REPRODUTION OF AMBULANCE RUNS OF THE CmCAGO FIRE DEPARTMENT

John Barry, Fire Department

ABSTRACT Map I. Ambulance Demand Zones The Reproduction of Ambulance Runs Program can assist decision making in the Department by suggesting more efficient deployments of present am­ bulance companies and optimal deployment of new com­ panies as they become available. The output from the program pinpoints the areas of the city with the most demand for emergency medical services and areas where all incidents may not be responded to within five minutes. The program can test various locations of new ambulance companies before deployment, and examine the effect on coverage throughout the city. The areas that are under­ covered. that is. not within five minutes of an ambulance company are also isolated in the output report.

INTRODUCTION The Chicago Fire Department presently receives 550 to 650 calls daily for emergency medical service, and has 54 advanced life support ambulances to respond to those incidents 24 hours a day. The Fire Department has enough spare ambulances so that if a particular rig is out of service for repair or maintenance that ambulance company can still be available and 54 ambulances will always be on the street. to that company's anival on the scene of the incident, was Chicago is approximately 220 square miles in area and determined by actual test runs of emergency vehicles made has 104 firehouses. An ambulance is assigned to about half on various kinds of streets using lights and sirens to simu­ of all firehouses. The Reproduction of Ambulance Runs late emergency travel conditions. Members of the Chicago Program attempts to show how various deployments of Fire Department's Research and Planning Division ob­ ambulance companies throughout the city affect response tained the travel times with a stopwatch while riding on times to all areas of the city. The deployment goal of the various apparatus (engines, trucks and ambulances) under Chicago Fire Department has two parts: first, to have an simulated emergency conditions. The travel times used in ambulance on the scene of every incident within five the progrnm are an average of these test nm times. minutes, and second, to have an ambulance within five minutes of every part of the city. The most efficient THE INPUT DATA deployment would have an ambulance on the scene of an The records with missing or unusable data were screened incident in the shortest possible time. The target of the out at the beginning of the program. These records were analysis was to find the most efficient deployment of the closely examined to ensure that they were randomly dis­ present fifty-four companies, and to find the most efficient tributed throughout the ftle. Most of the unusable records deployment of new ambulance companies. The optimal were missing either the response time or the return time for deployment can be suggested by testing different deploy­ that incident. A smaller proportion was missing the Am- ments in the program and studying the effect on citywide coverage. Figure I. Breakdown of UllIuable Records CONSTRUCTING THE INPUT RECORD PARAMETERS • Aeoponoe tim. & cia .. Chicago was geographically divided into 1,071 Am­ bulance Demand Zones (ADZ's) of approximately equal area (see map above). A network of major city streets has been formed on an Intergraph Interpro 32 system through a V AX lIngO computer with software from Public Tech­ nology, Inc. of Washington, DC. titled Fire Station Loca­ tion Package©. The network gave distance and time parameters from each firehouse to each demand rone and contains the city streets used by Fire Department vehicles as they travel from a firehouse to the scene of an incident. The travel time parameter, that is, the time from dispatch

470 Figure 2 Distribution of Unusable R.econb was released and available for another run. Ifthe held time was greater than the present incident's dispatch time that company was still at an incident and unavailable at this

July. 1986 time for another run. The second company and its return

A.... time was then held for the duration of the second incident. Subsequent incidents were handled in a similar manner. For example, the lOOOth incident was read, and its dispatch

Nwu_ ;;;;;;; time was compared to the return time of all the companies '=~ still held. Any held company that had a return time less than the present incident's (the lOOOth) dispatch time was released and available for another run. Ambulance com­ panies with a later time remained unavailable and were examined again when the 100lst incident was read. The program always attempted to dispatch the closest ambulance to an incident, but often that ambulance was responding to another incident, and the program went to the next closest ambulance until it found one that was bulance Demand Zone because of a bad or incomplete available. Every incident in the ftle had been merged by address. The unusable data for 1987 were found to be ADZ with arrays containing every ambulance and its travel distributed through 699 of the 1071 Ambulance Demand time to the ADZ of the incident so the program had only Zones (or 71 %), showing a random geographic distribution to loop through the array of companies already on a run of the bad data. (See figures 1 and 2 for the breakdown and the array of ambulance companies in order of and distribution of unusable records for the twelve month proximity tothe incident, until it found the closest available period of7/l/86 through 6/30/87.) ambulance to dispatch. The reproduction program was written on the SAS© Occasionally more than one ambulance was needed at system and run on an IBM 3090 mainframe. The data are an incident. The program handles the second through actual incidents. as recorded on dispatch forms by the fire seventh ambulances in the same way as the first; however. alarm operators (dispatchers) as they were taking emergen­ the accuracy of the data seems to break down after the third cy calls from the public. The relevant information for the ambulance because the times are often missing. The oc­ reproduction were the ambulance company involved in currence of incidents where more than three ambulances each incident, the dispatch time, the address of the incident, were sent was so rare that the-overall effect on the year's and the return time which was also the time that ambulance total activity was minimal. company was available to make another run. The dispatch and return times were necessary to detennine how long the OUTPUT ambulance was unavailable for other runs, and the address Output from the program included the information on of the incident is necessary to assign it to a specific ADZ. travel time (the time it takes to arrive at the incident) and The incidents have been transcribed into computer run time (the time from dispatch time to the time that readable form by a data entry contractor, then geocoded by company is available for another run). PROC incident address to ADZ by the Chicago Area Geographic UNIV ARIATE gave us the mean times for both these Information Study of the University of Illinois at Chicago's variables plus· various statistics on the range of the data. Geography Department. Geocoding is a computer process How the runtime data was skewed was of special interest that matches a certain address to some unit of geography. because it showed if there were large numbers of runs that The unit of geography could be neighborhoods, census took a long time but not enough to significantly alter mean tracks, zipcodes, or in this case, Ambulance Demand Zones. Finally, the incidents are merged with several arrays of ambulance companies and travel times arranged Figure 3. Run Tune - 10 Minute Inlervah by proximity to the ADZ of the incident, and sorted by date --....,.. and time. The data in the arrays of ambulance companies and travel times were detennined from the network on th& Intergraph system, and the arrays were created with SAS software. Each observation in the input recor~ was a single incident.

THE PROGRAM The study file represented a year of ambulance data (approximately 220,000 incidents). When the program reads the frrst incident of the study file the ambulance company and its return time was. held in an array, and the second incident was read. The response time of the second incident was checked against all held times (in this case, the return time of the first incident). If the held tin)e was run time. less than the second incident's dispatch time that company PROC CHART produced horizontal bar charts with

471 accompanying statistics for the frequency of the hour of the day that runs begin, the day of the week, and month of Map 3. Fifty-six. Ambulance Deployment the year. A horizontal bar chart was produced for the frequency of run times in ten minute intervals (see figure 3). This chart produced a ben curve that was skewed to the righL Frequency distributions were generated using PROC FREQ to show the ADZs in order of frequency of incidents, showing which areas of the city had the most demand, and also a frequency of dates to illustrate what days of the year had the heaviest activity and the lightesL Frequencies of runs made by each ambulance company were also collected in order of frequency, and gave us a list of companies in order of the number of incidents they responded to. Final output from the program, and most imponant for our planning, was the frequency of ADZ in which the incidents occur that have a travel time in excess of five minutes arranged by frequency of ADZ. This gave the exact areas • lin to ZJ3 tt1« !ncldenU' lla! .. 10 Il6 (2721 \ndd""IS) of the city that required more ambulance coverage. There 'Ia 39 HI.'1l lncld~llIsl were two reasons why an incident was not responded to in five minutes: first, some of the edges of the city were not within five minutes of an ambulance, and second, the ambulance closest to the incident was already out on a run and an ambulance from farther away had to be dispatched. Presently, an ambulance is on the scene within five minutes about 87% of the time. CONCLUSION Downloading the zone coverage data to a dBase III +© The Reproduction of Ambulance Runs Program is a tool file and using Maping Information Systems Corporation's used to examine the efficiency of various deployments of ambulance companies throughout Chicago. The output from the program isolates areas of the city that have the Map 2. Fifty-four Ambulance Deployment most demand for emergency medical services, and areas that consistently have incidents that are not responded to within five minutes. By isolating the areas of the city that have the most demand or the least coverage by present ambulance deployment it is possible to assist decision making by suggesting more efficient redeployment of present ambulance companies, deployment of new com­ panies and locations of new firehouses. The optimal deployment of new ambulance companies can be suggested by testing various locations for the new company before actual deploymenL The program can examine the effect of a test deployment on overall coverage throughout the city. SAS is a registered trademark of SAS Institute INC., Cary, NC, USA.

Author's Address: John Barry Senior Operations Research Analyst Chicago Fire Department Research & Planning Division Room 105 Mapinfo Desktop Mapping Software©, the wnes of the 121 North LaSalle Street most incidents with over five minutes travel time were then Chicago, Illinois 60602 mapped, and a visual representation was produced of Phone: 312-744-8621 where additional ambulance service was most needed. Map #2 shows incidents with over five minutes travel time with 54 ambulances deployed and Map #3 is a test deploy­ ment of two new ambulances on the South Side with the original 54 ambulances of Map #2. A comparison of the shaded areas of the two maps show the areas of the city most effected by the new deployment.

472