Predictive Risk Mapping of Leptospirosis for North of Iran Using Pseudo-absences Data

Leptospirosis is a common zoonosis disease with a high prevalence in the world and is recognized as an important public health drawback in both developing and developed countries owing to epidemics and increasing prevalence. Because of the high diversity of hosts that are capable of carrying the causative agent, this disease has an expansive geographical reach. Various environmental and social factors affect the spread and prevalence of the disease. The combination of epidemiology and Geospatial Information System plus using Ecological niche modeling provides the ability to identify areas at risk of disease, then predict the risk map of the disease for other regions by using relevant environment variables, and prevent and eventually eradicate the disease by conducting constructive activities such as increasing public awareness with education. In this study, using land use, environmental, and climate variables and taking advantage of the occurrences of the disease between 2009 and 2018, the risk level of Leptospirosis was modeled in three provinces of Gilan, Mazandaran, and Golestan based on ecological perspective. For modeling, highly correlated variables and also variables with high multicollinearity were identified and omitted. Because in ecological modeling regions to represent the absence of disease is required in addition to the presence and since these areas are not available, the second objective of this study was to evaluate the efficacy of different methods of generating pseudo-absence data in modeling leptospirosis. Finally, more accurate modeling of the prevalence of the disease in the northern provinces of the country can be obtained. Therefore, after selecting suitable variables for modeling, first, based on five methods (completely random generation of points in the study area, applying physical constraints with buffer at two radii of 5 and 10 km the generating points outside of designated buffer, applying environmental constraints by implementing two models of one-class support vector machine and BIOCLIM and generating points in unsuitable areas defined by these two models) pseudo-absence points representative of disease absence points in the study area were produced. Next, four models of Artificial Neural Network, Generalized Linear Model, Random Forest, and Gradient Boosting Machine were deployed to produce the disease risk in the study area. BIOMOD2 package in the R programming language was applied for modeling. The results showed that applying physical constraints with buffers yields the most reliable performance in comparison to the other three methods. Finally, the constructed model that performed best in TSS Statistics (with values of 0.76, 0.87, 0.84, 0.82 for Models of Artificial Neural Network, Generalized Linear Model, Random Forest, and Gradient Boosting Machine) was considered as the final output. Between all deployed models, Artificial Neural Network delivered the worst performance and had the most unstable results. Based on the risk-map of leptospirosis, central regions of Mazandaran and Gilan province, especially rural areas of Layl, Asalam, Eslam Abad, Chahar-deh, and Lafmejan have very high values of risk. Measures need to be made to reduce the high rate of Leptospirosis incidence in these regions. Furthermore, yearly precipitation was considered the most influential variable for the distribution of Leptospirosis.