ArticlePDF Available

Epidemiological geomatics in evaluation of mine risk education in Afghanistan

Authors:

Abstract and Figures

Evaluation of mine risk education in Afghanistan used population weighted raster maps as an evaluation tool to assess mine education performance, coverage and costs. A stratified last-stage random cluster sample produced representative data on mine risk and exposure to education. Clusters were weighted by the population they represented, rather than the land area. A "friction surface" hooked the population weight into interpolation of cluster-specific indicators. The resulting population weighted raster contours offer a model of the population effects of landmine risks and risk education. Five indicator levels ordered the evidence from simple description of the population-weighted indicators (level 0), through risk analysis (levels 1-3) to modelling programme investment and local variations (level 4). Using graphic overlay techniques, it was possible to metamorphose the map, portraying the prediction of what might happen over time, based on the causality models developed in the epidemiological analysis. Based on a lattice of local site-specific predictions, each cluster being a small universe, the "average" prediction was immediately interpretable without losing the spatial complexity.
Content may be subject to copyright.
BioMed Central
Page 1 of 10
(page number not for citation purposes)
International Journal of Health
Geographics
Open Access
Methodology
Epidemiological geomatics in evaluation of mine risk education in
Afghanistan: introducing population weighted raster maps
Neil Andersson*
1
and Steven Mitchell
2
Address:
1
Centro de Investigación de Enfermedades Tropicales (CIET), Universidad Autónoma de Guerrero, Acapulco, Mexico and
2
CIETcanada,
Ottawa, Ontario, Canada
Email: Neil Andersson* - neil@ciet.org; Steven Mitchell - steve@ciet.org
* Corresponding author
Abstract
Evaluation of mine risk education in Afghanistan used population weighted raster maps as an
evaluation tool to assess mine education performance, coverage and costs. A stratified last-stage
random cluster sample produced representative data on mine risk and exposure to education.
Clusters were weighted by the population they represented, rather than the land area. A "friction
surface" hooked the population weight into interpolation of cluster-specific indicators. The
resulting population weighted raster contours offer a model of the population effects of landmine
risks and risk education. Five indicator levels ordered the evidence from simple description of the
population-weighted indicators (level 0), through risk analysis (levels 1–3) to modelling programme
investment and local variations (level 4). Using graphic overlay techniques, it was possible to
metamorphose the map, portraying the prediction of what might happen over time, based on the
causality models developed in the epidemiological analysis. Based on a lattice of local site-specific
predictions, each cluster being a small universe, the "average" prediction was immediately
interpretable without losing the spatial complexity.
Background
Colourful and seductive, geographic information systems
(GIS) offers status symbol technology that researchers and
development agencies rush out to purchase. The imagery
conjured up by a map is immediate and compelling: if we
believe the map, we seem to know where the problems are
and, it should follow, where interventions should take
place to address them. In many aspects of public sector
planning, GIS can help information sharing. Analysis of
the spatial dimensions of public service programmes can
also enhance coordination and specificity of action to the
local conditions [1]. The hope that GIS might increase
programme efficiency and investment choices, however,
remains largely unfulfilled. This depends on the elusive
integration of GIS with formal analysis of risk and resil-
ience.
We developed the term epidemiological geomatics for an
approach reaching beyond visual overlays of spatial data,
to include modelled interactions between layers that
inform population based predictive models. Using this
approach, planners who need to know more than just the
locality of a particular problem can visualise epidemiolog-
ical models of programme performance.
Several problems obstruct the use of maps in epidemiol-
ogy. The inaccessible "black box" nature of GIS technol-
ogy has led some to believe that maps might solve a
problem they cannot solve. Like any presentation tool, the
Published: 03 January 2006
International Journal of Health Geographics 2006, 5:1 doi:10.1186/1476-072X-5-1
Received: 10 October 2005
Accepted: 03 January 2006
This article is available from: http://www.ij-healthgeographics.com/content/5/1/1
© 2006 Andersson and Mitchell; 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.
International Journal of Health Geographics 2006, 5:1 http://www.ij-healthgeographics.com/content/5/1/1
Page 2 of 10
(page number not for citation purposes)
information that maps convey can be no better than the
data that goes into them. The attractiveness of the imagery
is far from a guarantee that the evidence has meaning.
Meta-data of maps – where and how the evidence was col-
lected – are often taken for granted, yet they can change
completely the interpretation.
A map showing the location of a place with a high rate of,
say, a particular disease, is a very partial model of reality.
Most maps that show disease patterns cover the geo-
graphic domain where the evidence is generated. A case in
point is the National Center for Health Statistics (NCHS)
Atlas of United States Mortality [2] that portrays the
national rates of various conditions. The cases in County
X are assigned to a polygon demarcated by health service
areas, where it is related to the population to produce a
rate. The rate is colour coded and the resulting image
offers the planner an average indicator level across the pol-
ygon. It has a meaning only insofar as the average has a
meaning. But averages miss the extremes, often the most
vulnerable who are the focus of much public health plan-
ning. Averages are notoriously distorted by outliers, with-
out fully reflecting those outliers. Sometimes it is the
outlier that most merit the public health action.
One might also find a health service area or district with a
low mortality or coverage immediately adjacent to one
with a very high level of the same indicator. The "bound-
ary" between two polygons can be an unrealistic and
abrupt change. Since it does not reflect any gradient
between highest and lowest rates, the map loses coher-
ence and, with that, much of its use in planning.
Almost all countries have a routine data system that
makes some attempt at collecting cases or documenting
service activities. These can sometimes be useful for map-
ping. The challenge intensifies with data from sample sur-
veys. In order to fill in a polygon based on a sample, the
survey must be of a sufficient sample size and it must be
chosen in such a way as to permit generalisation to the
whole polygon. This must be done for each polygon.
Then there is the difficulty of compatibility of data sets.
Data come in different formats and they refer to different
domains, at different times. In the Afghanistan case, a
mine survey mapped the physical location of the mine
fields. Existing services were catalogued and mapped, and
these data can be related to where the population is
thought to live. But in Afghanistan the last census was
done several decades ago. Even the 1990 "update" was
altered drastically by several new phases of the war. Unre-
lated sample surveys with varying objectives provide data
from some places but not from others. This information
asymmetry is typical of situations where maps are needed
most. Even outside of emergency situations, data on envi-
ronmental risks, population distribution and services are
rarely coterminous.
Case study from Afghanistan
Land mines are an intrinsically spatial issue – land mines
alter geographic access – and integration of epidemiolog-
ical risk analysis with GIS should offer substantial benefits
for planning and evaluation of mine action. The harsh
conditions of Afghanistan offered an opportunity to test
the added value of GIS in epidemiological evaluation[3].
Even in the mid-1990s, Afghanistan was not without
existing data. Because of the strategic importance of the
country and its particular history in the last years of the
Cold War, its geography was unusually well documented.
For years, US spy satellites followed movement of people,
documenting every contour and cave. When the USA
pulled out of the region, they made available huge banks
of data accessible to the humanitarian organisations
working in Afghanistan.
The quality and comprehensiveness of the topographic
data were complemented, if not entirely matched, by
moderately up-to-date mapping of landmines. The survey
of mined areas offered some understanding of the distri-
bution and density of landmines or unexploded ordnance
(UXO) contamination. With all this wealth of spatial
information, the missing elements were the people and
their behaviour. Sample surveys of the impact of land-
mines and the evaluation of mine risk education have
contributed elements towards a comprehensive geo-
graphic information base on mines and mine action in
the country.
But these sources of evidence, each rich as it may be, were
uneven, fragmented and not integrated in a usable way for
decision-taking. Mine risk education in Afghanistan, prior
to a CIET evaluation conducted in 1997/98 [4], rested on
two main types of education. Direct training through lec-
tures at community level was conducted by three non gov-
ernmental organisations; they were supposed to establish
"mine committees" in each community where training
had occurred, to sustain and to continue the risk educa-
tion. The second mine risk education modality was
through a soap opera, called New Home New Life, on the
BBC's Afghanistan Service.
The CIET evaluation found many people who did not
even know they were listed as "trained" in the direct train-
ing scheme. There was some evidence of a positive impact
of direct training in the mine-affected areas, for example,
a higher likelihood of reporting a mine or UXO, but not
all results from the training programme were positive.
Those who said they received direct training were more
likely to declare high risk attitudes and practices (such as
searching for scrap metal and considering someone brave
International Journal of Health Geographics 2006, 5:1 http://www.ij-healthgeographics.com/content/5/1/1
Page 3 of 10
(page number not for citation purposes)
who goes into a mined area but is not a deminer). Among
trainees, there was a significantly higher risk of injury after
the training programme began. They suffered higher risk
of injury affecting the upper part of the body (hands,
arms, eyes and torso); injuries most likely resulting from
attempts to diffuse the mines rather than mines being
stepped upon. This suggests inappropriately heightened
confidence in handling mines among those who received
training. One explanation for these findings was that
much of the mine risk training was conducted by dem-
iners. They may have taught what they know about mines
– for example, where the detonator is housed and how it
works – thus engendering the false feeling of security.
The BBC radio soap-opera New Home New Life, had a
measurably positive effect on reinforcing appropriate
mine behaviour, frequency and type of injury.
Epidemiology behind maps, and using maps in
epidemiology
The planner's question
The fact that an action is planned and paid for does not
mean it is implemented. The fact that a service is offered
does not mean people make use of it. And, if they do take
up services that are offered, it does not mean these services
have the desired effect. Like almost any public sector inter-
vention, mine risk education does not work equally well
everywhere and, in some places, it does not work at all.
This is the issue at the heart of evidence-based planning:
what actually happens is what counts, not only what is
intended to happen.
The central exercise in evidence-based planning is to
quantify the gap between what is intended and what actu-
ally happens, and to identify the particular mix of circum-
stances under which an intervention is effective. Planners
need to know the coverage achieved by each service, but
they also need to know what allocation of resources it will
take to close the gap between intended impact and actual
impact. More than the precise identity of the localities
where services work or do not work, planners need to
know about the particular mix of circumstances under
which an intervention works.
A framework for epidemiological study
The cross design of techniques known as the CIETmethod
[5] – also known as sentinel community surveillance
(SCS) [6] or service delivery surveys (SDS) [7] – tries to
maintain epidemiological coherence of evidence intro-
duced into planning. The method relies on a panel of clus-
ters weighted to link the sample to the universe it
represents.
Cyclical contacts with these a representative panel of clus-
ters (Table 1) involve a concentration of measurement
resources in time and place, an intense focus of quantita-
tive and qualitative methods in a panel of mini universes.
The ability to repeat measurement in the same place
makes impact estimation relatively straightforward. These
households can be contacted in successive cycles, perhaps
a year or two years later, to measure differences over the
period. These differences can be related to programmatic
input and other factors that might be vary across different
clusters. The impact assessment is based on the time
sequence and the heterogeneity between clusters.
The contiguous households in each cluster to permit the
analysis of local factors in the context of household-level
occurrences. Some environmental factors might be quanti-
fied easily (for example, presence of school, cost of drugs)
or they may be more qualitative (adequacy of sanitation,
level of participation in community affairs). If these fac-
tors affect the whole cluster, comparisons can be made
between clusters or groups of clusters.
Costs of public services (and costs of not accessing public
services) can vary from place to place in such a way that a
particular solution may have a favourable cost-effective-
ness ratio in one place but not in another. Remoteness can
drive up transport costs or create monopoly conditions
for service providers. Household or cluster-specific cost
data can be collected from key informants and fed into the
Table 1: The CIET fact-finding/feedback cycle
1. Identification of the issue to be researched, for example, access to water and sanitation, access to and use of services, adequate food supply, etc.
2. Ordering and analysis of data from routine sources and previous studies, attempting to align data in three analytical categories: impact, coverage,
and costs.
3. Development of the instruments including precise objectives, questionnaire, key informant interviews, data entry format, and report outline.
4. Pilot testing including data entry and analysis.
5. Fieldwork including household questionnaires and qualitative techniques (key informants, observation, focus group discussions).
6. Data entry and preliminary analysis, identification of confounders and effect modifiers.
7. Feedback and interaction in sentinel communities for interpretation and strategy development.
8. Completion of analysis, refinement of programme options.
9. Development of the communication strategy that can be consultative process in the same clusters.
10. Communication of results to all communities, development of strategies of action to resolve issues.
International Journal of Health Geographics 2006, 5:1 http://www.ij-healthgeographics.com/content/5/1/1
Page 4 of 10
(page number not for citation purposes)
mix. Decisions on resource allocation can thus become
more flexible, adapting to local cost realities.
A model with predictive value
The task of epidemiological geomatics in Afghanistan was
to predict what might happen with mine education based
on the empirical evidence in those clusters.
We are used to hearing a statistic with predictive value
derived from a sample survey. For example, an opinion
poll about what people think prior to an election can have
predictive value regarding an election outcome. A service
delivery survey may attempt to establish the coverage of a
given public service, and to gain insight into why the serv-
ices are not utilised optimally.
Sampling, so familiar in epidemiology, has yet to find a
comfortable fit with GIS. Formal steps can link a sample
with a given domain. In Afghanistan, we stratified the
country into the four UN operational regions to ensure
the sample was balanced between those domains. Second
stage random selection of districts in each UN operational
region produced a sample of 48 districts. We stratified
each sample district by population density, urban and
rural, to ensure appropriate balance in population pro-
portion in these substrata. From a list of all candidate
communities in each substratum, giving each cluster a
specified chance to enter the sample, we randomly
selected survey communities. From a map of each survey
community, we randomly selected a cluster of house-
holds. We then calculated the population weight for each
fixed-size cluster (100–120 households), based on the
population it was supposed to represent in the four UN
regions.
The resulting sample of 86 sentinel clusters represented
the spread and proportions of different conditions under
evaluation. No sub-sampling was done within each clus-
ter or sentinel site (100–120 contiguous households, each
including 500–1000 people). This large and relatively
fixed cluster size derives from an epidemiological concern
to have sufficient data in each place for that "point" in the
model to be interpretable. In effect, each cluster is a mini
universe, and the sample of clusters is a lattice of mini uni-
verses.
Weighted clusters
The unit of mapping is usually physical space – square kil-
ometres. Planners are often concerned about physical
space but they are usually even more concerned about the
people who live and work there. We "weight" the clusters
to relate them to the population, not the land area. Some
segments of a country will be densely populated, and oth-
ers sparsely populated. It is possible to introduce this pop-
ulation dimension by applying the sample weights to link
the model to the "country", with the weight of each cluster
proportional to the population it represents.
This involved building a "friction surface" under the map,
rather like the underlay of a carpet, with circles radiating
outwards around each one of the 86 sentinel clusters (Fig-
ure 1, the lowest segment represents the location of the
"tent-poles", the second layer from the bottom is the
interpolated weight layer, and the third layer is the friction
surface). The friction surface for a large and small town
would look a bit like a big stone and a small stone
dropped some distance apart into a calm pool. The ripples
caused by the big stone go further and have more influ-
ence on the overall picture than the ripples from the small
stone. This friction surface, set to match the population
weight of the cluster in the overall sample, sets the influ-
Layers in a population-weighted raster, based on a cluster sampleFigure 1
Layers in a population-weighted raster, based on a
cluster sample. a) position of sentinel sites or "tent-poles",
b) population weights, c) friction layers and d) raster drape
(top) where the shade is set by the height of the tent-pole
and the extension of an interpolation based on the popula-
tion weights.
International Journal of Health Geographics 2006, 5:1 http://www.ij-healthgeographics.com/content/5/1/1
Page 5 of 10
(page number not for citation purposes)
ence each point has on the overall map that they make up
together.
The friction surface to introduce population weights was
initially created using the Cost function in Idrisi. The fric-
tion surface is an inverse of the interpolated cluster
weights: where clusters were under-represented in the
sample, having a higher population weight, friction levels
during interpolation were lower. Therefore under-repre-
sented clusters were allowed to "spread out further" on
the surface during the interpolation (acceleration). Con-
versely, over-represented clusters moved higher frictions
during interpolation and their overall effect thus reduced.
Population weighting transforms entirely the use of maps
for planning. The clusters in our Afghan sample were each
referenced to the population they represent by a weight.
On the map, this weight determined the spread of a col-
our or shade in proportion to the population. The net result is
that the map can show 40% of the country in a given
shade of green, when 40% of the people who live in that
country are affected by the exposure represented by that
shade. This makes immediate sense for planners, who
usually think about populations rather than square kilo-
metres.
The practical jump was to apply the facts from the mini
universe that was each cluster to the remainder of the sam-
ple.
Raster modelling
Most public sector planners have at least heard of vector
and raster mapping systems. Perhaps because of the long
history of the use of vector mapping in physical planning,
where exact locations of roads, sewers or archeological
remains are points and lines, vector-based packages (like
ArcView and MapInfo) are well known. A characteristic of
vector maps is that the exact location of every house in a
given cluster or sample can be identified in a database. In
some applications, this can be useful, allowing "drill-
down" to show the street number of the house where a sit-
uation X is located. Usually these are more data about
occurrence – and less about the relationships around
occurrence – than planners need. Planners need to know
more about sets of circumstances associated with impact.
They need occurrence data on the convergence of several
different factors, rather than exquisite accuracy on any sin-
gle occurrence.
The ethics of holding highly identifiable data like this
must be questioned. Confidentiality is a well-known com-
plication of a drill-down ability, especially in this age of
remote access to information. There are detailed research
protocols that govern this from a purely ethical stand-
point, but having intensely personal details on the identi-
ties and locations of cases or exposures does pull in the
opposite direction to transparency and participation in
planning. Data cannot be made widely accessible, which
defeats some of the exercise of bringing the maps into the
planning process – to increase the stake holding in the evi-
dence.
Sample size of a cluster sample is a particular issue in vec-
tor mapping. For each polygon, sufficient clusters are
needed to provide a reasonable representation of the
domain. Aggregated over a country with several hundred
districts, this means the survey must include a huge
number of clusters (or individual households).
A raster-based or fusion raster-vector GIS package (like
Idrisi or CIETmap) essentially creates surfaces that
"drape" over a lattice of identified places from where the
data are available. This fits comfortably with a cluster sur-
vey: clusters can be likened to a lattice of "tent-poles",
each pole potentially of a different height, depending on
the value of the indicator. Draping a tent over the grid of
tent poles produces a surface that shows the different pole
heights. This surface or raster offers an opportunity to
assign different colours to different areas of the canvas,
depending on the height of the pole holding up each por-
tion. A high pole, an elevated indicator, might have a dark
tone to indicate a high level of coverage with a given mine
action. A pale shade might illustrate a low level of cover-
age.
Figure 3, a schematic representation of two rasters or sur-
faces, depicts a high pole, representing a high level of an
indicator, and a low pole. Somewhere between these two
poles, at a point determined by the relative importance of
"Ortho" or three-dimensional view of population-weighted rasterFigure 2
"Ortho" or three-dimensional view of population-
weighted raster. Distribution of households that say land
mines affect their livelihood (Afghanistan 1997).
International Journal of Health Geographics 2006, 5:1 http://www.ij-healthgeographics.com/content/5/1/1
Page 6 of 10
(page number not for citation purposes)
each in the population (the friction surface), the colour
changes from dark to lighter.
This interpolation that sets the exact point of colour
change between two clusters is done by the mapping pack-
age, but the operator should have a say in how it happens.
The raster maps generated for the Afghanistan mine
awareness evaluation originally used Idrisi with global
inverse-distance weighting (IDW). The friction layer was
then used to adjust the interpolations to take into account
the population weights. In CIETmap, an open source epi-
demiological geomatics software, an adjusted IDW for-
mula takes into account population weights during
interpolation. Users can control the influence of distance
and population weights on the resulting image.
The friction surface generated in Idrisi using the Cost mod-
ule creates an isotropic effect around each cluster. This is
a limitation of these early population weighted maps as,
in reality, population patterns would not be symmetrical
in this way. A non-isotropic function, such as Idrisi's Var-
cost, allows for effects to vary in different directions if suf-
ficient data on population movement, roads and physical
boundaries (rivers, lakes or forests) are added. In the
Afghan case, regional inferences were more important
than specific local inferences and, consequently, the
cruder picture generated by the Cost module was probably
sufficient for the purpose.
Raster-based modelling also allows several surfaces to
interact, which is where epidemiological causality analysis
comes into the picture. One type of occurrence data from
a cluster can be set in a formula that interacts it with
another datum for the same cluster. For example, Figure 3
shows a schematic depiction of one surface being draped
over an existing surface. This is what happens when an
"intervention" coverage layer is added to an "outcome"
layer, for example, the effect on behaviour of BBC listen-
ership. The resulting new layer, the modified outcome, is
based on the interaction at each cluster; the effect of listen-
ing to the BBC on the behaviour is the actual cluster-spe-
cific effect, after taking account of other possible
influences (see below).
The surfaces interact cluster-to-cluster, based on real data
from each cluster. We can also allow for the other factors
we know, like literacy and food security, that also influ-
ence how the social costs of land mines will be felt and
how the mine action will be received differently in each
cluster. As each subsequent layer is "lowered" onto the
existing surface, the cluster-specific interaction between
the layers sets the new contours of the occurrence rela-
tions (Figure 4).
Although the household data from each cluster might be
exquisitely accurate for that place, raster modelling extrap-
olates this finding radially outwards. As the result from
each cluster radiates outward, it "seeks" and interpolates
with every other result radiating out from each other clus-
ter.
It may eventually be possible to condition the outward
radiation, with a function like Idrisi's Varcost. Thus, one
might show the influence of physical or cultural barriers.
However, the model does locate the proportion of the
population represented by the cluster as geographically
contiguous to the cluster. In reality, this may not be the
case and so the interpretation for planning is "in places
like this", rather than "in this specific place". Based on a
reliable sample, stratified by region and urban/rural char-
acter as in Afghanistan, this allows the planner to make
strategic choices and to focus resources accordingly.
Levels of indicators and epidemiological analysis for
planning
Different levels of indicators summarise what planners
need to know, progressing from simple description,
through analysis of causality to modelling programme
investment. A basic indicator gives a bland description of
occurrence, the corollary of an opinion prevalence in an
opinion poll (Level 0 in our nomenclature, Table 2). This
can be the proportion of the population directly affected
by a problem, in the landmine context those who have
lost a hand or a leg, or it can be the proportion who have
been trained in mine risk education or who listened to the
BBC soap opera. Overlays of Level 0 coverage indicators
permit identification of low cover areas and high risk
areas. To a very limited extent, Level 0 indicators permit
coarse interpolation of where there is convergence of cov-
erage and effect and, therefore, a first indication of where
to focus resources to make up the shortfall. The problem
of inappropriate inferences from this sort of data has been
referred to as the "ecological fallacy" [8].
The Level 1 indicator avoids this problem by combining
exposure and outcome, to estimate the average individual
risk (odds ratio) of inappropriate behaviour of an individ-
ual who had received mine risk education. In these higher
level indicators, the concern is causality: does the expo-
sure decrease the risk of mine accidents?
Table 3 shows the evidence of impact of the BBC pro-
gramme, reflected in terms of number of cases before and
after programme commencement in late 1994, in
response to an earlier study on social costs of land mines
[9]. The 1998 data refer to the type of injury sustained and
mine-smart behaviour in fixed settlements only (exclud-
ing nomads and camps). Although it is not clear why lis-
teners had higher risks than non listeners prior to 1994,
International Journal of Health Geographics 2006, 5:1 http://www.ij-healthgeographics.com/content/5/1/1
Page 7 of 10
(page number not for citation purposes)
the evidence suggests that listening in some way neutral-
ises part of the huge increase in risk associated with the
post-1994 dramatic increase in population movement.
In addition to the protection against mine events, there
was also evidence that listening to the BBC affected mine-
smart attitudes and practices. A BBC listener was only one
half as likely to support risky behaviour or attitudes com-
pared with a non listener (among those with easy access
to a radio 377/4207 compared with 65/427; among those
with difficult access to a radio, 2/83 compared with 311/
4410; unbiased estimate OR 0.52 95%CI 0.39–0.69,
unbiased RD 5.4% 95%CI 3.1–7.7). There is notable con-
founding by ease of listenership: those who have least
access to radios also have different pressure, probably eco-
nomic, to engage in risky behaviour. To be sure that the
association was not explained by chance or one of the
many other factors that could be linked with the risk edu-
cation, the Mantel-Haenszel procedure was applied
sequentially for each of the potential explanations [10].
The final model was confirmed using logistic regression
analysis [11].
In the Afghanistan project, there were more than 70 of
these factors to be excluded as possible explanations.
None of these could explain the apparently protective
effect of the BBC's New Home New Life programme. In
contrast with the BBC programme, no such protective
effect could be found for the direct training programmes.
Such evidence as could be obtained pointed to signifi-
cantly more upper limb injuries after the direct training,
compatible with inappropriate tampering with the
devices. Direct training was also associated with more
inappropriate attitudes and practices (odd ratio 1.73,
95%CI 1.48–2.02; 329/2917 in the programme areas
reported these attitudes and practices, compared with
426/6207 in the non programme areas). This association
could not be explained by any of the possible confound-
ers taken into account in the analysis.
The implication is that stopping or modifying the direct
training programme could be expected to have a measur-
able reduction in upper body injuries and a decrease in
inappropriate attitudes and practices (Table 4). When the
effect of stopping or changing the direct training is com-
bined with the BBC programme, there is an expected pro-
tective effect, greater than that of changing the training on
its own and of the BBC on its own.
Level 2 indicators or "expected gains" are based on the
same epidemiological approach to exclude other possible
explanations for the association. Derived from the risk dif-
ference (the incidence in the exposed minus the incidence
in the unexposed), the expected gain is the central param-
eter for the public heath planner. The gain, simply put, is
the number of cases that can be "saved" – after taking into
account all the other factors that could explain the associ-
ation – by a given public health intervention. The gain is
the risk difference divided by the "proportion requiring
intervention", the proportion of the population exposed
to a risk, or the proportion without coverage in the case of
a positive intervention. The expected gain has proved use-
ful in assessing the planning importance of different inter-
ventions. Table 4 shows a completed gains table for the
case of direct training and BBC listenership influencing
mine risk education.
Level 3 indicators model the combination of actions;
what, for example, would happen if the training was made
more appropriate and everyone could receive a battery-
Raster map showing Level 0 indicatorFigure 4
Raster map showing Level 0 indicator. Population-
weighted distribution of listenership to BBC soap opera New
Home New Life including ortho (three-dimensional) view.
Schematic portrayal of a raster being laid over a weight layerFigure 3
Schematic portrayal of a raster being laid over a
weight layer. The interpolation of the colour change is
based on the population weight.
International Journal of Health Geographics 2006, 5:1 http://www.ij-healthgeographics.com/content/5/1/1
Page 8 of 10
(page number not for citation purposes)
free windup radio, to listen to the radio soap-opera. But of
course these potential combined gains are seldom real-
ised. Rarely can one afford to invest fully in all pro-
gramme options, everywhere, at the same time. It is also
wasteful to do so, since some things only work when other
things have happened first.
Level 4 indicators are the modelled investment options,
asking "what if" questions. This highest level indicator re-
grounds the model in a reality of costs (how much money
is available to buy radios) and in possible failure rates (for
example, what if only 50% of the trainers give up training
about technical specifics of land mines and embrace
"mine-smartness" as an approach).
Part of the challenge of introducing maps into epidemiol-
ogy is their standardisation for consistent interpretation.
After considerable field testing, CIET settled on three col-
our gradients that show up adequately with grey scales,
where colour printers are not available. For level zero indi-
cators, coverage is portrayed by shades of green (Figure 4);
the level of an outcome indicator, like behaviour, is
reflected by shades of yellow through brown. Level 1 indi-
cators, reflecting relative risk, are shown by shades of blue.
CIET colour schemes rely on the principle that darker
shades refer to areas in need of attention or investment.
"Morphing" the maps
An interesting output of epidemiological geomatics is the
sequence of predictions based on a sample of local
impacts. One might begin with a baseline occurrence, for
example, the place-specific land mine incident rates, or
the rate of mine-smartness (conditioned risk-avoidance
behaviour). One might overlay this with a map of place-
specific impacts – essentially a risk difference – after tak-
ing account of the possible confounders. Each of these
maps is generated in the GIS software (initially Idrisi, then
CIETmap). The resulting prediction scenario shows the
revised population weighted rates of the occurrence in
question. Flipping between the scenarios allows the plan-
ner to visualise the likely shift in the occurrence rate with
a given investment strategy. Alternative strategies can be
modelled, and the predicted effects compared.
Conclusion: mapping is not enough
The Afghanistan evaluation produced evidence that the
BBC soap opera worked while the mine training pro-
gramme was in serious need of revision. The only way the
programme managers could know this, was being pre-
pared to ask the tough question. The evaluation revealed
that mine risk education programmes in Afghanistan
underachieved substantially, despite possibly being well
run. The traditional direct training curriculum and educa-
tional delivery approaches, and the supervisory approach
to project management, fell short of the challenge. By the
same token, there were measurable local successes to
build upon, and best practice cases that could be held up
as models to be reproduced elsewhere.
The evaluation has already had a positive effect in Afghan-
istan. The focus of mine risk education training has shifted
from technical description of mines to behaviour around
mines, direct training programmes have been overhauled,
and the concept of a mine committee has been reassessed.
The next step is to measure and to map the improvements
produced by all these changes. As an ongoing decision-
making tool, a place-based approach to planning mine
action could facilitate evolution of the Afghanistan pro-
gramme, helping it respond in a more agile way to accom-
modate the continuous changes in the mine/UXO
situation in the country – and the community responses
to these changes.
The epidemiological geomatics exercise in Afghanistan
was not an unqualified success. The factors influencing
the success of a place-based approach in other settings
[12] were also at work here. For example, the unifying
authority has to be able to recognise and to deal with the
occurrence relations in identified types of places. Use of
UN regions of humanitarian assistance to stratify the sam-
ple should have supported incorporation of the maps in
the planning by the contracting agency – but the UN
Table 2: Five levels of indicators for planning
Level 0: Descriptive frequencies (percentages, rates) and characteristics (averages, modal values) provide an overview of the occurrence a given
risk-taking behaviour, or the coverage of a given health programme.
Level 1: Individual risk estimates (unbiased odds ratios) reflect the average risk of an individual "exposed" to a given mine awareness programme, in
comparison with the average unexposed individual.
Level 2: The expected gains are the number of cases that can be "saved" – after taking into account all the other factors that could explain the
association – by a given intervention.
Level 3: Combinations of programmes can produce additive or multiplicative gains, which are enormously important in estimating the cost options
for planners.
Level 4: Since is it rarely possible to invest fully in all programme options simultaneously, this models programme structure based on partial
investment. It also anticipates partial uptake of the programme and local cost variability
International Journal of Health Geographics 2006, 5:1 http://www.ij-healthgeographics.com/content/5/1/1
Page 9 of 10
(page number not for citation purposes)
involvement in mine action was limited, despite its self-
proposed coordination role. An emerging lesson is that
the geographic domain of the maps must coincide with
the domain of the unifying authority.
Maps can be useful in demonstrating and promoting a
common stake holding, but their ultimate value depends
on the genuineness of the development process. Subse-
quent developments (and reversals) in Afghanistan indi-
cate show the relative unimportance of planning, with or
without epidemiological geomatics, in this turbulent set-
ting.
Despite these failings, we were able to fulfil the epidemi-
ological imperative to take account of other possible
explanations of association before beginning causal infer-
ences or making predictions based on the evidence. The
maps were based on high quality epidemiological evi-
dence, meaning data accuracy, resolution (the level of
detail it offers) and currency (the usefulness in getting the
job done). The sampling process allows one, within the
limits of scientific probity, to generalise the findings of a
sample survey beyond the sample to the rest of the popula-
tion. Our ability to hook the last-stage random sample of
the population to the exact spatial location of the cluster
gives a useful coherence to the epidemiological models.
The case study illustrates the feasibility of epidemiological
geomatics. Under better conditions, more reliable spatial
data should be available, they should be more easily
accessed and analytical skill levels may be higher. Hope-
fully the capacity of the contracting agencies to absorb the
results might also be higher in other settings.
There is nothing new about the idea that, just as a sample
can produce an interpretable model with summary pre-
dictive statistic, a sample can also be portrayed on a map
as a spatial model of what happens and in what type of
place it happens. The episode of the famous Broad Street
pump and London's cholera epidemic probably had more
to do with the careful mapping of cases and inferences
from the spatial distribution of cases about the source,
than it did with a seemingly definitive removal of a pump
handle.
In more recent years, we have become quite accustomed
to seeing maps as presentation and planning devices. A
well-known example linked to prediction is the weather
raster map, where colourful bands spread across a recog-
nisable geographic domain. These rasters are models
based on a sample of empirical observations, now made
more comprehensive with satellite technology, projected
forward in time in an attempt to predict the weather. With
population weighted raster maps we can include epidemi-
ological analysis in the presentation, allowing planners to
visualise specific models of affected populations.
Population weighted raster maps overcome many of the
challenges of using maps in epidemiology – and epidemi-
ology in maps. While vector maps rely on polygon bound-
aries to represent data, raster maps are a grid of cells
representing data independently from a usually arbitrary
boundary (like a district or region). Raster maps can be
used with a representative sample and the information
can be interpolated from the sample locations to create a
surface of an indicator on the map. The sample allows for
larger amounts of specific data to be collected and fol-
lowed through time at relatively low cost. The inclusion of
population weights (affecting how much influence each
sample location has upon the overall picture) allows the
resulting maps to represent populations, rather than area,
an important consideration for planners. Since the inter-
polation process is independent from arbitrary bounda-
ries the data on the maps is continuous in nature – a trend
which is more consistent with most population health
patterns.
Table 3: Evidence of impact of New Home New Life on the number of mine/UXO events before and after 1994, upper limb (tampering)
injuries after 1994 and inappropriate attitudes and practices
Listen to New Home New Life
yes no
Mine/UXO events
1. Cases 1979–93 (per 10000 person-years)
21 (6.802) 9 (1.898)
2. Cases 1994–97 (per 10000 person years)
Risk of a case occurring after 1994: OR 0.28, 95%CI 0.07–0.99 (in urban mined areas)
9 (14.577) 14 (14.763)
3. Upper limb injuries as proportion of all injuries among survivors of mine accidents after 1994
Risk of an upper limb injury after 1994: OR 0.12, 95%CI 0.02–0.82 (in urban areas)
3/14 (21%) 11/16 (69%)
Inappropriate attitudes and practices (search for metal, consider non-qualified person brave if go into mined
area)
4. With easy radio access (%) 377/4207 (9%) 65/423 (15.4%)
5. Difficult radio access (%)
Risk of inappropriate behaviour, stratified by ease of access to radios OR 0.61, 95%CI 0.44–0.84
2/83 (2.4%) 378/4216 (9%)
International Journal of Health Geographics 2006, 5:1 http://www.ij-healthgeographics.com/content/5/1/1
Page 10 of 10
(page number not for citation purposes)
The complexity of calculations behind the maps does not
make up for the quality of the original data. Instead of tak-
ing meta-data (where and how the evidence was col-
lected) for granted, an expert layer can to weight the
confidence in a given layer of data.
Epidemiological geomatics offer a powerful communica-
tion and mobilisation tool. Its power comes from the vis-
ual summary of (i) the population weighting of the
detailed place-specific data in a scientifically coherent
sample; (ii) higher level indicators that move beyond sim-
ple description to occurrence relations and (iii) the epide-
miological discipline of ruling out other explanations in
causality. Instead of simply reflecting the locations, epide-
miological geomatics maps the occurrence relations or cir-
cumstances for changing risk and resilience. This is what
planners need to know even when they may not have the
full training to understand all the science that goes on
behind the maps in a predictive sequence.
Until recently, a trained GIS technician was required even
to produce simple maps. Linking of maps with epidemio-
logical analysis was beyond the reach of most develop-
ment agencies. With the advent of user-friendly
epidemiological geomatics software, the situation now is
reminiscent of the field of statistics a few decades ago,
when the service of a statistician was the sine qua non of
almost all research. Since then, incorporation of statistics
courses in most university careers has transformed the dis-
cipline of statistics. Perhaps the same will be true of epide-
miological geomatics, as planners begin to demand it to
help them to address their most pressing needs. We hope
that population weighted raster maps will serve as a useful
tool for evolution of the field of epidemiological geomat-
ics.
Authors' contributions
NA developed the concepts of population weighted raster
maps and designed CIETmap, the software able to imple-
ment this. He is author of the CIETmethods, and designed
and oversaw implementation of the Afghanistan Mine
Awareness Evaluation.
SM implemented the maps initially with Idrisi and then
with CIETmap. He coauthored the technical specifications
of CIETmap and oversaw the production of the software.
Additional information is available on this from http://
www.ciet.org/cietmap.
Acknowledgements
The Afghanistan Mine Awareness Evaluation was financed by the United
Nations Office for Coordination of Humanitarian Assistance to Afghanistan
and conducted by CIET, with fieldwork coordinated by Charlie Whitaker
Dixon.
References
1. Busgeeth K, Rivett U: The use of a spatial information system
in the management of HIV/AIDS in South Africa. International
Journal of Health Geographics 2004, 3(13):.
2. NCHS: Atlas of United States Mortality Centers for Disease Control
Atlanta; 1997.
3. Andersson N, Whitaker C, Swaminathan A: Afghanistan, the 1997
National Mine Awareness Evaluation. Project report. CIET/UNO-
CHA Islamabad 1998 [http://www.ciet.org
].
4. Andersson N, Swaminathan A, Whitaker C, Roche M: Mine Smart-
ness and the community voice in minerisk education: lessons
from Afghanistan and Angola. Third World Quarterly 2003,
24(5):873887.
5. Andersson N, Martinez E, Cerrato F, Morales E, Ledogar RJ: The
Use of CommunityBased Data in Health Planning in Mexico
and Central America. Health Policy and Planning 1989,
4(3):197206.
6. Ledogar RJ, Andersson N: Impact Estimation Through Sentinel
Community Surveillance: An affordable epidemiological
approach. Third World Planning Review 1993, 15(3):263-272.
7. CIETinternational/EDI/CIDA: Survey of corruption in the police,
judiciary, revenue and lands services. Presidential Commission of
Inquiry against Corruption. Dar es Salaam 1996.
8. Morgenstern H: Uses of ecologic analysis in epidemiologic
research. American Journal of Public Health 1982, 72:1336-1344.
9. Andersson N, Palha C, Paredes S: Social costs of land mines in
four countries. British Medical Journal 1995, 311:718-721.
10. Mantel N, Haenszel W: Statistical aspects of the analysis of data
from retrospective studies of disease. Journal of the National Can-
cer Institute 1959, 22:719-748.
11. Hosmer DW, Lemeshow S: Applied logistic regression New York: John
Wiley & Sons; 1989.
12. US Environmental Protection Agency: Towards a placedriven approach:
the Edgewater Consensus on an EPA strategy for ecosystem protection
EPA; 1994.
Table 4: Expected gains table (reduction in inappropriate behaviour) of BBC listenership and direct training
Actions Odds Ratio
(average individual risk)
Risk
difference
Proportion requiring intervention Gain**/1000
households
Universal listenership to the BBC
New Home New Life soap opera*
0.61 0.046 53% current non-listeners 24.3
Stop or change current direct
training programme*
0.57 0.044 32% in communities now exposed 14.1
Universal listenership and
change
direct training
0.52 0.066 67% is either non-listener or is exposed to training 44.2
*each excluding the effect of the others and the confounding effect of 'easy access to radios'
** 'Gain' is the risk difference divided by the proportion requiring intervention
... Analysis relied on CIETmap software, 21 which provides a user interface with the R statistical language. We compared the frequency of socio-economic variables in the video and non-video groups of women (maternal outcomes), children aged 12 to 18 months (child health outcomes) and male spouses (male knowledge outcomes). ...
Article
Full-text available
Objective A trial of evidence-based health promotion home visits to pregnant women and their spouses in northern Nigeria found significant improvements in maternal and child health outcomes. This study tested the added value for these outcomes of including video edutainment in the visits. Methods In total, 19,718 households in three randomly allocated intervention wards (administrative areas) received home visits including short videos on android handsets to spark discussion about local risk factors for maternal and child health; 16,751 households in three control wards received visits with only verbal discussion about risk factors. We compared outcomes between wards with and without videos in the visits, calculating the odds ratio (OR) and 95% confidence interval (95%CI) of differences, in bivariate and then multivariate analysis adjusting for socio-economic differences between the video and non-video wards. Results Pregnant women from video wards were more likely than those from non-video wards to have discussed pregnancy and childbirth often with their husbands (OR 2.22, 95%CI 1.07–4.59). Male spouses in video wards were more likely to know to give more fluids and continued feeding to a child with diarrhoea (OR 1.61, 95%CI 1.21–2.13). For most outcomes there was no significant difference between video and non-video wards. The home visitors who shared videos considered they helped pregnant women and their spouses to appreciate the information about risk factors. Conclusion The lack of added value of the videos in the context of a research study may reflect the intensive training of home visitors and the effective evidence-based discussions included in all the visits. Further research could rollout routine home visits with and without videos and test the impact of video edutainment added to home visits carried out in a routine service context.
... Subsequently, we will calculate fuzzy transitive closure to establish the strength of the influence that one node has on others through direct or indirect relationships [14]. We will use the free software CIETmap to apply this model [15]. The fuzzy transitive closure maps will be transferred to the freeware yEd [16] for visualization and further analysis. ...
Article
Full-text available
Introduction Academic fraud is any behavior that gives a student an undeserved advantage over another student. Few studies have explored the causes of and possible solutions to academic fraud in Latin America. We aim to map the knowledge of stakeholders in a Colombian faculty of medicine about the factors that facilitate and prevent academic fraud. Methods Fuzzy cognitive mapping. We will use the approach proposed by Andersson and Silver to generate fuzzy cognitive maps representing stakeholder knowledge. This process consists of ten steps: (1) definition of the research question; (2) identification of participants; (3) generation of ideas; (4) rationalization of ideas; (5) organization and connection of ideas; (6) weighing; (7) pattern grouping; (8) list of links and digitization; (9) combination of maps and network analysis; and (10) deliberative dialogue. To draw the maps, we will invite medical students, interns, resident physicians, master’s students, and professors in the faculty of medicine. Four medical students will receive training to facilitate the sessions. Participants will identify the factors contributing to academic fraud and their causal relationships. We will use a combination of network analysis and graph theory to identify the chains of factors with greatest influence on academic fraud. Conclusion The maps will serve to discuss strategies to reduce academic fraud in the Faculty of Medicine and to identify factors that could be addressed in other contexts with similar problems. This research will allow the students who facilitate mapping sessions to learn about research techniques, fuzzy cognitive mapping and academic fraud. Study registration: Registered in OSF Registries on August 2nd, 2022. Registration number: osf.io/v4amz
... 46 For each individual map reporting the relationships identified in a study, we will calculate fuzzy transitive closure in the open access software CIETmap V.2.0 (step 3 in figure 2). 52 Fuzzy transitive closure is a mathematical model used to calculate the influence of each Open access relationship on community participation, considering all the possible relationships represented in the map. 53 54 After transitive closure, each relationship will have a value between 0 (having no influence) and 1 (having the strongest influence) to represent the relative strength of their influence on community participation, with positive and negative signs indicating whether the relationship is stimulative or inhibitive. ...
Article
Full-text available
Introduction As the number of people living in cities increases worldwide, particularly in low-income and middle-income countries (LMICs), urban health is a growing priority of public and global health. Rapid unplanned urbanisation in LMICs has exacerbated inequalities, putting the urban poor at increased risk of ill health due to difficult living conditions in cities. Collaboration with communities in research is a key strategy for addressing the challenges they face. The objective of this scoping review is, therefore, to identify factors that influence the participation of urban communities from LMICs in public and global health research. Methods and analysis We will develop a search strategy with a health librarian to explore the following databases: MEDLINE, Embase, Web of Science, Cochrane, Global Health and CINAHL. We will use MeSH terms and keywords exploring the concepts of ‘low-income and middle-income countries’, ‘community participation in research’ and ‘urban settings’ to look at empirical research conducted in English or French. There will be no restriction in terms of dates of publication. Two independent reviewers will screen and select studies, first based on titles and abstracts, and then on full text. Two reviewers will extract data. We will summarise the results using tables and fuzzy cognitive mapping. Ethics and dissemination This scoping review is part of a larger project to be approved by the University of Montréal’s Research Ethics Committee for Science and Health in Montréal (Canada), and the Institutional Review Board of the James P Grant School of Public Health at BRAC University in Dhaka (Bangladesh). Results from the review will contribute to a participatory process seeking to combine scientific evidence with experiential knowledge of stakeholders in Dhaka to understand how to better collaborate with communities for research. The review could contribute to a shift toward research that is more inclusive and beneficial for communities.
... Para el análisis de los datos se utilizó el programa estadístico CIETmap. 29 Se calcularon las frecuencias simples de las variables relevantes. Los factores de asociación se midieron mediante análisis bivariado con el procedimiento del Mantel-Haenzel 30 que calcula la razón de momios (RM) para estimar la magnitud del efecto y la asociación entre cada variable independiente. ...
Article
Full-text available
Background Anaemia is a significant global health problem, especially, in developing nations like Ethiopia. Despite increasing rates over the past two decades, there is limited research on the specific prevalence of anaemia among pregnant women in the country. Objective To identify hotspot areas of anaemia-associated factors among pregnant women in Ethiopia. Study design Cross-sectional. Setting Ethiopian demographic study from 2005 to 2016. Participants This study analysed 3350 pregnant women. Primary and secondary outcome measures Hotspot area of anaemia among pregnant women, trend of anaemia and associated factors. Results The prevalence of anaemia among pregnant women has shown significant fluctuations over the years. Between 2005 and 2011, there was a notable decrease from 30.9% to 21.5% while the prevalence increased from 21.5% in 2011 to 29.58% in 2016. The identified determinants of anaemia among pregnant women were female-headed household, belonging to the highest wealth quintile, being in the second or third trimester of pregnancy, being a working woman and residing in the Somalia region. Hotspot areas, where the prevalence of anaemia was particularly high, were identified in Somalia, Dire Dawa, Afar and Harari regions. Conclusion Anaemia during pregnancy is a major public health concern in Ethiopia, with a concerning increase between 2011 and 2016. Hotspot areas like Somali, Dire Dawa, Afar and Harari are particularly affected. Shockingly, nearly one in three pregnant women in Ethiopia suffer from anaemia. To address this issue effectively, targeted interventions prioritising economically disadvantaged households and pregnant women in their second and third trimesters are crucial. Monitoring spatial patterns and contributing factors is vital to develop tailored interventions and improve maternal health outcomes in these high-risk areas. By strategically targeting hotspot areas nationwide, significant progress can be made in reducing anaemia among pregnant women.
Article
Background Socio-economically disadvantaged women have poor maternal health outcomes. Maternal health interventions often fail to reach those who need them most and may exacerbate inequalities. In Bauchi State, Nigeria, a recent cluster randomised controlled trial (CRCT) showed an impressive impact on maternal health outcomes of universal home visits to pregnant women and their spouses. The home visitors shared evidence about local risk factors actionable by households themselves and the program included specific efforts to ensure all households in the intervention areas received visits. Purpose To examine equity of the intervention implementation and its pro-equity impact. Research design and study sample The overall study was a CRCT in a stepped wedge design, examining outcomes among 15,912 pregnant women. Analysis We examined coverage of the home visits (three or more visits) and their impact on maternal health outcomes according to equity factors at community, household, and individual levels. Results Disadvantaged pregnant women (living in rural communities, from the poorest households, and without education) were as likely as those less disadvantaged to receive three or more visits. Improvements in maternal knowledge of danger signs and spousal communication, and reductions in heavy work, pregnancy complications, and post-natal sepsis were significantly greater among disadvantaged women according to the same equity factors. Conclusions The universal home visits had equitable coverage, reaching all pregnant women, including those who do not access facility-based services, and had an important pro-equity impact on maternal health.
Article
Full-text available
Introduction Cultural safety, interculturality and antiracism are crucial concepts in addressing health disparities of minority and diverse groups. Measuring them is challenging, however, due to overlapping meanings and their highly contextual nature. Community engagement is essential for evaluating these concepts, yet the methods for social inclusion and protocols for participation remain unclear. This review identifies experimental studies that measure changes resulting from culturally safe, intercultural or antiracist healthcare. The review will describe outcomes and additional factors addressed in these studies. Methods and analysis The study focuses on epidemiological experiments with counterfactual comparisons and explicit interventions involving culturally safe, intercultural or antiracist healthcare. The search strategy covers PubMed, CINAHL, Scopus, Web of Science, ProQuest, LILACS and WHO IRIS databases. We will use critical appraisal tools from the Joanna Briggs Institute to assess the quality of randomised and non-randomised experimental studies. Two researchers will screen references, select studies and extract data to summarise the main characteristics of the studies, their approach to the three concepts under study and the reported effect measures. We will use fuzzy cognitive mapping models based on the causal relationships reported in the literature. We will consider the strength of the relationships depicted in the maps as a function of the effect measure reported in the study. Measures of centrality will identify factors with higher contributions to the outcomes of interest. Illustrative intervention modelling will use what-if scenarios based on the maps. Ethics and dissemination This review of published literature does not require ethical approval. We will publish the results in a peer-reviewed journal and present them at conferences. The maps emerging from the process will serve as evidence-based models to facilitate discussions with Indigenous communities to further the dialogue on the contributing factors and assessment of cultural safety, interculturality and antiracism. PROSPERO registration number CRD42023418459.
Article
Full-text available
The cycle community-based measurement of impact using epidemiological methods provides a dynamic and inexpensive option for use in planning outside the field of health. Seven years' experience with Sentinel Community Surveillance in Mexico, Central America and Africa provides the experimental basis for this paper. The application of the method to physical planning is described, and two recent African experiences illustrate its use in measuring some effects of structural adjustment and detecting gender differentials in education. By separating and analysing the components of risk, the methodology provides the detail necessary to make informed judgements concerning the potential impact of planning decisions. It provides a voice for communities in such decisions while also having sufficient scientific rigour that its conclusions can interact with existing formal measurement systems. It is particularly suited to decentralized planning. -Authors
Article
Full-text available
Introduction to the Logistic Regression Model Multiple Logistic Regression Interpretation of the Fitted Logistic Regression Model Model-Building Strategies and Methods for Logistic Regression Assessing the Fit of the Model Application of Logistic Regression with Different Sampling Models Logistic Regression for Matched Case-Control Studies Special Topics References Index.
Article
Full-text available
Sentinel site surveillance methodology is used in Nicaragua, Honduras, El Salvador, Guatemala, Panama and in the Mexican state of Guerrero to collect reliable planning-oriented health information at community level. Limited resources available for health measurement can be concentrated in these sites which, when taken together, reflect the diversity of health conditions; some sites include health facilities and others do not, in proportion with the coverage of the population at large. Changes in health status over time have been measured in relation to coverage with health care activities, using repeatable problem-specific cycles of household interviews. A key consideration is the development of methodologies and tools to facilitate the decentralization of capabilities of questionnaire design, and the organization and analysis of data. Infant and child mortality are key concerns, but the methodology has also been used to investigate other mortality and morbidity in relation to coverage and costs of health services, and knowledge, attitudes and practices related to priority diseases at the community level. These data cannot easily be obtained from sources other than community surveys, and the sentinel site framework allows for their collection in a manner that facilitates systematic planning under conditions of limited resources, including limited capabilities of handling data, and of making it available for planning.
Article
Full-text available
Mine-risk education programmes will fall short of their intended impact for as long as they fail to take into account local responses—knowledge, logic and everyday practices—to mine threats. Community information, systematically collected through household and institutional surveys, can help to define and understand endogenous ‘mine smartness’. The same evidence provides insight into the impact of mine-risk education, including its unintended consequences. Using six criteria of mine smartness, ciet carried out evaluations of mine-risk education in Afghanistan (1997) and Angola (1999). The first clear lesson to be drawn from these evaluations is that people in mine-affected areas do generate their own broadly effective means of facing the daily threat of mines. The second lesson is that people take risks for reasons that make sense to them: ‘education’ that landmines are dangerous probably adds little value for them. The third lesson is that mine-risk education that does not take into account these first two lessons can cause harm. The evaluations produced evidence of unintended risk-taking by people exposed to mine-risk education programmes.
Book
From the reviews of the First Edition."An interesting, useful, and well-written book on logistic regression models . . . Hosmer and Lemeshow have used very little mathematics, have presented difficult concepts heuristically and through illustrative examples, and have included references."—Choice"Well written, clearly organized, and comprehensive . . . the authors carefully walk the reader through the estimation of interpretation of coefficients from a wide variety of logistic regression models . . . their careful explication of the quantitative re-expression of coefficients from these various models is excellent."—Contemporary Sociology"An extremely well-written book that will certainly prove an invaluable acquisition to the practicing statistician who finds other literature on analysis of discrete data hard to follow or heavily theoretical."—The StatisticianIn this revised and updated edition of their popular book, David Hosmer and Stanley Lemeshow continue to provide an amazingly accessible introduction to the logistic regression model while incorporating advances of the last decade, including a variety of software packages for the analysis of data sets. Hosmer and Lemeshow extend the discussion from biostatistics and epidemiology to cutting-edge applications in data mining and machine learning, guiding readers step-by-step through the use of modeling techniques for dichotomous data in diverse fields. Ample new topics and expanded discussions of existing material are accompanied by a wealth of real-world examples-with extensive data sets available over the Internet.
Article
Despite the widespread use of ecologic analysis in epidemiologic research and health planning, little attention has been given by health scientists and practitioners to the methodological aspects of this approach. This paper reviews the major types of ecologic study designs, the analytic methods appropriate for each, the limitations of ecologic data for making causal inferences and what can be done to minimize these problems, and the relative advantages of ecologic analysis. Numerous examples are provided to illustrate the important principles and methods. A careful distinction is made between ecologic studies that generate or test etiologic hypotheses and those that evaluate the impact of intervention programs or policies (given adequate knowledge of disease etiology). Failure to recognize this difference in the conduct of ecologic studies can lead to results that are not very informative or that are misinterpreted by others.