Exposure misclassification in studies of agricultural pesticides: insights from biomonitoring

Abstract

Background: Epidemiologists often assess lifetime pesticide exposure by questioning participants about use of specific pesticides and associated work practices. Recently, Dosemeci and colleagues proposed an algorithm to estimate lifetime average exposure intensity from questionnaire information. We evaluated this algorithm against measured urinary pesticide concentrations for farmers who applied glyphosate (n = 48), 2,4-D (n = 34), or chlorpyrifos (n = 34).

Methods: Algorithm scores were calculated separately based on trained field observers' and farmers' evaluations of application conditions. Statistical analyses included nonparametric correlations, assessment of categorical agreement, and categorical evaluation of exposure distributions.

Results: Based on field observers' assessments, there were moderate correlations between algorithm scores and urine concentrations for glyphosate (r = 0.47; 95% confidence interval [CI] = 0.21 to 0.66) and 2,4-D (0.45; 0.13 to 0.68). Correlations were lower when algorithm scores were based on participants' self-reports (for glyphosate, r = 0.23 [CI = -0.07 to 0.48]; for 2,4-D, r = 0.25 [-0.10 to 0.54]). For chlorpyrifos, there were contrasting correlations for liquid (0.42; 0.01 to 0.70) and granular formulations (-0.44; -0.83 to 0.29) based on both observers' and participants' inputs. Percent agreement in categorical analyses for the 3 pesticides ranged from 20% to 44%, and there was appreciable overlap in the exposure distributions across categories.

Conclusions: Our results demonstrate the importance of collecting type of pesticide formulation and suggest a generic exposure assessment is likely to result in appreciable exposure misclassification for many pesticides.