Multi-Criteria Evaluation:
A Validated Methodology
for Modelling the Geography of Tuberculosis Risk
Summary
Multi-criteria evaluation (MCE) was used to develop
a model of the current tuberculosis risk in the Lower Mainland in order to
develop and validate Vancouver-specific factor weightings for use in future
risk mapping. Factors were deliberately selected to match the current distribution
of TB. This will provide the research community with a set of validated, Vancouver-specific
tuberculosis risk factors which can then be applied to future risk maps based
on subsequent census data.
Strategy
An index of socio-economic and medical factors
associated with higher rates of tuberculosis was compiled. These factors were
subsequently operationalized to power a multi-criteria evaluation (MCE) to
develop an overall tuberculosis risk map displaying the propensity of different
areas in the Lower Mainland to developing tuberculosis rates in the near future.
Although substantial research has been conducted liking individual socio-economic
factors to population-level tuberculosis rates, attempts to integrate these
factors through a weighting process has, to our knowledge, been thus far limited
if not inexistent. Additionally, even if such geo-demographic factor weightings
had been developed elsewhere, they would be location-specific and thus would
not be directly application to the Vancouver region.
Modelling At-Risk Populations
Using MCE
MCE is a decision support method that combines
multiple spatial factors to create a "suitability map" based on
the combined weightings of these factors. In terms of the spatial epidemiology
of tuberculosis in the Greater Vancouver regions, the "suitability"
of areas for tuberculosis risk will be modelled, with the most "suitable"
areas representing areas with the highest risk.
This exercise will be performed using Idrisi32's built-in MCE module.
The process of performing an MCE is theoretically
quite simple. Census variables (factors) contributing to tuberculosis risk
are indexed, with coverages converted to raster grids. These files, once standardized
to a set scale of 0-255, are 'weighted' according to the degree that they
influence actual tuberculosis rates. The result will be a surface describing
the geography of tuberculosis risk in Greater Vancouver. While operationalizing
certain census variables such as the incidence of low income or educational
attainment is straightforward, choosing non-direct census variables to stand
as indicators of things such as residential overcrowding or of recent homelessness
is much more complicated and arbitrary.
Factor Selection
Our literature review identified a number of medical
conditions which predispose populations to higher rates of tuberculosis. Medical
conditions identified as predisposing populations to higher tuberculosis incidence
tend to be conditions or diseases that serve to weaken the body's immune system.
Being underweight, under five years of age, or having undergone a recent transplant
all imply a weakened immune system, and thus greatly increase the chances
of contracting active TB. The incidence of AIDS makes this issue that much
more acute.
Obtaining georeferenced data on the incidence of
these medical conditions at a meaningful resolution, however, proved to be
an impossible task within the confines proscribed by this study. Also, many
of these conditions serving as factors suffer from severe undercounting problems,
such as HIV/AIDS infection rates, and their inclusion would make any resulting
data highly suspect.
It seems likely that the above-tabulated medical risk factors were developed
so as to assist medical staff in assessing individual likelihood of contracting
active TB, rather than to model entire populations. A physician knows whether
or not her patient has a history of, for example, renal conditions, or is
HIV positive. She could then use these medical factor weightings to assess
that individual's likelihood of contracting active TB. However, because of
the lack of spatial data on the incidence of these medical predisposing conditions,
medical factors contributing to tuberculosis had to be dropped from the MCE.
Selecting Socio-economic Factors
Socio-economic variables identified in the literature
as exposing individuals to heightened risk for contracting TB are
- foreign-born
- low income, "severe economic dislocation" (Cdn Tuberculosis
Standards)
- incidence of unemployment within 2 yrs prior to diagnosis (Bishai et al)
- high drug abuse and crime rates (Bishai et al)
- homelessness within 1 year of diagnosis (Bishai et al)
- "lack of adequate housing" (interpreted as 'overcrowding')
- "membership in low socio-economic brackets" (Bishai et al) (interpreted
as
income and educational status)
- a history of previous tuberculosis incidents (Cegielski et al)
These factors occur at different frequencies and
vary over time and space. Thus the importance of using the current distribution
of TB and validating them for Vancouver.
Operationalizing Factors
Operationalizing socio-economic factors proved
to be somewhat of a challenge, as this was accomplished by selecting certain
census variables to be used as indicators of the socio-economic factors we
are interested in. For example, "lack of adequate housing" was interpreted
as "overcrowding," and operationalized by mathematically manipulating
census variables related to population density, and the average number of
people per bedrooms in occupied dwellings. This process resulted in ten factors,
which were scaled on an index from 0 to 255.
Factor Reduction
Running an MCE based on ten factors was a monumental
undertaking, and proved to be too great a task in view of our limited computing
power. As opposed to a ten-variable MCE, certain factors were dropped in the
interests of conserving computing power and processing time. Using fewer variables
will also make the process easier to replicate by future users, contributing
to the overall utility of this section.
Several of the ten original factors indicated similar
causal pathways and are therefore redundant. In order to simplify the analysis,
factors displaying high correlations with tuberculosis incidence were chosen.
Effectively, the ten original factors were reduced to four which were simpler
to work with, and represented a good compromise between selecting a low enough
number of factors to make this operation computationally possible, while at
the same time remaining true to the initial socio-economic variables.
Factor Weighting
After selecting the four factors to be used in
the MCE, identified as [XLOEDUC], [XBRD], [XIM], [XAI], several multi-criteria
evaluations were iteratively performed to derive factor weightings that best
described the current distribution of the tuberculosis population in Greater
Vancouver. Ten iterations were necessary to arrive at the final result, discussed
at the end of this section.
Each run was tested for appropriateness, or 'fit.'
in several ways. Firstly, a brief visual inspection was made. Closer to the
final stages of the factor selection, the MCE surface was mathematically compared
with a surface derived from the density of cases in the Greater Vancouver
area.
Model Evaluation
Evaluating Run 10: The selected factor weightings
The factor weightings developed in run 10 of the
MCE modelling exercise provide an excellent approximation of tuberculosis
risk in the Greater Vancouver area, with the notable exception of Strathcona,
where, because of the lack of suitable census data for many variables, tuberculosis
risk was greatly underestimated.
The selected weightings are:
[XLOEDUC] 0.3
[XBRD] 0.3
[XIM] 0.6
[XAI] 1.0
These were arrived at through an iterative procedure
outlined in the previous section, and represent one of many potential tools
for the approximation of tuberculosis risk in Greater Vancouver. Because the
model is necessarily Vancouver-specific, it should not be directly applied
to other metropolitan areas verbatim.
Run 10: a generally excellent match, although
due to census undercounting, the model is unreliable on the Downtown Eastside.
Conclusions
Medical conditions contributing to high tuberculosis
rates may not vary significantly among populations, as there likely is not
as much variation in human physiology as in human cultural patterns. Acquisition
of medical data was not possible, and thus could not be integrated in the
development of a risk map. As a result, the best alternative path to developing
a validated model of tuberculosis risk for Greater Vancouver was to rely on
socio-economic factors, which (unlike medical factors) have a higher likelihood
of varying significantly between metropolitan areas.
Overall, the incidence of low income seemed to
be a strong contributing factor to the propensity of populations to develop
high rates of tuberculosis. However, reliance on income alone led to a very
weak model, and other factors, such as low educational attainment, recent
immigration, and residential overcrowding (as indicated by the average number
of people per bedroom) also contribute to rates of tuberculosis, and should
be included in the risk modelling. It is also important emphasize that socio-economic
indicators such as low income in and of themselves do not expose individuals
to tuberculosis risk; rather, low income, for example, is a proxy for lack
of access to proper medical care, the inability to secure adequate housing,
etc.
Note that the "risk model" produced in
this section is trivial. It was deliberately designed to match the current
dataset, so as to provide a validated model and a set of Vancouver-specific
factors from which to develop future risk maps upon the release of the 2001
census.
Methodology
& Analysis
Visualization | Demographic
Profiling | MCE | Spatial Statistics
Home
| Introduction | Background
Objectives | Data
| Analysis | Error | Conclusions
Links | References
| Contact