Date post: | 19-Sep-2016 |
Category: |
Documents |
Upload: | mohammad-ali |
View: | 212 times |
Download: | 0 times |
Health & Place 8 (2002) 85–92
Spatial filtering using a raster geographic information system:methods for scaling health and environmental data
Mohammad Alia,c,*, Michael Emchb, Jean-Paul Donnayc
a ICDDR,B, Mohakhali, Dhaka, BangladeshbDepartment of Geography, Portland State University, Portland, OR, USA
cDepartment of Geomatics, University of Liege, Belgium
Accepted 3 October 2001
Abstract
Despite the use of geographic information systems (GIS) in academic research, it is still uncommon for public health
officials to use such tools for addressing health and environmental issues. Complexities in methodological issues
for addressing relationships between health and environment, investigating spatial variation of disease, and addressing
spatial demand and supply of health care service, hinder the use of GIS in the health sector. This paper demonstrates
simple spatial filtering methods for analyzing health and environmental data using a raster GIS. Computing
spatial moving average rates reduces individual affects and creates a continuous surface of phenomena. Another
spatial analytical method discussed is computation of exposure status surfaces including neighbors’ influences
weighted by distance decay. These methods describe how health and environmental data can be scaled in order to better
address health problems. Spatial filtering methods are demonstrated using health and population surveillance data
within a GIS that were collected for approximately 210,000 people in Matlab, Bangladesh. r 2002 Published by
Elsevier Science Ltd.
Keywords: GIS; Health; Environment; Spatial analysis
Introduction
An important component of understanding disease is
the interaction of humans with their environment.
Human–environment interaction can help inform the
emergence, resurgence, and distribution of infectious
diseases. Environmental phenomena vary in space and
therefore when studying environmental determinates of
health one should consider the spatial variation of these
phenomena (Briggs and Elliott, 1995). Health is affected
by a variety of lifestyle and environmental factors
including where people live (Scholten and de Lepper,
1991). Characteristics of where people live, including
socio-demographic and environmental exposure quali-
ties, offer valuable insight into epidemiological investi-
gations. Because health and environmental issues are
complex, it is essential to choose appropriate methods to
adequately address these types of problems. Geographic
information systems (GIS) are important tools for
addressing human–environment health problems be-
cause diverse data sets can be analyzed and related to
one another. Because of this capability, GIS has
reinforced and strengthened an old theme in the medical
sciencesFit is important to know where a disease
occurs. A GIS provides the utility to map health
indicators and disease incidence, analyze spatial patterns
of disease, and investigate health delivery systems
(Garner et al., 1993). A GIS can be used to describe
the distributions of social and environmental factors
that affect human health (Loslier, 1996).
*Corresponding author. International Vaccine Institute,
Kwanak P.O. Box 14, Seoul, 151-600, South Korea.
E-mail address: [email protected] (M. Ali).
1353-8292/02/$ - see front matter r 2002 Published by Elsevier Science Ltd.
PII: S 1 3 5 3 - 8 2 9 2 ( 0 1 ) 0 0 0 2 9 - 6
Despite the use of GIS in academic research, it is
uncommon for public health officials to use a GIS to
analyze health data (Rushton, 2000). Defining the
relationship between health and environment is complex
because it is difficult to choose a suitable spatial scale to
represent the spatial variation of disease, health, and
health seeking behavior (Glenn and Burkett, 1999;
Elman and Myers, 1999; Egunjobi, 1993; Matos et al.,
1990: Pringle, 1986; Cook et al., 1999; Tursz and Crost,
1999; Kirscht et al., 1976). Various methods have been
developed for addressing health and environmental
problems at different spatial scales (Gatrell et al.,
1996; Bailey and Gatrell, 1995; Haining, 1990; Cliff
and Ord, 1981). However, the use of GIS in public
health has been hampered by a variety of conceptual and
methodological problems (Onsrud and Pinto, 1991). In
order for public health officials to benefit from such
tools, the methodology needs to be made more
accessible to users (Rushton, 2000). Also, quick, reliable,
and scientifically valid methods of rapid assessment are
needed to assist in health research (Scholten and de
Lepper, 1991). Raster GIS, which divides space into
discrete units called cells, can be a simple and useful
tool for analyzing spatially referenced data. Raster
GIS is efficient for managing and integrating diverse
data sets including satellite image data. It can be used to
scale data by means of spatial filtering methods
commonly used to enhance satellite imagery. Spatial
filtering can be used to create smoothed maps of health
data (Talbot et al., 2000; Rushton and Lolonis, 1996;
Kafadar, 1996).
There are several reasons to use spatial filters on data
in health-related research. Filters can remove random
noise caused by inaccurate records or mislocated cases.
Filters can also be used to address influences of
neighbors on disease processes. For instance, behavioral
practices of a household may be influenced by neighbors
(Twigg et al., 2000; Munshi, 1996), which can affect
human health. People are exposed to the affects of poor
sanitation in their neighborhoods, not just within the
household in which they live. Addressing these issues by
using simple raster spatial filtering techniques can make
a valuable contribution to future health research efforts.
This paper demonstrates the use of scaling health and
environment data by using several spatial filtering
techniques.
Spatial filtering
Spatial filtering is commonly used to enhance satellite
imagery for visual interpretation. It involves applying a
mathematical formula such as a mean or median to a
group of pixels in a raster image using a moving
window. For example, the mean of 9 pixels in a 3� 3
moving window would be calculated and the window
would then moved one pixel over and the mean would
be calculated for the next 3� 3 area in the image. The
moving window would be applied to all 3� 3 areas of
the image until the mean was calculated for the entire
raster data set. Spatial filtering techniques smooth data
and therefore average errors inherent in the data
(Kafadar, 1999). Field survey data gathering systems
usually generate some errors. Also, there are many
intervening factors at the individual scale that may
influence normal processes of environment and disease
phenomena. For instance, economic conditions or
educational status influences the prevalence of diseases
in different households. In practice, neighbors would
have similar incidence rates for environmentally related
diseases unless the processes are affected by socio-
economic conditions of the households or some other
intervening variable.
Variation in the size of population among households
may also influence health events. While individual level
data are needed when the epidemiology of the disease is
concerned with socioeconomic conditions or human
biological aspects, investigating relationship between
disease incidence and environmental factors may fail to
detect critical risk factors of the disease when conducting
individual level analyses (Haining, 1998). In these
situations, spatial filtering techniques can be used to
remove these individual effects for both environmentally
related diseases and environmental variables if the data
are obtained through a household survey. Various
filtering methods have been used so that data collected
from a field survey can be scaled in order to remove the
noise in the data used (Meijerink et al., 1994; Watkins
et al., 1993). This paper proposes the use of a low pass
filter (sum of kernel values is greater than 1) as a method
for integrating and addressing geographical scale issues
in health and environment data.
Filter size and type
The size of the moving window (neighborhood) within
the raster cell array must be defined as part of the spatial
filtering process. The amount of smoothing in the data
depends on the size of the filter. With a larger filter, local
level characteristics will be obscured and a smaller filter
will retain more local characteristics. If the filter is too
large, it may remove important small-scale variation
that could point to etiological associations, while if it is
too small it may not remove the random noise in the
data (Talbot et al., 2000). Defining an appropriate filter
size is, thus, critical. Several different scales of filters can
also be used on a particular data set. Another important
issue is the determination of spatial filter kernel values.
The kernel values determine the type of operation to be
performed on the data for a variety of data scaling
purposes.
M. Ali et al. / Health & Place 8 (2002) 85–9286
Applications of spatial filtering
Empirical study area
The ICDDR,B’s (Centre for Health and Population
Research) research site in Bangladesh, widely known as
Matlab, is used as a study area to demonstrate the
spatial filtering methods. Matlab, located in south
central Bangladesh, has a long and rich history of
community-based health and population research. A
vector spatial database of the study area (184 km2),
based on the geographic coordinates of households and
physical features, was created to support health and
population research efforts. Fig. 1 shows the study area
and components in the spatial database. The spatial
database includes points that represent the location of
baris (groups of patrilineally related households).
Approximately 210,000 people live in Matlab within
Fig. 1.
M. Ali et al. / Health & Place 8 (2002) 85–92 87
7691 baris. A longitudinal health and population
database has been integrated with the spatial database
so that health and population distributions can be
described by bari location. This data set is used to
demonstrate the usefulness of spatial filtering techniques
in scaling health and environment data.
Raster implementation of the study data
In order to demonstrate the benefits of spatial
filtering, a common raster GIS software package called
Idrisi (Idrisi, 1997) was used. Since the Matlab GIS
database is maintained in a vector format, it was
rasterized. We chose a raster system because it provides
simple solutions to the complex issues that are faced in
scaling health and environment data. Within the raster
system, the spatial resolution of the pixels was set to
30m to represent a bari by a single pixel. Each of the
7691 pixels representing baris in the raster GIS was
assigned a unique bari identification number so that
health and population attribute data could be linked to
the spatial database. Since a raster data set is a
continuous array of pixels, those that did not represent
baris were assigned a value of ‘‘zero.’’ The Matlab
surveillance databases are maintained at the individual
level and therefore these data sets were aggregated by
bari pixels for raster processing. Each data set including
population, sanitation conditions, and educational
status was represented by a separate raster layer.
The spatial distribution of baris was investigated by
filtering the database using several different filter sizes.
An average of two baris was present when using a 3� 3-
pixel filter. However, averaging values with only two
baris does not remove the roughness in the data. With a
5� 5-pixel filter, an average of five baris was present and
with a 7� 7-pixels filter, the average number of baris
was seven. An average of ten baris was observed when
choosing a 9� 9-pixel filter. A 7� 7-pixel filter was
found suitable for the study area as it provides an
optimum size of population for removing random noise
while retaining enough local level variation in the data.
However, filtering based on the fixed geographic size
may be a reasonable approach when the population is
homogeneously distributed in space. When there is a
heterogeneous spatial distribution of population, as in
the case of this study area, a spatial filter based on a
fixed population size is more appropriate than methods
that use a fixed geographical size. This requires a locally
adaptive filter size based on the distribution of popula-
tion.
Computing edge correction terms
Data scaling by means of spatial filtering is always
affected by boundaries, because there are no data
beyond the boundary. An adjustment is therefore
proposed for minimizing the effects of boundaries in
the data. Spatial filtering offers a simple solution by
computing edge correction terms for all pixels. The
correction term is defined as the proportion of cells that
fall inside the boundary of the study area for the
particular filter size. It is expressed as
ei ¼1
n
Xn
j¼1
Vj�kj ;
where ei is the edge correction term for pixel i; vj=1 if
cell j is in the study area otherwise the value is 0, kj the
kernel value of cell j for the filter, n the number of cells
in the kernel filter.
A unitary kernel filter was used in which all the values
within the filter window add up to 1. The sum of the
product of kernel values and the attribute values of
the image yield the number of pixels that fall inside the
study area. In Fig. 2, the selected pixel, in the center of
the image, is affected by the boundary when data are
scaled using a 7� 7-pixel window, because only 35 pixels
out of 49 are in the study area. Therefore, the edge
correction term of the target pixel is 0.7143 (35C49).
Fig. 3 is an edge correction image of all pixels of the
study area calculated for the 7� 7-pixels filter. In the
image, different edge correction terms are displayed in
different gray tones. The pixels near the boundary lines
have different gray tones indicting different edge
correction terms. In contrast, the pixels in the central
part of the image are a single gray tone indicating that
the boundary lines do not affect those pixels. Computing
edge correction terms using this method is relatively
simple and there are no data lost near the perimeter of
the original study area (Gatrell et al., 1996).
Computing spatial moving average rate
Standardized mortality ratios (SMRs) are often used
to map health data such as mortality rates by
geopolitical or census boundaries. These maps can be
Attributes of the study area (1=inside, 0=outside)
1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1
1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1
0 0 0 0 1 1 10 0 0 1 1 1 10 0 1 1 1 1 1
0 0 1 1 1 1 1
0 1 1 1 1 1 10 1 1 1 1 1 10 1 1 1 1 1 1
Mov
ing
Win
dow
(u
nita
ry k
erne
l val
ues )
Fig. 2. An example of image attributes used to complete edge
correction terms.
M. Ali et al. / Health & Place 8 (2002) 85–9288
useful in making etiological hypotheses about diseases
(Clayton et al., 1993; Cislaghi et al., 1995). Alternatively,
we can use spatial moving average rates (SMAR) in
representing the health status of an area. The SMAR
gives a spatially smoothed rate (Rushton, 1998; Bithell,
1990; Bailey and Gatrell, 1995), which can remove the
influence of spatial inhomogeneity of population struc-
ture in the data. By using a spatial filter, the SMAR at
each point can be computed by superimposing a filter
over the point of interest. For the area surrounding the
point, the number of cases and population at risk are
summed and the ratio yields the SMAR at that point.
The mathematical expression for computing spatial
moving average rate in a raster system is given as
mi ¼
Pnj¼1 Cj�kjPnj¼1 Pj�kj
�1000;
where mi is the spatial moving average rate for pixel i; cj
the number of cases in pixel j; pj the number of people in
pixel j; kj the kernel values of cell j of the filter, n the
number of cells in the kernel filter.
The edge correction term is not necessary because the
ratio is expressed per 1000 people. In Fig. 4, the SMAR
of the target pixel is 213.52 ((60C281)� 1000) per 1000
people when scaling the data with a 7� 7-pixel filter.
Using the conventional method, the rate would have
been 96.77 ((3C31)� 1000) per 1000 people, which is
less than the rate obtained from the moving average
method. The higher number of cases among the
neighbors increases the spatially scaled rates in this
case. The SMAR would have been lower if the neighbors
had a smaller number of cases. Thus, the SMAR gives us
a spatially smoothed measure, which is free from
dependency at individual locations.
Capturing neighbors’ influences
Hazardous elements of a source of pollution normally
diffuse as a distance decay function, (i.e., pollution
decreases as distance increases) (Briggs and Elliott,
1995). Friction caused by physical features influences the
diffusion process by accelerating, impeding, or changing
the diffusion direction. Similar to the diffusion process
of hazardous elements, adoption of an innovation is also
assumed to be higher in nearby places because peoples’
behavioral practices are influenced more by their near
neighbors than that by the distant neighbors (Munshi,
1996). Social relationships with neighbors may add
friction to the diffusion process. For instance, disputes
among neighbors may impede the diffusion process of
an innovation and in contrast a friendly relationship
with neighbors may accelerate the process of the
innovation. This indicates that the media through which
phenomena move is often spatially heterogeneous. A
raster GIS can facilitate the spatial modeling of the
spread of these phenomena (Xu and Lathrop, 1995).
Distance to sources of pollution/infection plays an
important role in measuring exposure to most diseases.
For example, a poorly constructed latrine, which is a
source of pollution, may pollute nearby areas by
spreading fecal matter. However, the spatial process of
pollution is usually lower as distance increases. The
mathematical expression for computing a household’s
exposure status within a raster system is
gi ¼1
ei
Xn
j¼1
bj�kj ;
where gi is the exposure status of the phenomena for
pixel i; bj the attribute of the phenomena in pixel j; kj the
kernel value of cell j of the filter, ei the edge correction
term for pixel i:
Fig. 3. Edge correction terms image.
Fig. 4. Event and population image used to compute a spatial
moving average rate.
M. Ali et al. / Health & Place 8 (2002) 85–92 89
The equation describes the filter that was used to
calculate the exposure status of a target pixel, which
represents decreasing neighbors’ influences with increas-
ing distances. A negative exponential distance function
(e�pixel lag) is proposed, where the kernel values for the
first, second, and the third lag pixels (away from the
target pixel) are 0.37, 0.14, and 0.05, respectively. The
edge correction term (ei) is necessary to obtain an
accurate representation of the exposure status near the
boundaries. In the equation, both the number and
strength of elements influence exposure at observed
locations. If more neighbors surround an observed
location, then exposure at that location will be higher.
We, therefore, suggest that the neighbors’ status be
aggregated freely in calculating the absolute exposure
status. For instance, in areas with high population
densities where sanitation is poor, people are more
exposed to poor sanitation. This can only be measured
by cumulating the influences of all of the neighbors. The
kernel values to be carried by the filter are also important
in capturing neighbors’ influences. It is very difficult to
determine the kernel values for the filter, as it requires
considering a number of factors including anisotropic
characteristics of the phenomena, because the diffusion
process of the phenomena may not be the same in all
directions. However, one may determine the values by
analyzing spatial variability of phenomena. There are
several geostatistical methods that can be used to
describe spatial variability. Determining true spatial
variability in a data set depends on knowledge of the
underlying phenomena being measured (Eastman, 1999).
In Fig. 5, the exposure status of the target pixel (gi)
obtained by using the spatial filtering method is 0.805.
The original attribute of the pixel is 0.6, which means
that neighboring values increase the value of the pixel by
34%. Fig. 6 is a surface map of poor sanitation of the
study area, which was created by calculating the
exposure status computed using the aforementioned
spatial filter. The map shows that the southwest part of
the study area has higher exposure to poor sanitation
conditions than other parts of the area.
Computing density of phenomena around observed
locations
Human settlement on the Earth’s surface can be
characterized as heterogeneous clusters of population.
These clusters have different shapes and sizes. Some
clusters have many people and others few. In some
clusters, people live close together while in others they
live far apart. Most traditional analyses examine density
at a given level of spatial resolution developed for other
purposes such as administrative units. Defining density
of a phenomenon by administrative units may be
insensitive to the relative location of the areas or points.
A spatial filter can be applied to compute population
density around clusters of households. The kernel values
of the window should be set to a unitary value resulting
in a cumulative sum of cases within the area covered by
the window. The edge correction term should be applied
to obtain the density of phenomena per unit area. The
expression for computing density around a particular
pixel location is,
yi ¼1
ei
Xn
j¼1
zj�kj ;Fig. 5. Image attributes used to compute filtered exposures
status.
Fig. 6. Spatial patterns of sanitation status.
M. Ali et al. / Health & Place 8 (2002) 85–9290
where yi is the density of the phenomena around pixel i;zj the attribute of the phenomena in pixel j; kj the kernel
value of cell j of the filter, ei the edge correction term for
the pixel i:A unitary kernel filter was used because it results
in the sum of the total values of the phenomenon
being measured and the edge correction term was
used to accurately represent the distribution near the
boundaries. In Fig. 7, the density for the target pixel
was calculated as 393.39. A 7� 7-pixel filter was
used, yielding a 0.0441 km2 area and a density of
8920/km2 (393.39C0.0441). Fig. 8 is the spatially
filtered population density surface map of the study
area. The map reveals the population density is
quite homogeneous, as a consequence of smoothing
the map throughout the study area, which would not
have been revealed if administrative boundaries were
used.
Conclusion
This paper demonstrates the use of spatial filtering for
scaling health, demographic, and environmental data.
Although data are often collected at the individual level,
the spatial distribution of households can influence an
individual’s health. Using spatial filtering methods,
spatial distributions of certain phenomena can be
exposed that otherwise would not be revealed. Creating
a spatially smoothed disease surface reveals a more
appropriate distribution than one created using geo-
political boundaries since pathogenic processes of
diseases are not usually associated with geopolitical
boundaries. Scaling spatial data sets reduces individual
effects when studying environmental diseases. Analyzing
variables by neighborhood rather than by households
can be accomplished using the aforementioned spatial
filtering technique. GIS methods can be used to better
understand environmental causes of the spatial distribu-
tion of diseases. In rural Bangladesh, households are
affected by the poor sanitation conditions of their
neighbors. Spatial filtering methods can also be used
model behavioral variables that affect human health.
For instance, neighbors influence health-seeking beha-
vior in rural Bangladesh.
Location is an important component of health and
environment studies and thus a GIS-based information
system is ideal. Implementing such a GIS is not as
expensive or difficult as it once was. Digital maps are
widely available in many places and a locational
component can be added to a health monitoring system
quite inexpensively using the global positioning system
(GPS). With the GPS, it is now possible to determine
accurate geographic coordinates (within 10m) of the
location of households and other features and integrate
these data into a GIS inexpensively. Once these data
have been collected and incorporated into a GIS
environment the methodological framework discussed
in this paper can be used to measure and integrate health
and environment variables.
Acknowledgements
This research was funded by Belgian Administration
for Development Cooperation and ICDDR,B: Centre
for Health and Population Research which is supported
by countries and agencies which share its concern for the
health problems of developing countries. Current
donors providing unrestricted support include: the aid
agencies of the Governments of Australia, Bangladesh,
Belgium, Canada, Japan, the Netherlands, Sweden, Sri
Lanka, Switzerland, the United Kingdom and the
United States of America; international organizations
include United Nations Children’s Fund.
Fig. 7. Image attributes used to compute filtered population
density.
Fig. 8. Population density surface.
M. Ali et al. / Health & Place 8 (2002) 85–92 91
References
Bailey, T.C., Gatrell, A.C., 1995. Interactive Spatial Data
Analysis. Longman, Harlow.
Bithell, J.F., 1990. An application of density estimation to
geographical epidemiology. Statistics in Medicine 9, 690–701.
Briggs, D.J., Elliott, P., 1995. The use of geographical
information systems in studies on environment and health.
World Health Statistics Quarterly 48, 85–94.
Cislaghi, C., Biggeri, A., Braga, M., 1995. Exploratory tools for
disease mapping in geographical epidemiology. Statistics in
Medicine 14, 2363–2381.
Clayton, D., Bernardinelli, L., Montomoli, C., 1993. Spatial
correlation in ecological analysis. International Journal of
Epidemiology 22, 1193–1202.
Cliff, A.D., Ord, J.K., 1981. Spatial processes: models &
applications. Pion, London.
Cook, J., Amevigbe, P.M., Crost, M., Gbetoglo, D., Tursz, A.,
Assimadi, J.K., 1999. Health care seeking behavior of
children in Togo. Revue d’Epidemiologie et de Sante
Publique 47 (Suppl 2), 2S93–113.
Eastman, J.R., 1999. Guide to GIS and Image Processing, Vol.
2. Clark Labs, Clark University, Worcester, MA.
Egunjobi, L., 1993. Spatial distribution of morality from
leading Notifiable disease in Nigeria. Social Science and
Medicine 36 (10), 1267–1272.
Elman, C., Myers, G.C., 1999. Geographic morbidity differ-
entials in the late nineteenth-century united states. Demo-
graphy 36 (4), 429–443.
Garner, B.J., Zhou, Q., Parolin, B.P., 1993. The application of
GIS in the health sector: problems and prospects. Fourth
European Conference on Geographical Information Sys-
tems, EGIS ‘93, Genoa, Italy, pp. 1350–1357.
Gatrell, A.C., Bailey, T.C., Diggle, P.J., Rowlingson, B.S.,
1996. Spatial point pattern analysis and its application in
geographical epidemiology. Transactions, Institute of the
British Geographers 21, 256–274.
Glenn, L.L., Burkett, G.L., 1999. Equal dependence of the high
prevalence of health problems on age and family income in
rural southern areas. Southern Medical Journal 92 (10),
981–988.
Haining, R., 1990. Spatial Data Analysis in the Social and
Environmental Sciences. Cambridge University Press,
Cambridge.
Haining, R., 1998. Spatial statistics and the analysis of health
data. In: Anthony C. Gatrell, Markku L .oyt .onen (Eds.),
GIS and Health (GISDATA Series 6). Taylor & Francis,
London.
IDRISI, 1997. Idrisi for Windows, Version 2.0. Idrisi Produc-
tion, Clark University.
Kafadar, K., 1996. Smoothing geographical data, particularly
rates of disease. Statistics in Medicine 15, 2539–2560.
Kafadar, K., 1999. Simultaneous smoothing and adjusting
mortality rates in us counties: melanoma in white females
and white males. Statistics in Medicine 18, 3167–3188.
Kirscht, J.P., Becker, M.H., Eveland, J.P., 1976. Psychological
and social factors as predictors of medical behavior.
Medical Care 14 (5), 422–431.
Loslier, L., 1996. Geographical information systems (GIS) from
a health perspective. IDRC: International Development
Research Centre, http://www.idrc.ca/books/focus/766/
loslier1.html.
Matos, E.L., Parkin, D.M., Loria, D.I., Vilensky, M., 1990.
Geographic patterns of cancer mortality. International
Journal of Epidemiology 19 (4), 860–870.
Meijerink, A.M.J., de Brouwer, H.A.M., Mannaerts, C.M.,
Valenzuela, C., 1994. Introduction to the Use of Geographic
Information Systems for Practical Hydrology. ITC, The
Netherlands UNESCO.
Munshi, K.D., 1996. Farmers as Econometricians: Social
Learning and Technology Diffusion in Indian Green
Revolution, Mimeo. Boston University.
Onsrud, H.J., Pinto, J.K., 1991. Diffusion of geographic
information innovations. International Journal of Geo-
graphic Information Systems 5 (4), 447–467.
Pringle, D.G., 1986. Disaggregating regional variations
in mortality by cause of death: a case study of the
republic of Ireland. Social Science and Medicine 23 (10),
919–928.
Rushton, G., 1998. Improving the geographic basis of health
surveillance using GIS. In: Anthony C. Gatrell, Markku
(Eds.), GIS and Health (GISDATA Series 6). Taylor &
Francis, London.
Rushton, G., 2000. GIS to improve public health. Transaction
in GIS 4 (1), 1–4.
Rushton, G., Lolonis, P., 1996. Exploratory spatial analysis of
birth defect rates in an urban population. Statistics in
Medicine 15, 717–726.
Scholten, H.J., de Lepper, M.J.C., 1991. The benefits of the
application of geographical information systems in public
environment health. World Health Statistics Quarterly 44,
160–170.
Talbot, T.O., Kulldroff, M., Teven, P.F., Haley, V.B., 2000.
Evaluation of spatial filters to create smoothed maps of
health data. Statistics in Medicine 18, 2399–2408.
Tursz, A., Crost, M., 1999. An epidemiologic study of health
care seeking behavior of children under 5 years of age by sex
in developing countries. Rev Epidemiol Sante Publique 47
(Suppl 2), 2S133–2156.
Twigg, L., Moon, G., Jones, K., 2000. Predicting small-area
health-related behaviour: a comparison of smoking and
drinking indicators. Social Science and Medicine 50, 1109–
1120.
Watkins, C.D., Sadun, A., Marenka, S., 1993. Modern Image
Processing: Warping, Morphing, and Classical Techniques.
Academic Press Professional, New York.
Xu, A., Lathrop Jr., R.G., 1995. Improving simulation
accuracy of spread phenomena in a raster-based geographic
information system. International Journal of Geographic
Information Systems 9 (2), 153–168.
M. Ali et al. / Health & Place 8 (2002) 85–9292