This work is part of a global project aiming to use medico-administrative big data from the whole French agricultural population (~3 millions), collected through their mandatory health insurance system (Mutualité Sociale Agricole), to highlight associations between chronic diseases and agricultural activities. At the request of the French Agency for Food, Environmental and Occupational Health & Safety (ANSES), our objective was to estimate which pesticides were probably used by each agricultural worker, in order to include this information in our analyses and search for association with diseases. We selected five databases to achieve this objective: the Graphical Land Parcel Registration (RPG), the French Agricultural Census, "Cultivation Practice" surveys from the Agriculture ministry, the MATPHYTO crop-exposure matrix and the Compilation of Phytosanitary Indexes from the French Public Health Agency. A geographical grid was designed to use geographical location while maintaining worker anonymity, dividing France into square tracts of variable surface each containing a minimum of 1500 agricultural workers. We developed an automated algorithm to predict each individual potential exposure by crossing her/his occupational activity, the geographical grid and the RPG to deduce cultivation practices and use it as a gateway to estimate pesticides use. This approach allowed drawing, from administrative data, a list of substances potentially used by each agricultural worker throughout France. Results of the algorithm are illustrated at collective level (descriptive statistics for the whole population), as well as at individual level (some workers taken as examples). The generalization of this method in other national contexts is discussed. By linking this information with the health insurance databases, this approach could contribute to the agricultural workers health surveillance.
Keywords: France; Occupational exposure; agricultural practices; big-data; health insurance data.