A Geographic Information System Simulation

Model of EMS: Reducing Ambulance

Response Time

KOBI PELEG, PHD,*‡ AND JOSEPH S. PLISKIN, PHD†‡

Response time is a very important factor in determining the quality of Consequently, various national strategies have been de-
prehospital EMS. Our objective was to model the response by Israeli veloped for increasing the quality of EMS, especially the
ambulances and to offer model-derived strategies for improved deploy- administration of life-saving procedures at the earliest time.
ment of ambulances to reduce response time. Using a geographic infor- Although citizens trying to give first aid can deliver such
mation system (GIS), a retrospective review of computerized ambulance
call and dispatch logs was performed in two different regional districts,
procedures sooner than the EMS,8 the organization of the
one large and urban and the other rural. All calls that were pinpointed EMS for timely and proper response remains the responsi-
geographically by the GIS were included, and their data were stratified by bility of local governments. Indeed, many models of orga-
weekday and by daily shifts. Geographic areas (polygons) of, at most, 8 nization of integrated EMS for delivery of trauma care exist
minutes response time were simulated for each of these subgroups to worldwide.8-11
maximize the timely response of calls. Before using the GIS model, mean In Israel, civilian ground prehospital EMS are the man-
response times in the Carmel and Lachish districts were 12.3 and 9.2 date of Magen David Adom (MDA, Red Star of David).12 It
minutes, respectively, with 34% and 62% of calls responded within 8 is a publicly funded nonprofit organization, the structure
minutes. When ambulances were positioned within the modeled poly- and modus operandi of which reflect its history, and is based
gons, more than 94% of calls met the 8-minute criterion. The GIS simu- on salaried and volunteer staff. It is organized in 12 regional
lation model presented in this study suggests that EMS could be more
effective if a dynamic load-responsive ambulance deployment is
districts, each having a single control center directing the
adopted, potentially resulting in increased survival and cost-effective- launch of ambulances and personnel from any of several
ness. (Am J Emerg Med 2004;22:164-170. © 2004 Elsevier Inc. All rights stations. MDA responds to approximately 300,000 calls per
reserved.) year with basic life support (BLS) ambulances that have
basic equipment and are operated by medics and with ad-
EMS focus on cardiac, vascular, respiratory, trauma, and vanced life support (ALS) ambulances outfitted with ad-
other emergencies. Many factors determine the quality of vanced resuscitation equipment and a team of a physician
EMS, and response time is an important EMS industry and a paramedic or a paramedic and an EMT. Each district
benchmark.1-4 This is also reflected in the term “The Golden had two ALS ambulances that operated during all shifts 24
Hour” that was derived from the observation that survival of hours a day for 365 days per year. The distribution of BLS
patients who receive in-hospital definitive treatment within ambulances varied by shift and day according to need. In
1 hour of injury was much higher than of those who re- 1997, in the Lachish district, ALS ambulances treated 5,020
ceived it later.5,6 This benchmark is driven by cardiac arrest cases, whereas the BLS ambulances treated 15,565. In the
survival and the likelihood of successful defibrillation of Carmel district, ALS treated 7,184 cases and BLS treated
fatal cardiac arrhythmia.7 Other factors such as proficiency 29,245. Deep historical and political roots are the founda-
of the EMS personnel, the quality of equipment and ambu- tion of the present structure and mode of operation of MDA.
lances, and the suitability of organization, communications, The main objective of the present study was to study and
and control are also important for patients who require apply the use of geographic modeling systems to deploy
EMS. ambulances to fulfill demand and quality and performance
requirements, and to offer model-derived strategies for im-
proving the response time through modified deployment of
From the *Trauma and Emergency Medicine Research Unit, The MDA ambulances potentially increasing the efficiency of
Gertner Institute for Health Policy Research, Sheba Medical Center, MDA services.
Tel-Hashomer, Israel; and the †Department of Industrial Engineering
and Management and Department of Health Systems Management, MATERIALS AND METHODS
Ben-Gurion University of the Negev, Beer-Sheva, Israel; and the
‡Department of Health Policy and Management, Harvard School of A retrospective review of local ambulance calls recorded
Public Health, Boston, Massachusetts. for a 12-month period during the years 1996 through 1997
Received December 15, 2002; accepted April 23, 2003. was performed on computer-logged data from two of the 12
Supported in part by grant 15/97 from the Israeli National Institute MDA regional districts. These two regions were selected to
for Health Policy and Health Services Research.
Address reprint requests to Kobi Peleg, PhD, Gertner Institute, represent the full range of needs for EMS and the whole
Sheba Medical Center, Tel-Hashomer 52621, Israel. Email: gamut of extrinsic factors affecting ambulance response
[email protected] time, namely distances, traffic conditions, and population
Key Words: Emergency medical service, ambulance, response load, and to enable testing a model of optimal ambulance
time, geographic information system
© 2004 Elsevier Inc. All rights reserved. deployment.
0735-6757/04/2203-0005$30.00/0 One district was Carmel, in the major metropolitan dis-
doi:10.1016/j.ajem.2004.02.003 trict of Haifa, Israel’s largest port and a heavy industry

164
PELEG AND PLISKIN ■ GIS MODEL FOR REDUCING AMBULANCE RESPONSE TIME 165

center. It includes urban residential neighborhoods and in- on any given weekday was determined after accounting for
dustrial zones of diverse characteristics, from heavy and the 73% inclusion of the logged calls and an extrapolation to
petrochemical industries to high-tech parks, and is afflicted the full call volume.
with typically heavy, yet variable, traffic loads. It is served The Geocode software was fed the exact address of every
by three major hospitals: one level I and two level II trauma call by shift and by day. Then, the density of calls was
centers. The second MDA district chosen is Lachish, a calculated by using the Spatial Analyst extension of the
southern rural district with three small and one medium Geocode software, and density maps were generated per
urban center interspersed with numerous agricultural com- district, per day, and per shift. A map showing various
munities. It contains one level II trauma center and a second shades for differing densities was generated. In the most
one is located in the adjacent district, and reaching a level I dense call areas, we initially positioned dispatch points for
center takes ambulances 30 to 60 minutes. ambulances so that travel to any point within the time line
(polygon) will not exceed 8 minutes, taking into account
Records were taken from each region’s computerized
road signs, one-way streets, and so on.
logbook, in which data are entered by the dispatcher in
Ambulance speeds were calculated from randomly cho-
real-time. In the Lachish district, it was available for an sen logs (at least 12 per subgroup), considering the distances
entire year, and in the Carmel district for a period of 9 using only routes permitted by traffic regulations. Travel
months. The entering calls and subsequent communications speeds were calculated using the actual travel times as taken
are also available on tape and could be verified against the from the logbook and the distance as calculated by the GIS
logbook. This verification was done only for the question- (which is very precise and takes into account one-way
able data. streets, and so on). When significant speed differences were
The Carmel district had 33,163 logs, but approximately observed in specific subregions, they were applied in the
2,600 were administrative rides (eg, refueling, service), and analysis.
several more belonged to neighboring districts and were A response time (from receipt of call until arrival at the
therefore ignored. Excluded were also calls whose exact scene) of 8 minutes or less was set as the working criterion,
address could not be pinpointed (first by software and then based on MDA’s own recommendations from 1994 and on
manually) and address-matched by a geographic positioning those of the Pre-Hospital Trauma Life Support (PHTLS).5
software (Geocode, ArcView 3.1, ESRI, Redlands, CA). Polygons in which this response time was maximally
Only 73% of the net logged calls were therefore included, achieved were simulated for each of the subgroups calcu-
resulting in 21,643 records. A similar yield was obtained for lated with the Network Analyst extension of the Geocode
the 27,085 logs of the Lachish district (after subtracting package. We wanted to assure that the derived polygons
approximately 1,700 administrative rides), resulting in a achieve the desired response time for at least 95% of calls.
database of 17,230 records (68%). The characteristics of Initial polygons had one ambulance per polygon, and then
calls that were removed from analysis were not different an interactive polygon determination was performed to
from those used, especially with regard to response times. maximize the number of calls with a response time of 8
From each included log, data were extracted on the type minutes or less, accounting for the number of ambulances
of ambulance and times of call, driver notification, take off, required for the call load. In a shift with a higher probability
arrival on scene, departure from scene, arrival at the hospi- than 5% for an additional call, we added a backup ambu-
tal, departure from the hospital, and return to base or to the lance using for it a polygon of 15 minutes response time.
next mission. The likelihood of ambulance unavailability was calculated
All data were categorized, by district, and into five daily using the Poisson distribution for probable calls. Figure 1
shifts representative of the degree of traffic load: 07:00- presents a flowchart of the analysis applied on the data for
09:00 (rush hour), 09:00-16:00 (business), 16:00-19:00 the determination of polygons.
(rush hour), 19:00-02:00 (going out), and 02:00-07:00 (off-
hours). Further stratification was by the days of the week, RESULTS
resulting in a total of 70 data subgroups. In Israel, Saturday The response times (from receipt of call to arrival on
is the day of rest, and business activity on Fridays is limited. scene) by MDA teams in the Carmel and Lachish districts
The number of ambulances needed for each shift in each are tabulated in Table 1. The mean response times in these
day was calculated from the turnaround time (from the districts were 12.3 and 9.2 minutes, respectively. The 95th
initial call until return to base) of calls in each subgroup. percentiles of response in these districts were in excess of
The turnaround time consists of the response time to the 26 and 23 minutes, respectively. Indeed, with 8 minutes set
call, travel time to the point of call, treatment time at point as the standard, only 34% of the calls were responded to in
of call, travel time to the hospital, time spent in the hospital, the Carmel district, implying that 66% of them were late
and travel time back to the dispatch point. Based on these (Fig 2A). In Lachish, 62% of calls were answered within 8
times for the various shifts, we could calculate the average minutes (Fig 2B).
number of calls that an ambulance can respond to during When mobile ALS ambulances, which respond only to
each shift. Then we calculated the probability of another call EM calls (as opposed to nonemergencies handled by MDA
in a given area when the ambulance was making a “round.” such as transporting nonendangered pregnant women or
We wanted to assure a situation in which this probability minor trauma) were analyzed separately, only a slight im-
will not exceed 5%. In the Lachish district, this was calcu- provement in the 8 minutes response fraction was found.
lated separately for each of the towns. From the mean Approximately 40% of these calls in the Carmel district
turnaround time, the number of ambulances needed per shift were answered within the standard amount of time.
166 AMERICAN JOURNAL OF EMERGENCY MEDICINE ■ Volume 22, Number 3 ■ May 2004

FIGURE 1. Flowchart of analysis.

In both districts, a substantial component of the response and in the Lachish district 2.4 ⫾ 4.7 minutes, accounting for
time was the activation time, the time from receipt of a call approximately 20% of the actual and more than 30% of the
to departure. In the Carmel district, it was 2.5 ⫾ 6.2 minutes desired response time. The time of travel to the scene, which
is the second major part of the response time, was also
spread out, and only 25% and 34% of MDA ambulances
TABLE 1. Response Time of Magen David Adom Teams traveled in less than 8 minutes to the scene in the Carmel
(minutes) and Lachish districts, respectively. This was reflected in the
Mean ⫾ mean ambulance traveling speed that was calculated for
Standard 5th 95th each shift in each district to allow the subsequent derivation
District Deviation Median Percentile Percentile of response polygons. In the metropolitan Carmel district,
the mean driving speed during the first shift (7 AM–9 AM) on
Carmel 12.3 ⫾ 9.5 10.1 3.9 26.5
working days (Sunday–Thursday) was 25 km/hr (15.6
Lachish 9.2 ⫾ 8.2 6.8 3.0 23.0
mph), whereas in the equivalent shift on Saturday morning,
PELEG AND PLISKIN ■ GIS MODEL FOR REDUCING AMBULANCE RESPONSE TIME 167

shift could be responded by arrival on the scene within 8

minutes, as illustrated by the data of the Sundays and
Saturdays in Table 2 (data of other weekdays are not
shown). For example, of the 341 calls received on the
first shift of Sunday, 98.5% were within the 8-minute
polygons (Table 2). If ambulances were positioned within
the polygons every shift of the entire week, more than
94% of incoming calls would have been responded to
within 8 minutes in the Carmel district, and similar data
were obtained for the Lachish district (not shown).
However, when only a single ambulance is placed
within each polygon, sometimes it might not be possible
to respond promptly to all the calls. The maximal number
of potential responses per shift and the probability that
demand will exceed this maximum is illustrated for shifts
on Sundays and Saturdays in the Carmel district in Table
3. For example, a mean of 8.9 ⫾ 4.36 calls were received
during the first shift of all Sundays examined, with a
minimum of two and a maximum of 23 calls. If seven
ambulances were on duty during this shift, as suggested
from the polygon model, approximately nine calls could
be answered in this shift, calculated from the turnaround
time. However, a likelihood of 40% existed that more
calls might be received. Of the 35 weekly shifts, 19 had
a potential inability to answer all potential calls with a
probability exceeding 0.05. To rectify this, additional
FIGURE 2. Distribution of response times in Carmel (A) and polygons, each with an additional ambulance of 15 min-
Lachish (B). utes response time, were constructed in all the shifts with

it was 40.7 km/hr (25.4 mph); in Lachish, the corresponding

speeds were 26.3 km/hr (16.4 mph) and 23.9 km/hr (14.9
mph). Subregional variations of driving speeds were noted
in each district. For instance, in the northern suburbs of
Haifa, long stretches of highway resulted in traveling speeds
of 34.2 km/hr (21.4 mph) and 57.1 km/hr (35.7 mph) in
these shifts, and these local speeds were applied when
appropriate. There were big differences in travel speeds
among shifts but the standard deviations within shifts were
small.
Figure 3A depicts the frequency of calls throughout the
day and Figure 3B during the week. These distributions
were statistically similar in both districts and display a
typical daily pattern of activity and traffic.
During shifts when more than 5% of calls were not
answered within 8 minutes, two backup ambulances were
added. These ambulances had a response time not exceeding
15 minutes, and they covered an area of more than one
polygon. Figure 4A depicts the layout of the 8-and 15-
minute response time polygons for the first rush hour shift
and Figure 4B the 8-minute polygon for the fifth, off hours
(2 AM–7 AM) shift in the Carmel district. Figures 5A and 5B
depict the polygons for the Lachish district in the first and
fifth shifts. Similar polygon maps were constructed for each
shift on every weekday using the density of calls and the
average driving speeds.
With the polygons so positioned, it was possible to cal-
culate what percentage of calls received in a particular daily FIGURE 3. Distribution of calls by hour (A) and by day (B).
168 AMERICAN JOURNAL OF EMERGENCY MEDICINE ■ Volume 22, Number 3 ■ May 2004

FIGURE 4. Polygon overlay of maps for the Sunday’s first (A) and last (B) shifts for the Carmel district (thin lines— 8 minutes polygons;
heavy lines—15 minutes polygons).

insufficient potential response capacity (Fig4A). As illus- DISCUSSION

trated in Table 3, this scheme enabled response to 13 The simulation model that was constructed and presented
calls in the previously mentioned example and lowered in this study suggests that EMS could become more effec-
the likelihood of potential excessive calls to 0.07 (this is, tive if a dynamic, load-responsive ambulance deployment
however, a short 2-hour shift). In all other 18 weekly model is adopted. This model can be tailored for any district
shifts that were previously potentially shorthanded, the and EMS system in which call data are available for retro-
addition of 15 minutes polygons remedied the potential spective analysis.
ambulance deficit. In the Lachish district, there was am- The response time is critical for the patient’s survival and
ple capacity to respond just with the 8-minute polygons. reduction of disability, and many have shown a direct rela-

FIGURE 5. Polygon overlay of maps for Sunday’s first (A) and last (B) shifts for the Lachish district.
PELEG AND PLISKIN ■ GIS MODEL FOR REDUCING AMBULANCE RESPONSE TIME 169

TABLE 2. Carmel District Polygon Analysis for Sundays ferred alternative would be for the EMS crew to stay in the
and Saturdays, Stratified by Shift* ambulance during the entire shift, except for patient care.
Responses Within Mean Number of We must remember that the data reflect the situation in
Day Shift Calls the Polygons (%) Calls Responded 1996 through 1997. Today, ambulances are equipped with
location technology. However, to be effectively used, loca-
Sunday 1 341 98.5 8.9
Sunday 2 1,402 97.2 37.0
tion and allocation have to be carefully planned.
Sunday 3 489 95.7 12.8 In our analysis, travel from one district to another is not
Sunday 4 823 94.9 21.5 relevant because the two districts are geographically remote
Sunday 5 358 95.5 9.2 from one another. Ambulances can, however, travel outside
Saturday 1 171 97.1 4.5 their designated polygon. One must remember that polygon
Saturday 2 914 95.8 23.8 construction was based on existing call patterns. If an am-
Saturday 3 422 94.8 10.8 bulance travels beyond the limits of its polygon, there is a
Saturday 4 978 95.5 25.0 probability of momentary shortage of ambulances to re-
Saturday 5 368 96.5 9.4 spond to calls and ambulances from adjacent polygons
*Actual calls over the study period and the extent of their coverage might have to respond.
by the 8 minutes polygons. Although this model has dealt only with surface EMS
services, the incorporation of airborne EMS teams for areas
not covered by the polygons should be considered, espe-
tionship between prehospital medical treatment or resusci- cially in distant regions where the “Golden Hour” cannot be
tation1-6 and survival.8,13 A rapid response time is also attained.
reflected in an overall shorter turnaround time for ambu- There are many other factors that contribute to the quality
lances, thus increasing their availability for additional calls. of EMS that might be added to any comprehensive model
The simulation model also resulted in a significant (alas, for the refinement of policies on EMS. They could include
only theoretical) savings of resources in personnel hours, analysis of the training levels required of ambulance teams,
ambulances, and equipment in each of the districts, permit- the ratio of ALS units to regular BLS ambulances, the
ting more cost-effective organization of the EMS services. equipment carried by either, the organization of the EMS
This is reflected in the reduced number of ambulances and under national or regional hierarchy,15 the public nature of
these services, and more.
personnel assigned to different shifts and the improved
We must be aware of some of the limitations of our study.
organization of working shifts (manuscript in preparation).
The conclusions are based on the results of the model, but
System savings from the implementation of a geographic its functionality has not been demonstrated. Implementation
positioning system (GPS) have been demonstrated else- could encounter political and operational problems that
where.14 However, this does not necessarily carry over to cannot be dealt with in a formal model. Sensitivity analyses
the present analysis, which relates to planning of services, were not performed on all parameters (eg, travel speeds);
whereas GPS relates to operational, dynamic aspects of the extrapolation to the full call volume might not be ap-
providing service. propriate. The model must be updated at regular intervals to
Presently, ambulances are dispatched from a few fixed account for changes in travel times and speeds as a result of
stations, but the simulation model suggests that additional changing urban and rural environments, changes in road
EMS posts should be established within the territory con- directions, expansion of lanes in roads, and so on. Opera-
tained in the polygons. This arrangement could perhaps also tional issues relating to frequent repositioning of ambu-
reduce the activation time spent between transmission of the lances and varying the on-duty personnel were not ad-
call to the EMS team and their actual departure. The pre- dressed in this study. In conclusion, a suitable model based

TABLE 3. Carmel District Polygon Analysis for Sundays and Saturdays, Stratified by Shift*

Only 8 Minutes Polygons Polygons of 15 Minutes Added

Maximum Possible Probability of Demand Maximum Possible Probability of Demand

Day Shift Responses Above Maximum Responses Above Maximum
Sunday 1 9 0.4 13 0.07
Sunday 2 35 0.35 50 0.02
Sunday 3 16 0.15 22 ⬍0.01
Sunday 4 30 0.032 45 ⬍0.01
Sunday 5 22 ⬍0.01 34 ⬍0.01
Saturday 1 7 0.1 12 ⬍0.01
Saturday 2 38 ⬍0.01 53 ⬍0.01
Saturday 3 16 0.05 23 ⬍0.01
Saturday 4 32 0.07 48 ⬍0.01
Saturday 5 24 ⬍0.01 36 ⬍0.01

*Calculated maximum potential responses per shift and the probability that demand will exceed it, under two scenarios: with only 8 minutes
polygons and with 15 minutes polygons added.
170 AMERICAN JOURNAL OF EMERGENCY MEDICINE ■ Volume 22, Number 3 ■ May 2004

on retrospective EMS data can help to potentially improve Committee on Trauma of the American College of Surgery:
these essential services and result in higher survival rates PHTLS—Basic and Advanced Pre-Hospital Trauma Life Support,
3rd ed. New York: Mosby, 1998
and reduced disability of their clients. 7. Feero S, Hedges JR, Simmons E, et al: Does out-of-hospital
EMS time affect trauma survival? Am J Emerg Med 1995;13:133-
The authors thank Joseph Rosenblum, MD, for his invaluable as- 135
sistance in the statistical and computing aspects of this work. 8. Eisenberg MS, Hallstrom AP, Copass MK, et al: Treatment of
ventricular fibrillation. Emergency medical technician defibrillation
REFERENCES and paramedic services. JAMA 1984;251:1723-1726
9. Bury G, Dowling J: Community cardiac awareness teaching in
1. Roberts MJ, McNeill AJ, MacKenzie G, et al: Time delays to a rural area: the potential for a health promotion message. Resus-
lytic therapy and outcome in 100 consecutive patients with a history citation 1996;33:141-145
suggestive of acute myocardial infarction in an area with access to 10. Braun O, McCallion R, Fazackerley J: Characteristics of mid-
a mobile coronary care unit. Eur Heart J 1994;15:594-601 sized urban EMS systems. Ann Emerg Med 1990;19:536-546
2. Trunkey DD: Trauma care at mid-passage—a personal view- 11. Jakobsson J, Nyquist O, Rehnqvist N: Cardiac arrest in
point. 1987 AAST Presidential Address. J Trauma 1988;28:889-895 Stockholm with special reference to the ambulance organization.
3. Sampalis JS, Denis R, Frechette P, et al: Direct transport to Acta Med Scand 1987;222:117-122
tertiary trauma centers versus transfer from lower level facilities: 12. Champion HR, Teter H: Trauma care systems: the federal
Impact on mortality and morbidity among patients with major role. J Trauma 1988;28:877-879
trauma. J Trauma Inj Infect Crit Care 1997;43:288-296 13. Sampalis JS, Lavoie A, Williams JI, et al: Impact of on-site
4. Bissel RA, Eslinger DG, Zimmerman L: Advanced life support: care, prehospital time, and level of in-hospital care on survival in
a review of the literature. Prehospital and Disaster Medicine 1998; severely injured patients. J Trauma 1993;34:252-261
13:69-79 14. Ota FS, Muramatsu RS, Yoshida BH, et al: GPS computer
5. Mullie A, Van Hoeyweghen R, Quets A: Influence pre-CPR navigators to shorten EMS response and transport times. Am J
conditions on EMS response times in circulatory arrest. The Cere- Emerg Med 2001;19:204-205
bral Resuscitation Study Group. Resuscitation 1989;17:199-206 15. Sampalis JS, Denis R, Lavoie A, et al: Trauma care region-
6. Pre-Hospital Trauma Life Support Committee of the National alization: a process– outcome evaluation. J Trauma 1999;46:565-
Association of Emergency Technicians in Cooperation with the 579