|Home | About | Journals | Submit | Contact Us | Français|
The aim was to develop quantitative estimates of farmers’ pesticide exposure to atrazine and to provide an overview of background levels of selected non-persistent pesticides among corn farmers in a longitudinal molecular epidemiologic study.
The study population consisted of 30 Agricultural Health Study farmers from Iowa and 10 nonfarming controls. Farmers completed daily and weekly diaries from March to November in 2002 and 2003 on pesticide use and other exposure determinants. Urine samples were collected at 10 timepoints relative to atrazine application and other farming activities. Pesticide exposure was assessed using urinary metabolites and diaries.
The analytical limit of detection (LOD) ranged between 0.1–0.2 μg/l for all pesticide analytes except for isazaphos (1.5 μg/l) and diazinon (0.7 μg/l). Farmers had higher geometric mean urinary atrazine mercapturate (AZM) values than controls during planting (1.1 vs. <limit of detection μg/g creatinine; p<0.05). AZM levels among farmers were significantly related to the amount of atrazine applied (p=0.015). Interestingly, farmers had a larger proportion of samples above the limit of detection than controls even after exclusion of observations with an atrazine application within 7 days before urine collection (38% vs. 6%, p<0.0001). A similar pattern was observed for 2,4-D and acetochlor (92% vs. 47%, p<0.0001 and 45% vs. 4%, p<0.0001, respectively).
Urinary AZM levels in farmers were largely driven by recent application of atrazine. Therefore, the amount of atrazine applied is likely to provide valid surrogates of atrazine exposure in epidemiologic studies. Elevated background levels of non-persistent pesticides, especially 2,4-D, indicate importance in epidemiologic studies of capturing pesticide exposures that might not be directly related to the actual application.
A wide variety of agricultural pesticides is used on farms, including herbicides, crop and livestock insecticides, fungicides, and fumigants (Curwin et al., 2005a). Farmers may be exposed to these pesticides when mixing, loading and applying. After application they may be exposed if they work in treated fields or if equipment, work clothing or the home environment is contaminated with pesticides.
One of the major challenges in epidemiologic studies of health effects from pesticide exposure is obtaining an accurate assignment of exposure. The application of pesticides by farmers is characterized by large seasonal differences in the types, amounts, and frequency of pesticide application (Kromhout and Heederik, 2005). This temporal variation in pesticide use may give rise to large within-worker variations in exposure over the course of a year. In such cases, a small number of measurements may lead to imprecise estimates of long-term average exposures. However, it is often infeasible to measure individual exposure levels over an entire year or across all workers when studying chronic health effects. In epidemiologic studies, individual long-term average exposures can be estimated by using (empirical) deterministic models based on known determinants of exposure (Preller et al., 1995). Such models may reduce non-differential misclassification by using detailed longitudinal information on important exposure determinants over time (i.e. taking into the account the effect of variation in the application pattern throughout the season) to predict exposure levels.
In this paper we describe application-specific exposure levels for atrazine, and background exposure levels for some selected non-persistent pesticides among corn farmers and controls over an entire growing season. The purpose of the exposure survey was to develop daily quantitative estimates of farmers’ pesticide exposure to atrazine during one year using measurements of atrazine mercapturate (AZM) in urine and detailed information on pesticide application from exposure diaries. The models will be used in a molecular epidemiologic study evaluating the relationship between exposure to atrazine and various immune effects in corn farmers (Vermeulen et al., 2005).
Corn farmers from Iowa were identified from the Agricultural Health Study (AHS) (Alavanja et al., 1996). To be eligible for this study, the farmers had to be male, age 40–60 years, non-smoking, and planning to personally apply atrazine to at least 300 acres of corn as part of their normal farm operation during the upcoming season. Nonfarmers serving as controls in the study were agricultural extension agents from Iowa State University in Ames, IA. The control inclusion criteria were the same as for farmers except that they had not applied pesticides occupationally within the last 5 years. To reduce the complexity of the field logistics, farmers and controls residing in counties closest to the Iowa Field Station in Iowa City, IA were selected. The inclusion of farmers and controls continued until the target population of 30 farmers and 10 controls was reached. Of the subjects who were eligible and invited to participate, the participation rate was 90% and 100%, among farmers and controls, respectively. Study participants received a compensation of $300 for completing the study protocol. This study was approved by the Human Subjects Office at the University of Iowa and the National Cancer Institute Institutional Review Board.
The field study periods were March/April 2002 through January 2003, and March/April 2003 through January 2004. Ten farmers and 5 controls were followed during the first year, and 20 additional farmers and 5 controls during the second year of the study. Two farmers participated in both years. Table 1 gives an overview of the data collection activities. The study population was followed throughout an entire farming year, which was divided into pre-planting (March/April), planting (April/May), growing (May – September), post harvest (October/November) and an off-season period (January). In total, we collected 10 urine samples from each farmer: 9 spot samples and one post-application overnight urine sample (average collection period ~12 hrs). Urine samples at timepoint 2–5 were self-collected according to a standard protocol and kept cold at 4 degrees Celsius. Other samples were collected during a field research visit. The two farmers who participated in both years had 18 spot samples and two post-application samples. We collected only spot urine samples from controls: 4 samples from controls in the first year (urine collection timepoints 1, 7, 9, and 10; Table 1) and 6 samples from controls in the second year (urine collection timepoints 1, 6–10). The timing of urine collections was scheduled relative to atrazine application and seasonal farming activities (pre-planting, pre- and post-atrazine application during planting, growing, post-harvest, and off-season). In farmers, we first collected a pre-planting spot urine sample in March/April prior to any pesticide application. The second sample (pre-application) was collected the morning of the first atrazine application of the year for each farmer. The third urine sample (post-application) was collected starting at the completion of the first atrazine application through the first morning void the next day. The next 7 urine samples were then collected at specific timepoints throughout the year after the first atrazine application (i.e. average 1, 2, 30, 100, 195, and 250 days) and therefore not linked to any specific pesticide application. Urine sample collection among controls occurred within the date range of farmers’ sample collection for any given timepoint (Table 1).
Several questionnaires were used to collect information about pesticide use throughout the study period (Table 1). Each farmer and control was administered a baseline questionnaire when the first urine sample was collected in March/April. The farmers then completed daily diaries during the planting season (roughly 4 weeks, April-May), followed by weekly diaries for the remainder of the growing and post-harvest season (roughly 28 weeks, May-November). Controls completed weekly diaries from April to November (for roughly 32 weeks). Finally, each farmer and control was administered an off-season questionnaire by a nurse in January of the following year.
At each urine collection timepoint, approximately 50 ml of urine was retained. The urine samples were stored in coolers with icepacks and transported to the clinical laboratory in the Department of Pathology at the University of Iowa Hospitals and Clinics where they were aliquoted and stored at −80 °C until analysis. External quality control was based on splits of selected field urine samples submitted as blind duplicates to the laboratory.
Urine samples were analyzed for a standard panel of 14 non-persistent pesticides (pesticide (measured compound)): 2,4,5-T (2,4,5-T), 2,4-D (2,4-D), pyrethroid (3-phenoxybenzoic acid), acetochlor (acetochlor mercapturate), atrazine (atrazine mercapturate), isazaphos (5-chloro-1,2-dihydro-1-isopropyl-[3H]-1,2,4-triazol-3-one), coumaphos (3-chloro-4-methyl-7-hydroxycoumarin), pirimiphos methyl (2-diethylamino-6-methyl pyrimidin-4-ol), diethyl-m-toluamide (diethyl-m-toluamide), diazinon (2-isopropyl-4-methyl-6-hydroxypyrimidinol), malathion (malathion dicarboxylic acid), metolachlor (metolachlor mercapturate), methyl parathion (para-nitrophenol), chlorpyrifos (3,5,6-trichloro-2-pyridinol (TCPY)). The urine analysis panel covered six of the reported pesticides including four of the major pesticides used by the farmers.
Pesticide metabolites (or parent compounds) were measured in urine using a modification of a high-performance liquid chromatography–tandem mass spectrometry method (HPLC-MS) with atmospheric pressure chemical ionization that included confirmatory ions as well as quantification ions (Olsson et al., 2004). The target analytes were quantified using isotope dilution calibration. Creatinine in urine was determined using a commercially available enzyme slide technology (Vitros 250 Chemistry system, Ortho-Clinical Diagnostics) (Olsson et al., 2004). The analytical limit of detection (LOD) ranged between 0.1–0.2 μg/l for all pesticide analytes except for isazaphos (1.5 μg/l) and diazinon (0.7 μg/l).
The daily and weekly diaries provided information on some potential predictors of pesticide exposure (e.g., the total amount of product applied, acreage treated, duration of application, application method, personal protective equipment, personal hygiene practices). This information was used to identify predictors of exposure. Reported product names were linked to their respective EPA labels and various databases (i.e., http://entweb.clemson.edu/pesticid/) to identify the amount of active ingredient(s) (a.i.) in the product. The amount of product used, as reported in the questionnaires, was converted to total kg of a.i. based on the percentage or amount of a.i. in each product. Summary variables were then created for total kg of a.i. used the day before and 7 days before the urine collection. These timeframes were selected because they coincided with information we had on pesticide use from daily diaries in the planting season and weekly diaries for the remainder of the year. Furthermore, given the relative short half-lives of non-persistent pesticides one would not expect applications carried out more than 7 days prior to urine collection to be directly related to the specific urine sample. Similarly, we calculated a summary variable for acreage treated and duration of application for the same time periods.
The laboratory analytical precision for each pesticide metabolite was estimated as the CV and the intraclass correlation coefficient (ICC) from pairs of duplicate urine samples (n=30) (from both farmers and controls), which had been aliquoted and assigned random identification numbers prior to shipment to the laboratory. The CV was estimated as CV(%) = √(exp(s2 ws)−1)*100 where s2ws is the estimated within sample error variance obtained from a one-way analysis of variance of the ln-transformed levels of each pesticide (PROC MIXED of SAS) (Kim et al., 2006). The ICC was estimated as the between-sample variation divided by the sum of the between-sample variation and within-sample variation over all subjects.
Using cumulative probability plots, the urinary analytes were found to be best described by lognormal distributions and were ln-transformed for the statistical analyses. Standard measures of central tendency and distributions (i.e. arithmetic mean (AM), median, geometric mean (GM), range, and geometric standard deviation (GSD)) were calculated. Urinary concentrations reported as below the LOD were replaced with values equal to the LOD divided by 2 (Hornung and Reed, 1990) and adjusted for creatinine before analysis.
Correlation between urinary levels of atrazine mercapturate and possible exposure determinants (i.e., the amount of atrazine a.i. applied, acreage treated, and duration of application) was evaluated using the Spearman correlation (RSpearman) coefficient.
Mixed effect models (PROC MIXED) were used to evaluate differences between farmers and controls and between seasons within farmers and controls, and to analyze the association between potential determinants of exposure and the amount of AZM excreted in the urine of farmers. Other pesticides were not modeled because the numbers of observations with a pesticide application within 7 days prior to the urine collections were limited. 17 farmers had urine collections at timepoints 5 (first morning void) and 6 (spot sample during the day) on the same day and hence these samples can not be regarded as independent observations given the high level of autocorrelation. For these farmers the urine collection at timepoint 6 was omitted from the mixed model analysis. In these models subject was introduced as a random effect to account for repeated measures on the same person and to estimate the between- and within-person variance components. The within-and between-worker variability was also expressed as the ratio between the 97.5th and 2.5th percentiles of the log-normal within- and between-worker exposure distribution, and computed as exp[3.92*variance component0.5] (Rappaport, 1991). Several structures of the covariance matrix were explored. A compound symmetric covariance matrix was selected based on the Akaike’s Information Criterion (AIC). An intercept-only model without any explanatory variables was calculated to estimate baseline variance components using restricted maximum likelihood (REML) estimation. Possible demographic (e.g., age, body mass index (BMI)), lifestyle (e.g. past smoking status, alcohol consumption) and occupational exposure determinants (e.g. fixed effects) were first investigated in univariate models. Occupational exposure determinants evaluated were: amount of atrazine a.i. applied, acreage treated, duration of application, type of mixing/loading method (direct tank filling via a hose coupling to a bulk container; pre-mixed in a secondary container and transferred by pouring into the primary spray tank; scooping a solid directly into the tank), type of application method (backpack; hand-held spray gun; boom on tractor/truck open or closed cab), use of personal protective equipment (dust mask; gloves; coveralls; long sleeved shirt; hat), personal hygiene practices (washing hands; washing body), urine collection timepoint/season, and type of urine collection (e.g., spot sample, post-application sample). Multivariate models were then constructed using a forward stepwise procedure (Kleinbaum et al., 1998). The model was built in steps beginning with the variable with the lowest p-value (variables with p-values greater than 0.2 were not included) in the univariate models and the largest decrease of the AIC in the univariate analyses, and adding variables until further additions did not result in a statistically significant p-value for the added variable or earlier variables lost their significance (p-value greater than 0.1). AIC was used to compare different models. Separate analyses were performed for atrazine on the post-application urine samples only. In these analyses the second post-application sample from the two farmers, who participated in both years, was not included. All models were evaluated for pesticide concentrations adjusted and unadjusted for creatinine. Results did not differ, however, and therefore we only present creatinine-adjusted urinary pesticide levels. Median urinary creatinine concentration was 1.36 g/l (range 0.1 – 4.2) in our study samples.
The proportion of urine samples above the LOD of farmers and controls were compared using Fisher’s exact test.
SAS version 9.1.3 (SAS Institute Inc, Cary, NC, US) was used for all statistical analyses.
In total, 97 different products, containing 61 different a.i., were reported in the diaries. Table 2 identifies the 12 most frequently reported pesticides (reported by at least 25% of the farmers) and the number of farmers who had applied the corresponding pesticides 1 day and within 7 days prior to urine collections. All farmers (by design) and none of the controls reported applying atrazine. Except for glyphosate (n=29) and 2,4-D (n=25), other pesticides were applied by fewer than 50% of the farmers. The four pesticides applied in the highest quantities (kg of a.i.) were acetochlor, atrazine, glyphosate, 2,4-D and chlorpyrifos.
One farmer did not provide urine samples for timepoints 3–5 (planting). Laboratory results were not available for 4 farmers at timepoint 10 (off-season) and one control at timepoint 1 (pre-planting). In total, 367 urine samples were successfully analyzed resulting in 5034 analytical results.
Of the pesticides analyzed, only atrazine, 2,4-D, chlorpyrifos and acetochlor were used regularly by the farmers; thus, we report detailed results for only these four pesticides. Results of other measured urinary analytes are reported in appendix A. Estimates of CVs for the four pesticide analytes based on duplicate samples from the study were as follows (analyte (CV)): AZM (6.6%); 2,4-D (12.1%); acetochlor mercapturate (8.3%) (after exclusion of one outlier); and TCPY (32.7%). The ICC was ≥ 80% for all analytes.
Table 3 gives an overview of urinary levels for AZM, before and after stratification by season for both farmers and controls. Farmers had significantly higher GM urinary levels in the planting season compared to controls (p<0.05), but there was no difference in urinary levels between farmers and controls in the pre-planting/off-season or in the growing season (p>0.05). Statistically significant increases in urinary AZM were seen among farmers in the planting season versus the pre-planting/off season (p<0.05). Among controls the mean urinary pesticide levels of AZM did not differ by season (p>0.05). The proportion of samples above the LOD was higher among farmers than controls even when observations with an atrazine application within the last week were not included.
Not unexpectedly, given the seasonality of pesticide application, the within-worker variance was considerably higher than the between-worker variance for AZM (R0.95ww =1881 vs. R0.95bw =21). After stratification by season, the within-worker variance was reduced considerably (range of R0.95ww: 20–397). The same change was seen for the between-worker variance except for the planting season, where the R0.95bw increased (R0.95bw=156).
Most of the information on pesticide application practices reported in the daily and weekly diaries did not show much variation in application practices between farmers. Most farmers used a spray boom with an enclosed cab for applying pesticides (n=26) and direct tank filling as the mixing and loading method (n=25). Farmers also were similar with regard to hygiene and use of personal protective equipment practices (i.e., most farmers washed their hands and body at the end of the day and most farmers used gloves when mixing and loading pesticides but not when spraying the pesticide). However, there were large variations among farmers in the amount of atrazine applied and acreage treated with atrazine (Table 2). Because atrazine has a presumed relatively short biological half-life (24–28h) (Gilman et al., 1998), only applications occurring one day prior to urine collection were included in the analysis (n=66). Correlation between the duration of application and amount of atrazine applied (rSpearman=0.55, p<0.0001); the duration of application and acreage treated (rSpearman= 0.69, p<0.0001); and acreage treated and amount of atrazine applied (rSpearman= 0.72, p<0.0001) was moderate to high.
Statistical modeling of the determinants indicated that the amount of atrazine applied the day before urine collection was the best predictor of urinary AZM levels when all spot and the post application samples were included in the model (Table 4). Separate analysis restricted to the post application samples gave similar results. Identified exposure determinants explained 29% and 38% of the between- and within-worker variance, respectively (Table 4). There were no significant effects of BMI, past smoking status, and current alcohol consumption on AZM urine levels (p>0.05, not shown).
Of the 12 most frequently reported pesticides (Table 2) only biomarkers of 2,4-D, acetochlor and chlorpyrifos were quantified in the urine in addition to atrazine. Table 5 gives an overview of urinary levels for 2,4-D, acetochlor mercapturate and TCPY before and after stratification by season for both farmers and controls. Farmers had significantly higher GM urinary 2,4-D levels compared to controls in each seasons (p<0.05). These differences remained significant even after exclusion of urines collected within 7 days of a 2,4-D application.
No statistically significant difference between farmers and controls in urinary levels for acetochlor mercapturate and TCPY were found in any season except a borderline significant difference for acetochlor mercapturate in the planting season (p=0.09). Although, no difference was seen in average levels of all samples analyzed across the entire year, for acetochlor mercapturate a striking difference was observed in the proportion of samples above the LOD between farmers and controls (48% vs. 4%; Fisher exact test p<0.0001). This difference remained after exclusion of urine samples collected within 7 days of an acetochlor application (45% vs. 4%; Fisher exact test p<0.0001). A similar pattern (after exclusion of urine samples with an application in the last 7 days) was observed for atrazine (38% vs. 6%; p<0.0001), and 2,4-D (92% vs. 47%; p<0.0001).
Statistically significant increases in 2,4-D, and acetochlor mercapturate levels were seen among farmers in the planting season versus the pre-planting/off season (p<0.05). These differences remained significant for 2,4-D even after exclusion of urines collected within 7 days of the corresponding application. Interestingly, this pattern seemed to be independent of whether the farmer himself had applied 2,4-D during the growing season (n=25) or not (n=5). No statistically significant differences between seasons in GM urinary levels for TCPY were seen for farmers (p>0.05). GM urinary pesticide levels of each agent among the controls did not differ by season (p>0.05).
In this study we have investigated exposure patterns of atrazine and selected nonpersistent pesticides throughout a whole growing season among a group of corn farmers from the AHS cohort (Alavanja et al., 1996). Atrazine exposure levels as measured by urinary atrazine mercapturate was significantly associated with the amount of atrazine applied but explained only part of the variability in metabolite levels. This information is being used to assess daily pesticide exposure throughout a year for an ongoing molecular epidemiologic study on various immune effects in corn farmers (Vermeulen et al., 2005).
The goal of the exposure assessment in an epidemiologic study is to develop estimates of exposure for every agent of interest. In this study, 61 a.i. were reported. A standard panel analysis of the urine samples provided quantitative information on 14 a.i. (of which 6 overlapped with the 61 a.i. reported). The panel was developed to measure widespread exposures in nonoccupationally exposed subjects (Olsson et al., 2004) and was selected for this study for practical reasons. The urine analysis panel covered four of the major pesticides used by the farmers including two of the most commonly used herbicides (atrazine and 2,4-D) and the most commonly used insecticide (chlorpyrifos) and provided some valuable data on background levels of other pesticides in a farming population (Appendix A). Atrazine was the target pesticide in our study; therefore, urine collection scheduling was partially non-random with respect to atrazine applications. For other pesticides, the strategy was essentially random as urine collection days were selected a priori and independent of pesticide applications. Thus, except for atrazine, the application of the reported pesticides was not directly associated with urine collections. The data therefore have limited use for estimating exposure levels from application of other pesticides. However, the exposure data provided information on pesticide levels during seasons when the farmers are not applying pesticides, i.e., background levels. Based on the results of this study, it appears difficult to capture the application of more than one pesticide within a study if the sampling strategy is not targeted to specific pesticides of interest. To overcome this problem, and to limit practical and cost-related constraints associated with scheduled farm visits, self-collection of urine samples may be an alternative when several pesticides are of interest. In this study all samples collected as part of the self-collection protocol were obtained successfully and stored and documented appropriately until collected by the field-staff.
For atrazine, we found that urinary metabolite levels were directly related to recent application of the pesticide. However, a significant part of the between- and within-farmer variability in AZM levels could not be explained by differences in the total amount of a.i. applied or by any of the other recorded determinants. Differences in actual work practices, behaviors, metabolism and/or meteorological conditions might account for this. Under the assumption that unexplained variance leads to random misclassification our models can be used to predict exposure levels. However, the significant residual between-farmer variability observed in our model might indicate that the prediction error is not random and that this factor (although unexplained) needs to be accounted for in the prediction of the individual exposure levels. This can be done in our study by obtaining the empirical best linear unbiased predictors (EBLUPs) of the individual’s random effects parameter. Furthermore, there was some indication that even if atrazine was not recently applied, background levels (as assessed by the proportion of samples above the LOD) were higher among farmers than controls. However, these differences seemed to be minor as compared to differences in exposure levels during application. Therefore, it seems reasonable to estimate long-term exposure levels based on application information collected in diaries. It should, however, be noted that AZM is only one of the urinary metabolites of atrazine that can be measured in urine (Barr et al., 2007). Results on background levels should therefore be interpreted with some caution.
For 2,4-D, we found that farmers had consistently higher urinary levels and larger proportions of samples above the LOD as compared to controls throughout the year, including when the pesticide was not recently applied. This suggests that farmers may be exposed to 2,4-D through other occupational, environmental or dietary sources that are not strictly related to the actual application of 2,4-D. This also seemed to hold true for the few farmers that did not apply 2,4-D themselves during this particular growing season. This observation might hint towards a more general environmental factor leading to 2,4-D exposure. However, we cannot exclude that applications were not reported accurately in the diaries, (e.g. applications actually performed by a contractor and not the farmer). Nonetheless, the elevated background levels of 2,4-D in farmers compared to controls indicates that the exposure to 2,4-D during non-application days needs to be investigated to develop an accurate estimate of long-term exposure to 2,4-D.
For acetochlor mercapturate, background levels were only slightly elevated among farmers as compared to controls while for TCPY, no differences in background levels were detected. For these compounds, it would appear that farmers’ exposure is more directly related to the application of these compounds as we observed for atrazine.
The farmers in this study experienced large day-to-day variability in exposure to atrazine because of differences in the application pattern throughout the year. This causes a challenge in epidemiologic studies because a small number of measurements may lead to imprecise estimates of long-term average exposures. Determinants of exposure were therefore identified for which information across the whole year was available. Estimation of long-term exposure based on empirical modeling of exposure using information from exposure diaries offers an advantage over using measurement data alone, as it will allow us to estimate the exposure level for each farmer for every day/week throughout the season. In a previous study among pig farmers, it was found that applying empirical models instead of the few exposure measurements available on each individual can compensate for the loss of information due to unmeasured factors affecting exposure (Preller et al., 1995).
This exposure assessment study was designed to support a study of seasonal differences in immune markers related to atrazine applications and not for identifying all important determinants of atrazine exposure. As a result, the farmers were homogeneous with regard to many application characteristics. This likely limited our ability to identify other significant determinants of exposure such as glove use, washing hands, etc. which have been found to be important in other studies (Hines et al., 2001; Stewart et al., 2001). The model for atrazine may therefore not be suitable for predictions of urinary metabolite levels in other study populations with more diverse farming activities. Furthermore, as the sampling did not follow a strict probability sample procedure results might not necessarily reflect exposure circumstances among corn farmers in Iowa at large. However, the measured exposure levels among farmers after an application in this study are comparable to levels found in another study among Corn farmers in Iowa (Curwin et al., 2005b).
In summary, the results of this study show that urinary atrazine metabolite levels were largely driven by recent application of the pesticide, and therefore the amount of pesticide applied, duration of application or acreage treated are likely to provide a valid surrogate of exposure in the molecular epidemiologic study. We observed that farmers, independent if they applied 2,4-D themselves or not have higher urinary levels of 2,4-D compared to controls throughout the active farming season even when no application of the pesticide has occurred within the last 7 days before a urine collection. Furthermore, since we were not able to identify sources of the increased background level of these farmers, these sources should be investigated in future studies of 2,4-D exposure.
This research was supported by the Intramural Research Program of the National Institutes of Health, National Cancer Institute and funding from the Environmental Protection Agency, US.
The authors would like to acknowledge Cynthia J. Hines (National Institute for Occupational Health and Safety), Jane Hoppin (National Institute of Environmental Health Sciences), and Kent Thomas (U.S. Environmental Protection Agency), for useful comments on an earlier version of this manuscript.
The views expressed are those of the authors and do not necessarily represent the official policy of U.S. Environmental Protection Agency and the Centers for Disease Control and Prevention.