A Dynamic Model to Support Surge Capacity Planning ina
Rural Hospital
William Manley, RN* *
Jack Homer, PhD°
Marna Hoard, MPA, MPH* +
Sanjoy Roy, BS*
Paul M. Furbee, MA*
Daniel Summers, RN, BSN*
Robert Blake, MDt
Marsha Kimble, RNt
*West Virginia University, Robert C. Byrd Health Sciences Center, Center for Rural
Emergency Medicine, Morgantown, WV
+ Presently a Lieutenant in the United States Public Health Service, Office of the U.S.
Surgeon General
°Homer Consulting, Voorhees, NJ
+St. Joseph's Hospital, Buckhannon, WV
1 Corresponding author
West Virginia University, Robert C. Byrd Health Sciences Center,
Center for Rural Emergency Medicine, PO Box 9151, Morgantown, WV 26506-9151,
Office (304) 293-6682, Fax (304) 293-0265
bmanley@ hsc.wvu.edu
This study was funded by the Department of Health and Human Services, Health
Resources and Services Administration, Grant #1D1ARH00981-01-2.
Abstract
A system dynamics model was developed to help hospitals assess their ability to handle
surges of demand during various types of disasters. The model represents all major
flows of patients through a hospital and indicates how specific responses to a surge may
ameliorate bottlenecks and their potentially harmful effects on patients. The model was
calibrated to represent a specific hospital in West Virginia and was tested under three
quite different surge scenarios: a bus crash, a chemical plant leak, and a SARS outbreak.
Under the difficult conditions of the SARS scenario, avoidable deaths of patients awaiting
emergency care could be effectively reduced by adding reserve nursing staff not in the
emergency department, as might be expected, but in the overloaded inpatient wards. The
model can help hospital planners better anticipate how patient flows may be affected by
disasters, and identify best practices for maximizing the hospital’s surge capacity under
such conditions.
Introduction
Despite the ever-present risk of natural and man-made disasters, the healthcare
system has never been well-prepared for their potential consequences (Levitin and
Siegelson, 1996; Ghilarducci et al., 2000). The recent events of 9/11 and the emergence
of virile infectious diseases, such as Severe Acquired Respiratory Syndrome (SARS),
emphasize the need for continued planning to ensure appropriate, sufficient, and timely
healthcare system response (JCAHO, 2003).
Sufficient healthcare response includes the ability of a healthcare system to ramp up
quickly to handle a large surge in patient load— an ability known as surge capacity.
Although hospitals and public health agencies are modestly more prepared now than prior
to 9/11 (GAO-03-373, 2003; Hoard et al., unpublished data; Becker et al., 2003),
adequate planning for surge capacity is an area that has been generally overlooked
(Heinrich, 2003; Franaszek, 2002).
Many hospitals, and in particular their emergency departments (EDs), already operate
near or over capacity much of the time (JCAHO, 2003). ED overcrowding is caused by a
number of inter-connected issues (Fatovich, 2002) and can compromise quality of care
(Fatovich, 2002) and safety (Gore and Hughes, 2003).
The Center for Rural Emergency Medicine (CREM) decided to pursue System
Dynamics modeling as a way to evaluate the dynamic complexities of disaster planning
in healthcare. Although the issue of surge capacity can span the entire healthcare
spectrum, the project team decided to focus initially on the surge capacity of a small rural
hospital. Our broader research agenda is outlined in the paper, “Systems modeling in
support of evidence-based disaster planning for rural areas” (Hoard et al., 2005).
Hospital Readiness
Hospitals of all sizes are now required to have disaster plans. Foremost in these plans
is a hospital's ability to effectively address surge capacity. If they cannot do this within
their own system, then they must take steps to shore it up, either with permanent or
temporary (reserve) resources, or perhaps through flow-control methods such as triage,
transfer, and early discharge.
A recent report by the General Accounting Office (GAO) suggests three indicators for
identifying an overcrowding event: the number of hours that an ED is on ambulance
diversion, the proportion of patients (so-called “walkouts”) who voluntarily leave the ED
because of the delay in receiving a medical evaluation, and the proportion of patients and
length of time that patients “board” in the ED. Ambulance diversion refers to the fact that
when the ED is overcrowded, a hospital may decide to divert new ambulance arrivals to
other hospitals (Fields, 2004). ED boarders are patients who have been completed ED
care and are waiting to get admitted to a staffed inpatient bed. Increased ED boarding
reflects an inability to move patients who were already screened and stabilized by
emergency staff out of the ED and into inpatient beds (GAO-03-60, 2003; Asplin, 2003).
Many hospitals experience a combination of fluctuating and unpredictable demands
of ED admissions and a persistently high inpatient census (Burt, 2004; Capwell, 1996;
Crossen- White, 1998). Researchers have identified the need to study the ways in which a
random surge of emergency admission may affect the utilization of ED and inpatient beds
(Bagust et al., 1999). A previous system dynamics model examined the impact of hour-
by-hour demand on an Accident and Emergency Department in the United Kingdom,
with particular emphasis on how waiting times for admission may lead to the
postponement or cancellation of elective procedures (Lane et al., 2000).
Traditionally, EDs have provided a community benefit as the one place available to
patients all hours of the day regardless of ability to pay, and they are critical for
responding to emergencies of any size or type (e.g., public health, terrorism and natural
disasters). Crowding limits the ability of EDs to fulfill their community benefit function
in responding to such emergencies (Cunningham and May, 2003). However, in recent
years, hospital administrators have had to respond to the financially-driven need for
greater operational efficiency by reducing bed capacity and increasing occupancy rates.
Consequently, even within the limits of normal day-to-day operations, hospitals and their
EDs may be stretched to the limits by modest surges that test their ability to manage and
streamline existing resources.
Rural America has its own particular challenges with healthcare delivery. Seventeen
percent of the U.S. population lives in non-metropolitan areas (ERS, 2003), and hospitals
in rural communities serve as a logical focal point for community reaction to a
bioterrorist event (Schur, 2004) or any multiple casualty incident that tests the hospital’ s
capacity to handle a surge in patients. Most of these rural areas have limited resources
and capacity (Moscovice, 2004; ORHP, 2002), a relatively small number of providers
(AHA 2003), and volunteer rather than professional Emergency Medical Services
personnel (ORHP, 2002). Also, the geographic isolation of rural areas can greatly extend
the time required for external help to arrive in extreme situations, particularly when
exacerbated by poor road conditions and weather. These limitations of rural areas detract
from their surge capacity and complicate response planning (CRHP, 2004).
In dealing with difficult medical cases, rural hospital EDs typically stabilize and then
transfer patients to larger facilities that can provide higher levels of care (Williams et al.,
2001). But the “safety valve” of transfers may not be available when a disaster causes the
tural hospital to become isolated, or when the disaster is so widespread (as in an
infectious disease epidemic) that even the large hospitals are operating at full capacity
and cannot accept transfers.
System Dynamics Approach
Why have we taken a System Dynamics (SD) approach to studying rural hospital
surge capacity planning? Given the many variables that may affect a hospital’ s ability to
handle even a minor surge in patient flow and that distinguish one type of surge from
another, it is impossible to assess surge capacity using one simple formula involving
numbers of staff, beds, and equipment. It is also not possible to use historical data on a
hospital’s operations to assess its surge capacity. First, historical data are inadequate to
cover all possible surge scenarios. Second, hospital resources and their configuration
frequently change over time, so that a look backward at surge response may not
accurately indicate how well the hospital would perform today under similar
circumstances. Third, if we not only want to measure surge capacity, but also test ways of
improving it, the historical method obviously will not work.
SD simulation has been used by other researchers to analyze patient flows in hospitals
and other health care facilities (Homer and Hirsch, 2005). It is well-suited for testing
alternative implementation strategies in the dynamically complex and constantly shifting
settings that characterize health care, and for considering many possible conditions and
scenarios imposed upon such settings (Sterman, 2000; Homer et al., 2004).
In contrast with other formal methodologies, SD modeling allows one to consider
self-correcting and self-reinforcing feedback loops that may become important in a surge
situation, and to do so in a model that can be easily and thoroughly tested and understood.
Self-correcting feedback loops, which help to mitigate the impact of a surge, appear
wherever there are reserve resources, staff and beds that can be called upon when the
need arises. Self-reinforcing feedback loops, on the other hand, can exacerbate the
impact of a surge. One such loop involves the clinical deterioration of patients awaiting
admission to the ED: The more that such patients deteriorate, the more load that
represents on the ED, thereby extending further the waiting times for other patients and
allowing their further deterioration.
Methods
This project was completed in three stages by a modeling team made up of
researchers from West Virginia University (WVU) Center for Rural Emergency Medicine
(CREM) — with expertise in emergency medicine, hospital systems, and public health—
and an expert SD practitioner.
In Stage 1 we constructed an initial “straw-man” proof-of-concept model based on in-
house expertise and literature review. We also considered the various sorts of surge
scenarios that a model should be able to evaluate. Through testing the straw-man model,
we gained an initial appreciation of what might be the important assumptions and
constraints differentiating one hospital from another in terms of its surge capacity, and
the key parameters that could differentiate one surge scenario from another.
Stage 2 involved further development of the model and its application to an actual
hospital in rural West Virginia. St. Joseph’s Hospital (SJH) was selected for its location
and size, and because hospital management was enthusiastic about the aims and approach
of the project and understood that modeling might help them to do a better job of disaster
planning. Several employees of SJH participated, including the managing physician of
the ED, nurse managers and schedulers, and hospital administrators. Three day-long
sessions at SJH (1) introduced these participants to the straw-man model and the SD
approach, (2) allowed us to revise the contents of the model to better fit their
circumstances, and (3) allowed us to ensure that the scenarios and policy options being
tested were realistic in their assumptions and results.
During Stage 2, “baseline” conditions were defined, based in part on recent SJH
patient log data over a two-month period during which no unusual surges occurred. The
baseline conditions describe average patient flows (and initial census levels), as well as
nurse staffing levels, by day-of-week and hour-of-day. Surge scenarios could then be
defined in terms of incremental change to patient inflows relative to the baseline.
During Stage 3 we identified and detailed three different plausible surge scenarios.
The three cases— a bus crash, a chemical plant leak, and an outbreak of Severe A cute
Respiratory Syndrome (SARS)— represent a range in terms of their magnitude, duration,
and severity, and in terms of the types of patient care required. Outcomes from each of
the three surge scenarios were compared with the baseline to identify their differential
impacts, and potential strategies were tested to determine their ability to improve surge
capacity and reduce the adverse consequences of surges.
An Institutional Review Board duly approved protocol for the study.
Results
Model Structure
Figure 1 shows, in simplified form, the model’s depiction of patient flow through the
hospital. For example, the figure simplifies by showing “Pts await/in surgery” and “Pts
await/in elective non-surgery” as single boxes, whereas in the model, there are actually
separate boxes for “await” and for “in.” Boxes represent accumulations or stocks of
patients. The arrows with valve symbols represent flows of patients. A cloud at the end
of a flow represents a source or sink of patients that starts or ends outside the hospital.
Flows into the hospital must be specified as time series inputs to the model. These
inflows include arrivals of patients to the ED, the scheduling of elective surgeries and
non-surgeries, and direct arrivals to inpatient wards. Flows out of the hospital include
discharges from all locations (ED, elective surgery and non-surgery, wards), walkouts of
patients awaiting ED admission, and transfers of patients to other hospitals from ED and
wards.
The model contains about 1,500 parameters, of 5 types:
- Assumed constants and initial values of patient stocks (about 150)
- Assumed time series inputs (28)
- Assumed X-Y lookup functions (4)
- Calculated patient stocks (about 300)
- Calculated flows and flow rates, waiting times and other outcome variables, and
situation-dependent decision variables (about 1,000)
Patients are differentiated along four dimensions, some subsets of which apply
depending on location within the hospital:
1. Triage type (for ED patients): Green, Y ellow, Red, Black (dead) (Pedrotti and Perry,
1999);
Acuity type (for surgery and ward patients): Regular, High;
Presence of trauma (for ED patients): No, Y es;
Risk of contagion/need for isolation (for all locations): No, Y es.
Pe Cob:
The model steps through all equations at simulated 15 minute intervals, tracking day-
of-week and time-of-day as it proceeds. Simulations described here are all 20 days, or
480 hours, in length, starting at 12 AM (midnight) on a Monday and ending at 12 AM on
a Sunday.
Many of the model's equations help to specify the determination of patient flows.
Among the most intricate equations are those dealing with admission to the ED and the
inpatient wards. ED admission is modulated by the availability of constraining resources,
which in the model include ED nurses and beds. If ED admission is constrained, then
patients are prioritized based on their evaluated triage status, with Red a higher priority
than Y ellow, and Y ellow higher than Green. However, this prioritization scheme may
break down if there is a glut of patients awaiting ED admission, and ED nurses end up
falling behind on required triage or hourly re-triage of patients.
Figure 1. Overview of model stock-and-flow structure
Elective surgeries
scheduled
ED arrivals (by .
acuity, trauma, and spp | Pts waitin - |
contagion status) Post-ED directed to surgery Elective surgeries
surgery (trauma) postponed
Pts awaitED <——P| PtsinED = Post-surgery
ED admits Post-ED discharges discharges
& facility transfers
ED deaths & Post-surgery
wakouts \* wards
A ED need Post-E D directed
i to wards
vile wating egnostk raging oes ig) Pts await
p| Ward admit <—_—_
Post-non-surgery Direct arrivals
PE 7] directed to wards tiwards
's await /in lL
elective ;) Ward admits
Elective non-Surgery
non-surgeries
scheduled . 7
Pts in wards Z
Possponsumely Ward discharges
neenenge® after full stay
Elective ——rO)
non-surgeries Ward early
postponed discharges & facility
transfers
Similar to ED admission, inpatient ward admission is modulated by the availability of
ward nurses and beds. If ward admission is constrained, then high acuity patients have
higher priority for admission than regular acuity patients.
Resource constraints may also affect intake for surgery and elective non-surgical
procedures. A lack of OR nurses or beds may limit intake into surgery and cause
postponement of some elective procedures. If surgery is constrained, then emergency
trauma surgeries have higher priority than elective ones, and elective surgeries may be
postponed to the following day. In regard to elective non-surgical procedures, a lack of
outpatient nurses may limit intake and cause postponement of some of these to the
following day.
Another possible reason for postponement of elective surgeries or non-surgeries is if
there is an extended delay for inpatient ward admission. A large fraction of surgical
patients are directed to wards post-surgery, so postponement of surgery can help to
alleviate the build-up of patients awaiting ward admission. Also, though only a very
small fraction of non-surgical outpatients are directed to a ward post-surgery, such
direction usually occurs because of an acute event during the outpatient procedure
requiring immediate ward admission, which could put the wards in a difficult situation if
they are already at capacity.
The postponement of elective procedures due to ward admission delay is an example
of a compensating feedback response, or “safety valve,” that can help to alleviate
bottlenecks that may develop during a surge. The model includes other realistic feedback
responses by the hospital and by patients. These include, (1) the cancellation of some
diagnostic imaging for ED patients if there is a long wait for the same (due to insufficient
technician capacity to meet the demand); (2) the early discharge of some ward patients if
there is a long wait for ward admission; and (3) ED patient walkouts by the “walking
wounded” (triage Greens and some Y ellows) if there is a long wait for ED admission.
In addition to these compensating feedback loops, the model includes a self-
reinforcing feedback loop in the ED: a potentially troublesome vicious cycle that is
related to the possibility of patient deterioration while waiting for ED admission. If ED
nurses become temporarily overloaded leading to extended ED waiting times, some
Greens may deteriorate to Y ellow, and some Y ellows to Red, as they await admission. In
addition, a small fraction of Reds may deteriorate to Black, indicating an avoidable death.
Although death from deterioration actually alleviates demand on the ED, it is the ultimate
indicator of inadequate surge capacity. Patients with more severe conditions require a
higher nurse-to-patient ratio than those with less severe conditions, so the deterioration of
patients place an even greater load on the ED, extending waiting times, therefore leading
to the possibility of further patient deterioration.
The model was calibrated to represent SJH, in terms of its nursing and bed resources,
its typical patient inflows, and the way it responds to overload conditions. This
calibration and associated assumptions are described in A ppendix Tables 3a, 3b, and 3c.
The baseline “no surge” scenario for ED arrivals assumes— for the sake of easy
comparison— a consistent volume, unchanged from day-to-day or week-to-week; Table
3a (item 1.1), Table 3b (3.1, 4.1), and Table 3c (5.1). Given all other assumptions in the
model, simulation indicates that this volume is well handled by the hospital and results in
no significant patient bottlenecks, and only a small amount of waiting in the ED during
the peak afternoon hours of each day. The project team at SJH inspected the model’ s
response to the baseline demand scenario, and declared that it presents a faithful picture
of the hospital’s typical patient flow and load situation.
Scenario Testing
Description of surge scenarios: Note these volumes are in addition to the normal 50-
per day ED volume of the no-surge baseline.
Bus crash: 45 patients arrive ED Day 3 (Wednesday) 6-8 PM; plus 15 dead on the
scene (impact of dead patients would effect utilization of available EMS resources).
Arrival distribution: 25 G, 9 Y, 6 R, 5 B; 100% with trauma injury (no contagion).
Chemical plant leak: 100 patients arrive ED Day 1 (Monday) 2-7 PM; plus 25 dead
on the scene. Arrival distribution: 45 G, 30 Y, 25 R, 0 B; 5% with trauma injury (no
contagion).
SARS outbreak: 837 pts arrive ED Days 2-14, according to Singapore model (CDC,
2004), with peak of 106 on Day 10. Distribution: 377 G, 333 Y, 106 R, 21 B; 100%
contagious (no trauma).
Key metrics for 20-day simulations are presented in Table 1. The bus crash and the
chemical leak are short-lived events, causing only temporary difficulty, while SARS
leads to more severe problems. Probably most significant indicator of difficulty are
deaths while awaiting ED admit (avoidable deaths): 0 for the baseline and bus crash, 3 for
the chemical leak, and 109 for the SARS scenario. This is related to the number of
patient- hours awaiting ED admission, which indicates an ED admission bottleneck.
Patient hours awaiting surgery and elective non-surgery indicate the postponement of
procedures due to ward admission bottleneck.
Note that although all 45 bus crash victims have trauma injury, only 6 surgeries are
required (3 Y ellows and 3 Reds). Because they arrive late in the day, when only the OR
call team is available, only one surgery is done at a time. When the regular team arrives
the following morning, the remaining surgeries are done immediately.
Whereas the bus crash puts stress primarily on the ED, the chemical leak disaster puts
stress on both the ED and on the wards. The reason is not so much the difference in total
ED arrivals (100 vs. 45) as it is the difference in more severe arrivals: the Reds (25 vs. 6)
and the Y ellows (30 vs. 9). Itis only the more severe arrivals who are directed to wards:
30% of Y ellows, and 50% of the Reds. The other 50% of Reds are transferred to other
hospitals for specialized treatment. Long waits for ED admission (up to 13 hours) in the
chemical leak case start on Day 1 and are eliminated by Day 2, thanks in part to walkouts
by many Greens and some Y ellows. Long waits for ward admission (up to 12 hours for
regular acuity patients) start on Day 3 and extend to Day 4, leading to some elective
procedure postponements and early ward discharges.
Similar to the chemical leak disaster, the SARS outbreak creates difficulties for both
the ED and the wards, but does so for many more days and with much more severe
consequences in terms of the number of avoidable deaths, patient hours spent waiting for
admission, early ward discharges, and postponement of elective procedures. Some of
these dynamics are reflected in Figure 2, which contains graphs for key patient stocks in
ED, surgery, and inpatient wards; in each graph, see the thick black line labeled
“1, SARS base.”
In the SARS scenario, ED waiting times start to increase on Day 3, and keep
elevating steadily until they reach 40 to 50 hours by Day 13, remaining at that level for 4
days until they finally return to zero during Day 17 (three days after the end of the
outbreak). This huge bottleneck persists despite the fact that more than half of the ED
arrivals over the 20 day period of simulation walk out. Also, in a seeming paradox,
significantly fewer patients are admitted into ED care in the SARS scenario than in any
of the other three scenarios, despite the greater level of demand (see Table 1). One
reason for the reduced ED intake in the SARS scenario is that the SARS patients absorb
25% more nursing time and spend an average of 50% more time in the ED by virtue of
their being in isolation. However, the more important reason for reduced ED admissions,
is that the wards reach capacity (in terms of what the nursing staff can handle), and
consequently, ward admissions slow to a trickle. (Note that isolation of SARS-infected
patients makes things more time-consuming in the wards as it does in the ED.) This ward
bottleneck leads to growth in the number of patients “boarding” in ED awaiting ward
admission, thereby absorbing a large amount of ED nursing time and detracting from the
Table 1. Surge scenario outcomes after 20 simulated days
No Surge Chemical
Baseline | Bus Crash Leak SARS
Max ED arrivals in a day 50 95 150 156
ED arrivals 1005 1050 1105 1842
Max pts awaiting ED admit 2 38 88 115
Patient hrs awaiting ED admit 226 430 1137 18712
Deaths while awaiting ED admit 0 0 3 109
(Avoidable deaths)
ED walkouts 9 29 79 940
ED admits 991 1010 1018 767
DI studies skipped in ED 0 2 2 9
ED nurse hours utilized 1200 1230 1273 2079
Max pts awaiting surgery 0 4 i 29
Patient hrs awaiting surgery 0 54 309 3476
Max pts awaiting elective 0 0 8 13
non-surgery
Patient hrs awaiting elective 0 0 127 519
non-surgery
Max pts awaiting ward admit 0 0 8 39
Patient hrs awaiting ward admit 0 0 258 8340
Ward admits 279 289 302 344
Ward early discharges 0 0 11 43
Ward nurse hours utilized 2880 2880 3015 4231
120
Patients
s
Figure 2. SARS scenario patient stock dynamics under alternative reserve-staffing
policies
A. Patients awaiting admit to ED
B. Patients in ED
Patients
Days elapsed
4 6 8 10 12 4 6 18 2
Days elapsed
C. Patients awaiting surgery
Patients
—— 1. SARS base
—— 2. More ED nurses
3. More ward nurses
cesses 4. More ED & ward nurses
2 4 6 8 10 12
Days elapsed
4 6 18 2
E. Patients in wards
Patients
4 6 8 10 12
Days elapsed
14 16 18 20 0 2
4 6 8 10 12 14 16 18 2
Days elapsed
time ED nurses can spend with new ED admits— and therefore limiting the number of
such admits.
Note also that the postponement of elective surgeries becomes a problem in the SARS
scenario; such postponement is triggered when ward admission delays have grown
sufficiently, more than halfway through the outbreak period. These postponements begin
on Day 8, and by Day 18, more than 25 patients are awaiting surgery, and then subsiding
gradually, as the OR call team works off the backlog one surgery at a time.By the end of
Day 20, there are still 8 patients awaiting postponed elective surgeries. Thus, some
repercussions of the SARS outbreak are still felt nearly a week after the surge of ED
arrivals has concluded.
Policy Testing
A variety of simulation experiments were performed, all designed to identify where
the hospital’s procedural policies should be modified, or its reserve resources bolstered,
in order to make the hospital better to cope with demand surges. In looking for ways to
improve hospital performance under surge conditions, we have taken the lead from the
project team members at SJH Hospital, who indicated what types of changes might be
feasible.
In regard to procedural policies, we have experimented with changes in the practices
of diagnostic imaging (DI) cancellation, elective surgery postponement, elective non-
surgery postponement, and early ward discharge. The model suggests that all of these
policies, with the possible exception of non-surgery postponement, can be helpful in
managing surges. Because only one percent or so of elective non-surgeries end up
impacting the wards (which is rather insignificant, even taking into account the fact that
the cases directed to wards are typically ones of high acuity), it is not clear that there is
enough benefit from postponing such procedures to justify the disadvantages of doing so.
In regard to reserve resources, the SJH team indicated the possibility of some
increases in reserve nursing staff, specifically in the ED and inpatient ward areas, and
also allowing for a reserve DI technician. Such reserve staff could be drawn from off-
duty or part-time hospital staff, from office-based nurses and technicians in the
community, or from retired nurses and technicians in the community.
e ED reserve nurses: The baseline model assumes that in addition to the normal
complement of 3 day-shift staff and 2 night-shift staff, another 3 day nurses and 2
night nurses are available as needed during a surge. The project team felt that the
reserves could be increased to 5 day and 4 night, to give a maximum total of 8 day
(=3 regular + 5 reserve) and 6 night (=2 regular + 4 reserve) nurses in the ED during a
surge.
e Ward reserve nurses: The baseline model assumes that in addition to the normal
complement of 6 ward nurses (all hours), 4 reserves nurses (all hours as well) are
available as needed during a surge. The project team felt that the reserves could
perhaps be increased to 8, to give a maximum total of 14 (=6 regular + 8 reserve)
nurses in the wards during a surge.
11
e DlIreserve technician: The baseline model assumes one technician or technician-
equivalent (FTE) is available at all hours specifically for ED and ward patient
imaging needs. Other DI technicians may be present to work on diagnostic imaging
associated with elective procedures. The project team felt that one reserve DI
technician position (all hours) could perhaps be created, to give a maximum total of
two (=1 regular + 1 reserve) DI technicians during a surge.
We have experimented with increases in reserve staff in conjunction with the bus
crash, chemical leak, and SARS surge scenarios. Because the benefits of such increases
are more obvious when the scenario creates more difficulty for the hospital, we will focus
here on the impacts of reserve staff on SARS scenario outcomes. Table 2 presents the
key outcome metrics from five 20-day simulations, as follows:
1. SARS base: Uses baseline assumptions for reserve staff (same simulation as in the
last column of Table 1).
2. More ED nurses: Increases the ED reserve nurse contingent as described above.
3. More ward nurses: Increases the ward reserve nurse contingent as described above.
4. More ED and ward nurses: Increases both reserve nurse contingents as described
above.
5. More nurses and DI techs: Increases both reserve nurse contingents and creates a
reserve DI technician position as described above.
Figure 2 shows graphically how the first four of these simulations differ, over time, in
terms of the accumulation of patients at different points in the hospital. The fifth
simulation is similar to the fourth in its dynamics, and has been left out of Figure 2 for
the sake of legibility.
It is evident from Table 2 and Figure 2 that by adding more reserve ward nurses, it is
more effective than adding more reserve ED nurses in reducing bottlenecks and
improving hospital performance under the SARS scenario. The boosting of ward nurses
dramatically reduces the number of patients awaiting admission, not only to the wards
themselves (Figure 2d), but also to surgery (Figure 2c) and to the ED (Figure 2a),
reducing the number of avoidable ED deaths from 109 to 52. By contrast, the boosting of
ED nurses is ultimately much less effective, reducing avoidable ED deaths only to 94,
and actually causing a large increase in the numbers of patients awaiting surgery and
ward admission.
Close inspection of the graphs reveals that boosting ED nurses is initially more
effective at reducing the ED backlog (Figure 2a) than is boosting ward nurses, but only
for the first four days of the outbreak (through Day Five). After Day 4, the addition of
ED nurses becomes increasingly ineffective at reducing the ED backlog, resulting in
virtually no improvement over the baseline SARS run from Day 10 through Day 16. This
sort of diminishing policy impact over time has been described as policy resistance
(Sterman 2000). The source of the policy resistance in this case is the greater build-up of
patients awaiting ward admission when the number of ED nurses is increased but the
number of ward nurses is not. Nearly all of these waiting patients are post-ED boarders
and are still the responsibility of the ED nurses. The net effect of this backlash effect is
to neutralize the benefit of adding more ED nurses in terms of their capacity to care for
12
new ED admissions. By Day Eight, the capacity of the ED to care for new patients
(Figure 2b) is no greater with additional ED nurses than it was under the baseline policy.
The story is quite different with the boosting of ward nurses. By raising the capacity
of the wards to care for patients (Figure 2e), the ward backlog is cut substantially, and
results in a much reduced load of boarders for the ED. Consequently, although there is
no increase in the number of ED nurses, the capacity of the ED to care for new patients
(Figure 2b) is increased. This improvement starts the third day of the SARS outbreak
(Day 4) and continues until one day after the end of the outbreak (Day 15). With the
additional ward nurses, the ED backlog is eliminated two days earlier than in either the
baseline SARS run or in the run with more ED nurses only.
Next consider what happens when the numbers of ED nurses and ward nurses are
both boosted. The result is smooth performance of the entire hospital for the first 6 days
of the outbreak (Days 2 through 7), with both ED and ward backlogs staying low (Figures
2a, 2d), and the patient throughput of both ED and wards remaining high (Figures 2b,
2e). As the number of new SARS arrivals to the ED continues to grow— peaking on Day
10 before it starts to subside— ward nursing capacity becomes strained. Recall again that
isolated SARS patients are significantly more time-consuming than non-isolated patients.
Although the number of ward beds may be sufficient to accommodate a large number of
such patients, the greater time requirements may cause even an increased number of
nurses to become overloaded. Consequently, the ward backlog climbs quickly through
Day 12. This, in turn, leads to a situation in which the ED backlog is no better off from
Day 11 onward with additional ED and ward nurses than it is with additional ward nurses
alone. Nonetheless, the combination of additional ED and ward nurses still results in
fewer avoidable ED deaths than with additional ward nurses alone (41 vs. 52).
Consider finally what happens when an additional DI technician supplements the
addition of ED and ward nurses. The most obvious benefit is that the number of deferred
DI studies is cut dramatically, from 89 to 26 (Table 2). Note that the simulation with
more ED and ward nurses is the only one in which the number of DI studies not
completed is relatively large. This greater skipping of DI studies indicates that DI has
become backlogged. With the downstream bottleneck of ward admission and the
upstream bottleneck of ED admission both largely removed by adding nurses, diagnostic
imaging becomes the new location where a backlog may develop. The DI bottleneck is
less damaging than either of the other two, however, because the option of skipping DI
keeps patients flowing through the ED rather than getting stopped up there. However, the
skipping of DI does increase the fraction of Y ellows directed to wards (from 30% to
60%), and places a greater burden on the wards. That is why, when a second DI tech is
added, the number of ward admits, as well as the number of patient hours awaiting ward
admits, is reduced. The reduction of ward admits backlog in tum permits less of a build-
up of patients awaiting ED admission, and so allows for a further reduction in avoidable
deaths (from 41 to 35) when a second DI tech is added on top of the additional ED and
ward nurses.
13
Table 2. SARS scenario outcomes under alternative reserve-staffing policies
More ED | More Ward | More ED & | More nurses
SARS base nurses nurses Ward nurses | & DI techs
Max pts awaiting ED admit 115 110 89 79 74
Patient hrs awaiting ED admit 18712 16622 12271 10146 8942
Deaths while awaiting ED admit 109 94 52 41 35
(Avoidable deaths)
ED walkouts 940 856 809 686 604
ED admits 767 866 955 1088 1176
DI studies skipped in ED 9 35. 21 89 26
DI tech hours utilized 480 480 480 480 568
ED nurse hours utilized 2079 2752 1985 2568 2502
Max pts awaiting surgery 29 42 6 12 12
Patient hrs awaiting surgery 3476 5582 400 823 771
Max pts awaiting elective 13 14 7 7 7
non-surgery
Patient hrs awaiting elective 519 531 243 176 176
non-surgery
Max pts awaiting ward admit 39 56 19 49 49
Patient hrs awaiting ward admit 8340 13344 2413 6370 6230
Ward admits 344 359 395 430 425
Ward early discharges 43 59 46 65 65
Ward nurse hours utilized 4231 4225 5124 5084 4916
14
Discussion
Hospitals have an essential role to play in community preparedness for emergencies
of all sizes. The ED in a hospital is often considered the safety net for health care,
accepting all comers in all conditions. Many believe that it is time for the hospital
community to develop new strategies to support this safety net before it collapses (Adams
and Biros, 2001; Richards and Hwang, 2001). Even a relatively small influenza outbreak
can overwhelm the resources of many rural hospitals. Hospitals should develop
standardized responses for both these smaller, more commonly experienced surges as
well as for the less likely but more consequential mass casualty incidents (MCIs) that
may also occur.
Published emergency preparedness scenarios have often been oriented to urban or
densely populated areas. Although 20% of the nation lives in rural areas (Quality
Through Collaboration, 2004), little action has been take to understand the dynamics of
tural preparedness, emergency care, and public health response specifically (Williams et
al., 2001; Treat et al., 2001). This study was developed to better understand the
requirements of surge capacity preparedness for a small rural hospital.
Using the SD approach, we have developed a modeling tool that could assist rural
hospital and emergency planners in preparing for surge events that take place regularly
(e.g., influenza), as well as those that have much less chance of occurring (e.g., bio-
terrorism).
The results of the SARS scenario policy testing seem to suggest that the first, best
place for St. Joseph’s Hospital to consider additional reserve nurses is in the inpatient
wards. Only after doing so does it make sense for them also to expand the number of
reserve nurses in the ED, and perhaps provide for reserve diagnostic imaging technicians.
Can these results be generalized to other surge scenarios and other hospitals? With a
few conditional statements, we believe the answer is yes. The most important aspect of
the SARS scenario that may differentiate it from other scenarios is that a large proportion
of the patients are of sufficient acuity to require inpatient care. The need for isolation is
another salient feature of the SARS scenario, but one that, along with the overall number
of ED arrivals and the duration of the surge, has implications for the proper scaling of the
hospital overall to deal with peak-load conditions, rather than for how the reserve
resources in the hospital are balanced between ED and the wards and other locations.
Moving from the nature of the surge to the hospital itself, a couple of salient points
about SJH need to be considered. The first point is that, for historical reasons, SJH has
plenty of reserve bed capacity. Consequently, beds were not a limiting constraint in our
SJH simulations, but might become constraining in other hospitals. A second point about
SJH is that they regularly transfer a large fraction of their highest-acuity and most
difficult ED and ward patients to other hospitals. In all of the scenarios tested, we
assumed that this ability to transfer—a significant safety valve for SJH and other rural
hospitals— was not interrupted. However, it should be recognized that most large, urban
hospitals do not have such an ability to transfer patients, and that under certain disaster
conditions smaller rural hospitals may lose much of that ability.
15
Conclusion
The focus of this study was on the internal processes of a small rural hospital and the
determination of what specific locations in the hospital should receive higher priority for
reserve capacity than others. The SD model that was developed is general enough to help
hospitals of any size to plan and allocate resources more effectively and rapidly, and
possibly mitigate loss of life and prevent further spread of infection or disease. We
realize, however, that disasters can impact more parts of a healthcare system than only its
hospitals. Hirsch (2004) and other researchers have recently begun to apply SD to
investigate the implications of disasters for entire healthcare systems. We, too, would
like to extend our work on surge response in this direction, tying in the roles of
emergency medical services, as well as that of office-based physicians. We have
previously described a research agenda for studying disaster response that starts with the
hospital and works outward to encompass an entire community (Hoard et al., 2005).
As hospitals and other health planners struggle with plan development in regard to
surge capacity, SD models can provide a more realistic view of what is likely to work and
what is not, and under what circumstances. Simulations can provide useful information
to hospitals in readying themselves to confront not only catastrophic mass casualty
incidents, but also the more frequent surges that can overwhelm a small rural ED.
SD could provide healthcare professionals and policymakers the tools to better plan
for surge capacity needed for public health emergencies of any size and could assist in
developing best practices for rural healthcare preparedness.
Acknowledgements
We would like to thank the participants from St. Joseph’s Hospital— Sandra Knotts
RN, Annamarie Fidler RN, Denise Campbell RN, Kevin Stingo RN, Robert Coholich and
John Currence— for their enthusiasm and expertise in completion of this project.
In addition, we would like to thank David Fulaytar, CREM Senior Office
Administrator for his help in review and editing this paper, and Arshadul Haque, former
Principal Investigator, for his support in this endeavor.
References
Adams JG, Biros, MH. The endangered safety net: Establishing a measure of control.
Academic Emergency Medicine 2001; 8:1013-1015.
AHA. Rural Hospital Disaster Readiness, AHA Update. American Hospital Association
Section for Small or Rural Hospitals. November 2001.
Asplin BR et al. A conceptual model of emergency department crowding. Ann Emerg
Med 2003; 42:173-80.
Bagust A, Place M, Posnett JW. Dynamics of bed use in accommodating emergency
admissions: stochastic simulation model. BMJ 1999; 319:155-8.
Becker S et al. Ready or Not? Protecting the Public’s Health in the Age of Bio-
Terrorism. Trust for America’s Health Issue Report. Available at:
http://www.healthyamericans.org . Accessed on 01/08/04.
16
Burt CB, Schappert SM. Ambulatory care visits to physician offices, hospital outpatient
departments, and emergency departments: United States, 1999-2000. National Center
for Health Statistics. Vital Health Stat (in preparation).
Capewell S. The continuing rise in emergency admissions. BMJ 1996; 312:9912.
CrossenW hite H et al. The rise in emergency admissions. Report to the West Midlands
NHS Executive, 1998.
CDC. Preparedness and Response in Healthcare Facilities: Supplement C of Public
Health Guidance for Community-Level Preparedness and Response to SARS. Centers
for Disease Control and Prevention. Jan 8, 2004.
CRHP. Bridging the Health Divide: The Rural Public Health Research Agenda.
University of Pittsburgh Center for Rural Health Practice. April, 2004
Cunningham P and May J. Insured Americans drive surge in Emergency Department
Visits. Issues Brief: Center for Studying Health System Change. Washington, DC.
No. 70. October, 2003.
Economic Research Services. United States Department of Agriculture, Rural Population
Migration. Available at http://www.es.usda.gov/Briefing/Population/. Accessed
December 15, 2004.
Fatovich DM. Emergency Medicine. British Medical J ournal. 2002;324:958-962.
Fields WW. Emergency Care In California: Robust Capacity Or Busted Access? Health
Affairs web exclusive. March 24, 2004.
Franaszek JB et al. Responding to Emergency Department Crowding: A Guidebook for
Chapters. American College of Emergency Physicians (ACEP), 2002
GAO. Hospital Emergency Departments: Crowded Conditions Vary Among Hospitals
and Communities. U.S. General Accounting Office (GA O-03-60),Washington, D.C.
March, 2003.
GAO. Preparedness Varied Across State and Local Jurisdictions. GAO Report to
Congressional Committees. U.S. General Accounting Office (GAO-03-373) 2003.
Ghilarducci DP, Pirrallo RG, Hegmann KT. Hazardous materials readiness of United
States Level 1 Trauma Centers. Journal of Occupational and Environmental
Medicine. 2000;42:683-692.
Gore L, Hughes C. Eighty Percent of Emergency Physicians Surveyed Say Emergency
Departments Lack Surge Capacity. ACEP Press Release. Available at:
http://www.acep.org/1,33103,0.html . Accessed on: 01/02/04.
Heinrich J. SARS Outbreak: Improvements to Public Health Capacity are Needed for
Responding to Bio-Terrorism and Emerging Infectious Diseases. U.S. General
Accounting Office Report (GA O-03-769T) 2003.
Hirsch GB. Modeling the consequences of major incidents for health care systems. In:
22nd International System Dynamics Conference. Oxford, England: System
Dynamics Society; 2004. Available from:
http://www.systemdynamics.org/publications.htm.
17
Hoard ML et al. Systems Modeling in Support of Evidence-based Disaster Planning for
Rural Areas. International J ournal of Hygiene and Environmental Health.
2005;208:117-125.
Hoard ML et al. Local Health Department and Hospital Collaboration Study of Public
Health Emergency Preparedness Activities: Post 9/11. (Unpublished Data - Poster
Presentation)
Homer J, Hirsch G, Minniti M, Pierson M. Models for Collaboration: How System
Dynamics Helped a Community Organize Cost-Effective Care for Chronic Illness.
System Dynamics Review. 2004; 20:199-222.
Homer JB, Hirsch GB. System Dynamics for Public Health: Background and
Opportunities. American Journal of Public Health. 2005 (in press).
Joint Commission on Accreditation of Healthcare Organizations (JCAHO). Health Care
at the Crossroads: Strategies for Creating and Sustaining Community-wide
Emergency Preparedness Systems. JCAHO. Illinois 2003.
Lane DC, Monefeldt C, Rosenhead JV. Looking in the wrong place for healthcare
improvements: a system dynamics study of an accident and emergency department. J
Operational Research Society. 2000; 51:518-531.
Levitin HW, Siegelson HJ.: Hazardous Materials: Disaster Medical Planning and
Response. Emergency Medicine Clinics of North America. 1996;14:327-348.
Lipsitch M et al. Transmission Dynamics and Control of Severe A cute Respiratory
Syndrome. Science, 300:5627. pp 1966-70. 20 June, 2003.
Moscovice I et al. Measuring Rural Hospital Quality. Working Paper #53. Rural Health
Research Center, Division of Health Services Research and Policy. School of Public
Health, University of Minnesota. A pril 2004.
Office of Rural Health Policy, U.S. Department of Health and Human Services, Health
Resources and Service Administration. Rural Communities and Emergency
Preparedness, April 2002. Available at
ftp://ftp.hrsa.gov/ruralhealth/RuralPreparedness.pdf . Accessed March 17, 2004.
Pedrotti D, Perry R. Triage: A Comparison of Triage Tags in Common Use in the United
States. Prehospital Care Research Forum. May 1999.
Quality Through Collaboration: The Future of Rural Health Care (2004). Available from
www.nap.edu/openbook/03099094399/html/30.html. Accessed on February 15,
2005.
Schur CL. Understanding the Role of the Rural Hospital Emergency Department in
Responding to Bioterrorist Attacks and Other Emergencies: A Review of the
Literature and Guide to the Issues. NORC Walsh Center for Rural Health Analysis.
April, 2004.
Sterman JD. Business Dynamics: Systems Thinking and Modeling for a Complex World.
Boston: Irwin McGraw-Hill. 2000.
18
Task Force on Guidelines Society of Critical Care Medicine. Guidelines for
categorization for services for the critically ill patient. Critical Care Med 1991;
19:279-285.
Treat KN et al. “Hospital preparedness for weapons of mass destruction incidents: An
initial assessment,” Annals of Emergency Medicine, Vol. 38(5) Nov 2001 pp 562-565.
Williams JM, Ehriich PF, and Prescott JE. “Emergency medical care in rural America,”
Annals of Emergency Medicine, Volume 38(3), September 2001, pp 323-327.
19
APPENDIX: BASELINE MODEL ASSUMPTIONS
Table 3a. Baseline model assumptions— Emergency Department and Diagnostic Imaging
1. Emergency Department (ED)
dak,
12;
1.3.
1.4.
1.5.
1.6.
1s
1.8.
1:9;
1.10.
A111,
1.12.
1.13.
ED arrivals per day: 26 Green, 22 Yellow, 2 Red, 0.25 Black (die before admission);
40% with trauma (potentially requiring surgery), 3% contagious (requiring isolation).
ED arrivals go through daily cycle with minimum at 5 AM to 7 AM, building up to
maximum at 11 AM to noon, then falling off gradually through afternoon and night.
Minimum is 7% of maximum.
Normally 3 ED nurses work day shift (incl. 1 reserved for triage), 2 night shift; may
increase to as many as 6 day and 4 night during a surge.
Normally 5 ED beds available; may increase to as many as 15 during a surge.
Triage capacity of 15 patients per triage nurse per hour.
Red has highest priority for ED admission, then Y ellow, then Green. Re-triage
required after each hour of waiting. If ED nurses insufficient to do all required triage
and re-triage, then to that extent ability to prioritize patients is compromised.
Deterioration per hour of waiting: 10% of Greens deteriorate to Y ellow, 5% of Y ellows
to Red; 1% of Reds to Black (death).
In the event of an ED bottleneck, some Greens walk out: 10% per hour if the wait is 2
hrs., 25% if 4 hrs., 60% if 8 hrs., 92% if 12 hrs, 100% if 16 hrs. or more. Y ellow
walkout rate is 20% of Green. No Red walkouts.
Average time to complete ED care, aside from diagnostic imaging, if not isolated:
Green 1 hr, Y ellow 1.5 hrs, Red 2 hrs. Isolation increases time by 50%.
Required nurse-to-patient ratio in ED if not isolated: Green 0.25, Y ellow 0.5, Red 1.
Isolation increases ratio by 25%.
50% of Reds transferred out to other hospitals following ED evaluation (0.5 hour).
30% of Y ellows normally directed to wards (starting with surgery, if trauma) after ED
care; 100% of Reds. But if a bottleneck in diagnostic imaging (D1) causes some
Yellows to skip needed DI while in the ED (see 2.4 below), then 60% are directed to
wards for testing and observation.
Required nurse-to-patient ratio for post-ED patients awaiting surgery or ward
admission: Y ellow 0.125, Red 0.25.
2. Diagnostic Imaging (DI)
2h
ids
2.3.
2.4,
1 DI technician-equivalent is available at all hours for ED and ward patient imaging
needs, and can do up to 2 cases per hour. Ward cases have priority over ED cases.
Fraction of ED patients needing DI: Reds 100%; Y ellows 60-80% if not contagious,
100% if contagious; Greens 5-15% if not contagious, 85-95% if contagious.
Fraction of ward patients needing DI: 60% of regular acuity within 36 hours of
admission; 80% of high acuity within 24 hours of admission.
In the event of a DI bottleneck, some needed DI for ED cases is skipped: 20% if the
wait is 1 hr., 43% if 2 hrs., 70% if 4 hrs., 92% if 8 hrs., 100% if 12 hrs. or more.
20
Table 3b. Baseline model assumptions— Surgery and Elective Non-Surgery
3. Surgery/Operating Rooms (OR)
3:1;
3.2.
333;
3.4,
330s
3.6.
3.7.
3.8.
Elective surgeries scheduled: 8 per weekday, 0.5 per weekend day; scheduled to arrive
at uniform rate 7 AM to 2 PM.
In the event of a bottleneck at ward admission, some elective surgeries are rescheduled
to the next day: 40% if the waiting time for a regular-acuity admission is 2 hrs., 90% if
A hrs., 100% if 6 hrs or more.
Average time to complete OR procedure: 2.0 hours elective, 2.75 hours emergency
(trauma)
15 OR nurses work weekdays 7 AM to 4:30 PM; 5 OR call team nurses available all
other hours and all day Saturday and Sunday.
3 OR beds available at all times.
Required OR nurse-to-patient ratio during surgery: 5.0 if not isolated; Isolation
increases ratio by 25%.
Required OR nurse-to-patient ratio for post-surgery (boarding) patients: 0.25 regular
acuity, 0.5 high acuity; Isolation increases ratios by 25%.
75% of elective surgery patients directed to wards post-surgery, of whom 75% are
regular acuity and 25% high acuity. 100% of emergency surgery patients directed to
wards post-surgery; acuity indicated by triage color (Y ellow: regular, Red: high).
4. Elective Non-Surgical Outpatient Procedures
4.1,
4.2.
4.3,
4.4,
4.5.
4.6.
4.7.
Elective non-surgeries scheduled: 8.4 per weekday, 0.8 per Saturday, none on Sunday;
scheduled to arrive at uniform rate 7 AM to 2 PM.
In the event of a bottleneck for ward admission, some elective non-surgeries are
rescheduled to the next day: 40% if the waiting time for a regular-acuity admission is 2
hrs., 90% if 4 hrs., 100% if 6 hrs or more.
Average time to complete non-surgical procedure: 1 hr.
3 outpatient procedure nurses work weekdays 7 AM to 4:30 PM; 1 outpatient nurse on
call at all other hours and all day Saturday. (Outpatient beds not a limiting factor.)
Required nurse-to-patient ratio during outpatient procedure: 1.5.
Required outpatient nurse-to-patient ratio for post- procedure (boarding) patients: 0.25
regular acuity, 0.5 high acuity.
Only 1% of non-surgical outpatients directed to wards post-procedure, of whom 10%
are regular acuity and 90% high acuity.
21
Table 3c. Baseline model assumptions— Inpatient Wards
5. Inpatient Wards
5.1.
5.2.
5:3:
5.4,
5.05
5.6.
O7.
5.8.
Direct ward arrivals: 1.4 per day including weekends. They go through a daily cycle
with minimum at 5 PM through 6 AM, building up to maximum at 11 AM to 1 PM,
then falling off gradually through afternoon. Minimum is 50% of maximum.
Usual length of stay in ward if not isolated: regular acuity 79 hrs., high acuity 99 hrs.
If isolated: regular acuity 108 hrs., high acuity 128 hrs. Normal ward discharges occur
8 AM to noon.
In the event of a bottleneck for ward admission, some regular-acuity, non-isolated ward
patients are discharged early: 5% per hour if the waiting time for a regular-acuity
admission is 1 hr., 40% if 2 hrs., 70% if 3 hrs., 90% if 4 hrs, 100% if 6 hrs. or more.
Normally 6 ward nurses are available at all hours; may increase to as many as 10
during a surge.
Normally 28 ward beds available; may increase to as many as 47 during a surge.
Required nurse-to-patient ratio in wards if not isolated: Regular acuity 0.1, high acuity
0.5. Isolation increases ratio by 25%.
High acuity has higher priority for ward admission that regular acuity.
30% of high acuity patients transferred out to other hospitals following initial 24 hours
of testing and observation.
22
CC BY-NC-SA 4.0