We log10 transformed continuous fumigant use variables to reduce heteroscedasticity and the influence of outliers, and to improve the fit of the models. We used logistic regression models to estimate odds ratios of respiratory symptoms and/or asthma medication use with residential proximity to fumigant use. Our primary outcome was respiratory symptoms defined as positive if during the previous 12 months the mother reported for her child any respiratory symptoms or the use of asthma medications, even in the absence of such symptoms . We also examined asthma medication use alone. The continuous lung function measurements were approximately normally distributed, therefore we used linear regression models to estimate the associations with residential proximity to fumigant use. We estimated the associations between the highest spirometric measures for children who had one, two or three maneuvers. We fit separate regression models for each combination of outcome, fumigant, time period, and buffer distance. We selected covariates a priori based on our previous studies of respiratory symptoms and respiratory function in this cohort. For logistic regression models of respiratory symptoms and asthma medication use, we included maternal smoking during pregnancy and signs of moderate or extensive mold noted at either home visit . We also included season of birth to control for other potential exposures that might play a causal role in respiratory disease , pollen , dryness , and mold. We defined the seasons of birth as follows: pollen , dry , mold based on measured pollen and mold counts during the years the children were born . In addition,fodder systems for cattle we controlled for allergy using a proxy variable: runny nose without a cold in the previous 12 months reported at age 7.
Because allergy could be on the causal pathway, we also re-ran all models without adjusting for allergy. Additionally, for spirometry analyses only, we adjusted for the technician performing the test, and child’s age, sex and height. We included household food insecurity score during the previous 12 months , breastfeeding duration , and whether furry pets were in the home at the 7 year visit to control for other factors related to lung function. We also adjusted for mean daily particulate matter concentrations with aerodynamic diameter ≤ 2.5 µm during the first 3 months of life and whether the home was located ≤150m from a highway in first year of life determined using GIS, to control for air pollution exposures related to lung function. We calculated average PM2.5 concentration in the first 3 months of life using data from the Monterey Unified Air Pollution Control District air monitoring station. In all lung function models of postnatal fumigant use, we included prenatal use of that fumigant as a confounder. To test for non-linearity, we used generalized additive models with three-degrees of-freedom cubic spline functions including all the covariates included in the final lung function models. None of the digression from linearity tests were significant ; therefore, we expressed fumigant use on the continuous log10 scale in multi-variable linear regression models. Regression coefficients represent the mean change in lung function for each 10-fold increase in wind-weighted fumigant use. We conducted sensitivity analyses to verify the robustness and consistency of our findings. We included other estimates of pesticide exposure in our models that have been related to respiratory symptoms or lung function in previous analyses of the CHAMACOS cohort. Specifically, we included child urinary concentrations of dialkylphosphate metabolites , a non-specific biomarker of organophosphate pesticide exposure using the area under the curve calculated from samples collected at 6-months, 1, 2, 3.5 and 5 years of age .
We also included agricultural sulfur use within 1-km of residences during the year prior to lung function measurement . We used similar methods as described above for fumigants to calculate wind-weighted sulfur use, except with a 1-km buffer and the proportion of time that the wind blew from each of eight directions during the previous year. The inclusion of these two pesticide exposures reduced our study population with complete data for respiratory symptoms and lung function . Previous studies have observed an increased risk of respiratory symptoms and asthma with higher levels of p, p’– dichlorodiphenyltrichloroethylene or p, p’-dichlorodiphenyldichloro-ethylene measured in cord blood . As a sensitivity analysis, we included log10- transformed lipid-adjusted concentrations of DDT and DDE measured in prenatal maternal blood samples . We also used Poisson regression to calculate adjusted risk ratios for respiratory symptoms and asthma medication use for comparison with the ORs estimated using logistic regression because ORs can overestimate risk in cohort studies . In additional analyses of spirometry outcomes, we also excluded those children who reported using any prescribed medication for asthma, wheezing, or tightness in the chest during the last 12 months to investigate whether medication use may have altered spirometry results. We ran models including only those children with at least two acceptable reproducible maneuvers . We ran all models excluding outliers identified with studentized residuals greater than three. We assessed whether asthma medication or child allergies modified the relationship between lung function and fumigant use by creating interaction terms and running stratified models. To assess potential selection bias due to loss to follow-up, we ran regression models that included stabilized inverse probability weights . We determined the weights using multiple logistic regression with inclusion as the outcome and independent demographic variables as the predictors. Data were analyzed with Stata and R .
We set statistical significance at p<0.05 for all analyses, but since we evaluated many combinations of outcomes, fumigants,fodder sprouting system distances and time periods we assessed adjustment for multiple comparisons using the Benjamini-Hochberg false discovery rate at p<0.05 .Most mothers were born in Mexico , below age 30 at time of delivery , and married or living as married at the time of study enrollment . Nearly all mothers did not smoke during pregnancy. When cohort participants were 6 and 12 months old, most households showed signs of moderate or extensive mold at either visit. At age 7, based on maternal report, the majority of families was living below the Federal Poverty Level, 15.7% of cohort children experienced a runny nose without a cold within the past year, 16.3% displayed asthma symptoms, and 6.1% were currently taking asthma medication. Table 2 shows the distributions of wind-weighted fumigant use within 8 km of CHAMACOS residences during the prenatal and postnatal exposure periods. Methyl bromide and chloropicrin were the most heavily used fumigants during the prenatal period, with mean ± SD wind-adjusted use of 13,380 ± 10,437 and 8,665 ± 6,816 kg, respectively. Reflecting declines in methyl bromide use, the use of chloropicrin was greater than the use of methyl bromide during the postnatal period, with median values of 127,977 and 109,616 kg during the 7 years, respectively. When we examined correlations within each fumigant, use within 3, 5, and 8 km from the home was highly correlated for each fumigant . Fumigant use during the prenatal and postnatal periods was also highly correlated for methyl bromide and chloropicrin, but was not correlated for metam sodium use and was inversely correlated for 1,3-DCP use . We also examined correlations among fumigants and observed high correlations between prenatal methyl bromide and chloropicrin use and between prenatal metam sodium and 1,3-DCP use . There were negative correlations between prenatal methyl bromide and chloropicrin use with prenatal metam sodium and 1,3-DCP use .In sensitivity analyses using multi-variable models including other pesticide exposures that have been previously related to respiratory symptoms and lung function including childhood urinary DAP metabolites , proximity to agricultural sulfur use during the year prior to lung function assessment and prenatal DDT/DDE blood concentrations , the results were very similar to those presented in Tables 3 and 4. For example, the relationships between prenatal methyl bromide use within 8 km were very similar for FEV1 and FEF25–75 . Prenatal fumigant use was generally not correlated with other pesticide exposures that we found to be associated with lung function in this cohort, except for weak correlations between agricultural sulfur use within 1 km during the year prior to spirometry and prenatal use of metam sodium and 1,3 – DCP with r = 0.14 and r=0.26 respectively. The results were very similar when we only included children with two acceptable reproducible maneuvers in the analyses . The results were also similar when we excluded those currently using asthma medication, excluded the one outlier for FEV1 models or used inverse probability weighting to adjust for participation bias . Risk ratios estimated for asthma symptoms and medication using Poisson regression were nearly identical to the ORs presented in Table 3 and Supplemental Table 2. We did not observe effect modification by asthma medication use. Maternal report of child allergies modified the relationship between FEV1 and prenatal proximity to methyl bromide use and we only observed higher FEV1 among children without allergies . After adjusting for multiple comparisons, none of the associations reached significance at the critical p-value 0.002 based on the Benjamini-Hochberg false discovery rate.This is the first study to examine lung function or respiratory symptoms in relation to residential proximity to agricultural fumigant use.
We found no significant evidence of reductions in lung function or increased odds of respiratory symptoms or use of asthma medication in 7-year-old children with increased use of agricultural fumigants within 3 – 8 km of their prenatal or postnatal residences. We unexpectedly observed a slight improvement in lung function at 7 years of age with residential proximity to higher methyl bromide and chloropicrin use during the prenatal period and this improvement was limited to children without allergies. Although these results remained after adjustment for other pesticide exposure measures previously related to respiratory symptoms and lung function in our cohort, they do not remain significant after adjustment for multiple comparisons. There is a strong spatial pattern of methyl bromide and chloropicrin use during the pregnancy period for our study because of heavy use on strawberry fields near the coast at the northern portion of the Salinas Valley . There could be other unmeasured environmental or other factors that are confounding the relationship we observed between higher prenatal fumigant use and improved lung function. We explored the use of location as a proxy variable but the results remained similar. Previously published studies of prenatal exposure to air pollutants and lung function have generally observed links to alterations in lung development and function and to other negative respiratory conditions in childhood, and plausible mechanisms include changes in maternal physiology and DNA alterations in the fetus . Improved lung function was associated with higher estimates of recent ambient exposure to hydrogen sulfide in a study of adults living in a geothermal area of New Zealand . However, hydrogen sulfide has been shown to be an endogenously produced “gasotransmitter”, with anti-inflammatory and cytoprotective functions , and is being explored for its use for protection against ventilator-induced lung injury . In previous studies of this cohort, we found increased odds of respiratory symptoms and lower FEV1, and FVC per 10-fold increase of childhood average urinary concentrations of metabolites of organophosphate pesticides . Other studies of prenatal pesticide exposure and respiratory health in children have mostly evaluated exposure using cord blood concentrations of DDE, a breakdown product of DDT, and have observed an increased risk of respiratory symptoms and asthma with higher levels of DDE . Most studies of postnatal pesticide exposure and respiratory health in children have utilized self-reported information from mothers to assess pesticide exposure and have observed higher odds of respiratory disease and asthma with reported pesticide exposure . None of the previous studies of pesticide exposure and respiratory health have specifically evaluated fumigants. Another strength of the study is that CHAMACOS is a prospective cohort followed since pregnancy with extensive data on potential confounders of respiratory health and other measures of pesticide exposure. Our study also had some limitations. We did not have information on maternal occupational exposure to fumigants or the geographic location of maternal workplaces during pregnancy, and we did not have the location of schools during childhood. These limitations likely resulted in some exposure mis-classification during both the prenatal and postnatal periods.