2. Past Projects in Health Care
• Health and Wellness Related Services for VON Clients in Kings County and
Annapolis County by Greg Hunt (VON Annapolis Valley) in 2012/2013
• Analyzing Client Use of Fixed and Mobile Breast Screening Units from 2003 to
2010 by Kelly Gerrow (Nova Scotia Breast Screening Program) in 2010/2011
• Analyzing Accessibility to Breast Screening Services in Nova Scotia by Faye
Welsh (Nova Scotia Department of Health) in 2006/2007
3. Project Sponsor
• Annapolis Valley Health
• Brad Osmond,
Health Planner
• “Working together to
promote & improve the
health of individuals,
families and
communities". – AVH
Website
• Among many things,
they are responsible for
ensuring the residents
within the Annapolis
Valley have reasonable
access and coverage to
Health Care services
The Five Community Clusters in the Region
Communities are considered part of 5 larger clusters, Annapolis Royal, Middleton,
Berwick, Kentville and Wolfville
4. Project Objectives
• Display the usefulness of GIS within the Health Care sector
• By evaluating the concept of spatial accessibility within the Annapolis
Valley
• By evaluating the effect of physician retirement on the spatial
accessibility in projected years
• Making use of:
• Python Scripting
• Thematic/Choropleth Mapping
• Graduated Symbols
• Statistical Analysis: Grouping Analysis, Clustering, Obtaining Rules,
HotSpot Analysis and Mean Summary Statistics
• GIS alone does not solve the issue of spatial accessibility, it allows planners
and professionals to better assess the situation and decide upon solutions
5. Concept of Rurality, Annapolis Valley
2013 PRIZM Segmentation Groups
Majority of region presenting rural communities with an elderly population
PRIZM Segmentation Groups were developed by Environics Analytics for a better
description of the people within communities at the Dissemination Area level
• Rural communities
present unique
constraints when
assessing spatial
accessibility
• Majority of the
region can be
considered rural
• Majority of the rural
communities
contain large elderly
populations
6. Concept of Rurality, Higher Physician
Demand
2013 K-Means Clustering with Six Clusters
The more rural the community, the more demand for physician services can be
expected.
• Rural communities also
present a higher
demand for physician
services
• These communities are
similar in the fact they
present Elderly
Populations, Lower
Levels of Income,
Lower Degrees of
Education, etc.
7. Concept of Rurality, Low Spatial
Accessibility
2013 Grouping Analysis with Six Groups and Accessibility Scores
Rural communities also experience lower spatial accessibility scores
• The rural communities
which have less
desirable demographic
statistics also
experience lower
spatial accessibility
• Rural communities
must travel further for
physician services on
average than those
near the Kentville-New
Minas-Wolfville core
8. Concept of Rurality, Population Trends
2013 Total Population Across the Annapolis Valley
The majority of the population is centered around the Kentville-New Minas-Wolfville
core, with the rest following the HWY 101, main transportation route. The
population drops off as you get further away from these areas (rural communities)
• Rural communities
have less population to
service
• Resources are
allocated to areas
where it will affect the
majority of the
population
• Important to
understand this reality
• Low accessibility scores
can be expected in
more rural areas, but
does not necessarily
mean they are being
neglected
9. Data Sources
• Data was obtained and provided by numerous
public and private sources
• Annapolis Valley Health (Private)
• Physician Office Data
• Community Counts (Public)
• Community Boundary Shapefiles
• 2011 Population and Demographic
Data
• Environics Analytics (Private)
• Sourced from Business Analyst
• 2013 and 2016 demographic data
• 2013 Census Block Points
• PRIZM Segmentation Groups
• GeoBase (Public)
• Road Network Data for Nova Scotia
10. Data Preparation: Demographic Data
• Data was received as absolute and needed to be made relative
• RELATIVE TO TOTAL POPULATION
• Population 65 +
• Visible Minority
• Aboriginal Identity
• RELATIVE TO POPULATION 15 +
• Below High School Education
• High School Education
• Post-Secondary Education
• RELATIVE TO TOTAL NUMBER OF FAMILIES
• Lone Parent Families
• ALREADY RELATIVE
• Avg. Household Income
• Participation Rate in the
Workforce
• This task was performed in IBM
Modeler using the Derive node
• Editing null and zero values was also
necessary for the 2011 data
11. Data Preparation: Weighted Aspatial
Demand
• Idea was to consider the demand for physicians based on “Aspatial” factors and
not just the total population
• The demographic data discussed was broken into 7 quantile intervals (ranks) and
multiplied by a factor of importance.
Rank POP 65
PLUS %
AVG
HHDINC $
PART
RATE %
VM
%
LONE
PARENT %
NOEDU % HSCERTI % PSEDU %
1 7.3-14.7 72586-
96614
68.3-74.5 0.01-0.9 5.99-10.2 15.8-24.1 24.4-27.2 56.5-66.1
2 14.7-17.2 66734-
72586
66.7-68.3 0.9-1.6 10.2-11.4 24.1-27.9 23.2-24.4 50.4-56.5
3 17.2-17.7 63524-
66734
64.4-66.7 1.6-2.0 11.4-12.9 27.9-31.2 22.9-23.2 46.0-50.4
4 17.7-19.6 57605-
63524
62.8-64.4 2.0-2.5 12.9-14.3 31.2-33.0 21.7-22.9 43.7-46.0
5 19.6-21.2 55273-
57605
59.6-62.8 2.5-3.9 14.3-16.5 33.0-34.9 20.6-21.7 42.9-43.7
6 21.2-25.6 51914-
55273
54.5-59.6 3.9-6.6 16.5-18.6 34.9-37.8 17.8-20.6 37.9-42.9
7 25.6-32.5 30667-
51914
15.2-54.5 6.6-13.6 18.6-90.9 37.8-73.9 14.6-17.8 26.1-37.9
• The higher the rank, the
more demand is expected
• Statistical values are then
replaced with 1-7 ranks
• The variable Avg.
Household Income was
weighted at 0.5
• The rest of the variables
were weighted at
(0.5/7)= 0.07
2013 Aspatial Ranking Intervals with Highest Demand Highlighted
12. Data Preparation: Road Network
Speed Limit (km/hr) Type
110 Freeway
100 Expressway / Highway
90 Arterial
80 Collector
50 Local / Street
25 Ramp
5 Resource / Recreation
5 Service Lane
• Road Network Data came with road class types but not distance, time or speed
limits
• Speed Limit Added
• Speed limits added for each road class type
• Distance Added
• Using “Calculate Geometry” distance is obtained for each segment (metres)
• Time Estimated
• Distance / Speed to estimate time
• 3.6 * Distance (metres) / Speed (km/hr) = Time (seconds)
13. Data Preparation: Physician Office
Data
• The number of physicians at each office was an important statistic for
measuring spatial accessibility
• Office data was provided per physician and not per office
• The data was edited to include an Office ID number (1-21) and the number of
doctors at that office were included at each location
• In this way, each point feature contained one office with many doctors instead
of many doctors at one location
The table shows how doctors who
practice within the same location were
assigned to the same Office ID number
• The number of doctors was
adjusted for 2013 and 2016 to
represent the possible
retirement of physicians at
each location
14. Two-Step Floating Catchment Area
Model
• The model used to measure spatial accessibility was the Two-Step
Floating Catchment Area (2SFCA)
• This model is widely accepted and present in many case studies
• How The Model Works?
• First, considers the ratio of
[# of physicians at a
location / community
demands within buffer]
• Second, considers the same buffer but around the community
centroids, summing the previously calculated ratios at each office
15. Two-Step Floating Catchment Area
Model: An Example
Step 1: a 15 km buffer around Office 18, sum the
demand of communities within,
5.5 + 4.6 + 5.1 + 6.1 = 21.25,
the # of physicians at site 18 = 7,
therefore ratio at this office = 7/21.25
Ratio = 0.329
Step 2: same 15 km buffer, except
around Cornwallis, sum the ratio of the
office that is within = 0.329 + 0
Accessibility = 0.329 for Cornwallis
16. Python Scripting (2SFCA): What needed
to be done?
• 30 minute driving time buffer done for each physician office
• Sum the demand from each community within that buffer
• Divide the # of physicians at each office by the sum of the demand, this is
the RATIO
• 30 minute driving time buffer done for each community centroid
• Sum the RATIO of each office found within that buffer
• This sum is the SPATIAL ACCESSIBILITY SCORE for the community
• How?
• PYTHON SCRIPTING
17. Python Scripting (2SFCA): How can
scripting help?
• Python scripting will allow for…
• The ability to perform the previous steps for 21 physician office
locations and for 36 community centroid locations efficiently
• Approximately 10 minutes to run this with scripting
• The ability to perform these steps over-and-over again allowing easy
repetition
18. Python Scripting (2SFCA): ArcGIS Tools
Used?
• Tools used within the script…
• New Service Area Solver in Network Analyst (30 minute driving time buffer
polygons)
• Clip (obtaining the necessary demand and ratio information within the buffer)
• Summary Statistics (to sum the demand and ratios)
• Add Field, Calculate Field (for recording the ratio and accessibility score for
each record)
• Join Field, Append (for attaching the ratios and accessibility scores to
appropriate tables)
• Select by Attributes (to consider each office and community individually)
• SearchCursor, getValue functions (to search within the attribute tables
and grab identifying values)
19. Python Scripting (2SFCA): Results
2013 Spatial Accessibility Scores for the Annapolis Valley
Scores were produced using the python script AccessDriveTime.py created by Alex
Zscheile, utilizing ArcGIS tools
• Map displaying the
results of the
Python script for the
year 2013
• Red/orange identify
the lower scores
• Beige/blue identify
the average and
higher scores
20. Observations: Low Spatial Accessibility
in Rural Communities
• The more rural
communities have
lower accessibility as
expected
• These communities
are further from the
Kentville-New Minas-
Wolfville core and the
main transportation
route (HWY 101)
• This trend is also true
with the 2013 and
2016 data
2011 Spatial Accessibility Scores Across the Annapolis Valley
The pattern of areas with lower accessibility also displays here
21. Observations: Kingston/Greenwood
• Kingston and
Greenwood present
unexpectedly low
scores
• These communities
have one of the
largest populations
within the
Annapolis Valley
and are near HWY
101
• These communities
have the population
and location to
expect greater
access to physician
offices
2013 Spatial Accessibility Scores
Kingston and Greenwood showing unexpected low scores given their population and location
Remember - Concept of Rurality: Population Trends, displaying population makeup
22. Observations: Areas With Zero Spatial
Accessibility
Closest Physician Office to Communities with Zero Spatial Accessibility
The communities of Port Royal, Milford and Lake George have zero spatial accessibility
• Port Royal, Milford and Lake
George present zero spatial
accessibility
• Closest Facility Solver in
Network Analyst: identified
the estimated driving times
to the nearest phsyician
office
• Neighbourhood Census
Blocks were used to
consider a more realistic
evaluation
• Most neighbourhoods
presented driving times
above 30 minutes
• Averaging the driving times
for neighbourhoods in each
community resulted in:
Port Royal approx. 50 minutes
Milford approx. 90 minutes
Lake George approx. 70
minutes
Port Royal
Milford
Lake George
23. Observations: Berwick
2013 Spatial Accessibility Scores
Berwick and surrounding area show the larger scores
Close enough to the main service hub to have access to their services
• Berwick and nearby
areas present the highest
spatial accessibility
scores
• Upon closer look, their
proximity to Kentville
allows for this
• With smaller demand
than the Kentville-New
Minas-Wolfville area, and
similar access to their
services, this results in a
larger accessibility score
• Berwick is connected to
the services of Kentville
and
Kingston/Greenwood,
therefore allocating
funds to these
population hubs would
also influence Berwick
24. Effect of Physician Retirement
• The number of doctors who were expected to retire by 2013 by community
cluster:
• Wolfville Cluster – 2 doctors
• Kentville Cluster – 1 doctor
• Annapolis Royal Cluster – 1 doctor
• The number of doctors who are expected to retire by 2016 by community
cluster:
• Wolfville Cluster – 3 doctors
• Kentville Cluster – 5 doctors
• Middleton Cluser – 2 doctors
• Total doctors lost due to expected retirement
• Wolfville Cluster – 5 doctors
• Kentville Cluster – 6 doctors
• Annapolis Royal Cluster – 1 doctor
• Middleton Cluster – 2 doctors
• From the Kentville-New Minas-Wolfville core, a total of 11 doctors are
expected to be lost due to retirement by 2016
• This area and the surrounding area should see the most effect of physician
retirement
25. Pattern of Change in Spatial
Accessibility Due to Retirement
• HotSpot Analysis
within ArcMap
identifies extreme
cluster patterns
• The red shows
areas that present
a high
concentration of
significant negative
change
• This pattern
analysis shows that
the Kentville-New
Minas-Wolfville
area is most
affected
26. Percentage Change in Spatial
Accessibility due to Retirement
• The most
affected
communities
across the 5 year
period are:
• Alton
• Blomidon
• Gaspereau
• Glooscap
• The communities
Kentville,
Coldbrook,
Cambridge,
Somerset and
Waterville all
experience a
-20% change or
worst
27. Limitations for Analysis
• Time
• Competing with class work
• Gathering and preparing data
• Network Analyst being offered later in the term
• 2011 Census and Household Survey Data
• Rural and First Nations communities did not have much data
• Lack of participation across the country
• Wait-Time for Patients
• Was not considered in this assessment
• In reality community members may be near a physician office but that
office or doctor may have longer patient wait times effecting availability
at that office
28. Improvements for Analysis
• The use of the Enhanced Two-Step Floating Catchment Area (E2SFCA) Model could
further enhance the analysis of this project
• Operates the exact same way as the 2SFCA except, it considers the fact that those,
for example, who are 30 minutes away, are less likely to attend an office than those
that are 10 minutes away
• How does E2SFCA do this?
• Weights are applied to intervals within the overall buffered area
• These weights represent the likelihood of someone wishing to travel
• Example: Those that reside within the interval…
• 0-10 minutes away (multiply by 1)
• 11-20 minutes away (multiply by 0.8)
• 21-30 minutes away (multiply by 0.5)
• How might it affect the analysis?
• Consider the observation of Berwick, those from Berwick are assumed to have equal
accessibility as those from Kentville, therefore by presenting less demand than
Kentville, Berwick’s accessibility is higher
• In reality, although those from Berwick can obtain services from Kentville, it is not as
easily accessible as if they lived in Kentville, therefore you would expect the
accessibility of Berwick to be less than it is presented with the 2SFCA model
29. Conclusions: The Usefulness of GIS in
Health Care
• Python scripting within a GIS environment allows for the use of
mathematical models such as the 2SFCA
• Maps help to locate and predict patterns that can help find solutions
• Maps and statistical reports can help to describe the communities and
people, allowing better understanding of the study area
• Network Analyst an extension of ArcGIS allows for the use of road
network datasets in order to consider a truer travel impedance
experienced by community members
• With the appropriate background knowledge, GIS can further
enhance analysis and predictions to discover more efficient solutions
30. Expected Driving Time for Communities
Choropleth Map Displaying Predicted Driving Time Throughout the
Annapolis Valley