Spatial Distribution of Seropositive Dogs to Leptospira spp. and Evaluation of Leptospirosis Risk Factors Using a Decision Tree

Abstract

Background: Leptospirosis is an important zoonotic disease associated with poor areas of urban settings of developing countries and early diagnosis and prompt treatment may prevent disease. Although rodents are reportedly considered the main reservoirs of leptospirosis, dogs may develop the disease, may become asymptomatic carriers and may be used as sentinels for disease epidemiology. The use of Geographical Information Systems (GIS) combined with spatial analysis techniques allows the mapping of the disease and the identification and assessment of health risk factors. Besides the use of GIS and spatial analysis, the technique of data mining, decision tree, can provide a great potential to find a pattern in the behavior of the variables that determine the occurrence of leptospirosis. The objective of the present study was to apply Geographical Information Systems and data prospection (decision tree) to evaluate the risk factors for canine leptospirosis in an area of Curitiba, PR.Materials, Methods & Results: The present study was performed on the Vila Pantanal, a urban poor community in the city of Curitiba. A total of 287 dog blood samples were randomly obtained house-by-house in a two-day sampling on January 2010. In addition, a questionnaire was applied to owners at the time of sampling. Geographical coordinates related to each household of tested dog were obtained using a Global Positioning System (GPS) for mapping the spatial distribution of reagent and non-reagent dogs to leptospirosis. For the decision tree, risk factors included results of microagglutination test (MAT) from the serum of dogs, previous disease on the household, contact with rats or other dogs, dog breed, outdoors access, feeding, trash around house or backyard, open sewer proximity and flooding. A total of 189 samples (about 2/3 of overall samples) were randomly selected for the training file and consequent decision rules. The remained 98 samples were used for the testing file. The seroprevalence showed a pattern of spatial distribution that involved all the Pantanal area, without agglomeration of reagent animals. In relation to data mining, from 189 samples used in decision tree, a total of 165 (87.3%) animal samples were correctly classified, generating a Kappa index of 0.413. A total of 154 out of 159 (96.8%) samples were considered non-reagent and were correctly classified and only 5/159 (3.2%) were wrongly identified. on the other hand, only 11 (36.7%) reagent samples were correctly classified, with 19 (63.3%) samples failing diagnosis.Discussion: The spatial distribution that involved all the Pantanal area showed that all the animals in the area are at risk of contamination by Leptospira spp. Although most samples had been classified correctly by the decision tree, a degree of difficulty of separability related to seropositive animals was observed, with only 36.7% of the samples classified correctly. This can occur due to the fact of seronegative animals number is superior to the number of seropositive ones, taking the differences in the pattern of variable behavior. The data mining helped to evaluate the most important risk factors for leptospirosis in an urban poor community of Curitiba. The variables selected by decision tree reflected the important factors about the existence of the disease (default of sewer, presence of rats and rubbish and dogs with free access to street). The analyses showed the multifactorial character of the epidemiology of canine leptospirosis.