The alternate parsimonious non-index model allowed for further comparison and validation of the WADI model by producing a regression coefficient that was significant. The non-index model consisted of all the components of the WADI from the social, ecological and
environmental factors at equal value, unlike the WADI model which weighted each of the components. The non-index multivariate model highlights whether by use of an index model explanatory power is lost as the index uses a methodology which takes each component and normalizes them. As a result, with all of the inputs are combined within exposure and susceptibility, it is difficult to determine which component or components can explain the results. Using a non-index multivariate regression produces a coefficient for each independent variable and which uncovers significant components and the level of the significance. The results from the regression also suggested a moderate fit of the non-index model, there was an association between the index values which evaluated vulnerability using the incidence of disease as a proxy for vulnerability, like the WADI model. As all the components in the non- index were weighted equally, they were not all of equal value to the outcome.
The non- index model provided further information regarding the components that were non- essential to assessing vulnerability in Dominica. The components such as temperature, land use, age, poverty, and unimproved water source were deemed to be essential in the literature (Dickin
et al 2013; Dickin and Schuster 2014; Fullerton et al 2014) and creators of the WADI to assessing vulnerability to an increased risk of dengue transmission in Dominica, whereas the precipitation, population density, household dengue control (education), were deemed to be non- essential (Dickin et al 2013; Dickin and Schuster 2014; Fullerton et al 2014). The model was tested using the same statistical methods as the WADI and was found to also be moderately significant with the new configuration of the social, ecological and environmental factors.
The non-index model was created to serve as a comparison to the index model, as mentioned above. The main critique of index models like the Water Associated Disease Index is in the computing of the index, which places the variables into a normalized or categorized value between zero and one and in the a priori selection of variables without analysing the data. This may seem like it does not allow for determining which factors are actually the major players and the categorization and normalization may be seen as altering the data.
The results of the multivariate regression of the non-index model gave a number of revelations that were contrary to the literature with the values for land use, population density, age, and poverty having negative coefficients. The negative coefficients convey that as there is an
increase in the value of these variables, there would subsequently be a decrease in the number of cases of dengue.
This indicates that rural areas with low population density should have the largest number of dengue cases, as opposed to the expected high density, highly urban areas posited by researchers into the dengue virus. As a result, this is contrary to the findings in the literature and,
interestingly, the actual number of cases per parish in the study area. The answer to the anomaly may be due to the data sets used to populate the models. The percent urbanization data set was
the proxy indicator for the land use per parish component and was calculated using GIS technology. This process, as mentioned in the methodology section, is well documented in research papers for health geography such as Chang et al 2009. Drawing a polygon over the Google Earth image to determine the size of a populated area is a widely accepted practice for generating land use data sets also according to Sherman et al 2014; however, the data set is only as accurate as the Google Earth image. The numerator of the equation for percent of urbanized area per parish was the total urbanized square kilometers, the sum of urbanized polygons in that parish. The denominator was the total square kilometers for the parish. For parishes where a large portion of the area is uninhabited, the number would be more correct if the uninhabited square meters could be excluded from the denominator. Population density may also be improved by a more sophisticated denominator due to the fact it also incorporates the total square kilometers of the parish. As GIS technology continues to improve with further
refinement, data sets computing population density and percent urbanized can also be improved. (Chang et al, 2009; Sherman et al, 2014.)
As discussed, age is a parameter which is seen as a social determinant of health that can increase the risk of dengue transmission and progression to the more severe states of dengue infection including dengue haemorrhagic fever and dengue shock syndrome, which can be fatal if there is inadequate medical intervention. The results of the non-index regression suggested that as the per cent of people inside of the vulnerable age range decreased, so would the number of cases of dengue per parish. The literature states that a vulnerable age definitely entails increased risk to developing serious symptoms. However, increased risk of serious symptoms may be conflated with risk of infection/dengue transmission. This requires further exploration in general and in this setting in particular.
The multivariate coefficient for poverty was also found to be contrary to what the literature traditionally states, with the results suggesting that lower poverty is associated with a lower number of incidence counts of dengue per parish. However, lately the notion of a link between dengue and poverty has become contested. For example, a research study by Muhmmad Azami et al 2011 and a systematic review by Mulligan et al 2015 attempt to challenge the inclusion of this social determinant of health as a driver for increased rate of infection with the idea of the ‗urbanization only‘ theory, a point of view that was included in the literature review. Neither research paper was able to make a convincing case but suggested switching the narrative from poverty to opportunity. However, the literature overwhelmingly states that although there are cases in wealthy densely populated areas, the areas of lower economic status with an
infrastructure that does not promote an improved water source, with low uptake of public health promotional material are at greater risk. In fact, the Horstick et al in 2015 study determined that dengue was still a threat to public health as the number of infections continues to increase worldwide, especially in the developing world where poverty is a major risk factor for
transmission of the Neglected Tropical Diseases (Horstick et al, 2015; Muhmmad Azami et al, 2011; Mulligan et al 2015)
In the end, the results of the thesis showed that the WADI, which provided equal weighting to the social and the environmental/ecological determinants, was more effective at determining vulnerability as measured by comparing actual versus predicted cases of dengue per parish than the multivariate model. However, the value of examining the issues raised by the results of the multivariate regression of the non-index model provides the impetus for further exploration and for challenging the accepted tenets of risk factors in the literature.