A machine learning approach for modelling the occurrence of Galba truncatula as the major intermediate host for Fasciola hepatica in Switzerland.

Fasciolosis caused by the trematode Fasciola hepatica is an important parasitosis in both livestock and humans across the globe. Chronic infections in cattle are associated with considerable economic losses. As a prerequisite for an effective control and prevention of fasciolosis in cattle fine-scale predictive models on farm-level are needed. Since disease transmission will only occur where the mollusc intermediate host is present, the objective of our research was to develop a regression model that allows to predict the local presence or absence of Galba truncatula as principal intermediate host for Fasciola hepatica in Switzerland. By implementing generalized linear mixed models (GLMMs) a total amount of 70 variables were analysed for their potential influence on the likelihood π of finding Galba truncatula at a certain site. Important site-specific features could be considered by selecting suitable modelling procedures. The statistical software R was used to conduct regression analysis, performing the grplasso and the glmmLasso method. The selection of parameters was based on 10-fold cross validation and the Bayesian Information Criterion (BIC). This yielded a total number of 19 potential predictor variables for the grplasso and 13 variables for the glmmLasso model, which also included random effects. Nine variables appeared to be relevant predictors for the occurrence of Galba truncatula in both models. These included reed/humid area, spring water, water bodies within a 100 m radius, and trees/bushes as powerful positive predictors. High soil depth, temperatures frequently exceeding 30 °C in the year preceding the search for snails and temperatures below 0 °C especially in the second year before were identified to exert an adverse effect on the occurrence of Galba truncatula. Temperatures measured near ground level proved to be more powerful predictors than macroclimatic parameters. Precipitation values seemed to be of minor impact in the given setting. Both regression models may be convenient for a fine-scale prediction of the occurrence of Galba truncatula, and thus provide useful approaches for the development of future spatial transmission models, mapping the risk of fasciolosis in Switzerland on farm-level.