Investigating Lagged County-Level Associations Between Agricultural Pesticide Application and Parkinson’s Disease Mortality in the Conterminous United States

Abstract

Pesticide exposure is among the most studied environmental risk factors for Parkinson’s disease (PD), but most evidence comes from individual-level designs, and national ecological studies rarely account for confounding or for the long interval between exposure and disease. This study examined whether county-level agricultural pesticide application from 1998 to 2002 was associated with later countylevel PD mortality, and whether any association was robust to exposure definition, spatial structure, multivariable adjustment, and count-based modeling. County agricultural pesticide estimates from the U.S. Geological Survey EPest series were linked by Federal Information Processing Standard code to underlying-cause PD mortality (ICD-10 G20) from CDC WONDER for 2003–2007 and 2008–2012. In unadjusted and population-only models, log pesticide application was weakly and positively associated with age-adjusted PD mortality in 2008–2012 (r = .096; population-only β = 0.240, p = .002) and null in 2003–2007. This weak association was robust to normalizing exposure by land area and to removing high-use counties. However, it did not survive further scrutiny. Exposure and regression residuals showed strong positive spatial autocorrelation (Moran’s I ≈ 0.47 and ≈ 0.20, p = .001), violating the independence assumption; adjustment for population density, region, county age structure, median household income, and racial composition reduced the 2008–2012 coefficient to essentially zero (β = 0.003, p = .97; partial R² ≈ 0.000); and a negative-binomial count model with a population offset showed no significant association (incidence rate ratio ≈ 1.06, p = .14). The data therefore do not support a robust county-level association between agricultural pesticide application and later PD mortality. The apparent weak positive gradient is explained by geographic, demographic, and socioeconomic confounding. The study is ecological and hypothesis-generating; stronger inference will require longer exposure windows, incidence data, individual-level exposure, and chemical-specific histories.