6533b7d9fe1ef96bd126d3ba

RESEARCH PRODUCT

Probabilité d'apparition d'un phénomène parasitaire et choix de modèles de régression logistique

subject

description

Epidemiological processes are now using spatial statistics and modelling tools. The main objective of most health risks studies consists in identifying potential contamination sources and factors capable of explaining their localization. Health data often prove binary (typically presence/absence) and specific methods such as binary logistic regression have to be used. This method's output consists in a probability for the pathogen of interest. A posterior classification of each sample is then conducted using a probability threshold. The method used to maximize this threshold is called the ROC curve which consists in giving a representation of the behaviour of the model and then to choose the optimal value to discriminate between negative and positive predictions. Using a point data epidemiological base, several models were generated and tested. Landscape indices have been derived in the environment of the points at three scale levels. Probability values allocated to each sample were then spatially represented, giving an insight on the expected geographical dispersion of contaminated samples. Variables identified in the models were then used to establish hypothesis as to which landscape factors might play a role in epidemiological processes. The outlining of potential high risk areas is a result of first importance in the geography of health.