Community Participatory Approach to Measuring Farmworker Pesticide Exposure: PACE3 (R01 ES08739), is an ongoing translational research program addressing the health of Latino farmworkers and their families in eastern North Carolina. The primary community partner for this project is the North Carolina Farmworkers Project (NCFP) (Benson, NC); additional partners include Greene County Health Care, Inc. (Snow Hill, NC), and Columbus County Community Health Center, Inc. (Whiteville, NC). PACE3 used a longitudinal design in which data were collected from participants up to four times at monthly intervals in 2007. All sampling, recruitment, and data collection protocols, including signed informed consent, were approved by the Wake Forest University School of Medicine Institutional Review Board.
The design of PACE3 is based on the conceptual framework proposed by Quandt and colleagues 
. Farmworker pesticide exposure can be understood as the result of proximal, distal, and moderating forces. Proximal determinants are workplace and household behaviors that bring workers into contact with pesticides. These behaviors are themselves the result of the corresponding environments. Thus, work environments that have regular, effective safety training and have an organization of work in which workers are able to exercise some degree of judgment in their work practices are more likely to have workers who exhibit good pesticide safety behaviors (e.g., regular hand washing). However, beliefs or psychosocial stressors may reduce the effects of a supportive environment on actual behaviors. Community factors may affect the work and home environments, as well. If pesticides are ubiquitous, the work and home environments may have levels of pesticides that are beyond the control of workers. While the conceptual framework recognizes multiple pathways for pesticide exposure, these multiple pathways are not addressed in this descriptive analysis.
Data collection was completed in 11 counties with large farmworker populations, including Brunswick, Columbus, Cumberland, Greene, Harnett, Johnston, Lenoir, Pitt, Sampson, Wayne, and Wilson Counties. For these counties in 2007, conservative estimates by the North Carolina Employment Security Commission put the number of migrant farmworkers without H2A visas at 13,675 (36.2% of those in North Carolina), the number of migrant farmworkers with H2A visas at 2,995 (34.3%), and the number of seasonal farmworkers at 5,800 (22.8%). The agricultural production in these counties varies, but the major hand-cultivated and hand-harvested crops include tobacco, sweet potatoes, and cucumbers.
PACE3 used a two-stage procedure to select farmworkers. First, the three partnering agencies prepared lists of farmworker camps for the counties that they served. Camps were approached in order until each agency recruited a minimum number of camps and a specified number of participants. All camps that were approached agreed to participate. At the end of the first round of data collection NCFP had recruited 131 participants at 20 camps, Greene County Health Care, Inc., had recruited 64 participants at 11 camps, and Columbus County Community Health Center, Inc., had recruited 66 participants at 10 camps, for a total of 261 participants at 41 camps. NCFP encountered problems with three of the initially recruited camps, and these camps were replaced. Therefore, 44 farmworker camps participated in the study.
Second, participants in each of the camps were recruited. In camps with seven or fewer residents, all farmworkers were invited to participate. In camps with more than seven residents, eight to ten farmworkers were recruited with interviewers recruiting participants as they became available. A total sample of 287 farmworkers was recruited: 261 at the first round of data collection, and 26 at the second round of data collection. Of all farmworkers approached by the interviewers, 13 chose not to participate, for a participation rate of 95.7%. At the second round of data collection, 41 participants were lost to follow-up, 20 were lost at the third round, and 12 were lost at the fourth round. Four rounds of data collection were completed with 197 farmworkers, three rounds with 27, two rounds with 14, and one round with 49.
Data collection included four components: a detailed interview, a finger stick blood sample to measure cholinesterase, a saliva sample for genetic analysis, and a first morning urine void to measure pesticide metabolites and metals. Participants were given an incentive valued at $20 when they completed data collection for each round.
Data collection was completed from May through September 2007. Data collectors included eight fluent Spanish speakers, divided into three teams. One team was affiliated with the camps served by each of the three partnering agencies. A detailed interview was completed with the farmworker participants at each round of data collection. At every contact the questionnaire included items on living conditions and recent (in the 3 days before the interview) risk factors for pesticide exposure, including workplace activities and behaviors, household behaviors, psychosocial stressors, work environment, and household environment. At the first contact, the questionnaire also included items on participant personal characteristics (e.g., age, educational attainment) and current health status. The questionnaire used in these interviews was developed in English and translated by an experienced translator who was a native Spanish speaker familiar with Mexican Spanish. Validated Spanish language versions of scales were used. The translated questionnaire was reviewed by four fluent Spanish speakers familiar with farm work, and then pre-tested with 16 Spanish-speaking farmworkers and revised as needed.
Blood samples to measure cholinesterase were collected on blotter paper at each of the four interviews. Saliva samples for genetic analysis and first morning urine samples to measure metals were collected at the initial interview with each participant. Cholinesterase, metals, and genetic data are not discussed in this paper.
For the measurement of pesticide urinary metabolites, at the end of each interview the interviewer gave the participants urine collection containers with labels attached. Participants were instructed to fill the containers with their first void upon rising the next morning. They were assured that the urine samples would be tested for agricultural chemicals and metals only, and not for the use of alcohol, drugs, or any health conditions. They were asked specifically to only provide their urine in the containers, not that of any other workers in the camp. They were asked not to put any other fluid or chemicals in the urine containers. Participants placed their urine containers in a cooler with blue ice that was provided to them. Each morning a project interviewer stopped by the camp and retrieved the containers, transported them to the nearest of the three collaborating community partners, alloquoted the samples into labeled containers, and placed them in a laboratory freezer where they were stored at −20°C. The urine samples were shipped on dry ice to the Centers for Disease Control and Prevention in Atlanta, Georgia, using an overnight delivery service. These samples were analyzed for pesticide metabolites.
Six urinary DAP metabolites of OP pesticides (DMP, DMTP, DMDTP, DEP, DETP, and DEDTP) were measured in urine samples using the method of Bravo et al. 
. Urine samples were thawed to room temperature. A 2-mL aliquot of each sample was fortified with isotopically labeled internal standards, and then mixed. The urine samples were lyophilized overnight to remove all traces of water. The residue was dissolved in acetonitrile and diethyl ether and the DAP metabolites were chemically derivatized to their respective chloropropyl phosphate esters. The reaction mixture was concentrated, and the phosphate esters were measured using gas chromatography-positive chemical ionization-tandem mass spectrometry in the multiple reaction monitoring mode. Unknown analyte concentrations were quantified using isotope dilution calibration with calibration plots generated with each sample run. The reported LODs were 0.6 μg/L for DMP, 0.2 μg/L for DMTP, 0.1 μg/L for DMDTP, 0.2 μg/L for DEP, 0.1 μg/L for DETP, and 0.1 μg/L for DEDTP. To ensure quality data, additional quality control materials, fortified samples, and blank samples were analyzed in parallel with all unknown samples.
The agricultural season is divided into four periods. Period 1 was May 1 to June 8, Period 2 was from June 9 to July 7, Period 3 was from July 8 to August 5, and Period 4 was from August 6 to September 4. These periods were selected as they roughly corresponded to the major periods of the eastern North Carolina agricultural season, with the major activities being tobacco and sweet potatoes being planted in Period 1; cucumbers being harvested, tobacco being topped, and sweet potatoes being planted in Period 2; tobacco being topped and harvested in Period 3; and tobacco being harvested and cured in Period 4.
Two outcome measures are the detection and concentration for the six DAP urinary metabolites of OP pesticides: DMP, DMTP, DMDTP, DEP, DETP, and DEDTP. Detection indicates if any of the metabolites was found in a urine sample. Concentration is the amount of the metabolite measured in μg/L.
Three sets of measures are used to describe the participants and their environments. The first set includes the individual characteristics sex; age in the categories 18 to 24 years, 25 to 29, years, 30 to 39 years, and 40 years and older; educational attainment in the categories 0 to 6 years, 7 to 9 years, and 10 or more years; country of birth with the values United States, Puerto Rico, Mexico, and other; country of residence with the values United States, Puerto Rico, Mexico, and other; the three dichotomous measures of language speaks English, speaks Spanish, and speaks indigenous (American Indian) language; seasons in US agriculture in the categories, 1 year or less, 2 to 3 years, 4 to 7 years, and 8 or more years; worker type in the categories migrant worker or seasonal worker; and H2A visa status. The values for these measures did not change during the agricultural season.
The second set of measures describes crops in which participants worked for at least one of three days before the data collection for each of the four periods. Farmworkers are coded as to whether they worked at least one of the three days preceding the data collection, and whether or not they worked in tobacco, sweet potatoes, cucumbers, blueberries, strawberries, cabbage, sod, other vegetables (beans, green beans, chilies, corn, squash, tomatoes, potatoes, watermelon, melons), or other crops (cotton, peanuts, flowers, hay, soybeans, wheat). The three day look-back period corresponds with the period in which most OP pesticides are metabolized.
The third set of measures focuses on pesticide safety. Participants indicated if they had never received pesticide safety training, if they had not received training in the current year, and if they received training in the in the current year, whether they understood the pesticide safety training (some or none, most, all). Participants indicated if pesticides had been applied in their room or in the camp in the week before the interview. They indicated the number of days in the previous three days (0, 1, 2, 3) that they had mixed, loaded or applied pesticides, worked in fields in which pesticides had been applied in last 7 days, or worked next to fields in which pesticides were being applied.
Descriptive statistics, counts and percentages, were used to describe farmworker characteristics for the season as a whole (287 farmworkers in the sample) and for each time period (per the number of farmworkers interviewed in the respective period). Pesticide metabolites were described with counts and frequencies for detection and percentiles for concentrations. Concentrations were not adjusted for creatinine. A large number of values were below the LOD for each metabolite. To determine percentile values below the LOD, we used a robust regression on order statistics (ROS) approach to obtain percentile estimates for each metabolite [Helsel 2005
]. The ROS assumes that values below the LOD follow a lognormal distribution. A regression equation based on the normal scores of the values above the LOD was used to predict the censored observations. Finally, the estimated non-detected values were combined with the detected values to compute summary statistics as if no censoring had occurred. The robust ROS approach is different from a simple substitution method in the sense that only the corporate collection of estimates below the LOD is used to compute the percentiles. The 95% confidence intervals for the medians of the metabolites at each time period (except for DEDTP due to the lack of sufficient number of detects) were obtained using a bootstrapping method [Efron & Tibshirani 1993