EXPLORATORY ANALYSIS OF THE ENVIRONMENTAL FACTORS ATTRIBUTED TO PODOCONIOSIS
The study was conducted over a 30 x 30 km area in northern Ethiopia. Data was collected on: disease magnitude (prevalence), soil composition and characterisation, and meteorology. House-to-house visits were made to find case counts and population at risk. Surface soils were collected from designated sampling sites and fully characterised (including chemical oxides and trace elements, mineral phase identification and particle size). The meteorology data (i.e. rainfall and temperature) were extracted from WorldClim.org and triangulated with data from the Ethiopian Meteorology Agency. We analysed data using R and geo-statistical packages.
The distribution of each predictor covariate and disease distribution was plotted in a histogram and summarized using mean, standard deviation, skewness and kurtosis. The presence of outliers was tested using box plots and Cleveland dot plots. Since our outcome variable is count of cases and we needed a probability distribution that allows greater variation for mean values, we opted for Poisson distribution. Collinearity between covariates was evaluated using scatter plots and correlation coefficients, and variables were discounted based on their correlation coefficient and theoretical understanding of the covariates. Next, we explored the relationship between the disease distribution and the remaining predictor covariates using univariate analysis. Variables that showed statistically significant relationship with the disease and returned the lowest AIC (Akaki Information Criterion) were selected for further investigation. The remaining variables, which included soil oxides, trace elements and clay minerals were assessed for interaction, spatial dependence and homogeneity of variance.
The final outputs will be a model for the environmental correlates of podoconiosis using stepwise model selection techniques, and one that accounts for the individual correlates.