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: mali@ivi.int (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