Objectives Pooling epidemiological studies can be beneficial when studying rare outcomes and subtypes of diseases. However, pooling of exposure information is a challenge since standard approaches are not employed. The objective of this undertaking was to enable pooling of agricultural cohort studies at level of exposure to active ingredients.

Methods Initial efforts in AGRICOH focused on cancer outcomes in three large cohort studies, the Agricultural Health Study (AHS-USA), Agriculture and cancer (AGRICAN-France) and Cancer in the Norwegian Agricultural Population (CNAPNorway) with data from >300000 individuals. Each study employed different methods, including Crop-Exposure Matrices (CEM), (AGRICAN and CNAP), and self-report (AHS), which were compared.

Results The CEM approach led to much higher prevalence of exposures. The CEM approach as expected generated false positive exposures. Lack of specificity is less of an issue for pesticides applied relatively frequently, but will lead to more exposure misclassification when prevalence of use is low. Given that assignment of exposure will be independent of disease status, misclassification is non-differential and will result in bias towards the null, especially when strength of association is modest, as for most agricultural exposures and health effects. Within AGRICOH our ability to detect associations are for now at best limited without further improvement of exposure assessment and assignment.

Conclusion Clear differences in farming systems, crops and animals, climate and agricultural inputs do exist and may result in large qualitative and quantitative differences in exposure when accurately assessed and assigned. Consequently, it may provide opportunities to evaluate exposure contrasts.To enable pooling in more informative ways, further thought is given within AGRICOH to collect agricultural occupational histories in more detail and to harmonize exposure assessment and assignment. Standardization of exposure assessment approaches within AGRICOH and future studies is a necessity and will make pooling easier and may result in more informative studies.