Using GIS to Evaluate Rural Emergency Medical Services (EMS) Zhaoxiang He Graduate Research Assistant Xiao Qin Ph.D., P.E. Associate Professor
Outline Introduction Literature Review Study Design Data Collection & Processing Methodology K Function Cross K Function Getis Ord G* Statistic Results Conclusions
Introduction Emergency Medical Service (EMS) is defined as the personnel, vehicles, equipment and facilities used to deliver medical services to those who need immediate care but outside a hospital; and therefore, it is considered as the vital expansion of emergency care from the emergency room to the community.
Introduction Wide disparity exists in the delivery of EMS in rural areas, compared with urban areas due to reasons such as geographic barriers, lack of professional and paraprofessional, inadequate financial resources, aging or inadequate equipment, absence of specialized EMS care and local medical facilities. The objective of the study is to assess the EMS station locations and recommend service improvements.
Literature Review In most studies, response time, the time interval from when an ambulance is en-route until it arrives at the scene, is used as the indicator to evaluate the EMS performance. Many researchers focus on the geographic distribution of 911 calls, but limited studies provide the spatial association between EMS stations and incident locations. Few studies investigated contributing factors to the EMS performance, probably due to the lack of data. A great amount of research deals with similar datasets (e.g. crime dataset, crash dataset) and spatial data analysis methods were developed and used.
Incident Data Point Shapefile K- Function Analysis Cluster Yes Getis- Ord G* Analysis High Incident Density Area Clustering Analysis Improper EMS Station Location at a State Level Cross K- Function Analysis No Cluster Yes Proper EMS Station Location at a State Level Spatial Relationship analysis EMS Station Point Shapefile Service Coverage Rate at a State Level Incident Data outside Converge Additional EMS Service Generate a Service Coverage Incident Data within& outside Coverage Incident Data within Converge Contributing Factor Analysis Contributing Factor Analysis Service Coverage Polygon Figure 2 Study Design Flow Chat Study design Provide Possible Department of Civil and Environmental contributing factor Engineering And Corresponding South Dakota Recommendation State University
Data Collection & Processing SD EMS data was collected through the National EMS Information System (NEMSIS) during 2012. Step Objective Criteria Data Percentage (%) Filtered Remained 1 Complete dataset N.A. 50,396 0 100 2 911 Response only with valid location coordinates Dispatch Type = 911 response Valid Location Coordinates 29,042 42.37 57.63 3 Filter missing or invalid odometer data Mile_Scene, Mile_Dest, Mile_In = 0 or blank 17,972 22.59 35.66 4 Filter missing or invalid time intervals ERTime, ERHTime, Total Time = 0, blank,or > 240 min 16,472 2.98 32.68 5 Filter missing or invalid distance data ERDistance, ERHDistance, Distance_Back = -, 0, or >400 miles 15,540 1.84 30.84 6 Filter invalid speed data ERSpeed, ERHSpeed >120 mph 14,899 1.28 29.56 EMS ambulance station data was found from the home page of South Dakota Emergency Medical Services and 125 stations in total.
Methodology K Function Ripley s K function is a spatial statistical method for point pattern analysis: whether the points appear to be clustered, dispersed, or randomly distributed. K(r)= λ 1 E( number of extra points within distance r of a randomly chosen point ) (1) L(r)= K(r)/π (2) Where λ= Observed density (number per unit area) of points; E ( ) = expected value;
Methodology Cross K Function Similar to K function, cross K function is to analyze the co-location pattern between two kinds of points, for example A ( a 1, a 2,, a i ) and B ( b 1, b 2,, b j ) : whether the two kind of points appear to be clustered, dispersed, or randomly distributed. K ba (r)= λ a 1 E( number of points A within distance r of a point b i in B ) (3) L ba (r)= K ba (r)/π (4) Where B are fixed locations; λ a =
Methodology Getis-Ord G* Getis-Ord G* statistic indicates locations surrounded by a cluster of high or low values, aka hot spots or cold spots. G i (d)= j=1 N w ij (d) x j x j=1 N w ij (d) /S N j=1 N w ij 2 (d) ( j=1 N w ij (d)) 2 /N 1 (5) Where x = j=1 N x j /N And S= j=1 N x j 2 /N ( x ) 2 ; x j = the attribute value for feature j; w ij =the spatial weight between feature i and j;
K Function Clustering Analysis K function was first applied to examine the point pattern for all the incidents in SD. The figure shows a significant clustering pattern at the 95% confidence interval. The corresponding r for summit of the observed curve indicates that incidents exhibit the most obvious clustering pattern at a distance of 3 mile. Figure 4 K Function expressed as L(r)-r for all incidents
Getis-Ord G* Clustering Analysis Incident points were aggregated to 5*5(mile) grids and the Getis-Ord G* analysis was conducted using the incident count. 22 clusters indicate 22 high incident density areas and they are used in the following analysis.
Spatial Relationship btw EMS Stations and Incidents Cross-K function was applied to examine colocation pattern of incidents and EMS stations. The Figure indicates incidents cluster around EMS stations at close distances(<25 mile) at the 95% confidence interval. This finding shows that the EMS station are reasonably well distributed at the state level.
Figure 7 Coverage Area of 8 min for each EMS Station Eight-minute response time was chosen as the service coverage time base on National Fire Protection Association (NFPA) 1710 and 90% of the 2012 incidents are covered.
Descriptive Analysis on Service performance Group 1 (9073 ) Group2 (4451) Statistics Response Time (min) Travel Distance (mile) Travel Speed (mph) Mean 6.17 4.10 32.04 STD. Dev. 7.45 7.45 18.36 Median 4 1.6 27.60 Range [1,178] [0.04,175] [0.67,120] Mean 7.66 6.32 37.69 STD. Dev. 8.26 8.55 23.41 Median 4 1.4 33.75 Range [1,88] [0.02,58] [0.17,120] 26.29 (%)>10 min 14.35 (%)>10 min Response Time (min)
Contributing Factors for Service Performance The incidents with the longest response time (top 2.5%) are 21.06 min and 24.17 min, respectively. Regress analysis shows that in high incident density areas, travel distance can explain most of the variation in the response time. Dispatch type and location type are also statistically significant. In low incident density areas, travel distance cannot be the main contributing factor, and also dispatch type shows statistically insignificant. Definition R Square Group 1 (298) Group 2 (210) Response Mode If Light& Sirens for the ambulance is on 0.02 0.02 Dispatch Type Incident type when dispatching 0.2 0.14 Location Type Location type such as residence 0.12 0.08 Travel Distance Distance from EMS station to incident 0.65 0.06
Conclusions and Future Work State wide assessment for EMS station locations is conducted both qualitatively (co-location pattern) and quantitatively (service coverage). An overall good coverage rate of the SD EMS was found. However, longer response times corresponding to longer travel distance were found in high incident density areas within the coverage area. Why? In low density areas, none of the variables (e.g., travel distance, location type, dispatch type) seem to be relevant to the long response time cases. Why? Future work includes seeking answers for the whys and developing tactical plans for the deployment of EMS stations.
Questions?