Separating Measurement Error and Signal in Environmental Data: Use of Replicates to Address Uncertainty.

Measurement uncertainty has long been a concern in the characterizing and interpreting environmental and toxicological measurements. We compared statistical analysis approaches when there are replicates: a Naïve approach that omits replicates, a Hybrid approach that inappropriately treats replicates as independent samples, and a Measurement Error Model (MEM) approach in a random effects analysis of variance (ANOVA) model that appropriately incorporates replicates. A simulation study assessed the effects of sample size and levels of replication, signal variance, and measurement error on estimates from the three statistical approaches. MEM results were superior overall with confidence intervals for the observed mean narrower on average than those from the Naïve approach, giving improved characterization. The MEM approach also featured an unparalleled advantage in estimating signal and measurement error variance separately, directly addressing measurement uncertainty. These MEM estimates were approximately unbiased on average with more replication and larger sample sizes. Case studies illustrated analyzing normally distributed arsenic and log-normally distributed chromium concentrations in tap water and calculating MEM confidence intervals for the true, latent signal mean and latent signal geometric mean (i.e., with measurement error removed). MEM estimates are valuable for study planning; we used simulation to compare various sample sizes and levels of replication.

Censoring Trace-Level Environmental Data: Statistical Analysis Considerations to Limit Bias.

Trace-level environmental data typically include values near or below detection and quantitation thresholds where health effects may result from low-concentration exposures to one chemical over time or to multiple chemicals. In a cook stove case study, bias in dibenzo[a,h]anthracene concentration means and standard deviations (SDs) was assessed following censoring at thresholds for selected analysis approaches: substituting threshold/2, maximum likelihood estimation, robust regression on order statistics, Kaplan-Meier, and omitting censored observations. Means and SDs for gas chromatography-mass spectrometry-determined concentrations were calculated after censoring at detection and calibration thresholds, 17% and 55% of the data, respectively. Threshold/2 substitution was the least biased. Measurement values were subsequently simulated from two log-normal distributions at two sample sizes. Means and SDs were calculated for 30%, 50%, and 80% censoring levels and compared to known distribution counterparts. Simulation results illustrated (1) threshold/2 substitution to be inferior to modern after-censoring statistical approaches and (2) all after-censoring approaches to be inferior to including all measurement data in analysis. Additionally, differences in stove-specific group means were tested for uncensored samples and after censoring. Group differences of means tests varied depending on censoring and distributional decisions. Investigators should guard against censoring-related bias from (explicit or implicit) distributional and analysis approach decisions.

The Biomarkers of Exposure and Effect in Agriculture (BEEA) Study: Rationale, Design, Methods, and Participant Characteristics.

Agricultural exposures including pesticides, endotoxin, and allergens have been associated with risk of various cancers and other chronic diseases, although the biological mechanisms underlying these associations are generally unclear. To facilitate future molecular epidemiologic investigations, in 2010 the study of Biomarkers of Exposure and Effect in Agriculture (BEEA) was initiated within the Agricultural Health Study, a large prospective cohort in Iowa and North Carolina. Here the design and methodology of BEEA are described and preliminary frequencies for participant characteristics and current agricultural exposures are reported. At least 1,600 male farmers over 50 years of age will be enrolled in the BEEA study. During a home visit, participants are asked to complete a detailed interview about recent agricultural exposures and provide samples of blood, urine, and (since 2013) house dust. As of mid-September 2014, in total, 1,233 participants have enrolled. Most of these participants (83%) were still farming at the time of interview. Among those still farming, the most commonly reported crops were corn (81%) and soybeans (74%), and the most frequently noted animals were beef cattle (35%) and hogs (13%). There were 861 (70%) participants who reported occupational pesticide use in the 12 months prior to interview; among these participants, the most frequently noted herbicides were glyphosate (83%) and 2,4-D (72%), and most commonly reported insecticides were malathion (21%), cyfluthrin (13%), and permethrin (12%). Molecular epidemiologic investigations within BEEA have the potential to yield important new insights into the biological mechanisms through which these or other agricultural exposures influence disease risk.

Assessing lead exposure in U.S. pregnant women using biological and residential measurements.

There is strong scientific evidence for multiple pathways of human exposure to lead (Pb) in residential settings, particularly for young children; however, less is known about maternal exposure during pregnancy and children's exposure during early lifestages. A robust, multi-faceted secondary analysis was conducted using data collected by the National Institute of Child Health and Human Development in the 2009-2014 National Children's Study Vanguard Studies. Descriptive statistics summarized Pb concentrations of maternal blood, maternal urine, and house dust vacuum samples collected during pregnancy and residence surface wipes collected both during pregnancy and six months post-partum. The maternal blood Pb level geometric mean was 0.44 µg/dL (n = 426), with no women having values ≥ 5 µg/dL; creatinine-adjusted maternal urinary Pb geometric mean was 0.43 µg/g (n = 366). These blood and urine concentrations are similar to those observed for females in the general U.S. population in the National Health and Nutrition Examination Survey 2010-2011 cycle. A modest correlation between maternal blood Pb and surface wipe measurements during pregnancy was observed (Spearman r = 0.35, p < 0.0001). Surface wipe Pb loadings obtained in mother's homes during pregnancy (n = 640) and from areas where children spent the most time at roughly 6 months of age (n = 99) ranged from 0.02 to 71.8 ng/cm2, with geometric means of 0.47 and 0.49 ng/cm2, respectively, which were relatively low compared to other national studies. Survey responses of demographic, lifestyle, and residence characteristics were assessed for associations with blood concentration and surface wipe loading. Demographic (e.g., race/ethnicity, income, education, marital status) and housing characteristics (e.g., year home built, paint condition, own or rent home, attached garage) were associated with both maternal blood and surface wipe loadings during pregnancy. The availability of residential environmental media and extensive survey data provided enhanced understanding of Pb exposure during pregnancy and early life.

Mortality in the agricultural health study, 1993-2007.

Comparing agricultural cohorts with the general population is challenging because the general healthiness of farmers may mask potential adverse health effects of farming. Using data from the Agricultural Health Study, a cohort of 89,656 pesticide applicators and their spouses (N = 89, 656) in North Carolina and Iowa, the authors computed standardized mortality ratios (SMRs) comparing deaths from time of the enrollment (1993-1997) through 2007 to state-specific rates. To compensate for the cohort's overall healthiness, relative SMRs were estimated by calculating the SMR for each cause relative to the SMR for all other causes. In 1,198,129 person-years of follow-up, 6,419 deaths were observed. The all-cause mortality rate was less than expected (SMR(applicators) = 0.54, 95% confidence interval (CI): 0.52, 0.55; SMR(spouses) = 0.52, 95% CI: 0.50, 0.55). SMRs for all cancers, heart disease, and diabetes were significantly below 1.0. In contrast, applicators experienced elevated numbers of machine-related deaths (SMR = 4.15, 95% CI: 3.18, 5.31), motor vehicle nontraffic accidents (SMR = 2.80, 95% CI: 1.81, 4.14), and collisions with objects (SMR = 2.12, 95% CI: 1.25, 3.34). In the relative SMR analysis for applicators, the relative mortality ratio was elevated for lymphohematopoietic cancers, melanoma, and digestive system, prostate, kidney, and brain cancers. Among spouses, relative SMRs exceeded 1.0 for lymphohematopoietic cancers and malignancies of the digestive system, brain, breast, and ovary. Unintentional fatal injuries remain an important risk for farmers; mortality ratios from several cancers were elevated relative to other causes.

Using multiple imputation to assign pesticide use for non-responders in the follow-up questionnaire in the Agricultural Health Study.

The Agricultural Health Study (AHS), a large prospective cohort, was designed to elucidate associations between pesticide use and other agricultural exposures and health outcomes. The cohort includes 57,310 pesticide applicators who were enrolled between 1993 and 1997 in Iowa and North Carolina. A follow-up questionnaire administered 5 years later was completed by 36,342 (63%) of the original participants. Missing pesticide use information from participants who did not complete the second questionnaire impedes both long-term pesticide exposure estimation and statistical inference of risk for health outcomes. Logistic regression and stratified sampling were used to impute key variables related to the use of specific pesticides for 20,968 applicators who did not complete the second questionnaire. To assess the imputation procedure, a 20% random sample of participants was withheld for comparison. The observed and imputed prevalence of any pesticide use in the holdout dataset were 85.7% and 85.3%, respectively. The distribution of prevalence and days/year of use for specific pesticides were similar across observed and imputed in the holdout sample. When appropriately implemented, multiple imputation can reduce bias and increase precision and can be more valid than other missing data approaches.

An updated algorithm for estimation of pesticide exposure intensity in the agricultural health study.

An algorithm developed to estimate pesticide exposure intensity for use in epidemiologic analyses was revised based on data from two exposure monitoring studies. In the first study, we estimated relative exposure intensity based on the results of measurements taken during the application of the herbicide 2,4-dichlorophenoxyacetic acid (2,4-D) (n = 88) and the insecticide chlorpyrifos (n = 17). Modifications to the algorithm weighting factors were based on geometric means (GM) of post-application urine concentrations for applicators grouped by application method and use of chemically-resistant (CR) gloves. Measurement data from a second study were also used to evaluate relative exposure levels associated with airblast as compared to hand spray application methods. Algorithm modifications included an increase in the exposure reduction factor for use of CR gloves from 40% to 60%, an increase in the application method weight for boom spray relative to in-furrow and for air blast relative to hand spray, and a decrease in the weight for mixing relative to the new weights assigned for application methods. The weighting factors for the revised algorithm now incorporate exposure measurements taken on Agricultural Health Study (AHS) participants for the application methods and personal protective equipment (PPE) commonly reported by study participants.

Urinary biomarker, dermal, and air measurement results for 2,4-D and chlorpyrifos farm applicators in the Agricultural Health Study.

A subset of private pesticide applicators in the Agricultural Health Study (AHS) epidemiological cohort was monitored around the time of their agricultural use of 2,4-dichlorophenoxyacetic acid (2,4-D) and O,O-diethyl-O-3,5,6-trichloro-2-pyridyl phosphorothioate (chlorpyrifos) to assess exposure levels and potential determinants of exposure. Measurements included pre- and post-application urine samples, and patch, hand wipe, and personal air samples. Boom spray or hand spray application methods were used by applicators for 2,4-D products. Chlorpyrifos products were applied using spray applications and in-furrow application of granular products. Geometric mean (GM) values for 69 2,4-D applicators were 7.8 and 25 microg/l in pre- and post-application urine, respectively (P<0.05 for difference); 0.39 mg for estimated hand loading; 2.9 mg for estimated body loading; and 0.37 microg/m(3) for concentration in personal air. Significant correlations were found between all media for 2,4-D. GM values for 17 chlorpyrifos applicators were 11 microg/l in both pre- and post-application urine for the 3,5,6-trichloro-2-pyridinol metabolite, 0.28 mg for body loading, and 0.49 microg/m(3) for air concentration. Only 53% of the chlorpyrifos applicators had measurable hand loading results; their median hand loading being 0.02 mg. Factors associated with differences in 2,4-D measurements included application method and glove use, and, for hand spray applicators, use of adjuvants, equipment repair, duration of use, and contact with treated vegetation. Spray applications of liquid chlorpyrifos products were associated with higher measurements than in-furrow granular product applications. This study provides information on exposures and possible exposure determinants for several application methods commonly used by farmers in the cohort and will provide information to assess and refine exposure classification in the AHS. Results may also be of use in pesticide safety education for reducing exposures to pesticide applicators.

Assessment of a pesticide exposure intensity algorithm in the agricultural health study.

The accuracy of the exposure assessment is a critical factor in epidemiological investigations of pesticide exposures and health in agricultural populations. However, few studies have been conducted to evaluate questionnaire-based exposure metrics. The Agricultural Health Study (AHS) is a prospective cohort study of pesticide applicators who provided detailed questionnaire information on their use of specific pesticides. A field study was conducted for a subset of the applicators enrolled in the AHS to assess a pesticide exposure algorithm through comparison of algorithm intensity scores with measured exposures. Pre- and post-application urinary biomarker measurements were made for 2,4-D (n=69) and chlorpyrifos (n=17) applicators. Dermal patch, hand wipe, and personal air samples were also collected. Intensity scores were calculated using information from technician observations and an interviewer-administered questionnaire. Correlations between observer and questionnaire intensity scores were high (Spearman's r=0.92 and 0.84 for 2,4-D and chlorpyrifos, respectively). Intensity scores from questionnaires for individual applications were significantly correlated with post-application urinary concentrations for both 2,4-D (r=0.42, P<0.001) and chlorpyrifos (r=0.53, P=0.035) applicators. Significant correlations were also found between intensity scores and estimated hand loading, estimated body loading, and air concentrations for 2,4-D applicators (r-values 0.28-0.50, P-values<0.025). Correlations between intensity scores and dermal and air measures were generally lower for chlorpyrifos applicators using granular products. A linear regression model indicated that the algorithm factors for individual applications explained 24% of the variability in post-application urinary 2,4-D concentration, which increased to 60% when the pre-application urine concentration was included. The results of the measurements support the use of the algorithm for estimating questionnaire-based exposure intensities in the AHS for liquid pesticide products. Refinement of the algorithm may be possible using the results from this and other measurement studies.

Adult and children's exposure to 2,4-D from multiple sources and pathways.

In this study, we investigated the 2,4-dichlorophenoxyacetic acid (2,4-D) herbicide exposures of 135 preschool-aged children and their adult caregivers at 135 homes in North Carolina (NC) and Ohio (OH). Participants were randomly recruited from six NC and six OH counties. Monitoring was performed over a 48-h period at the participants' homes. Environmental samples included soil, outdoor air, indoor air, and carpet dust. Personal samples collected by the adult caregivers concerning themselves and their children consisted of solid food, liquid food, hand wipe, and spot urine samples. All samples were analyzed for 2,4-D (free acid form) by gas chromatography/mass spectrometry. 2,4-D was detected in all types of environmental samples but most often in carpet dust samples, with detection frequencies of 83% and 98% in NC and OH, respectively. The median level of 2,4-D in the carpet dust samples was about three times higher in OH homes compared to NC homes (156 vs. 47.5 ng/g, P<0.0002). For personal samples, 2,4-D was more frequently detected in the hand wipe samples from OH participants (>48%) than from NC participants (<9%). Hand wipe levels at the 95th percentile were about five times higher for OH children (0.1 ng/cm(2)) and adults (0.03 ng/cm(2)) than for the NC children (0.02 ng/cm(2)) and adults (<0.005 ng/cm(2)). 2,4-D was detected in more than 85% of the child and adult urine samples in both states. The median urinary 2,4-D concentration was more than twice as high for OH children compared to NC children (1.2 vs. 0.5 ng/ml, P<0.0001); however, the median concentration was identical at 0.7 ng/ml for both NC and OH adults. The intraclass correlation coefficient of reliability for an individual's urinary 2,4-D measurements, estimated from the unadjusted (0.31-0.62) and specific gravity-adjusted (0.37-0.73) values, were somewhat low for each group in this study. The variability in urinary 2,4-D measurements over the 48-h period for both children and adults in NC and OH suggests that several spot samples were needed to adequately assess these participants' exposures to 2,4-D in residential settings. Results from this study showed that children and their adult caregivers in NC and OH were likely exposed to 2,4-D through several pathways at their homes. In addition, our findings suggest that the OH children might have been exposed to higher levels of 2,4-D through the dermal and nondietary routes of exposure than the NC children and the NC and OH adults.