Midnight Midnight to 8 AM

Site Site Site Site Site Site Site Site Site


Crowding Factors A B C A B C A B C
Waiting room time

Median 66 24 26 90 89 51 43 15 12
No. waiting, 90% 98 59 76 120 175 141 68 37 20
Pts. being treated, 90% 74 38 42 95 104 72 53 22 16
No. boarding, 90% 71 32 48 98 109 83 45 19 18
Treatment time

Median 258 289 230 196 251 198 224 260 217
Pts. being treated, 90% 275 296 224 206 260 208 246 282 221
No. boarding, 90% 244 279 221 194 254 204 226 261 228
Inpatient medicine, 90% 265 300 247 200 252 199 227 260 220
Boarding time

Median 244 353 728 192 157 288 159 123 389
No. boarding, 90% 291 442 1082 213 204 537 174 180 1072
Inpatient medicine, 90% 272 494 1013 198 159 322 167 131 656
*Predictions were calculated for patients who were between 35 and 54 years old, were female, had commercial insurance coverage, walked in, and had acuity level 3,
with a chief complaint of general symptoms.

Bolded values are statistically significant (P⬍.05).

All crowding factors estimated at median level.

503.e1 Annals of Emergency Medicine Volume , .  : October 


Figure E1. Sample R code to predict median time to completion of ED care.

Volume , .  : October  Annals of Emergency Medicine 503.e2


Figure E1. (Cont’d)

503.e3 Annals of Emergency Medicine Volume , .  : October 


Figure E1. (Cont’d)

Volume , .  : October  Annals of Emergency Medicine 503.e4