Tools from ecology: useful for evaluating infection risk models?

Abstract

Despite the increasing number of models to predict infection risk for a range of diseases, the assessment of their spatial limits, predictive performance and practical application are not widely undertaken. Using the example of Schistosoma haematobium in Africa, this article illustrates how ecozonation and receiver-operator characteristic analysis can help to assess the usefulness of available models objectively.The resources targeted at parasite control are finite and often limited. Consequently, when designing control programs, it is essential to know the distribution and abundance of a disease, to devise and target intervention strategies and to optimize the use of available resources. In many African countries, the paucity of epidemiological data hinders the quantification of disease burden for basic planning. In an effort to overcome this problem, environmental data (often derived from satellite sensors) are increasingly used to predict infection risk [1][2][3]. There is a need, in such approaches, to evaluate objectively the predictive accuracy of the models used to generate risk maps and to consider the spatial extent to which models can be reliably extrapolated [4]. Developing predictive modelsReliable maps of infectious diseases require an understanding of whether models developed for one location can be applied to another because the environmental factors that influence disease transmission are unlikely to be uniform over large geographical areas [5]. Political boundaries are routinely used to define the spatial extent of risk maps, but the ecological heterogeneity, which is usually independent of political boundaries, is ignored. Alternatively, remotely-sensed (RS) derived environmental data can be used to develop ecological zone maps that identify areas of ecological similarity * (Box 1), and are better at defining where existing predictive models can and cannot be applied [6].The statistical methods commonly used to predict the occurrence or distribution of disease in relation to environmental variables are logistic regression and discriminant analysis, and these approaches have been used to map filariasis [3], malaria [6,7] . The predictive performance of these models, which is a prerequisite for their refinement [9], is evaluated by examining the agreement between predictions and observations [10] by using data collected from sites other than those used in model development. In logistic regression models, predictions are based on the model outputs that measure the probability of infection occurrence. To assess the predictive performance of a model, a probability threshold needs to be identified that can differentiate locations of relative risk. Often, however, the researcher is confronted with the question: which threshold to select for when discriminating between these different populations. A commonly used method in medical diagnostics, and more recently in ecological studies, is receiver-operator characteristic (ROC) analysis, which can provide an overall measure of model accuracy and c...

Geographical analysis of the role of water supply and sanitation in the risk of helminth infections of children in West Africa

“…Model selection for prediction was based on the DIC (42). The ability of the final model to discriminate prevalence of natural floor type, water supply, and toilet availability and helminth infection thresholds was assessed using the area under the curve (AUC) of the receiver operating characteristic (ROC) (43). The proportion of geographical variation in helminth infections explained by WASH indicators was estimated by comparing the proportional decrease in the spatial variance parameter of the geostatistical random effect with a model containing only the environmental covariates.…”

Climate change and livestock production: A review with emphasis on Africa

“…Abiotic factors, such as temperature and day length, impose constraints on when and how ticks quest for hosts (Randolph, 2008b). Beyond vector-borne diseases, intestinal nematodes develop in soil, and factors such as soil humidity and temperature have a strong influence on developmental rates (Brooker et al, 2002). Climate change is bound to have further impact on heat-related mortality and morbidity and on the incidence of climate-sensitive infectious diseases (Patz et al, 2005).…”

The epidemiology and small-scale spatial heterogeneity of urinary schistosomiasis in Lusaka province, Zambia

“…1), after obtaining ethical clearance from the University of Zambia Ethics Committee. The two districts were selected on the basis of their ecological representativeness of the country in general (Abell et al, 1998;Brooker et al, 2002;Meteorology Department, 2003). In each of these districts 10 primary schools were selected.…”

“…However, because of small-scale focality of the disease, implementation of any control strategy should understand where the population at risk is located for spatial targeting of limited resources, for efficient and cost-effective control (Brooker et al, 2002). The spatial variability of schistosomiasis is partly explained by a range of climatic, ecological and socioeconomic factors.…”

“…We also examined small-scale spatial heterogeneity in prevalence, an area less commonly explored despite spatial focality of the disease (Utzinger et al, 2003a, Brooker, 2007. Recent advances in spatial statistical tools have seen an increase in the application of spatial methodologies in analysing the prevalence of schistosomiasis and soil-transmitted helminths (Kristensen et al, 2001;Brooker et al, 2002;Raso et al, 2005;Brooker, 2007). We applied a random-effects spatial logistic model to account for spatial heterogeneity in prevalence, adjusting for age, sex and other explanatory variables.…”