Multivariate spatial modelling of prevalence of Chagas disease in Argentina

C. DÍAZ-ÁVALOS; P. JUAN; J. MATEU; JORGE RABINOVICH

Valencia

Conferencia; IX Conference on Geostatistics for Environmental Applications (geoENV201); 2012

Universitat Politécnica de Valencia

Chagas disease iscaused by infection with the protozoa Trypanosomacruzi.  The disease has beenconsidered  an autochtonous disease of 22countries in the continental Western Hemisphere (WHO, 2002).  The estimated rate of prevalence fordifferent countries ranges between 0.7 and 15.5%.  For Argentina, the estimated prevalence rateof Chagas disease is 8.2% (Schmunis and Yadon, 2010). Among the 17 most commonspecies related to Chagas disease, the ones that belong to the genera Triatoma,Panstrongylus and Psammolestes are perhaps the most important and widespreadvectors of Trypanosoma cruzi, thecausative agent of Chagas disease.  Thesevectors are widely distributed in Argentina  and other South American countries, where theyprobably contribute to more than a half of the estimated 24 million cases ofthis disease. However, knowledge of the population characteristics of thesevectors is limited to laboratory studies and partial field observations. One ofthese field studies records data on presence-absence of the above-mentioneddisease vectors for a variety of  speciesover a fine grid covering part of Argentina. In addition, climatic andtopographical variates were also compiled. We aim at describing the spatialdistribution of prevalence of the vectors of Chagas disease to enablepredictive mapping of univariate and  multivariateprevalence of the species vectors and the disease risk. We analyze the binaryvariable of presence-absence of five species vectors of Chagas disease inArgentina, with data obtained from a long term field survey on a  grid covering the northern part of thecountry, in combination with meteorological and topographical covariatesassociated to  the grid. We use severalstatistical techniques to produce distribution maps of presence-absence,including a hierarchical Bayesian framework within the context of multivariategeostatistical modelling. The fitted logistic regression models range fromsimple logistic regression to models with a spatial term. We also exploremodels to  test the possibility ofinteraction between different vector species. Our results show that, as expected, the inclusion of covariates  improves the quality of the models fitted, andthat there is spatial interaction between variates in neighboring pixels in thegrid.