The ability to characterize long-term health outcomes, exposure variables, and confounders and the ability to follow subjects over time present specific challenges in studies of chronic health effects in farmworkers. Zahm and Blair (1993), in discussing the feasibility of cancer studies in this population, touched on many of the obstacles facing epidemiologic research in this realm. Subsequent work by these and other investigators explored several interesting approaches to several of these obstacles.

The mobility of the migrant and seasonal farmworkers presents a significant challenge to investigations of long-term health effects (Quandt et al. 2002). Establishing a true cohort has been overcome by some through the use of fixed housing opportunities such as the Northern California Migrant Family Housing Centers (McCurdy et al. 2002). Such fortuitous situations represent the exception rather than the rule for migrant farmworkers. The difficulty of such studies in these populations is rarely described in the literature (Quandt et al. 2002 is an exception). Follow-up studies that fail to achieve successful follow-up either appear in publication as cross-sectional studies or fail to appear at all. Publication bias toward successful studies tends to weed out the reports that would illustrate the difficulty of cohort follow-up in any population. Information on the difficulty of following migrant farmworker populations no doubt suffers the same fate. However, a pair of methodologic articles provides limited insight into the challenge of conducting long-term follow-up in farmworkers. In a study of Wisconsin workers, Nordstrom et al. (2001) attempted to locate 100 randomly selected farmworkers 10 years after their registration in a Wisconsin clinic. Only 6 of 100 could be located in Wisconsin, and the vital status could be ascertained on only 56% of 46 subjects after a moderately intensive search in the registered home-base state of subjects. Cooper et al. (2001) had remarkable success in relocating farmers enrolled in chronic disease clinic studies in Starr County, Texas, with 90.8% of subjects relocated 10 years after enrollment. These two follow-up studies may demonstrate the difference in success because of the start point for the follow-up search but, as Zahm and Blair (2001) point out, are probably a reflection of a variety of factors, including permanent residence, the quality of community contact, and the presence of a long-standing community-based research program.

There are currently several studies of farmworker populations that are attempting to study exposures during sensitive periods of childhood development and health outcomes in this population. Researchers from the University of California have established a birth cohort and are following the children longitudinally. Although the farmworker population they are targeting is not as mobile as all farmworkers, they have described the extensive methodologic issues in accessing and maintaining the cohort, including the need for culturally appropriate assessment tools and the use of community workers in recruitment and maintenance of the study sample (Eskenazi et al. 2005).

In the conduct of studies on factors related to occupational exposures, such as pesticide studies, the work-place is often the best and most important venue for data collection. However, studies of farmworker populations rarely use the work-site as the source for study recruitment (Arcury et al. 2001; Eskenazi et al. 1999; McCauley et al. 2001; Thompson et al. 2003). The migratory nature of farmworker often precludes the possibility of conducting traditional occupational health studies in agricultural settings. The power differential between employer and the nonunionized farmworker (most farmworkers) plays an important role in determining the feasibility of worksite studies. Workers will rarely volunteer for a study if they perceive that such participation threatens their jobs.

Language preference among affected populations can be a substantial barrier to efficiently conducting population epidemiologic research on pesticide health effects. In the United States, the large majority of farmworkers are Spanish speaking, but increasing numbers of workers do not speak Spanish as their primary language. Farmworkers of other nationalities are also being seen McCauley et al. 2002; McDonald 2001). The National Center for Farmworker Health (2000) estimates that 84% of the migrant and seasonal farmworker populations speak Spanish as their primary language. A greater challenge is the presence of predominantly indigenous Mexican language speakers in the migrant workforce. Villarejo (2003) found that 8% of California farmworkers studied reported being of indigenous origin, and a small percentage spoke only indigenous languages. McCauley et al. (2002) found that a surprising 36% of adolescent farmworker subjects in Oregon spoke indigenous Mexican languages. Of particular importance is that several of these languages do not have a written form. Even when workers speak Spanish, low literacy limits a worker’s ability to respond to written material and interferes with participation.

Thus, multiple factors, some unique to the population and the work and others generalizable to deficiencies in our national surveillance systems, make study of health effects of pesticides, long and short term, difficult in farmworker populations. Hence, investigators have designed studies addressing these difficulties and have identified health effects in this population. Studies of neurobehavioral health effects are excellent examples of strategies that can be taken to conduct these studies.

Assessment of neurobehavioral performance requires specific laboratory support and resources. Specific training is needed so that the tests are administered in a standardized way to each participant. The World Health Organization Neurobehavioral Core Test Battery (NCTB) (Johnson et al. 1987) and the Adult Environmental Neurobehavioral Test Battery (Anger et al. 1994) rely on individually administered tests and require extensive training of the examiner to ensure standard administration across participants. Computerized tests batteries such as the Neurobehavioral Evaluation System (Letz 1990), the Behavioral Assessment and Research System (BARS) (Anger et al. 1996), and the Swedish Performance Evaluation System (Iregren et al. 1996) also require examiner training.

During the actual data collection period, quality-control checks are necessary to determine if the examiners follow the correct protocol as trained. A checklist that examines whether the correct protocol is being followed, forms are filled out correctly, and appropriate interactions are occurring between the testers and the study participants should be developed for each protocol and administered periodically throughout the study.

The resources available to implement neurobehavioral testing can also affect the number of participants that can be assessed in a study, a crucial factor of studies of farmworkers. Noncomputerized or paper-and-pencil tests can be administered only one-on-one to study participants. Furthermore, at the end of the study these data need to be manually entered into a database. Computerized testing allows the possibility of increasing the participant to examiner ratio. Ten adults with at least some high school education can be tested with only one examiner with the computerized BARS (Anger 2003; Rohlman et al. 2003). However, when testing farmworkers with < 6 years of education, it is possible to test only up to four participants with one examiner. With young children, a one-to-one ratio is necessary to help maintain motivation throughout data collection (Rohlman et al. 2001a). As children become older, their reading ability improves and they have more school and computer experience, so this ratio may be increased.

Neurobehavioral outcome protocols typically study people at one point in time in a cross-sectional design, comparing the performance of exposed populations with either the performance of a control group or established normative data (e.g., Anger 1990; Anger et al. 1999). Finding a comparable control group is essential for an accurate interpretation of these results (e.g., Blair et al. 1996; Kelsey et al. 1986). Demographic variables such as age, education, and cultural background or ethnicity influence performance on neurobehavioral tests (Anger et al. 1997).

Years of education in one country may not be equivalent to years of education in another country (Puente 1990). A group’s familiarity with testing protocols or computers can also affect the validity of study findings.

There are several ways of handling these variables. During recruitment it is important to try to select groups that are comparable on age, education level, gender and ethnic background, and computers and to control for these variables in data analysis. Learning or practice effects may also confound interpretation of results. Studies that have assessed participants more than once have found improvements from the initial and subsequent testing that makes the determination of the effect of pesticide exposure alone very difficult (Bazylewicz-Walczak 1999; Daniell et al. 1992; Maizlish et al. 1987; Rohlman et al. 2001a). Strategies should be implemented to flatten the learning or practice effect before the exposure. Rohlman et al. (2001b) has done repeated testing of cognitive and motor performance with migrant farmworker children and found that there is significant practice effect between the first and second session but minimal change between the second and third session. Therefore, the use of the second pre-exposure measure would appear to be a valid baseline for comparison with performance postexposure.

To build the case that specific functions are affected by pesticide exposure, it is important to have converging evidence. Although doubt may exist that it is possible to have reliable data that can be compared across studies, evidence suggests that such comparisons are possible. Anger et al. (1993) studied performance on neurobehavioral batteries conducted in 10 countries from four continents using the NCTB protocol. Although differences emerged, they could be explained by educational differences in the populations. However, performances on the six performance tests were remarkably similar across the nine countries in Europe, North America, and Asia that had similar educational levels.

Consistent findings of deficits on similar neuropsychological functions have been shown in studies of the same chemical exposure conducted by different investigators in different countries, languages, and cultures (Anger 1990, 2003). Most studies examining neurobehavioral performance and pesticide exposure have found that pesticide exposure is associated with deficits in cognitive and psychomotor function [see Kamel and Hoppin (2004) for a review]. However, an examination of the literature shows that some discrepancies do exist, different tests were affected in different studies, and in some cases no relationship between exposure and performance deficits was found.

These inconsistencies may be due to methodologic differences or to differences in exposure. Methodologic issues include the different formats and protocol used to assess neurobehavioral functioning in different populations. For example, 14 studies examining pesticide exposure included a variant of the digit span test (Table 2). Of these studies comparing exposed populations (defined by exposure or by occupational group), 4 showed significant deficits between exposed and control populations, 4 showed decrements in performance related to exposure, and 6 showed no decrements in performance. The digit span test demonstrates the difficulties when variants of the same test are used in different studies. The methodology used in this research needs to be standardized. Similar tests and protocols should be used to allow comparisons across studies.

Another important factor in explaining inconsistencies among studies is how exposure is defined (Alavanja et al. 2004). A range of exposure metrics, including living in an agricultural community (Cole et al. 1997), working on a farm (Fiedler et al. 1997; Kamel et al. 2003), specific job title (Bazylewicz-Walczak et al. 1999; Farahat et al. 2003; London et al. 1997; Roldan-Tapia et al. 2005), work history (Baldi et al. 2001), or acute exposure history (Rosenstock et al. 1991; Savage et al. 1988; Wesseling et al. 2002) have been used. The link between these classifications and actual exposure is often unknown, and it is possible that the amount of exposure in any given population varies considerably. To date, there have been no reports of an association between neuropsychological performance and any biological marker of exposure; however, organophosphate pesticides have a short half-life, and it is not likely that a short-term biomarker of exposure would be correlated with a cumulative effect on neurobehavioral performance. Biomarkers of effects of more cumulative exposures are needed.

Although there are many challenges in studying the farmworker population, using strict protocols, standardized measures, and quality-control procedures can help strengthen conclusions drawn from these studies and help to develop converging evidence. Equally important are the methods used for defining exposure. The variability in these methods can lead to incorrect conclusions. New and emerging biological techniques are being developed that will help identify exposed populations and allow accurate conclusions to be drawn.

Exposure to pesticides has been associated with cancers, degenerative neurologic diseases, and altered immune response, but the mechanism of action is unclear. Genotoxic potential is a primary risk factor for long-term health effects such as cancer and reproductive health outcomes (Bolognesi 2003) Hagmar et al. (2001) reviewed the usefulness of cytogenetic biomarkers as intermediate end points in carcinogenesis and concluded that chromosomal aberration (CA) frequency predicts overall cancer risk in healthy subjects, but such associations have not been found for sister-chromatid exchanges and micronuclei (Mn). Although the genotoxic potential of pesticides is believed to be low, genotoxic monitoring in farmworker populations could be a useful tool to estimate genetic risk from exposure to complex pesticide mixtures over extended lengths of time. To date, genotoxic biomarker studies of workers exposed to pesticides have focused on cytogenetic end points, including CAs, Mn frequency, and sister-chromatid exchanges. In the last decade, single-cell gel electrophoresis or the comet assay has been established as a sensitive and rapid methods for the detection of DNA single-strand breaks and incomplete excision repair (Fairbain et al. 1995). These biomarkers have been well developed with high interlaboratory reliability, but they are not specific to pesticide exposure and to date have not been associated with a risk for human cancers or other disease outcomes.

In a review of the literature of pesticide exposure and DNA damage, Bolognesi (2003) reported a positive association between occupational exposure to complex pesticide mixtures and the presence of CAs, sister-chromatid exchanges, and Mn in most of the studies, but a number of studies failed to detect excess cytogenetic damage compared with control populations. The conflicting results from cytogenetic studies were attributed to the nature of the agricultural study populations and the type of exposure to pesticides. In general, data from one study in one particular occupational setting cannot be used to draw conclusions on genetic risk in another occupational setting. However, most studies on cytogenetic biomarkers in pesticide-exposed workers have indicated some dose-dependent effects, with increasing duration or intensity of exposure. The type of exposure can affect the results such as use versus nonuse of personal protective equipment and greenhouse workers versus open-field workers (Carbonell et al. 1993; Dulout et al. 1985; Falck et al. 1999). Studies have indicated that the persistence of chromosomal damage is short-lived for acute exposure (Eastmond 2000), and that damage may drop during low exposure periods for seasonal workers (Scarpato et al. 1996). However, multiple studies have indicated increased chromosomal damage associated with years of agricultural employment and year-round employment (Bolognesi et al. 1993; Gomez-Arroyo et al. 1992; Scarpato et al. 1996). Many of these new biomarkers will not provide a definitive answer linking exposure to disease; however, the use of these biomarkers could provide additional information to the weight of evidence that suggests a particular exposure is a potential health risk (Toraason et al. 2004). In some cases a specific genetic lesion that can be identified in an exposed population may be found, but this will not always be the case.

The internal dose of a pesticide can be measured by concentration of the pesticide, its metabolites, or its reaction products. Reaction products (e.g., hemoglobin, albumin, and DNA adducts) can be viewed as an early biological effects or reactions that could lead to a potential health effect (Pirkle et al. 1995). Adducts can also be considered biomarkers of exposure (Costa 1996; Grissom 1995). These biomarkers reflect the dose of a certain agent or its metabolites that escapes detoxification and reaches its target protein or DNA. Adducts may form between blood components and toxicants such as pesticides when the toxicant reacts with the nucleophilic centers of nucleic acids (e.g., DNA) and proteins (e.g., hemoglobin and albumin) (Needham and Sexton 2000). These biochemical modifications precede structural or functional damage.

Oxidative damage is thought to be an important mechanism of damage for organophosphate pesticides (Banerjee et al. 2001; Halliwell 2002). Organophosphate pesticides can generate reactive oxygen species and alter cellular antioxidant systems (Bagchi et al. 1995; Delescluse et al. 2001; Flessel et al. 1993). Levels of products of oxidative damage in urine reflect overall damage to all tissues and organs in the body. More than 100 different oxidative modifications to DNA have been described (Loft and Poulsen 2001), and several DNA base oxidation products are known to be mutagenic, including 8-oxo-7,8-dihydro-2′-deoxyguanosine and thymine glycol (Halliwell 2002). The most studied and most abundant is the C-8 hydroxylation of the guanine base and glycol. Another potentially important mechanism for DNA damage, and ultimately cancer, is the generation of reactive species through peroxidation of lipids (Halliwell 2002; Marnett 2002). Malondialdehyde, one of the most abundant carbonyl products, can react with DNA to form adducts with deoxyguanosine, deoxyadenosine, and deoxycytidine that are mutagenic in bacterial and mammalian cells. As is the case with most of these biological markers of effect, specific oxidative modifications have not been associated with specific organophosphates and can be induced by multiple agents. Nonetheless, they offer significant potential in understanding the mechanism of action of these toxicologic agents and to make useful comparisons of exposure and potential health effects among exposure groups.

The greatest potential for new biomarkers of early effect lies in toxicogenomics, a field of study that examines how the entire genome responds to toxicants or other hazards (Toraason et al. 2004). The ability to monitor changes in gene expression as a result of environmental exposure holds great promise in our understanding of the effect of environmental toxicants on human health. Ideally, gene expression studies will allow scientists to identify changes in transcription associated with exposure and subsequent risk of developing disease. Studies of the reaction of genes to an exposure usually results in thousands of genes showing altered expression patterns. Analysis of these changes in expression requires sophisticated data analysis techniques, storage, and mining strategies. These methods need to be developed before markers of changes in gene expression can be widely used in epidemiologic studies, but studies have been reported. For example, Infante-Rivard et al. (1999) conducted a case–control study of childhood leukemia and residential exposure to pesticides and examined gene–environment interactions, finding increased interaction odds ratios among carriers of the cytochrome P4501A1m1 (CYP1A1m1) and CYP1a1m2 mutations when the mother during pregnancy or the child had been exposed to certain indoor insecticides. Longitudinal epidemiologic studies are needed to fully establish the predictive value of a change in gene expression and subsequent development of disease. Researchers who gain access to farmworker populations and who are able to follow them longitudinally should be encourage to bank genetic samples for future analyses as this area of science becomes more developed. These procedures would require that researchers be sensitive to the ethical, social, and legal issues related to obtaining genetic tissue for vulnerable, minority populations.