J Med Sys (2006) 30(1): 23–32
DOI 10.1007/s10916-006-7400-5
RESEARCH ARTICLE
Comparing GIS-Based Methods of Measuring
Spatial Accessibility to Health Services
Duck-Hye Yang · Robert Goerge · Ross Mullner
Received: May 2005 / Accepted: June 2005


C Springer Science+Business Media, Inc. 2006
Abstract The inequitable geographic distribution of
health care resources has long been recognized as a problem
in the United States. Traditional measures, such as a simple
ratio of supply to demand in an area or distance to the closest
provider, are easy measures for spatial accessibility. How-
ever the former one does not consider interactions between
patients and providers across administrative borders and the
latter does not account for the demand side, that is, the com-
petition for the supply. With advancements in GIS, how-
ever, better measures of geographic accessibility, variants
of a gravity model, have been applied. Among them are (1) a
two-step floating catchment area (2SFCA) method and (2) a
kernel density (KD) method. This microscopic study com-
pared these two GIS-based measures of accessibility in our
case study of dialysis service centers in Chicago. Our com-
parison study found a significant mismatch of the accessi-
bility ratios between the two methods. Overall, the 2SFCA
method produced better accessibility ratios. There is room
for further improvement of the 2SFCA method—varying
the radius of service area according to the type of provider
or the type of neighborhood and determining the appropri-
ate weight equation form—still warrant further study.
Keywords Spatial accessibility · GIS · Floating catch-
ment · Kernel density
Introduction
One of the main goals for health care facility planning is
to achieve the equitable geographic distribution of health
care resources. An equitable distribution of health care re-
sources can achieve two main goals set for facility planning:
D.-H. Yang (B)· R. Goerge
Chapin Hall Center for Children at the University of Chicago,
Chicago, Illinois
e-mail: duckhye@uchicago.edu
R. Mullner
Health Administration, School of Public Health at the University
of Illinois at Chicago, Chicago, Illinois
(1) cost containment by decreasing oversupply and (2) eq-
uity of access by increasing supply to underserved areas.
One approach to health care facility planning is to com-
pare the supply of a certain type of service with the de-
mand within a service planning area, for example, a county.
However, this measure does not identify variations in local
areas. Moreover, it does not take into account the interac-
tion across borders that may have been drawn mainly for
planning purposes [1].
A number of studies examine the impact of distance to
health care services on utilization of those services [2–5].
Examples of distance-related measures include the distance
to the nearest facility, the average distance to a certain num-
ber of closest facilities, the average number of facilities
within a certain distance from a specific residence or loca-
tion. Distance-related measures are one of several aspects
of access to health care [6]. A long waiting list or crowded
waiting rooms may be a better indicator of accessibility to
health care, particularly in urban areas.
Combining the two components—distance and demand
relative to the supply—is known as the best measure of
spatial accessibility to health care resources. This measure
has been developed and enhanced in other fields over the
decades; for example, a gravity model, a method of predict-
ing the distribution of resources on a surface, has been used
for location analysis in many fields. However, such a model
requires intensive computation resources. With advances in
the geographic information systems (GIS), researchers in
the health care field have started to use the combined spatial
accessibility measure to examine the equitability of health
care resources, mainly because GIS-based methods are eas-
ier to implement and to map the accessibility scores. How-
ever, little work has been done to critically evaluate these
methods. Thus, we selected the two most recent studies that
used GIS-based methods: the two-step floating catchment
area method (2SFCA) and the kernel density method (KD).
We implemented the two methods in a case study of mea-
suring spatial accessibility to dialysis centers treating pa-
tients with end-stage renal disease (ESRD), the permanent
loss of kidney function, and compared the weaknesses and
strengths. We begin with descriptions of the two methods.
23
24
Fig. 1 A hypothetical example for two-step floating catchment area method
Two-Step floating catchment area method (2SFCA)
To predict access to social programs, Radke and Mu [7]
developed the two-step floating catchment area method
(2SFCA), the term coined by Luo and Wang [8]. The
2SFCA method is a special case of a gravity-based method
(see Luo and Wang [8] for the proof). Figure 1 shows a hy-
pothetical example of the 2SFCA method. The two-county
area contains three facilities with two physicians each and
15 census tracts representing population or demand for
physician services. For simplicity, there are four patients
needing dialysis in each of the tracts except for three tracts:
Tracts 5, 8, and 11 are assumed to have two patients. In
the first step, a service area for each facility is defined in
terms of census tracts. Figure 1 illustrates two service areas
centered at each facility point, drawn based on the prede-
termined distance (or time) between the facility and census
tracts. For each census tract within a service area, a ratio is
computed by dividing the number of physicians by the sum
of all discrete population values within the service area. For
example, each of the eight census tracts within the Facil-
ity A service area is assigned the service area ratio that is
obtained by dividing 2 (physicians) by 32 (patients). Sim-
ilarly, tracts within the Facility B service area are assigned
the service area ratio, 2/10. In the second step, Tract 4 in
the overlapping area is assigned a new ratio, 37/80, the sum
of the Facility A ratio and the Facility B ratio, 2/32 + 2/10.
Three observations are noteworthy. First, the accessibil-
ity ratio assigned to each tract is in fact an average ratio
within a service area, not within a tract itself. This suggests
that there is a smaller variability of ratios than those that
would be calculated within each tract. Second, not only the
supply itself, but also the location (i.e., overlapping area)
and the low demand contribute to high accessibility. Third,
distance does not affect the accessibility ratio once tracts
are included within a service area. For example, Tract 7
is given the same ratio as Tract 1, even though Tract 7 is
closer to Facility A. If a planner identifies potential under-
served areas as those with ratios lower than the entire area
J Med Sys (2006) 30(1): 23–32
ratio, ratios based on the 2SFCA method are problematic.
For example, Tract 7 where Facility A is located, has a
lower ratio than the two-county average ratio, 3/25. This
illustrates a potential drawback of the 2SFCA method.
The 2SFCA method has two properties that provide cer-
tain advantages for interpreting ratios. The accessibility
ratio obtained for each census tract is interpreted as the
amount of the total services shared by one patient living in
the tract. Using Fig. 1 as an example again, because there
are six physicians in total, the ratio for Tract 1, 1/16, indi-
cates that 1/16 physician is available to one patient in Tract
1. The share of total services available to all residents in
Tract 1, therefore, is obtained by multiplying its accessibil-
ity ratio (1/16) by the total number of residents (4) in the
tract, resulting in one-fourth physician. Note that the unit
for the ratio keeps the unit for the supply, the physician.
The aggregate supply per tract is summed across all tracts,
resulting in the total sum, which is equal to the total supply
or 6 physicians in this example. In short, the total supply
is reserved. Because of this reservation of the fixed total
supply, if the ratio for Tract 1 is 1/5 and the ratio for Tract 2
is 2/5, we not only say that Tract 2 has greater accessibility
relative to Tract 1, but also can say more specifically that
in Tract 2 heath care is twice as accessible as in Tract 1.
Kernel density (KD) method
Kernel function has been applied to address geographi-
cal issues. GIS software, such as the Spatial Analyst ex-
tension of ArcGIS, provides tools for kernel distribution
function estimation. First, a study area should be repre-
sented by grids, an array of equally sized square cells (or
pixels) arranged in rows and columns. The kernel density
estimator starts with facility location points. Each point
value represents service capacity, for example, the number
of physicians. Kernel function is used to create a surface
showing the predicted distribution of the service availability
throughout the study area. In this sense, the kernel density
J Med Sys (2006) 30(1): 23–32
Fig. 2 A Two-modal kernel density estimation. B One-modal kernel
density estimation
(KD) method is one variant of gravity modeling, although
density analysis is not usually referred to in those terms.
Figure 2A shows a 2-modal distribution profile showing
the construction of kernel density based on 10 facility lo-
cations (+). Facility service capacity (e.g., physicians) is
distributed on the surface based on 10 Gaussians (typically,
quadratic approximations) centered at 10 facility locations.
The width of Gaussian, or bandwidth, is supposed to re-
flect the radius of a facility’s potential service area. Cells
near the facility location receive higher values of service
capacity (i.e., accessibility), and those near the periphery
receive very little. Because facility service areas overlap
in practice, the density value in cells in these overlapping
areas is the sum of contributions from all overlapping ser-
vice areas. As a result, Fig. 2A shows two modes occurring
where facilities concentrate.
The number of modes in kernel distribution function esti-
mation depends on the bandwidth. Figure 2B shows a kernel
estimation with one mode that shows flattened Gaussians
due to the use of a larger bandwidth. Figure 6A (in Result
section) shows a map version of one-modal kernel density,
compared to a multimodal distribution shown in Fig. 6B (in
Result section). Although Fig. 6A shows only one peak of
contour lines, Fig. 6B shows multiple peaks, indicating the
25
use of a smaller bandwidth. Selection of optimal bandwidth
is the hardest part of kernel density estimation.
Method
Case study: planning for renal care in illinois
Illinois is one of the states that continue to keep Certifi-
cate of Need (CON) programs for renal care resources.
Renal dialysis centers must apply for permits before start-
ing projects, including establishment or modification of
facilities (including relocation/reduction of services). As
required by the CON statute, the Illinois Department of
Public Health publishes the inventory of existing dialysis
facilities and the projected needs in terms of the number of
stations for each planning area. The document titled Inven-
tory of Health care Facilities and Services and Need Deter-
minations (Edition 2002) projected a need of 309 additional
dialysis stations for the year 2005 [9]. The projected needs
concentrate mostly in Chicago and its suburbs.
We chose renal care facility planning as a case study
for three reasons. First of all, whereas most other ser-
vice categories (e.g., surgical care beds) show surplus
resources, renal care resources are in continuous shortage.
Second, proximity to a dialysis facility is critical for
dialysis patients because they require at least 3 visits a
week for dialysis treatments. Third, unlike most other
service categories, ESRD patients cannot delay or skip
lifesaving dialysis, and public programs such as Medicare
and Medicaid provide generous coverage for dialysis
treatments. The Illinois State Chronic Renal Disease
Program, established by statute in 1967, assists patients
who do not qualify for Medicaid or KidCare or cannot meet
spenddown requirements. All things considered together,
the revealed demand for renal care (i.e., actual utilization)
is most likely the same as the potential demand.
The number of dialysis patients is expected to increase.
The need determination formula for renal care uses a factor
1.33, meaning a 33% increase in the number of dialysis
patients annually [9]. There may be an upward adjustment
to the projected demand if there is an increase in diabetes
and hypertension, the two principal causes of ESRD.
Data
The CON statute of Illinois requires that facility-specific in-
formation be available to the public. This study uses the ad-
dress, the number of stations, and the number of patients for
each dialysis center reported for the year 2001–2002 [9]. A
total of 71 facilities were inside the Cook county boundary,
which includes the city of Chicago. Although we focused
on Chicago, we included its suburban area as well to deal
with the border problem. That is, accessibility measures
near the edge of an area are not correct because residents
may seek resources outside of the area and residents out-
side of the area may seek resources inside of the area. Thus,
we incorporated dialysis centers and patients in suburban
26
Table 1 Comparison of the two methods
Comparison Luo and Wang [7]
Software ArcGIS
Approach Vector-based
Population-weighted point Yes
Population data Tract-level
Distance measure Network
Study area Entire Illinois
Physician data Aggregated at zipcode
Radius of population residence Not applicable
Radius of physician service area 30-minute car travel time
Cell size, number of cells Not applicable
areas into the study to account for the contribution of dialy-
sis centers (supply) and patients (demand) outside Chicago.
The census block-level population data were extracted
from the 2000 Census Summary File 1 [10]. Based on
block-level population, weighted centroids of census block
groups were calculated. These centroids were used for the
KD method. Similarly, based on blockgroup-level popula-
tion, weighted centroids of census tracks were calculated
and were used for the 2SFCA method. Weighted centroids
are better choices when the census tracks consist largely of
nonresidential areas–such as park or airport.
Apparicio’s findings in his study of comparing the dif-
ference between distance approximations (Euclidean or
Manhattan distances on one side) and network-based dis-
tance on the other side suggest that network-based distance/
time needs to be used for small area analyses [11]. To cal-
culate travel time in the network, we used high-resolution
road network data that is based on street data used for
geocoding with ArcGIS Streetmap extension. Travel time
is estimated as the shortest time through road networks
between a resident location (census tract centroid) and
a facility [12]. Speed limit data were used to correctly
calculate travel time that should vary by various road
types.
Implementation of the two methods
Implementing the two methods in our case study was based
on two recent studies done by Luo and Wang [8] and
Guagliardo et al. [13]. Table 1 presents key comparisons
between the two methods. Detailed discussions on the two
studies follow.
Luo and Wang
Luo and Wang [8] used the 2SFCA method to identify
areas in Illinois that have a shortage of physicians. The
census tract was chosen as the analysis unit for population,
or demand. Zip code area was chosen as the analysis unit
for physicians, or supply. Population-weighted centroids of
census tracts (based on blockgroup-level population) were
used instead of simple geographic centroids to represent
J Med Sys (2006) 30(1): 23–32
Guagliardo et al. [12]
ArcGIS and its extension, Spatial Analyst
Grid-based
No
Block level
Euclidean
Washington, DC
Aggregated at zipcode
1 mile
3.9 miles
1/10 sq mile, 5,617
population locations more accurately. Luo and Wang used
a 30-min car road-network travel time in drawing a service
area centered at each physician location zip code [14].
Guagliardo et al.
Following the approach made by Guptill [15], Guagliardo
et al. [13] used the Gaussian KD method to identify areas
with a low ratio of primary care physicians to children in
Washington, DC. They chose the option of the quadratic
approximation Gaussian KD method from ArcGIS Spa-
tial Analyst. Three steps were taken to calculate accessi-
bility ratios at the level of census tracts. First, a physi-
cian density surface was created on a grid with a total
of 5617 cells (1/10 square mile). The density values in
the cells were expressed as the number of physicians per
square mile. The radius of a physician service area was set at
3.9 miles. Next, a child population density surface was cre-
ated. Guagliardo et al. used 1 mile as the radius for smooth-
ing population at the block level over the entire block group
area with overlaps with adjacent blockgroups. Note that
the radius for population smoothing is a lot smaller, that
is, 1/4 of the radius for the provider smoothing. Next, a
provider-to-child population ratio surface was created by
dividing the provider density surface by the child popula-
tion density surface. Finally, census tract boundaries were
overlaid on top of the ratio surface cells, and the mean ratio
for each tract was calculated. Guagliardo et al. observed that
calculating ratios within each cell made the variability of
ratios very volatile. Although the mean ratio was calculated
based on cells within a tract, some ratios were extremely
high, and were excluded from the subsequent analyses.
Our study
Figure 3 depicts our study area. There were 44 dialysis cen-
ters with 867 stations and 3722 dialysis patients in Chicago
during 2000–2001. As mentioned earlier, to account for the
border problems, the two methods were implemented in the
larger area–Cook County that Chicago belongs to. There
were 71 dialysis centers with 1440 stations in Cook County.
Figure 3 also indicates the rectangle, or extent, bounding
J Med Sys (2006) 30(1): 23–32 27

Fig. 3 Study area

Cook County. The rectangle is divided into 5.19 million
cells with an area of 900 m2 (30 m × 30 m), as imple-
mented by KD method.
Table 2 contains detailed information on the two meth-
ods implemented in our case study. Demand is expressed
as the number of patients who need dialysis treatment.
The number of patients is obtained for each census tract
by multiplying the 2000 census tract population by a con-
stant factor, 0.12%–the percent of population that requires
dialysis treatments observed for Chicago in 2000 [9]. Sup-
ply is expressed as the number of stations in each dialysis
center.

Table 2 Comparisons of two methods as implemented in our case study

2SFCA KD
Software Unix ArcInfo Station PC ArcGIS and Spatial Analyst

Population-weighted point Yes Yes

Population data Tract-level Block-level
Distance Network Euclidean
Study area Cook County Cook County
Dialysis Center data Street Address Street Address
Radius of patient residence Not applicable 1 km
Radius of facility service area 30-min car travel time 20 km
Cell size Not applicable 30 m × 30 m
Number of cells 5.19 million
28
Results
Comparisons of accessibility ratios
As expected, accessibility ratios calculated from the KD
method show greater variability than the 2SFCA method
(Table 3). First of all, the range of KD accessibility ratios is
larger–by a multiplicative factor of 7.5 (0.98 versus 0.13).
The maximum number of stations that one person in one
census tract can get access to, when it was estimated by the
2SFCA method, was approximately 0.31, compared with
slightly more than 1 station when it was estimated by the
KD method. The minimum number of stations calculated
by the KD method was 0.03, compared with 0.18 by the
2SFCA method. This large difference in the range is also
reflected in the large difference in standard deviation. A
greater variability in the KD ratios was expected because
the KD ratios were calculated first within a tiny cell, while
2SFCA ratios were calculated as a local average within a
service area defined using a 30-min travel time.
Not only does the range of ratios differ, but also the
frequency distributions of accessibility ratios are also strik-
ingly different. Figure 4A shows that the distribution by the
2SFCA method is skewed to the left. In contrast, Fig. 4B
shows that the distribution by the KD method is skewed
to the right. Figure 4C shows that the KD method tends to
deflate the ratios in census tracts that the 2SFCA method
identified as high-accessibility areas. The mean of the KD
ratios is smaller than the average ratio for Chicago.
We mapped out the accessibility ratios for comparison of
their positions relative to the average Chicago ratio (Fig. 5).
The census tracts were grouped according to the ratio of
its accessibility ratio to the Chicago ratio, which was 0.23.
Five groups of a total of 863 census tracts in Chicago were
made using the following categorization: less than 80, 80–
90, 90–100, 100–120, and more than 120%. The first three
groups represent census tracts that had less access to sta-
tions than the average patient in Chicago. The next two
groups account for census tracts that had more access than
the average patient.
Three observations from Fig. 5 are noteworthy. First of
all, the first three groups–having the below-average acces-
sibility as estimated by the 2SFCA method–accounted for
30% of all census tracts, contrasted with about 80% as es-
timated by the KD method. This difference was expected
as shown by the difference in the direction of skewness
between the two frequency distributions above. Second,
Table 3 Comparison of accessibility ratios
Mean SD Minimum Maximum Range
KD 0.180 0.108 0.030 1.006 0.976
2SFCA 0.264 0.019 0.176 0.307 0.132
Chicagoa 0.233
a
Indicates an average ratio of the number of stations to the number
of patients in Chicago.
J Med Sys (2006) 30(1): 23–32
the 2SFCA map shows a smoothing pattern of accessibility
ratios, compared with a somewhat disruptive pattern in ra-
tios on the KD map. The southern part of the city, which
includes two community areas, Hegewisch and Riverdale,
was identified as the shortage area by the 2SFCA method,
but as the super surplus area by the KD method. How-
ever, those two areas had adjacent census tracts that had
the lowest accessibility, showing a disruptive pattern, when
estimated by the KD method.
To examine the factors that might account for the mis-
match between the ratios produced by the two methods, we
examined a kernel density map showing stations distributed
to 5.9 million cells (Fig. 6A). The cells were divided into
nine groups, showing eight contour lines. Station values are
expressed in density, that is, the amount per square mile.
Figure 6A shows that two factors–the raster data approach
of the KD method and the use of the large radius of catch-
ment area (20 km)–make a significant number of stations
distributed outside of the land, that is, Lake Michigan. On
the other hand, the demand side, the number of patients,
was not affected as much because of its small radius (1 km)
for smoothing (Fig. 6B). Therefore, a great loss of stations
relative to a small loss of patients makes the KD accessi-
bility ratios smaller. By contrast, the 2SFCA method does
not lose any station and patient. This difference may ex-
plain the difference in the direction of the skewness of
accessibility ratios between the two methods. By the same
token, this explains why the KD method tends to deflate
the ratios in areas that the 2SFCA method identified as
high-accessibility areas.
Discussion
Our comparison study found a significant disparity in the
accessibility ratios between the two methods. We found
more problems with KD method than the 2SFCA method.
First of all, the supply side of the ratio calculation can be
significantly underestimated, as shown in case of Chicago,
which is adjacent to Lake Michigan. The extent to which
underestimation occurs depends on the bandwidth, or the
radius of a service area. A smaller radius allows less loss
of supply data. Second, the KD method calculates acces-
sibility ratios within a tiny cell. If these cell-based ratios
are subsequently averaged in a yet another small area, such
as census tracts in urban areas, accessibility ratios can be
extremely volatile.
These two problems, when combined, may make it hard
for a local facility planner to identify local areas that have
below-average accessibility. Our case study showed that ra-
tios calculated by the KD method were not useful in identi-
fying low-accessibility areas. Census tracts with ratios that
were less than 80% of the average ratio accounted for as
many as 65% of all census tracts. Furthermore, there was a
disruptive pattern of ratios. Some local areas identified as
super-surplus areas had adjacent areas that belonged to the
lowest ratio group, and some border areas that belonged
J Med Sys (2006) 30(1): 23–32 29

Fig. 4 A Percent distribution of accessibility ratio: 2SFCA method. B Percent distribution of accessibility score: Kernel density method. C
Scatter plot of accessibility ratios 2SFCA versus KD

to the highest-accessibility group belonged to the lowest-
accessibility tier on the 2SFCA map.
Interpretation of accessibility ratios obtained by the
2SFCA method is more straightforward. The ratio assigned
to a census tract is interpreted as the number of stations
available to each patient in the tract. Therefore, patient-
weighted ratios sum to the total number of stations avail-
able to the city, which makes interpretation straightforward.
Patients in a census tract with a ratio of 0.5 can be inter-
preted as having two times more resources than those in a
30
Fig. 5 Comparison of accessibility ratios on maps
census tract with a ratio of 0.25. Unlike the 2SFCA ratios,
ratios obtained from the use of the KD method cannot be
interpreted as straightforward. Only relative interpretation
is possible.
The 2SFCA method was a powerful tool for identify-
ing areas for expanding dialysis services. Mostly border
areas were identified as severe shortage areas. Note that
this is not due to a border/edge problem often observed
in other studies, because we included dialysis centers and
patients in suburban areas in calculating accessibility ratios
for Chicago. The northern part of Chicago was identified
as an area of severe shortage by both the 2SFCA method
and the KD method. Specifically, it includes Rogers Park,
Edison Park, West Ridge, and Edgewater among 77 offi-
cially defined community areas. Although dialysis centers
were located right in those areas, the patients living in those
areas appeared to compete for the stations with residents in
areas outside of Chicago that usually have high population
density.
The fact that census tracts with 2SFCA ratios higher than
the average accounted for about 70% of all census tracts
suggests that the 2SFCA method tended to determine cen-
sus tracts with lower population as areas with higher acces-
sibility than the average. This indicates a lower competition
for a fixed number of stations, boosting accessibility ratios
in census tracts with a small population. However, some
census tracts with lower population than the average can
J Med Sys (2006) 30(1): 23–32
be poorer neighborhoods. More importantly, the patients
living in those areas are more likely to rely on public trans-
portation. Because our study used a 30-min car travel time
in determining a service area, some dialysis centers may
not be reachable for residents who cannot afford automo-
biles. Therefore, the radius of the service area, 30-min car
travel time, may need to vary depending on the area’s auto-
mobile ownership rate or the proportion of the population
with elderly residents.
Further improvements for the 2SFCA method may be
possible. The actual service capacity can vary depending
on the type of provider. For example, a station in Cook
County Hospital, a major safety net provider in Illinois,
generated 20 times as many procedures as an average sta-
tion. A simple count of stations cannot reflect the true ca-
pacity that meets the demand. In addition to refinements
of the supply side for better accessibility measure, the de-
mand side should be refined as well. For convenience, we
assumed that the number of patients in census tracts was
proportional to the population size in census tracts using
the Chicago average. Better measures of demand should
incorporate the analysis of utilization by gender, race, and
age groups so that the number of patients in an area can
be adjusted by those important characteristics. It is well
known that African American and Hispanic populations
tend to concentrate and that they have particularly high
rates of diabetes [16].
J Med Sys (2006) 30(1): 23–32 31

Fig. 6 A Stations distributed by Kernel density method. B Patients distributed by Kernel Density method
32
References
1. Guagliardo MF, Ronzio CR, Cheung I, Chacko E, Joseph JG
(2004) Physician accessibility: An urban case study of pediatric
primary care. Health Place 10:273–283
2. O’Neill L (2003) Estimating out-of-hospital mortality due
to myocardial infarction. Health Care Manage Sci 6:147–
154
3. Fortney J, Rost K, Zhang M, Warren J (1999) The impact of ge-
ographic accessibility on the intensity and quality of depression
treatment. Med. Care 37(9):884–893
4. Hadley J, Cunningham P (2004) Availability of safety net
providers and access to care of uninsured persons. Health Serv
Res 39(5):1527–1547
5. Lovetta A, Haynesa R, Sunnenberga G, Galeb S, (2002) Car
travel time and accessibility by bus to general practitioner ser-
vices: A study using patient registers and GIS. Soc Sci Med
55:97–111
6. McLaughlin CG, Wyszewianski L (2002) Access to care:
Remembering old Lessons. Health Serv Res 37(6):1441–
1443
7. Radke J, Mu L (2000) Spatial decomposition, modeling and
mapping service regions to predict access to social programs.
Geogr. Inform Sci 6:105–112
8. Luo W, Wang F (2003) Measures of spatial accessibility to
health care in a GIS environment: Synthesis and a case study
J Med Sys (2006) 30(1): 23–32
in Chicago region. Environment and Planning B 30(6):865–
884
9. Center for Health Statistics, (2002) Inventory of Health Care
Facilities and Services and Need Determinations, Hospitals and
Associated Services, vol. 1, Parts I-V., Illinois Department of
Public Health, Illinois
10. US Bureau of Census (2000) Census Summary File 1
11. Apparicio P, Shearmur R, Brochu M, Dussault G (2003) The
measure of distance in a social science policy context: Advan-
tages and costs of using network distances in eight Canadian
Metropolitan areas. J Geogr Inform Decision Anal 7(2):105–
131
12. Wang F (2005) Measuring distances and time. In Wang F, Albert
DP, Albrecht J (eds.) Quantitative Methods and Applications in
GIS, Chap. 2, CRC Press, Boca Raton, FL
13. Guagliardo MF (2004) Spatial accessibility of primary care:
Concepts, methods and challenges. Int J Health Geogr 3(3):1–
13
14. Wang F (2005) GIS-based measures of spatial accessibility and
application in examining health care access. In Wang F, Albert
DP, Albrecht J (eds.) Quantitative Methods and Applications in
GIS, Chap. 5, CRC Press, Boca Raton, FL
15. Guptill SC (1975) The spatial availability of physicians. Proc
Assoc Am Geogr 7:80–84
16. United States General Accounting Office (2004) Medicare dialy-
sis facilities: Beneficiary access stable and problems in payment
system being addressed, June