See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/51519522

A Spatial Analysis of Variations in Health Access: Linking Geography, Socio-

Economic Status and Access Perceptions

Article  in  International Journal of Health Geographics · July 2011

DOI: 10.1186/1476-072X-10-44 · Source: PubMed

CITATIONS READS

135 357

3 authors:

Alexis Comber Chris Brunsdon

University of Leeds National University of Ireland, Maynooth
165 PUBLICATIONS   3,350 CITATIONS    172 PUBLICATIONS   12,491 CITATIONS   

SEE PROFILE SEE PROFILE

Robert Radburn

12 PUBLICATIONS   205 CITATIONS   

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Developing a Migration Model View project

PhD research: Mapping of salt marsh vegetation habitats with EO View project

All content following this page was uploaded by Chris Brunsdon on 06 June 2014.

The user has requested enhancement of the downloaded file.

Comber et al. International Journal of Health Geographics 2011, 10:44 INTERNATIONAL JOURNAL
http://www.ij-healthgeographics.com/content/10/1/44 OF HEALTH GEOGRAPHICS

RESEARCH Open Access

A spatial analysis of variations in health access:

linking geography, socio-economic status and
access perceptions
Alexis J Comber1*, Chris Brunsdon2 and Robert Radburn3

Abstract
Background: This paper analyses the relationship between public perceptions of access to general practitioners
(GPs) surgeries and hospitals against health status, car ownership and geographic distance. In so doing it explores
the different dimensions associated with facility access and accessibility.
Methods: Data on difficulties experienced in accessing health services, respondent health status and car ownership
were collected through an attitudes survey. Road distances to the nearest service were calculated for each
respondent using a GIS. Difficulty was related to geographic distance, health status and car ownership using
logistic generalized linear models. A Geographically Weighted Regression (GWR) was used to explore the spatial
non-stationarity in the results.
Results: Respondent long term illness, reported bad health and non-car ownership were found to be significant
predictors of difficulty in accessing GPs and hospitals. Geographic distance was not a significant predictor of
difficulty in accessing hospitals but was for GPs. GWR identified the spatial (local) variation in these global
relationships indicating locations where the predictive strength of the independent variables was higher or lower
than the global trend. The impacts of bad health and non-car ownership on the difficulties experienced in
accessing health services varied spatially across the study area, whilst the impacts of geographic distance did not.
Conclusions: Difficulty in accessing different health facilities was found to be significantly related to health status
and car ownership, whilst the impact of geographic distance depends on the service in question. GWR showed
how these relationships were varied across the study area. This study demonstrates that the notion of access is a
multi-dimensional concept, whose composition varies with location, according to the facility being considered and
the health and socio-economic status of the individual concerned.
Keywords: Accessibility, Geographically Weighted Regression

1. Introduction body of research has examined service accessibility by

The subject of health facility access has long been of considering the socio-economic aspects of access related
concern to community and public health planners [1-4]. to cost, insurance provision etc, with data collected
Previous research on public health access has been in using opinion or attitudes surveys [9-13]. In both cases
two distinct and usually non-overlapping areas. One the objective is usually to inform spatial planning and
tranche has considered the spatial dimensions related to health policy making. This paper presents an analysis
geographic access (distances, travel times, catchments, that straddles these different types of accessibility
etc), with data being manipulated and geographically research. It uses a local regression analysis (as opposed
analysed using geographical information systems (GIS) to a global one) to explicitly link the experiential and
before subsequent statistical analyses [5-8]. Another geographical dimensions of access in order to provide a
more nuanced and comprehensive analysis of health
* Correspondence: [email protected] facility access. It combines analyses of public percep-
1
Department of Geography, University of Leicester, Leicester, LE1 7RH, UK tions of service accessibility from an attitudes survey
Full list of author information is available at the end of the article

© 2011 Comber et al; licensee BioMed Central Ltd. This is an Open Access article distributed under the terms of the Creative Commons
Attribution License (http://creativecommons.org/licenses/by/2.0), which permits unrestricted use, distribution, and reproduction in
any medium, provided the original work is properly cited.
Comber et al. International Journal of Health Geographics 2011, 10:44 Page 2 of 11
http://www.ij-healthgeographics.com/content/10/1/44

with an analysis of geographic road distance to those whilst these various analyses have in some cases recog-
services. The attitudes survey captured information on nised the different dimensions of accessibility [31], they
the difficulty experienced by respondents in their access have generally adopted a specifically spatial or geo-
to different medical facilities, respondent health status graphic definition of ‘access’,-i.e. one based on quantita-
and car ownership. tive analyses of distances and travel times to services to
The primary aim of this study was to determine local define service accessibility.
spatial variations in the statistical relationships between The concepts of ‘Access’ and ‘Accessibility’ are more
perceptions of health facility access with geographical complex than simple distance measures [32,33]. They
distance to the nearest facility, health status and car encompass a wider set of factors relating to behaviours
ownership. Examining the spatial non-stationarity in and perceptions which relate to a range of highly quali-
these relationships identifies locales where mismatches tative factors such as perceived service quality, opening
between access perceptions and geographic access exist, hours and previous experiences. From the social
thereby allowing community health planners to target sciences literature Farrington and Farrington note that
different activities in those specific areas. For example, accessibility can be viewed as “the ability of people to
areas where negative perceptions of access are not pre- reach and engage in opportunities and activities” [[34],
dicted by geographic distance and health status may be p2] and therefore improving access outcomes involves
indicative of some underlying problem in service deliv- overcoming the social dimensions of access and separa-
ery. A secondary aim was to explore the different tion, as well as spatial constraints. Multi-dimensional
dimensions associated with the concept of ‘accessibility’ approaches in health planning have been recommended,
that ought to be considered in health planning. This ones that consider aspects other than distance and cost,
was done by analysing access attitudes in combination in order to identify different barriers to health care ser-
with access geographies. Hitherto, much of the health vices [35]. However, in only a few cases where qualita-
geographics literature has only been concerned with tive and quantitative access dimensions have been
physical or geographic distance. The use of local regres- considered, were the local spatial variations in the rela-
sion techniques to accommodate any spatial non-statio- tionships examined. For example, Maroko et al [33]
narity builds on and extends previous work that has used Geographically Weighted Regression to explore the
considered the different dimensions associated with ser- spatial relationships between the variables associated
vice access [14]. with models of park acreage and density of physical
activity sites.
2. Background There has been little research that has explicitly exam-
The ‘spatial’ or geographic aspects of health provision ined the spatial variation of factors related to access per-
and access to facilities have been considered in much ceptions against geographic factors. The purpose of this
previous research. Typically in such studies distances to research was to address such gaps. First, it examined
services or facilities are measured (straight line or road how the perceptions of access to health facilities, as cap-
distance) and analysed in order to quantify differences tured by an attitudes survey, related to geographical or
in access, gaps in service provision, to model optimal spatial measures of access and health status globally,
facility location and to identify inequalities in service using a generalized linear model (GLM). Second, the
provision. Recent examples of these purely spatial spatial variations in these relationships were analysed
approaches in health science include identifying health using Geographically Weighted Regression (GWR), a
catchments [5,15], examining equity of access for differ- local regression analysis which allows the spatial non-
ent social groups [8,16,17] and modelling spatial pat- stationarity of relationships between variables to be
terns of facility usage and access [18-21]. Additionally, a examined. The models resulting from these two
number of reviews of the use of GIS based technologies approaches were used to predict respondent perceptions
to evaluate geographic or physical access to health ser- over service access from stated health status, network
vices have been published [22-25]. This body of research distance to the nearest facility and car ownership. By
applies geographic and spatial statistics to determine analysing access perceptions and access distances to
how best to allocate resources in order to minimise gaps hospitals and GPs this research also compares how
in provision and to identify service users with low levels these relationships vary for different types of health
of access. Increasing sophistication in analysis is also service.
evident with evaluations of different distance measures
relating to access [26,27], alternative statistical models 3. Methods
[21], exploration of geographical variation in access Data and Study Area
models [27-29] and advanced heuristic search techni- An attitudes survey in the UK county of Leicestershire
ques for optimising facility locations [1,30]. However, was conducted in 2008 by Leicestershire County Council
Comber et al. International Journal of Health Geographics 2011, 10:44 Page 3 of 11
http://www.ij-healthgeographics.com/content/10/1/44

(LCC) as part of the UK government’s Department of addresses at random from the Post Office small users
Communities and Local Government’s ‘Place Survey’. Address File database. For each of the 7 districts in Lei-
Leicestershire is a rural county, with the City of Leice- cestershire, sampling was stratified with the aim of
ster (a separate local authority) forming a hole in centre reaching a sample size of at least 1,100 in each district,
of the county (see Figure 1). The Place Survey is a postal regardless of population size. Central government pro-
survey designed to collect data to support national indi- vided the sample of addresses. The questionnaire was
cators. Individual local government authorities were sent to households only and was completed by any resi-
responsible for administering the survey and were able dent aged 18 or over living at the address. A total of
to include additional questions if they so wished. 20,260 questionnaires were sent out and the response
Because of this, LCC included questions that asked rate for each district was between 41% and 43%. The
respondents to describe their perceptions of their access survey response rates by demographic factors are sum-
to a range of health services (GP surgeries, dentists, hos- marised in Table 1. Leicestershire Statistics and
pitals and pharmacies) using a 5-point scale that allowed Research Online provide detail of the Place Survey in
respondents to indicate whether they found access ‘Very Leicestershire 1 and an interactive visualisation of the
easy’, ‘Fairly easy’. ‘Neither easy nor difficult’, ‘Fairly dif- results2.
ficult’ or ‘Very difficult’. Respondents were also asked to In the UK GP surgeries provide free access to a medi-
indicate their general health (a 5-point scale from very cal practitioner who treats acute and chronic illnesses,
good to very bad), whether they had any long-standing provides preventive care and health education for all.
illness, disability or infirmity (yes or no) and whether Data for GP surgeries and major National Health Ser-
they owned a car or not. In Leicestershire there were vice (NHS) hospitals, with and without Emergency
8530 responses to the Place Survey, with 4.9% indicating Department (ED) facilities, were downloaded from the
difficulty (i.e. replying either ‘difficult’ or ‘very difficult’) NHS website http://www.nhs.uk and spatially located
in their access to GPs and 20.2% indicating difficulty in from their postcodes. In the UK there are an average of
their access to hospitals. Of the respondents, 4.6% stated ~15 residential addresses per postcode providing a fine
that they had ‘bad health’ or ‘very bad health’ (hence- level of geographical detail. The locations of GP sur-
forth ‘Bad Health’), 33.1% indicated that they had some geries, hospitals and Place Survey respondents are
Long Term Illness and 16.0% stated that they did not shown in Figure 1. The road data was the Ordnance
own a car (henceforth ‘Non-Car Ownership’). The sam- Survey MasterMap Integrated Transport Network layer
pling frame for the Place Survey selected household provided via the EDINA data library http://edina.ac.uk/.
A GIS-based network analysis (ArcGIS 9.3) was used to
calculate the road distance from each Place Survey post-
code location respondent to the nearest GP surgery,
hospital and hospital with ED facilities. All of the statis-
tical analyses and mappings were performed in R ver-
sion 2.13.0, the open source statistical software http://
cran.r-project.org/.

Analysis
The Place Survey data were analysed using Generalized
Linear Models (GLMs), which predict the response coef-
ficients from a linear predictor generated from the inde-
pendent terms. A logistic GLM was used to analyse the
extent to which different variables predict difficulty in
access to GPs and Hospitals. The logit function is
defined by
exp(Q)
logit(Q) = (1)
1 + exp(Q)
The dependent variable was the survey response to the
appropriate access question. A value of 1 was given for a
Figure 1 The study area, Leicestershire UK, and the locations response of ‘difficult’ or ‘very difficult’, a value of 0 for
of the attitude survey respondent postcodes, GP surgeries,
any other response. The first independent variable tested
hospitals and hospitals with Emergency Departments (ED).
was whether the respondent had Long Term Illness. The
Comber et al. International Journal of Health Geographics 2011, 10:44 Page 4 of 11
http://www.ij-healthgeographics.com/content/10/1/44

Table 1 The summary of the Place Survey response rates.

Factors
Age Count Health Count Disability Count Gender Count Ethnicity Count
18 to 24 145 Very good 2377 Limiting 1913 Female 4816 White British 7949
25 to 44 1839 Good 3622 Non-Limiting 911 Male 3530 BME 416
45 to 64 3187 Fair 1958 None 5425
65 to 74 1561 Bad 333
75 to 84 1104 Very bad 60
85 + 348
Not provided 346 180 281 184 165
Totals 8530 8530 8530 8530 8365
BME is ‘black and minority ethnicity’.

GLM was then extended to include respondent state- pr(y2 = 1) = logit(b0 + b1 x1 + b2 x2 + b3a x3a ) Model 8 (9)
ments on their health status (Bad Health), car ownership
and their distance to the nearest facility as measured and
using a GIS-based network analysis. pr(y2 = 1) = logit(b0 + b1 x1 + b2 x2 + b3a x3a + b4 x4 ) Model 9 (10)
For ease of access to GPs, three models were consid-
ered: where x1, x2 and x4 are as above, y2 is a 0/1 indicator
variable showing whether the respondent stated that
pr(y1 = 1) = logit(b0 + b1 x1 ) Model 1 (2) they experienced difficulty in their access to hospitals,
x 3 is the distance from the nearest hospital to the
pr(y1 = 1) = logit(b0 + b1 x1 + b2 x2 ) Model 2 (3) respondent’s address, and x3a is the distance from the
nearest hospital with an Emergency Department to the
pr(y1 = 1) = logit(b0 + b1 x1 + b2 x2 + b3 x3 ) Model 3 (4) respondent’s address. As before, coefficients were esti-
mated, odds ratios computed and different models were
and compared using AICs.
pr(y1 = 1) = logit(b0 + b1 x1 + b2 x2 + b3 x3 + b4 x4 ) Model 4 (5)
Geographic Variation
where y 1 is a 0/1 indicator showing whether the The use of linear regression is common in many areas
respondent expressed difficulty in their access to GPs, of science. Ordinary linear regression implicitly assumes
x 1 is an 0/1 indicator variable showing whether the spatial stationarity of the regression model-that is, the
respondent said they had a Long Term Illness, x2 is an relationships between the variables remain constant over
indicator variable stating whether the respondent con- geographical space. It is self evident that global averages
sidered they were in Bad Health, x3 is the distance from of spatial data are not always helpful, whether they are
the respondent to their nearest GP surgery based on related to health, or other domains (e.g. unemployment
road network distance and x4 is a 0/1 indication of car or climate). Spatial non-stationarity occurs when a rela-
ownership. tionship (or pattern) that applies in one region does not
The quantity exp(b i) gives the odds ratio associated apply in another. Global models are statements about
with a unit increase in xi - that is the ratio between the processes or patterns which are assumed to be station-
odds of a y-value of 1 for xi and a y-value of 1 if xi is ary and as such are location independent-i.e. are
replaced by xi + 1. These values, together with 95% con- assumed to apply in all locations. In contrast, local mod-
fidence intervals, are given in Table 2. All three models els are spatial disaggregations of global models, the
were compared using Akaike’s Information Criterion results of which are location-specific. The template of
(AIC). the model is the same: the model is a linear regression
For access to hospitals, similar models were fitted, as model with certain variables, but the coefficients alter
below. geographically. The above is essentially a description of
Geographically Weighted Regression [36-38] (GWR).
pr(y2 = 1) = logit(b0 + b1 x1 ) Model 5 (6)
One of the fundamental tenets of geographical analyses
is to evaluate the potential existence of spatial variability
pr(y2 = 1) = logit(b0 + b1 x1 + b2 x2 ) Model 6 (7) of statistical models. GWR allows one to consider and
test for the possibility that relationships vary geographi-
pr(y2 = 1) = logit(b0 + b1 x1 + b2 x2 + b3 x3 ) Model 7 (8) cally. It is an approach that deals with spatial non-
Comber et al. International Journal of Health Geographics 2011, 10:44 Page 5 of 11
http://www.ij-healthgeographics.com/content/10/1/44

Table 2 Results of the GLM analyses of dissatisfaction over access to doctors/GP (Models 1 to 4) and hospitals (Models
5 to 9)-for each set of models, the best AIC value is highlighted in bold.
Analysis Model Variable Odds Lower Upper AIC
ratio 95% CI 95% CI
Access to GP surgeries Model 1 Long Term Illness 2.27 1.86 2.76 3257.9
Model 2 Long Term Illness 2.00 1.62 2.46 3242.9
Bad Health 2.10 1.49 2.90
Model 3 Long Term Illness 2.07 1.68 2.56 3181.1
Bad Health 2.10 1.49 2.92
Geographic Distance (to nearest GP) 1.29 1.22 1.36
Model 4 Long Term Illness 1.80 1.45 2.24 3049.5
Bad Health 1.69 1.18 2.37
Geographic Distance (to nearest GP) 1.34 1.27 1.42
Non-Car Ownership 3.81 3.06 4.72
Access to Hospitals Model 5 Long Term Illness 1.42 1.27 1.58 8549.2
Model 6 Long Term Illness 1.32 1.18 1.48 8535.5
Bad Health 1.61 1.28 2.02
Model 7 Long Term Illness 1.32 1.18 1.48 8537.5
Bad Health 1.61 1.28 2.02
Geographic Distance (to nearest Hospital)** 1.00 0.99 1.01
Model 8 Long Term Illness 1.32 1.18 1.48 8532.7
Bad Health 1.61 1.28 2.03
Geographic Distance (to nearest ED Hospital)* 0.991 0.982 0.999
Model 9 Long Term Illness 1.26 1.12 1.42 8488.4
Bad Health 1.50 1.19 1.89
Geographic Distance (to nearest ED Hospital)* 0.991 0.982 0.999
Non-Car Ownership 1.61 1.41 1.84
All variables significant at the 1% level except where indicated at 5% (*) and as not significant (**).

stationarity in multivariate regression by estimating Access to GPs

regression coefficients locally using spatially dependent Model 1 shows that Long Term Illness is a significant
weights, under the assumption that the effect of the pre- predictor of experiencing difficulty in access to GPs.
dictor variables on the dependent variable will vary con- The inclusion of additional terms (Models 2, 3, 4) each
tinuously over space. The logistic regressions of Model improved the model as shown by the decreasing AIC
4 and Model 9 were extended to a GWR analysis as fol- score. AIC is minus twice the maximized log-likelihood
lows: plus twice the number of parameters, as computed by
the AIC component of the family. For the binomial
pr(y2 = 1) = logit(b0(ui ,vi ) + b1 x1(ui ,vi ) + b2 x2(ui ,vi ) + b3 x3(ui ,vi ) + b4 x4(ui ,vi ) ) (11) family of models, the dispersion is fixed at one and the
number of parameters is the number of coefficients.
pr(y2 = 1) = logit(b0(ui ,vi ) + b1 x1(ui ,vi ) + b2 x2(ui ,vi ) + b3a x3a(ui ,vi ) + b4 x4(ui ,vi ) ) (12) The inclusion of health status (Model 2), distance to the
nearest GP surgery (Model 3) and Non-Car Ownership
with the coefficients for each of the predictor variables
(Model 4) significantly improved the model. Non-Car
assumed to vary across the two-dimensional geographi-
Ownership was significant at the 99% level and the AIC
cal space defined by the coordinates (u, v). Consequently
decreased between Models 3 and 4 by around 132
the coefficients in GWR can be considered as functions
points. The analysis of deviance tests between Model 1,
of these coordinates, rather than single-valued variables.
Model 2, Model 3 and Model 4 confirm the significance
of these variables (Table 3).
4. Results
Analysis of the exponentials of the coefficient esti-
The results of applying the GLM are shown in Table 2,
mates (Table 2) allows the odds ratios and confidence
with Models 1 to 4 relating to access to GPs and Mod-
intervals associated with different factors to be calcu-
els 5 to 9 relating to hospital access.
lated. The odds ratios calculated from Model 4
Comber et al. International Journal of Health Geographics 2011, 10:44 Page 6 of 11
http://www.ij-healthgeographics.com/content/10/1/44

Table 3 Analysis of Deviance of the terms associated with dissatisfaction over access to doctors, *** indicates
significance at the 0.1% level, * indicates significance at the 5% level.
Analysis Terms Df Residual Df Residual Deviance Deviance
Reduction
Access to GP surgeries NULL 8529 3318.6
Long term Illness 1 8528 3253.9 64.764***
Bad Health 1 8527 3236.9 16.975***
Distance to nearest GP surgery 1 8526 3173.1 63.813***
Non-Car Ownership 1 8525 3039.5 133.56***
Access to Hospitals NULL 8529 8583.8
Long term Illness 1 8528 8545.2 38.611***
Bad Health 1 8527 8529.5 15.655***
Distance to the nearest ED Hospital 1 8526 8524.7 4.826*
Non-Car Ownership 1 8525 8478.4 46.295***

coefficients suggest the following statements: The odds ratios associated with different factors and
models in relation to difficulty in accessing hospitals are
- For respondents with Long Term Illness the rela- shown in Table 2. The odds ratios calculated from the
tive odds of experiencing difficulty in access to GPs Model 9 coefficients suggest the following statements:
are around 1.8 times greater than for those who not
have Long Term Illness; - For respondents with Long Term Illness the rela-
- For respondents with Bad Health the relative odds tive odds of experiencing difficulty in access to hos-
of experiencing difficulty are around 1.7 times pitals are around 26% greater than for those who
greater than for those who not have Bad Health; not have Long Term Illness;
- The relative odds of experiencing difficulty in - For respondents with Bad Health the relative odds
access to GP surgeries increases by 34% per extra of experiencing difficulty are around 50% greater
km to the nearest GP surgery; than for those who not have Bad Health;
- Non-Car Ownership was found to have a profound - Whilst distance to hospitals was not found to be a
impact on GP access perceptions. For those who not good predictor of difficulty in hospital access, dis-
own a car, the relative odds of experiencing difficulty tance to hospitals with EDs was significant but nega-
over access to GPs are 3.8 times more than for those tive. The relative odds decreased slightly (1%) with
who do own cars. each extra km distance to the nearest ED hospital;
- The impact of Non-Car Ownership was again pro-
found: for those who do not own a car the relative
Access to hospitals odds of experiencing difficulty over access to hospi-
Model 5 shows Long Term Illness to be a significant tals are 61% greater than for those who do own cars.
predictor of experiencing difficulty in access to hospitals.
The model was extended to include the additional terms
of Bad Health, distance to the nearest hospital and to Geographic Variation
the nearest Emergency Department facility, as measured To complement the logistic regression above and to
using a GIS-based network analysis, and car ownership. examine the spatial variation in these relationships,
The inclusion of the health terms (Model 6) improved GWR was used to generate spatially explicit logistic
the model, but distances to the nearest hospital did not regression models. Table 4 summarises the results of
(Model 7). Distance to the nearest hospital with an ED the two GWR analyses (Equations 11 and 12) and
improved the model slightly (Model 6 to Model 8)-ED describes the variation of the odds ratios for the differ-
hospital distance was significant only at the 95% level- ent independent variables. The Inter-Quartile Range of
whilst Non-Car Ownership (Model 9) again significantly the odds ratios provides a good indicator of the spatial
improved the model. In this case the AIC decreased by variation. For Access to GPs, there was little spatial var-
around 44 points between Models 8 and 9. The results iation in Distance and Long Term Illness as predictors
of analysis of deviance tests between Model 5, Model 6, of access difficulty, whilst Bad Health showed some var-
Model 8 and Model 9 confirm the significance of these iation, with the relative odds of experiencing difficulty in
variables (Table 3). access to this service ranging from 69% to 81% greater
Comber et al. International Journal of Health Geographics 2011, 10:44 Page 7 of 11
http://www.ij-healthgeographics.com/content/10/1/44

Table 4 The variation in the odds ratios of the independent variables from the GWR models of access dissatisfaction,
with the Inter-Quartile Range (IQR) providing a measure of the spatial variation the relationships.
GWR model Variable Minimum 1st Quartile Median 3rd Quartile Maximum Global IQR
GPs Distance to nearest GP 1.32 1.33 1.34 1.37 1.41 1.34 0.04
Bad Health 1.61 1.63 1.68 1.77 1.90 1.69 0.14
Long Term Illness 1.69 1.77 1.80 1.81 1.82 1.80 0.04
Non-Car Ownership 3.41 3.58 3.64 3.94 4.26 3.81 0.36
ED Hospital Distance to nearest Hospital 0.798 0.981 0.987 0.992 1.240 0.991 0.011
Bad Health 0.553 1.35 1.55 1.64 3.92 1.50 0.29
Long Term Illness 0.800 1.21 1.24 1.27 1.99 1.26 0.06
Non-Car Ownership 1.08 1.47 1.55 1.73 3.50 1.61 0.26

than for who do not have Bad Health. There was more Illness, Bad Health and Non-Car Ownership. Geo-
variation in the effects of Non-Car Ownership, which graphic distance was a significant predictor of perceived
ranged from 3.58 to 3.94 times greater than for those difficulty in access to GP surgeries but not for hospitals
with cars, although the 25th percentile is close to the with or without EDs. A GWR analysis identified consid-
median, indicating a positive skew in the distribution of erable geographic variation in the relationships between
the variation. For access to ED hospitals the relative perceived difficulty in access to GPs and hospitals with
odds of experiencing difficulty with Bad Health ranged Bad Health and Non-Car Ownership but not with Long
from 35% to 64% greater than for those without Bad Term Illness or geographic distance. For instance, diffi-
Health. The effects of Non-Car Ownership were greater culty in accessing GP surgeries in relation to Bad Health
but with similar spatial variation, and the relative odds was greater in the West and South West of the study
of experiencing difficulty ranged from 47% to 73% area. Whilst difficulty in accessing hospitals in relation
greater than for those who owned a car. to Bad Health was greater in the South West and North
The spatial distribution of the relationships between East.
access difficulty and the predictor variables showing These results reflect the variation in perceived ease of
high variation can be mapped. Figure 2 shows the spa- access to services within and between different groups
tial variation in the predictive strength of Bad Health defined on health status, socio-economic attributes, dis-
and Non-Car Ownership on perceptions of access to tance etc and that different factors are correlated with
GPs and Hospitals with ED. Table 3 shows that the access difficulties, depending on the service. They sug-
other variables, whilst significant, did not vary spatially- gests the following statements for this study area:
i.e. the global model for these variables can be assumed
to be unaffected by spatial non-stationarity. For access 1) Distance is a significant factor in perceived diffi-
to GPs there is a clear trend of increasing perceived dif- culties in access to GPs but with little local variation.
ficulty in access for those with Bad Health and who do The notion of GP accessibility is strongly related to
not own cars running from the North East to the South geographic distance.
West. For access to Hospitals, the spatial variations in 2) Distance is not a factor in perceptions of hospital
the relationship with Bad Health on and Non-Car Own- accessibility.
ership are not so even. There is much more short-range 3) Long Term Illness and Bad Health are significant
variation in the trends and clusters are evident in differ- predictors of perceived difficulties in accessing GPs
ent parts of the study area, in contrasts to the general and hospitals, indicating that for people in these
trend observed in GP access difficulty. The impact of groups, the notion of accessibility is also related to
Bad Health is greatest in a band running to the South their health status.
and East of the study area and lowest in the North 4) Non-Car Ownership was found to be a significant
West and South East. The impact of Non-Car Owner- predictor of perceived difficulties in accessing GPs
ship is greatest in the North West and least in a band and hospitals indicating that the notion of accessibil-
running from the South and East. ity is also related to the choices afforded by socio-
economic status.
5. Discussion 5) Additionally, the impacts of health status (Bad
In this study area, perceptions of difficulty in access to Health) and socio-economic status (Non-Car Own-
different types of health services (hospitals and GPs) ership) on the perceived difficulties in access services
was found to be significantly related to Long Term were found to vary spatially, suggesting that other
Comber et al. International Journal of Health Geographics 2011, 10:44 Page 8 of 11
http://www.ij-healthgeographics.com/content/10/1/44

GPs

ED Hospitals

Figure 2 Spatial variation in the relationships between perceived difficulty in access to hospitals and GPs with Bad Health and Non-
Car Ownership.

local factors may also be contributing to perceptions economic and health barriers to services as well as geo-
and notions of accessibility. graphic factors [41,42]. The perception of access to any
given facility or service will also be related to other fac-
The results highlight some important issues related to tors, and the choices made by individuals over which
service accessibility for consideration in health planning services to access (if they exist) will reflect these: cost,
research. These relate to the choices available to and previous experience, reputation (first and second hand),
made by individuals over the services they use: percep- perceived quality of service, convenience etc. The extent
tions of access will be influenced by their personal to which we have active choices in the services we
(health) circumstances and experiences. Whilst distance access will also vary depending on the service in ques-
has been found to an important and significant determi- tion: there may be little choice over the hospital ED we
nant in health outcomes [39,40], patient access to and use, a bit more choice over which specialist hospital
use of services will depend on a number of interacting clinic we are (or choose to be) referred to and yet
factors, including socio-economic ones [41] and health further choice over which GP we visit. These levels of
status. For these reasons, other research has sought to choice are reflected in the results of this work: distance
combine a range of different measures in order to gen- was significant factor in respondent perceptions over
erate access indices that incorporate potential socio- their access to GPs and not to hospitals. For these
Comber et al. International Journal of Health Geographics 2011, 10:44 Page 9 of 11
http://www.ij-healthgeographics.com/content/10/1/44

reasons, Place Survey respondents with ongoing health do provide an indication of the exclusion experienced
problems may be more likely to be concerned about by a robust sample of the population in the study area.
access to their regular outpatient or inpatient treatment Ongoing work will seek to unpick the underlying causes
centres than to the emergency department. The variable of the negative perceptions of access. The data used in
impact of distance over perceptions of access highlights the study did not capture any information about use
an important point: the concepts of facility ‘access’ and and access to private facilities, such as are available
‘accessibility’ involve much more than just geographic or under personal health insurance schemes.
spatial access [32,33]. Much spatial planning in public This analysis did not account for age and gender for a
health and other domains is predicated on the assump- number of reasons. First, the main messages from this
tion that geographic distance is important per se regard- research are that local statistical techniques can add
less of the nature of the facility, whereas this research considerable value to accessibility (and other) analyses
has shown that this may not be the case. Notions of by identifying spatial variation in relationships, and that
‘Access’ or ‘Accessibility’ in a health planning context access perceptions are driven by different factors (dis-
has been shown to involve considerations of: tance, health status, etc) depending on the facility under
Financial access relating to measures that describe the consideration. Second, whilst in epidemiological studies
financial ability of people to access health care through age and gender are included to adjust for the ‘popula-
health insurance and other cost-related barriers [43,44]; tion at risk’, with certain kinds of illness more likely to
Behavioural access, describing actual utilization of dif- occur in age groups than in others, this work does not
ferent health services such as visits, prescription uptake, assess the relative occurrences of diseases over geogra-
ambulatory calls [45-47]; phical space but changes in perceptions of accessibility.
Spatial access relating to geographic distance, trans- From a policy viewpoint it is useful to consider this for
port and travel times and describing service catchments, populations as a whole, even if the composition of such
optimal spatial arrangement of resources [1,5,8,30,48]. populations varies geographically. Third, a larger sample
Whilst some public health studies incorporate multi- would be required to calibrate the GWR model reliably
ple definitions of access [32,33,47], in general the litera- to analyse population subsets and to allow for the geo-
ture describes access as relating to the cost barriers graphical factor. Therefore a pooled analysis was carried
associated with health [35]. For example, guidelines pro- out because of the geographical detail required and the
duced by the National Academy of Sciences include danger of small sample problems. It may be appropriate
access as a quality criterion and as an objective of health at a later stage to consider the factors of age and gender
practice but adopt a cost based concept of access related in the light of the findings outlined here.
to insurance [35]. In other research domains notions of
access relate to social justice, social inclusion, environ- 6. Conclusions
mental justice, public participation and public engage- This study demonstrates that the notion of access is a
ment. For example, in the health GIS literature access is multi-dimensional concept, whose composition varies
described in relation to geographic distance with location, according to the facility being considered
[1,5,15,22-25,48], whilst in the social sciences it is and the health and socio-economic status of the indivi-
related to access perceptions [34] and notions of social dual concerned. Some conclusions from this study (and
capital [49-51]. This study has highlighted the need for this study area) can be drawn:
research on access to public health facilities to accom- • For some types of health facilities geographic dis-
modate the different dimensions of access, that relate to tance is a significant predictor of experiencing access
geography, behaviours and perceptions as well as finan- difficulty (GPs), whilst for others it is not (hospitals);
cial and cultural barriers. • Those with Long Term Illness and Bad Health status
Some limitations to this study should be noted. The are much more likely to experience difficulty in their
hospital data was downloaded from the NHS website to access to health facilities;
include ‘major’ NHS Hospitals. However, survey respon- • Non Car Ownership was found to be significantly
dents were simply asked about their perceptions of related to access difficulty;
access to ‘hospitals’ which, depending on their personal • Some of these relationships vary spatially indicating
experiences may include children’s hospitals and long the need for accessibility analyses to include spatially
stay psychiatric facilities. The attitudes survey captured nuanced statistical methods that accommodate local var-
the degree of difficulty experienced in accessing services iations, such as are afforded by GWR.
but not the underlying reasons for that difficulty. Simi- • Identifying the spatial variations in relationships, by
larly, the analysis uses geographic distance to the nearest estimating local regression parameters, allows the spatial
facility, which may or may not be the facility actually distribution and interaction of predictor variables to be
used by the survey respondents. However, the responses explored. Analysing the local variation in relationships
Comber et al. International Journal of Health Geographics 2011, 10:44 Page 10 of 11
http://www.ij-healthgeographics.com/content/10/1/44

provides those concerned with public health policy with 8. Shen YC, Hsia RY: Changes in Emergency Department Access Between
2001 and 2005 Among General and Vulnerable Populations. Am J Public
the ability to target resources and to achieve improved Health 2010, 100:1462-1469.
outcomes through location-specific activities. 9. Nandi A, Galea S, Lopez G, Nandi V, Strongarone S, Ompad DC: Access to
• Future analyses of facility access and accessibility and Use of Health Services Among Undocumented Mexican Immigrants
in a US Urban Area. Am J Public Health 2008, 98:2011-2020.
should seek to include the different dimensions related 10. Yu SM, Huang ZJ, Kogan MD: State-Level Health Care Access and Use
to service access including: public perceptions, beha- Among Children in US Immigrant Families. Am J Public Health 2008,
viours, geographical access and service quality. These 98:1996-2003.
11. Kon ZR, Lackan N: Ethnic Disparities in Access to Care in Post-Apartheid
were found to provide a more comprehensive analysis of South Africa. Am J Public Health 2008, 98:2272-2277.
health service access when considered together. 12. Huang ZJ, Yu SM, Ledsky R: Health Status and Health Service Access and
Use Among Children in U.S. Immigrant Families. Am J Public Health 2006,
96:634-640.
Endnotes 13. Lasser KE, Himmelstein DU, Woolhandler S: Access to Care, Health Status,
1
http://www.lsr-online.org/leicestershire-place-survey- and Health Disparities in the United States and Canada: Results of a
2008.html Cross-National Population-Based Survey. Am J Public Health 2006,
2 96:1300-1307.
http://www.lsr-online.org/placesurvey.html 14. Fone DL, Christie S, Lester N: Comparison of perceived and modelled
geographical access to accident and emergency departments: a cross-
sectional analysis from the Caerphilly Health and Social Needs Study. Int
Acknowledgements J Health Geogr 2006, 5:16.
The authors would like to thank the Policy, Performance and Research 15. Patel AB, Waters NM, Ghali WA: Determining geographic areas and
Group at Leicestershire County Council, for use of the (anonymised) Place populations with timely access to cardiac catheterization facilities for
Survey data. The comments of the anonymous reviewers greatly improved acute myocardial infarction care in Alberta, Canada. Int J Health Geogr
the original draft. 2007, 6:47.
16. Rosero-Bixby L: Spatial access to health care in Costa Rica and its equity:
Author details a GIS-based study. Social Science & Medicine 2004, 58:1271-1284.
1
Department of Geography, University of Leicester, Leicester, LE1 7RH, UK. 17. Ghirardelli A, Quinn V, Foerster SB: Using Geographic Information Systems
2
Department of Geography, University of Liverpool, Liverpool, L69 7ZT, UK. and Local Food Store Data in California’s Low-Income Neighborhoods to
3
Research and Insight Team, Leicestershire County Council, Leicester, LE3 Inform Community Initiatives and Resources. Am J Public Health 2010,
8RA, UK. 100(11):2156-2162.
18. Casas I, Delmelle E, Varela A: A Space-Time Approach to Diffusion of
Authors’ contributions Health Service Provision Information. Int Regional Sci Rev 2009,
AC conceived designed and implemented the analysis including exploratory 33(2):134-156.
data analysis and visualization of data and data processing. RR developed 19. Alcaraz KI, Kreuter MW, Bryan RP: Use of GIS to identify optimal settings
the accessibility components to the questionnaire and initiated this for cancer prevention and control in African American communities. Prev
investigation. AC led the writing. CB supervised the statistical analysis. All of Med 2009, 49:54-57.
the authors have read and approved the final manuscript and contributed 20. Al-Taiar A, Clark A, Longenecker JC, Whitty CJM: Physical accessibility and
to the writing. utilization of health services in Yemen. Int J Health Geogr 2010, 9-38.
21. Messina JP, Shortridge AM, Groop RE, Varnakovida P, Finn MJ: Evaluating
Competing interests Michigan’s community hospital access: spatial methods for decision
The authors declare that they have no competing interests. support. Int J Health Geogr 2006, 5:42.
22. McLafferty SL: GIS and health care. Annu Rev Public Health 2003, 24:25-42.
Received: 7 May 2011 Accepted: 25 July 2011 Published: 25 July 2011 23. Higgs G: The role of GIS for health utilization studies: literature review.
Health Serv Outcomes Res Method 2009, 9:84-99.
References 24. Guagliardo MF: Spatial accessibility of primary care: concepts, methods
1. Sasaki S, Comber AJ, Suzuki H, Brunsdon C: Using genetic algorithms to and challenges. Int J Health Geogr 2004, 3:3.
optimise current and future health planning-the example of ambulance 25. Boulos NMK: Towards evidence-based, GIS-driven national spatial health
locations. Int J Health Geogr 2010, 9:4. information infrastructure and surveillance services in the United
2. Walsh SJ, Page PH, Gesler WM: Normative Models and Healthcare Kingdom. Int J Health Geogr 2004, 3:1.
Planning: Network-Based Simulations Within a Geographic Information 26. Gibson J, Deng X, Boe-Gibson G, Rozelle S, Huang J: Which households are
System Environment. Health Services Research 1997, 32:243-260. most distant from health centers in rural China? Evidence from a GIS
3. Patel AB, Waters NM, Ghali WA: Determining geographic areas and network analysis. GeoJournal 2010, 1-11.
populations with timely access to cardiac catheterization facilities for acute 27. Apparicio P, Abdelmajid M, Riva M, Shearmur R: Comparing alternative
myocardial infarction care in Alberta, Canada. Int J Health Geogr 2007, 6:47. approaches to measuring the geographical accessibility of urban health
4. Parker EB, Campbell JL: Measuring access to primary medical care: some services: Distance types and aggregation-error issues. Int J Health Geogr
examples of the use of geographical information systems. Health and 2008, 7:7.
Place 1998, 4:83-193. 28. Beyer KMM, Saftlas AF, Wallis AB, Peek-Asa C, Rushton G: A probabilistic
5. Schuurman N, Fiedler RS, Grzybowski SCW, Grund D: Defining rational sampling method (PSM) for estimating geographic distance to health
hospital catchments for non-urban areas based on travel-time. Int J services when only the region of residence is known. Int J Health Geogr
Health Geogr 2006, 5:43. 2011, 10:4.
6. Mclafferty S, Grady S: Immigration and Geographic Access to Prenatal 29. Bagheri N, Holt A, Benwell GL: Using Geographically Weighted Regression
Clinics in Brooklyn, NY: A Geographic Information Systems Analysis. Am to Validate Approaches for Modelling Accessibility to Primary Health
J Public Health 2005, 95:638-640. Care. Appl Spatial Analysis 2009, 2:177-194.
7. Hossain MM, Laditka JN: Using hospitalization for ambulatory care 30. Sasaki S, Igarashi K, Fujino Y, Comber AJ, Brunsdon C, Muleya CM, Suzuki H:
sensitive conditions to measure access to primary health care: an The impact of community-based outreach immunization services on
application of spatial structural equation modeling. Int J Health Geogr immunization coverage with GIS network accessibility analysis in peri-
2009, 8:51. urban areas, Zambia. J Epidemiol Commun H .
 Comber et al. International Journal of Health Geographics 2011, 10:44 Page 11 of 11
http://www.ij-healthgeographics.com/content/10/1/44

31. Ray N, Ebener S: AccessMod 3.0: computing geographic coverage and

accessibility to health care services using anisotropic movement of
patients. Int J Health Geogr 2008, 7:63.
32. Boulos NMK: Geographic information systems and the spiritual
dimension of health: a short position paper. Int J Health Geogr 2003, 2:6.
33. Maroko AR, Maantay JA, Sohler NL, Grady KL, Arno PS: The complexities of
measuring access to parks and physical activity sites in New York City: a
quantitative and qualitative approach. Int J Health Geogr 2009, 8:34.
34. Farrington J, Farrington C: Rural accessibility, social inclusion and social
justice: towards conceptualisation. J Transp Geog 2005, 13:1-12.
35. Rutherford ME, Mulholland K, Hill PC: How access to health care relates to
under-five mortality in sub-Saharan Africa: systematic review. Trop Med
Int Health 2010, 15(5):508-519.
36. Fotheringham AS, Brunsdon C, Charlton ME: Geographically Weighted
Regression: The Analysis of Spatially Varying Relationships Chichester: Wiley;
2002.
37. Brunsdon CF, Fotheringham AS, Charlton M: Geographically Weighted
Regression-A Method for Exploring Spatial Non-Stationarity. Geogr Anal
1996, 28:281-298.
38. Nakaya T, Fotheringham AS, Brunsdon C, Charlton M: Geographically
weighted Poisson regression for disease association mapping. Stat Med
2005, 24:2695-2717.
39. Maheswaran R, Pearson T, Jordan H, Black D: Socio-economic deprivation,
travel distance, location of service, and uptake of breast cancer
screening in North Derbyshire, UK. J Epidemiol Commun Health 2006,
60:208-212.
40. Bin Huang MS, Dignan M, Han D, Johnson O: Does Distance Matter?
Distance to Mammography Facilities and Stage at Diagnosis of Breast
Cancer in Kentucky. J Rural Health 2009, 266-271.
41. Wang F, Luo W: Assessing spatial and nonspatial factors for healthcare
access: towards an integrated approach to defining health professional
shortage areas. Health and Place 2005, 11:131-146.
42. McGrail MR, Humphreys JS: The index of rural access: an innovative
integrated approach for measuring primary care access. BMC Health
Services Research 2009, 9:124.
43. Quinn SC, Kumar K, Freimuth V, Musa D, Casteneda-Angarita N, Kidwell K:
Racial Disparities in Exposure, Susceptibility, and Access to Health Care
in the US H1N1 Influenza Pandemic. Am J Public Health 2011,
101(2):285-293.
44. McCarrier K, Zimmerman FJ, Ralston JD, Martin DP: Associations Between
Minimum Wage Policy and Access to Health Care: Evidence From the
Behavioral Risk Factor Surveillance System, 1996-2007. Am J Public Health
2011, 101(2):359-367.
45. Gulley SP, Rasch EK, Chan L: Ongoing Coverage for Ongoing Care: Access,
Utilization, and Out-of-Pocket Spending Among Uninsured Working-
Aged Adults with Chronic Health Care Needs. Am J Public Health 2011,
101(2):368-375.
46. Soleimanpour S, Geierstanger SP, Kaller S, McCarter V, Brindis CD: The Role
of School Health Centers in Health Care Access and Client Outcomes.
Am J Public Health 2010, 100(9):1597-1603.
47. Yu SM, Huang ZJ, Singh GK: Health Status and Health Services Access
and Utilization Among Chinese, Filipino, Japanese, Korean, South Asian,
and Vietnamese Children in California. Am J Public Health 2011,
100(5):823-830.
48. Comber AJ, Sasaki S, Suzuki H, Brunsdon C: A modified grouping genetic
algorithm to select ambulance site locations. Int J Geogr Info Sci 2011,
25(5):807-823.
49. Gray D, Shaw J, Farrington J: Community transport, social capital and
social exclusion in rural areas. Area 2006, 38(1):89-98. Submit your next manuscript to BioMed Central
50. Urry J: Mobility and proximity. Sociology 2002, 36:255-74. and take full advantage of:
51. Mohan G, Mohan J: Placing social capital. Prog Hum Geog 2002,
26(2):91-210.
• Convenient online submission
doi:10.1186/1476-072X-10-44 • Thorough peer review
Cite this article as: Comber et al.: A spatial analysis of variations in
health access: linking geography, socio-economic status and access • No space constraints or color figure charges
perceptions. International Journal of Health Geographics 2011 10:44. • Immediate publication on acceptance
• Inclusion in PubMed, CAS, Scopus and Google Scholar
• Research which is freely available for redistribution

Submit your manuscript at

www.biomedcentral.com/submit

View publication stats