Decade of residential pesticide exposure tied to blood pressure and cholesterol changes

A decade of living near agricultural pesticide applications is associated with measurably higher blood pressure and lower HDL cholesterol among midlife women in one of California’s most intensively farmed regions, according to a new longitudinal study published in the Journal of Exposure Science & Environmental Epidemiology. Researchers followed 459 women from the Salinas Valley and estimated each participant’s average residential proximity to agricultural use of seven pesticide classes over the ten years preceding health assessments conducted between 2022 and 2024. Using two complementary statistical mixture models, the team found that a joint one-quartile increase in exposure across all seven classes was associated with systolic blood pressure 1.82 mm Hg higher, pulse pressure 1.59 mm Hg higher, and HDL cholesterol 1.35 mg/dL lower, along with a 12 percent higher risk of metabolically defined low HDL cholesterol. Joint associations with body mass, blood sugar, and inflammatory markers were null. The study is, to the authors’ knowledge, the first to examine long-term joint exposure to current-use pesticides in association with a wide range of cardiometabolic outcomes.

The findings come from the Center for the Health Assessment of Mothers and Children of Salinas, or CHAMACOS, a cohort of low-income, predominantly Mexican American women recruited beginning in 1999 and 2000, when they sought prenatal care at a Salinas Valley safety-net hospital and federally qualified health centers. A demographically similar refreshment sample joined in 2009 through 2011. For the CHAMACOS Maternal Cognition Study, ongoing participants who had completed a visit since 2014 were invited back, and after exclusions for missing address history, incomplete covariate information, pregnancy, or missing measurements, the analytical sample comprised 459 women, with 445 contributing blood-based outcomes. At outcome assessment, participants averaged 48.4 years of age. Most were married, nearly 90 percent were foreign-born, 44.2 percent had completed sixth grade or less, and 73.9 percent lived in households at or below the federal poverty level; 42.7 percent worked in agriculture themselves. The cohort carried a heavy cardiometabolic burden: 64.9 percent had obesity, 39.2 percent had hypertension, 34.3 percent had diabetes, and 53.5 percent met criteria for metabolic syndrome.

California’s Pesticide Use Reporting database, which records the active ingredient, amount, date, and location of every agricultural application, made the long-term exposure estimates possible. The researchers geocoded each participant’s residential history, drew a one-kilometer buffer around every home, and calculated the kilograms of active ingredient applied within that radius, weighting reported use by the share of land area inside the buffer. They chose a one-kilometer radius because prior validation work found the strongest correlations between proximity-based estimates and measured pesticide levels in house dust and ambient air at that distance. Exposure was averaged over the ten years before outcome assessment to capture chronic exposure. Nearly all participants, 92.8 percent, had exposure to all seven pesticide classes during the window, with at least one exposure in an average of 8.7 to 9.5 of the ten years for six of the seven classes and 5.7 years for paraquat herbicides. Estimated use ranged from a geometric mean of 1.9 kg/year for paraquat to 57.3 kg/year for glyphosate-based herbicides, and correlations between classes spanned 0.37 to 0.93.

Because agricultural communities are exposed to many pesticides simultaneously, as different crops peak in different seasons and application practices produce overlapping exposures to multiple active ingredients, the researchers turned to environmental mixture methods rather than modeling each pesticide class in isolation. Quantile g-computation, or QGCOMP, estimates the effect of a simultaneous one-quartile increase in all exposures, treating the mixture as a single coefficient under an assumption of approximately additive, linear effects; the team bootstrapped 1,000 iterations to obtain robust confidence intervals. Bayesian kernel machine regression, or BKMR, flexibly models nonlinear and interactive exposure-response surfaces and handles correlation among exposures implicitly through its kernel, using Markov chain Monte Carlo with 10,000 iterations. Models adjusted for confounders identified through a directed acyclic graph, including age at delivery, nativity and age of arrival in the United States, educational attainment, household poverty, agricultural work in the household, and marital status, all measured before the exposure window. Comparing the two frameworks allowed the team to test whether conclusions held under different assumptions about the shape of the exposure-response relationship.

The QGCOMP models showed that a one-quartile increase in the total pesticide mixture was associated with systolic blood pressure 1.82 mm Hg higher, with a 95 percent confidence interval of 0.38 to 3.25, and pulse pressure 1.59 mm Hg higher, with a 95 percent confidence interval of 0.56 to 2.63. The same joint increase was associated with HDL cholesterol 1.35 mg/dL lower, with a 95 percent confidence interval of negative 2.61 to negative 0.10, and, for the binary metabolic syndrome indicator of low HDL cholesterol, defined as below 50 mg/dL or use of cholesterol medication, a risk ratio of 1.12, with a 95 percent confidence interval of 1.02 to 1.23. Associations with hypertension, at a risk ratio of 1.11 with a 95 percent confidence interval of 0.97 to 1.26, and the metabolic syndrome high blood pressure indicator, at a risk ratio of 1.12 with a 95 percent confidence interval of 0.98 to 1.28, were suggestive but crossed the null. Results for other continuous outcomes, including body mass index, waist circumference, triglycerides, glucose, glycated hemoglobin, and the inflammatory markers C-reactive protein and interleukin-6, were null. Sensitivity analyses excluding non-fasting blood draws, excluding participants with prevalent hypertension, and applying inverse probability of attrition weights yielded generally similar estimates.

The BKMR results were broadly consistent and added a nonlinear dimension. Increasing the entire mixture from the 50th to the 75th percentile was associated with systolic blood pressure 3.88 mm Hg higher, with a 95 percent credible interval of 0.76 to 6.99, and pulse pressure 3.46 mm Hg higher, with a 95 percent credible interval of 1.35 to 5.57, whereas the shift from the 25th to the 50th percentile produced smaller, less precise estimates; for pulse pressure, just 0.32 mm Hg, with a 95 percent credible interval of negative 1.65 to 2.29. Both exposure-response functions were approximately monotonic, with steeper increases at higher mixture levels. For the low HDL cholesterol indicator, the association was strongest at the lower end: a risk ratio of 1.09, with a 95 percent credible interval of 1.01 to 1.18, for the 25th-to-50th percentile shift, attenuating to 1.03 at the higher shift. Suggested associations with hypertension and the high blood pressure indicator had credible intervals crossing the null. The authors note that QGCOMP’s reparameterization may increase precision for nearly linear effects, while BKMR’s flexible kernel may trade power for flexibility, which could explain the divergence for continuous HDL cholesterol.

Disentangling which pesticide classes drive the joint signal proved difficult because the exposures are highly correlated. In exploratory analyses, QGCOMP weights pointed to manganese fungicides and, to a lesser extent, neonicotinoid insecticides as the largest contributors to the blood pressure associations, with carbamate insecticides and glyphosate-based herbicides contributing most to the HDL cholesterol association. BKMR’s hierarchical variable selection gave glyphosate-based herbicides a posterior inclusion probability of 0.89 for the low HDL cholesterol indicator, the highest support of any group, while group probabilities for the correlated insecticide and manganese fungicide cluster were modest. The authors emphasize that these component-level summaries should not be interpreted as causal or independent effects. Still, the glyphosate pattern aligns with earlier findings in this same cohort, where two-year average residential exposure to glyphosate and paraquat was linked to higher blood pressure in adult women, and with external evidence, including NHANES analyses identifying glyphosate as the main contributor to cardiovascular disease risk within a herbicide and insecticide mixture, and an occupational study reporting higher odds of coronary artery disease among glyphosate-exposed workers.

Experimental evidence offers a plausible biological thread: many pesticide classes share the capacity to generate oxidative stress, producing reactive oxygen and nitrogen species that overwhelm endogenous antioxidant defenses. Oxidative injury to the endothelium can promote vasoconstriction and elevated blood pressure, while lipid peroxidation can impair HDL function and reverse cholesterol transport. Notably, the study found no association between joint pesticide exposure and systemic inflammatory biomarkers, but the authors point out that C-reactive protein and interleukin-6 reflect short-term systemic inflammation and may not capture chronic or localized oxidative processes. The population-level stakes are considerable. The observed 1.8 to 3.8 mm Hg rise in systolic blood pressure corresponds, based on a meta-analysis by the Prospective Studies Collaboration, to an estimated 6 to 14 percent higher risk of cardiovascular disease mortality in similar age groups. Even small shifts in blood pressure and HDL cholesterol across a population can translate into substantial increases in cardiovascular disease burden, the researchers note.

The authors caution that several limitations shape interpretation. Pesticide applications were assigned probabilistically within land sections rather than measured at individual residences, addresses were reported prospectively for only about half the exposure period, and the estimates capture ambient exposure at the home but not dietary, occupational, or para-occupational routes, though agricultural work in the household was adjusted for. Averaging over ten years obscures whether specific timing windows matter, and aggregating active ingredients into classes may mask differences in persistence and toxicity. Reverse causation cannot be fully excluded, although excluding women with hypertension prevalent before the exposure window generally strengthened the binary blood pressure associations, suggesting pre-existing disease is unlikely to explain the findings. The results may not generalize to populations with lower baseline cardiometabolic risk or different exposure patterns. Future work, the team writes, should repeat outcome assessments over time, probe alternate exposure windows and metrics such as peak or cumulative exposure, and integrate higher-resolution exposure models and repeated biomonitoring to better separate the contributions of individual pesticides within highly correlated mixtures.

Subject of Research: Long-term residential proximity-based exposure to agricultural pesticide mixtures and cardiometabolic risk factors among women in an agricultural region

Article Title: Joint associations between 10-year average residential proximity-based pesticide exposures and cardiometabolic markers among women from an agricultural region

Article References: Calderon, L. E., Gunier, R. B., Warner, M., Kogut, K., Perez-Rocha, Y., MacCuish, K., Rauch, S., Eskenazi, B., & Torres, J. M. (2026). Joint associations between 10-year average residential proximity-based pesticide exposures and cardiometabolic markers among women from an agricultural region. Journal of Exposure Science & Environmental Epidemiology. https://doi.org/10.1038/s41370-026-00994-0

Image Credits: AI Generated

DOI: 10.1038/s41370-026-00994-0

Keywords: pesticide exposure, cardiometabolic risk, blood pressure, HDL cholesterol, agricultural communities, CHAMACOS cohort, environmental mixtures, glyphosate, quantile g-computation, Bayesian kernel machine regression, Salinas Valley, oxidative stress