Study setting
This study was approved by the Conjoint Health Research Ethics Board at the University of Calgary and was conducted within and surrounding a large Canadian city, Calgary. At the time of the study, the Calgary Health Region (CHR) administered all publicly funded hospital care, including emergency services in three tertiary care adult hospitals, to the residents of the cities of Calgary and Airdrie and approximately 20 nearby small towns, villages, and hamlets (population 1 million) in the Province of Alberta, Canada. The City of Calgary Emergency Medical Services was the sole provider of ambulance services to the City of Calgary and to the surrounding regions, which include the Town of Chestermere, the Tsuu T’ina Nation and sections of the Municipal District of Rockyview.
This EMS system had approximately 44 response units, all of which were Advanced Life Support equipped and staffed. In 2006, this service recorded 107,562 EMS unit responses[13]. Based on information provided by the caller, and interpreted by a registered emergency medical dispatcher using the Medical Priority Dispatch System (MPDS), emergency situations were identified and given the designation of Alpha, Bravo, Charlie, Delta, or Echo level events. The MPDS rates the emergency from least serious (Alpha) to most critical (Echo)[14]. The dispatch of EMS units in this jurisdiction using the MPDS is consistent with industry-accepted quality standards. In the jurisdiction for this study, Alpha level calls receive an ambulance response without lights and sirens (non-emergency). Bravo, Charlie, Delta, and Echo calls receive an ambulance response using lights and sirens, and Delta and Echo level calls also receive a response by the fire department, who provide Basic Life Support with defibrillation medical first-response and scene assistance[13].
Study sample
The study sample contained 31,385 patient trips (for adults 18 years of age or older) within the Calgary area between January 1 and December 31, 2006. A full year of data allowed for the daily and seasonal fluctuations in EMS trips to be accounted for over the entire study period. All calls contained in this sample were either a Bravo, Charlie, Delta or Echo level call, which means that the caller provided key information that led the dispatcher to conclude that the call was for a time-dependent emergency[14]. This ensured that the travel times considered were for ‘emergency’ response by EMS. Functionally two ambulances dispatched from the same location to the same location would arrive at the same time even if one was a Delta and the other Bravo, as the response is the same. The ‘lights and sirens response’ is the only difference in the priority of dispatch, but if two calls are made at the same time, the closest ambulance will be dispatched to the higher level call.
Study variables
Each record contained a patient location at the time of call. The location was recorded as an address, an intersection or a common place name. Each patient location was converted to an x/y coordinate by CHR analytic support staff within the administrative data group to ensure precise representation of the patient event location. Each record also contained the hospital address to which patients were transported. This data was prepared for a prior study focused on the associations between emergency response time and mortality in an urban setting[13]. Records that did not include patient event or transport location information were excluded from this study. Records for scene locations within the city boundary were categorized as urban, while those for scene locations outside of the city were categorized as rural. In previous studies, categorizations of urban, suburban and rural have been made based on tertiles of population density[5, 10]. Our method of dividing the scene locations by inside and outside the city limits yielded comparable categorizations to the methods used from these previous studies.
Each record also contained a time stamp that identified the start of the activation interval, response interval, on-scene interval and transport interval for each patient trip. The activation interval was defined using the time stamps from when the time the emergency call was received to the time the ambulance was en-route to an event. The response interval was defined using the time stamps from the time the vehicle was en-route to the time it arrived on scene. The on-scene interval was defined using the time stamps from the time the ambulance arrived on scene to the time it left the scene. Finally, the transport interval was defined using time stamps from the time the vehicle left the scene to the time it arrived at the hospital. The time stamp corresponding to when the emergency call was received was automatically generated by the EMS computer aided dispatch (CAD) system upon receipt of the call. All other timestamps were automatically recorded by the CAD system when the responding paramedic pressed the appropriate button on a mobile data terminal in the vehicle. Records with any missing time stamps were excluded from this study.
GIS data and travel time analysis
The transport interval is the time interval that can be modeled using GIS. This time component is the travel time along the road network from patient scene location to the hospital. Travel time along a road network requires data representing origins of travel, destinations and the linear features along which travel occurs. The origins in this study were the recorded scene location of patients; the destinations were the geocoded hospital emergency department locations. The road network we used was the CanMap® RouteLogistics file (DMTI Spatial, Markham, Ontario). This file can be used for shortest route analyses of both time and distance. In addition to containing detailed street names and address locations along each segment of road, fields are included for the length and speed limit along each segment of road.
This GIS derived transport time from scene to hospital was subsequently used to model the response interval (time from dispatch to patient scene). In many study areas, the location of the ambulance at the time of dispatch is unknown. For example, the EMS database that we used does not consistently include the location of the ambulance at the time of dispatch. In past studies, in the absence of information on ambulance locations at time of dispatch, empirically derived constants based on the literature were used to account for the time for an ambulance to reach a patient[5, 10, 12]. Using actual ambulance data, these empirical constants were derived by determining the relationship between the response and transport intervals. GIS derived travel times from scene to hospital were multiplied by 1.6 to obtain overall travel times in urban areas and 1.4 in rural areas[5, 6]. This means that the response interval was modeled to be 60% and 40% of the transport interval in urban and rural areas, respectively.
Objective 1: Validation of EMS modeling assumptions
The overall time from emergency call to hospital requires a representation of the various time factors that constitute pre-hospital time. Below is a description of how comparisons between the actual EMS time factors captured in our database were made with the time factors as described in previous literature or derived using GIS. All statistical analyses were conducted using Stata 10.0[15].
Activation interval and On-scene interval
Descriptive analysis was conducted with the entire study sample to determine the actual median activation and on-scene intervals. The sample was divided into within city and outside city in order to understand urban/rural differences in activation and on-scene times. A comparison with the average activation and on-scene interval from past studies was made using boxplots.
Response interval
The actual response interval is contained within the EMS data. Descriptive analysis was conducted with the entire study sample to determine the median time to patient. The sample was divided into within city and outside city in order to understand urban/rural differences in the response interval. A comparison was made using scatterplots between the actual response interval and the times derived from the GIS model using empirical multipliers.
Transport interval
We used the Network Analyst extension of Esri ArcGIS 10.0[16] to estimate travel time by ground from the scene location to the hospital destination for each trip in the EMS database. Travel cost matrices were used to determine the shortest route in minutes from each patient scene location (geocoded from their x/y coordinates) to one of the three tertiary care centers as described in the database.
The actual transport time from scene to hospital is contained within the EMS database. Descriptive analysis was conducted with the entire study sample to determine the median transport time. The sample was divided into within city and outside of city in order to understand urban/rural differences in the transport time. Using scatterplots, a comparison was made between the actual transport intervals and those derived from the GIS model.
Total pre-hospital time
The overall pre-hospital time is derived as a sum of the four above-mentioned time components and can be described as follows:
1) Activation interval + Response interval + On-scene interval + Transport interval
The EMS database contains the actual total pre-hospital time from emergency call to hospital. The model for overall time is as follows where ‘GIS transport’ is the time modeled from scene to hospital using GIS and defined assumptions are from published literature[4, 6]:
2a) Urban: 1.4 minutes + (0.6*GIS transport) + 13.5 minutes + GIS transport
2b) Rural: 2.9 minutes + (0.4*GIS transport) + 15.1 minutes + GIS transport
The actual and modeled overall pre-hospital times were evaluated for statistically significant differences using the Wilcoxon signed-rank test. Using scatterplots, a comparison was made between the actual times from emergency call to hospital and these times derived using GIS and published literature. The differences between these two times were mapped to gain an understanding of the underestimation and overestimation of the model over the study area. All GIS analyses and map production were conducted using Esri ArcGIS 10.0[16].
Objective 2: Revising EMS modeling assumptions for a Canadian setting
The second objective of this study was to create a model that more accurately reflected total EMS pre-hospital time in a Canadian context. We used the information acquired through the analysis of the EMS trip data as outlined above, to develop a model that could be used to estimate travel time in the absence of actual EMS trip records. The creation of such models has been useful for estimating the access of a population across a large study area[5, 10, 12].
The actual EMS median dispatch time and time at patient scene were used to represent the dispatch time and time at scene respectively. The transport time from scene to hospital was modeled using GIS network analysis from patient scene location to hospital using the shortest time algorithm. The response interval was subsequently modeled using linear regression with no y-intercept to determine the relationship between the GIS modeled transport interval and the actual EMS response interval. This created empirical constants that could be used to derive the response interval as had been done in previous studies[5, 6]. These four time components were then added together to create a model that could estimate total EMS pre-hospital time in the absence of actual trip data. The differences between the actual EMS trip times and times derived from the updated model were mapped to gain an understanding of any spatial patterns in the underestimation and overestimation of the updated model over the study area.