Farms, Findings, and False Causation: A Systematic Review of Allergy Prevalence

Background: A lower prevalence of allergy in children has been observed across numerous studies in the farming environment. However, establishing a causal relationship is contentious due to the typical farm versus non-farm study design, which may have introduced a stratification collider bias. Methods: We conducted a quantitative meta-analysis of published risks from farming studies, categorizing them into subgroups based on representative or stratified sampling. The two risk ratios are subjected to another fixed effects meta regression model and the difference is quantified. Additionally, we analyzed the correlation across generations to indicate either change of exposure or of disease status in children. Finally we simulate outcomes under various diseases prevalences, while drawing further qualitative insights from previous migration studies within the farming population. Results: The results indicate that the lower prevalence of allergic diseases in farming environments depends on the sampling method used. The studies with random sampling provided an OR of 1.1 which is significantly different from the OR of 0.7 in studies with stratified sampling (P=0.001). Children of diseased farmers are not inherently "protected" at farms as the association is found in their parents as well. They change exposure which is in line which is inline with migration studies in the farming communities who described a healthy worker effect. A simulation under different prevalence scenarios shows that the observed ORs can already be observed if 1 in 100 children per generation moves out of the farming environment. Conclusion: The combined findings suggest that a selection bias is the most likely reason for the observed lower allergy prevalence in some farming environments.