First International Symposium of Health GIS
Bangkok, Thailand
December 1-2, 2005
CONTEXTUALIZING THE URBAN HEALTHCARE SYSTEM
METHODOLOGY FOR DEVELOPING A GEODATABASE OF
DELHI’S HEALTHCARE SYSTEM
Pierre CHAPELET, Bertrand LEFEBVRE
University of Rouen, France – Centre de Sciences Humaines, India
chapelet@altern.org, bertrand.lefebvre@csh-delhi.com
ABSTRACT:
This communication introduces the setting up of a Geographical Information System on Delhi for studies in the Social
Sciences focusing on healthcare system organization (developed at Centre de Sciences Humaines, New Delhi, India).
Through an explanation of the methodological procedure and demonstration of thematic applications focusing on the
healthcare system’s spatial organization, the author lead us through the inherent difficulties of building a GIS in an
emerging country like India. They also attempt to demonstrate that this kind of tool remains, however, a relevant support
for research in the Social Sciences as long as it is used with care and knowledge of the dataset frame. From this
perspective, Exploratory Data Analysis coupled with the play of scales provide powerful ways to assess socio-spatial
and healthcare system dynamics taking place in the Indian capital.
KEY WORDS: GIS, Healthcare system, Data Exploratory Analysis, Multiscalar, Delhi, Census 1991/2001.
INTRODUCTION
This Paper presents the setting-up of a Geographical
Information System (GIS) about Healthcare system of the
Delhi agglomeration. This GIS is still under development
and currently used by different researchers working on
health care system in Delhi. Instead of presenting results of
these research programs (which could be very boring for the
non specialist or indianist), we though it would be more
relevant to briefly introduce this tool, and then to focus on
few specific methodological points, which are in fact
relevant for each research project dealing with GIS. For
further informations about research projects using this tool,
I invite the reader to go to http://www.cshdelhi.com/electronicop11/start.htm to discover in-depth
interactive presentations of its application fields.
1.
OBJECTIVES OF THE PROJECT
This GIS has been primarily developed in the context of
a health geography research project on access to
pharmaceuticals in the Indian capital, Delhi.
At that time, the purpose of implementation of such a
tool was manifold:
Firstly, at the beginning of the project in January 2002,
there was an acute lack of documentation precisely
describing the spatial organisation and functioning of the
city. Thus, integrating detailed census data in a GIS was a
first step to better assess socio-spatial dynamics taking place
in the Indian capital. Furthermore, the idea of comparing the
recently released census 2001 datasets with already
available census 1991 datasets was clearly a very promising
way to obtain a dynamic picture of the city, which has not
been attempted till now.
Secondly, in order to work on access to pharmaceuticals and
more generally on health topic as geographers, it was necessary to
set up a spatial database enabling us to visualize the spatial
settings of healthcare infrastructures in the city and its
periphery, identifying actors, their localisation and spatial
organisation. Cross-checking this layer of information with census
data was seen as a way to better understand the location strategies
of healthcare system actors. In the field of health geography, our
posture was thus essentially to rely on the spatial analysis of
healthcare infrastructure locations.
Thirdly, given the two above-mentioned objectives, this tool
seemed to be very promising for sampling purposes:
To select samples of health infrastructure, in particular those
distributing medicines, for our fieldwork survey according to their
location in the urban agglomeration;
To select urban areas according to their profile (socio-economic,
demographic), as well as according to the spatial repartition of the
healthcare actors for further investigations (survey of households);
As soon as we started conceptualizing our spatial database, given
the variety of data available through the census database it was soon
obvious that it could be very useful for any researcher working on
Delhi, particularly in the field of social sciences, urban development,
or health studies. This led us to design this GIS in order to ensure
that it could be re-used. That is why we decided to publish a
detailed presentation of this work on a digital format (CDRom and
internet web site). This allowed us to incorporate interactive maps
(flash), helping the potential user in understanding the way he can
use this tool. Moreover, since the main limitation in the use of this
kind of tool often stems from the difficulty to feed the system with
data, we also had to ensure that it could be quickly updated and
completed with fresh datasets coming from other research projects.
This led us to base our databases on a common georeferenced
framework and to develop tools for easy updates. Besides, the
inevitable imperfections of datasets led us to resort to
First International Symposium of Health GIS
Bangkok, Thailand
December 1-2, 2005
methodological devices to check the bias generated by
relatively poor data quality.
are not listed centrally, or listings are incomplete, such as for
general practitioners, pharmacies, or private nursing homes.
2.
After this brief overview of statistical data availability, what
about cartographic data?
RESEARCH POSTURE AND METHODOLOGY
The research posture and the resulting framework of our
GIS emerged from different constraints.
2.1 Constraints to the use of GIS
There are in fact many constraints to build-up and use a
GIS in a developing country such as India, especially when
working in the field of health and, furthermore when
focusing on intra-urban problematics. As we will see, these
constraints often shape the design of the GIS as such, and
lead to resort to specific methodologies to work around
them.
The main constraints are of course related to Data
(statistical AND cartographic data):
Albert, Gesler & all point out the ”4-I” Rules of every
GIS: Intensive, Inaccurate, Inaccessible, Incomplete (Albert,
Gesler & Al). In order to restrict the unavoidable inaccuracy
of a system that seeks to simplify reality through modelling
(GIS), one must intensively feed the system with fresh
datasets. The processing power of such a system partly
comes from data wealth. Now, datasets are often
inaccessible, inaccurate or simply incomplete.
India is not an exception to this assumption. The Indian
situation is in fact quite paradoxical regarding data
availability. Indeed, on the one hand India is an important
producer of statistics. On the other hand, these statistics are
not easily accessible, especially when looking for detailed
data. As soon as statistical information below the district
level in needed, it becomes very difficult to procure it.
(WHAT IS A DISTRICT FOR A NON SPECIALIST ?)
2.1.1 Statistical datasets
One of the main sources of information, when one wants
to analyse relations between the healthcare system and
deserved population is of course the Census of India, which
provide detailed datasets about demographic, social and
economic statistics every decade. Moreover, it is available
on a digital format since 1991. However, when working on
intra-urban dynamics, many limitations emerge. Firstly, the
number of spatial units in a city is quite low for an in-depth
understanding of intra-urban dynamics. Secondly, in Delhi
the shape of these units has been modified between the last
two censuses.
We could also cite another data sources, such as the
National Sample Survey (NSS), or the National Family
Health Survey (NFHS), which are especially focusing on
health and healthcare thematic. Again, data is not available
on a sufficient small level for integration in a GIS. For
example, the NFHS statistical sample has been merged in 2
spatial units in Delhi (rural/urban)…
The situation is the same when one wants to even only
locate healthcare infrastructures. All government agencies
(central, federal, municipal), maintain its own list of health
infrastructures but there is no coordination between these
agencies. Furthermore many actors of the healthcare system
2.1.2 Cartographic data
Cartographic data are in fact highly controlled by the
government and thus it is very difficult to obtain a full coverage of a
given area. Moreover, many maps are outdated, or simply
unavailable. This is a real paradoxical situation given that very
accurate tools now become available on the market. For the big
Indian metropolises, private companies such as Eicher City Maps
now provides detailed coverages (but they are very costly for small
research projects). We could also cite Google Map, which allow the
user to zoom deeply inside a city (this tool was not available when
we started our project, but is now used to extend our database).
Finally the picture of data availability does not seem to be very
motivating for a researcher starting a GIS project on healthcare
system in Delhi. However, it is possible to work around these
problems using specific tools and methodologies, which are often
missing in GIS…
2.2 Contextualising data through the play of scales
If GIS manages the question of scale in spatial continuity, we
have seen previously that thematic data are often dependent on
administrative divisions, which are on the contrary not spatially
continuous (thematic discontinuity). When working with GIS, this
obliges us to choose perception levels adapted to the scale of
studied spatial units. (See figure 1). However, selecting only one
perception level, such as census administrative divisions to study
the spatial distribution of healthcare infrastructures in Delhi or to
compute catchment area assessments (Desserte rate), can disguise
the spatial organisation of the studied phenomenon because of the
internal heterogeneity of each object (Are census units really
pertinent to compute this rate?).
Source: Roudier Daval, 2004
Figure 1 – Comparison Criteria between Exploratory Data Analysis
and Confirmatory Data Analysis
Thus we decided to develop a multi-level framework based on
different perceptions levels. The idea is to say that a phenomenon
can be properly analysed only if studied at different scales in order
to understand in which context its spatial organisation takes place.
Furthermore, one can work around the problem of data weakness
(limited accuracy or limited spatial range) when a dataset is
contextualised using other working scales, themselves unveiling
other levels of spatial organisation. The results of a study at a given
scale will reveal a trend in the spatial organisation of a phenomenon
which may or may not be confirmed when contextualised with other
scales.
2.3 Cartomatic and Exploratory Data Analysis
Associated to this idea of contextualisation, we used Exploratory
Data Analysis techniques (EDA) and specific cartomatic tools to
analyse collected datasets and unveil there underlying structure.
Table 1 resumes the main differences between exploratory and
classical confirmatory analysis.
Though an increase in the number of views, this approach allows
not only to try to choose a representation amongst a set of solutions,
but also to favour the emergence of hypotheses regarding the
First International Symposium of Health GIS
Bangkok, Thailand
December 1-2, 2005
underlying spatial organisation of the studied phenomena.
EDA finally appears as a kind of “Interative System of
Thoughts Assisted by Computer” (Antoni & Klein, 2003).
Exploratory Analysis
Confirmatory Analysis
Descriptive Approach
Inferential Approach
Robust Statistics
Sensitive Statistics
Flexible Research Program Rigid Research Program
Graphic Expression
Numerical Expression
Intuitive Vision
Deductive Vision
Source: (WANIEZ 2002)
Table 1 – Comparison Criteria between Exploratory Data
Analysis and Confirmatory Data Analysis
Graphic expression remains the key factor of such an
approach. However, visualisation is of course not limited to
the simple representation of a given data. Indeed, many
classical tools were used in our projects to help visualize the
data.
Amongst the methods used to process statistical variables,
we can mention:

Factor analysis;

Hierarchical Agglomerative Cluster Analysis;

Linear Regression and Scatter plot analysis;

Spatial Autocorrelation coefficients such as Moran and
Geary.
These methods have been used using different free
softwares such as Philcarto and GeoDa. Non-geographic
statistical tests are now conducted using R.
It is only after this first “radiography” of data series (this
contextualisation - Exploratory methods), that we built up
our research hypothesis about the organisation of Delhi
health care system. We then conducted specific spatial
analysis tests and surveys to answer our hypotheses
(Confirmatory methods). These results were in turn
integrated and generalised in our database allowing new
interactions with datasets and new hypotheses.
3.
RESULTS AND FUTURE DEVELOPMENTS
At present, our GIS contains four perception levels, each
one corresponding to an individual geographic database. Of
course, georeferencing of each database warrants
compatibility between them.
As mention earlier, in order to help researchers in
discovering the potential of such multilevel approach, we
published few illustrations of possible treatments in our
CDrom.
Selected perception levels are as follow (from smaller to
larger scale):
health care infrastructures at the Census Charge level (the more
accurate spatial unit available from Census of India). It aims at
catching up the full extends of Delhi agglomeration and its satellite
cities.
3.3 Delhi (National Capital Territory):
This perception level contains 1991 and 2002 Census Datasets.
We also collected the complete list of health infrastructure from
various government agencies. The tables pertaining to public
hospitals and dispensaries are as exhaustive as we could expect but
as far as private nursing homes are concerned, most of them were
still unregistered in 2002. Instead of attaching these infrastructures
to census spatial units, a method which implied a loss of accuracy in
location given the size of the concerned units), we preferred attach
them to a new layer containing the location of each locality-place in
the city (2300 georeferenced punctual entities located at the
geometric centre of each locality). This method still allows the user
to agglomerate data by census unit for a rapid comparison, without
loosing location accuracy during geocoding. This layer is also a
very useful tool for researchers doing fieldwork surveys, allowing
them for example to exactly locate the different places visited by
patients, and then study variations of spatial mobility in relation to
socio-economic status, morbidity profile or health service used.
Since the shape of spatial units has been modified between the
two censuses, we plan to rely on interpolation techniques in order to
allow comparisons between 1991 and 2001 and study demographic
trends. We already have done this treatment for the 1991 census
data. For instance, instead of mapping density calculated on the
basis of census units, we used a remote sensing image to
automatically calculate population density based on real land use.
For each census unit, we assigned a point entity at the barycentre of
each built-up area. The number of inhabitants has then been divided
between each built-up area according to its spatial share in the
census unit. We calculated density by dividing the total number of
inhabitants by the surface measurement of each built-up area (and
not by census charge area). Finally we generated a trend surface
(map 2).
This method can be applied to various demographic and health
data. Generated trend surfaces can then be overlaid, allowing study
of population dynamics.
If the user can of course map each demographic indicator by
census unit to compare it with the spatial organisation of healthcare
system actors, or combine few variables for deeper analysis, we
though it would be interesting to already synthesise census datasets
for the user. In order to do so, we selected different sociodemographic variables (15) and executed a Factor Analysis. The
table 2 presents composition of the four first principal components
(which sum up 64% of the total information).
3.1 North West India:
Contains demographic data and number/type of public
healthcare infrastructures attached to cities (up to 20 000
inhabitants). It allows the user to contextualise Delhi
situation regarding healthcare provision and population
evolutions in much larger dynamics. (See figure 2)
3.2 Delhi and its periphery (Delhi Metropolitan Area):
Covering around 11 000 km2 and more than 17 000
inhabitants, this layer is still incomplete and is actually
extended to cover a much more larger area around the city.
It will contain Census demographic data (2001) and public
Principal Components
Variabl
Name
e
V01
Density
V02
% Workers
V03Persons per Household
V04
% SC
Sex Ratio
V05
Population
Workforce Sex
V06
Ratio
Child Women
V07
Ratio
V08
% Literate
CP1
CP2
CP3
CP4
-629
191
221
240
219
786
-704
371
405
-310
397
-69
172
323
275
-92
-327
-578
121
-29
-113
164
-583
453
701
32
265
-392
-837
-325
-187
37
First International Symposium of Health GIS
Bangkok, Thailand
December 1-2, 2005
V09
V10
V11
V12
V13
V14
V15
Women
% Households
Manufacturing
% Manufacturing
(others)
% Construction
%
Trade/commerce
%
Transport/storage
% Other Services
% Primary sector
-158
92
580
-24
-247
673
318
45
39
362
-186
-623
-768
339
208
40
-11
-444
22
-560
-327
754
-378
-437
-591
24
-290
418
Table 2 – Principal Component Analysis Axe Saturations
(*1000)
Source: Census of India, 1991
Then, we launched a Hierarchical Agglomerative Cluster
Analysis based on the four first principal components and
mapped the results. The figure X presents the resulting
typology of NCT space.
Finally, comparing this typology of NCT space with
location of each type of healthcare system actor available in
the database gives strong evidences to users of the different
location strategies deployed. (mettre carte de localisation
public-privé sans typologie des charges ?)
3.4 Specific Intra-urban Zones (Gurgaon, NewDelhi,
Shadhara):
The use of the three previous perception levels already
allows to build up an initial picture of the healthcare system
organisation. However, in order to better grasp the reality,
we strengthened the analysis by zooming on a large-scale
level. Since there is no legislation constraining private
structures such as general practitioner or medical shop in
their location choice, do we observe spatially specific
location strategies? Does place matter?
Yet, only specific census units have been selected
according to their socio-demographic profile and
infrastructure availability (a peri-urban unit and four central
ones).
This time, we digitalised built-up areas and land use
(commercial, residential, industrial…) using the Delhi
Eicher City Map (the only map precisely indicating the land
use) and went for fieldwork investigations to locate each
and every health infrastructures in selected census units.
Then, we generated different distance calculations (see
figureX):
Firstly, crossing these two layers of information, we
generated graphs showing the attractiveness of each kind of
land use for a given actor, enabling us to gauge to which
extent urban environment can influence its establishment.
Results showed clear evidences of specific location
strategies for each actor. While public actors such as
dispensaries equally serve each type of urban area (spatial
equity), private infrastructures such as pharmacies prefer
particular areas such as commercial ones.
Secondly, we generated graphs showing the
attractiveness of each actor for others infrastructures. This
work clearly unveiled cooperation or substitution strategies
between actors. The case of pharmacies is again very
interesting on this matter since their location closely
depends on their functional need to cooperate with other players
such as doctors. As observed on the graph this cooperation leads to
the creation of spatial clusters.
CONCLUSION
Contextualising Datasets through a multiplicity of perception
levels is really helpful to avoid data weaknesses.
… A finir…
First International Symposium of Health GIS
Bangkok, Thailand
December 1-2, 2005
REFERENCES
Journal
Albert, Gesler & Al
Dale, P.E.R., Chandica, A.L., and Evans, M., 1996,
Using image substraction and classification to
evaluate change in subtropical intertidal wetlands.
International Journal of Remote Sensing, 17, 703719.
Books
Barret, E.C., and Curtis, L. F., 1992, Introduction to
Environmental Remote Sensing, 3rd edition,
(London: Chapman and Hall).
Edited Books
Strum, B., 1981, The atmospheric correction of
remotely sensed data and the qualitative
determination of suspended matter in marine water
surface layers. In Remote Sensing in
Oceanography and Meteorology, edited by A. P.
Cracknell (Chichester: Ellis Horwood).
References from websites:
Nakhapakorn, K., 2005. Proceeding on the Health
GIS symposium “Analysis of Spatial Factors
affecting dengue epidemics using GIS”, Bangkok,
Thailand.
http://www.jgeoinfo.net/HealthGIS/HG001.html
3.5 Acknowledgements (optional)
Acknowledgements
of
support
project/paper/author are welcome.
for
the