Pesticide Residue Transfer in Thai Farmer Families: Using Structural

Nov 19, 2014 - *Phone: +44.(0)117.928.7239; e-mail: ... A conceptual model for take-home workplace exposures. Rachael M. Jones , Igor Burstyn. Journal...
0 downloads 0 Views 2MB Size
Article pubs.acs.org/est

Pesticide Residue Transfer in Thai Farmer Families: Using Structural Equation Modeling To Determine Exposure Pathways Hanhua Liu,† Chalalai Hanchenlaksh,†,‡ Andrew C. Povey,† and Frank de Vocht*,†,§ †

Centre for Occupational and Environmental Health, The University of Manchester, Manchester M13 9PL, United Kingdom Institute of Medicine, Suranaree University of Technology, Nakhon Ratchasima 30000, Thailand § School of Social and Community Medicine, University of Bristol, Bristol BS8 2PS, United Kingdom ‡

ABSTRACT: Use of pesticides in agriculture may lead to downstream exposure of farmers’ families to pesticide residues inadvertently taken home. Identification of the independent contribution of different exposure pathways from the farmer to their children can provide clear targets to reduce exposure of farmers’ children. Individual contributions of different pesticide transfer exposure pathways were investigated using structural equation modeling methods, and the benefits of these methods compared to standard multiple regression are described. A total of 72 Thai families, consisting of a farmer, a spouse, and a child, participated in this study. Family members completed a questionnaire and self-collected three spot morning urine samples in the spraying season. Urine samples were analyzed for diethyl phosphate, diethyl thiophosphate, diethyl dithiophosphate, dimethyl phosphate, dimethyl thiophosphate, and dimethyl dithiophosphate. A path model was developed based on an a priori hypothesized framework to examine the individual contributions of different exposure pathways that may directly or indirectly affect transfer of pesticide residues from farmers to their children. Transfer from the farmer to the child occurs indirectly, primarily through transfer to the spouse in the first instance, but also through contamination of the home environment. Clear targets for interventions are directly the reduction of farmers’ take-home exposures and indirectly frequent cleaning of the home to avoid buildup of pesticide residues.



INTRODUCTION Pesticides are used to prevent, destroy, repel, or mitigate any pest ranging from insects to animals, weeds, and microorganisms1 and, although also used in home and garden environments, are mainly used in large quantities in agriculture.2 Despite this, it has been shown that people can get exposed to pesticides at work, in the general environment near fields where spraying occurs,3 and also at home and from the consumption of contaminated foods.4 Farmers and farm workers who use pesticides at work can also contaminate their home environment and their families by transport via their skin, their working clothes, and their footwear.5,6 Although pesticides can have both acute and chronic health effects,7,8 chronic health effects are of particular concern to farmers’ families because acute effects are typically caused by experiencing a high dose on a single occasion, which may occur when working with pesticides but hardly ever happens in the home environment (except deliberately9) . The magnitude of any acute effect depends on both the toxicity of the product and the quantity absorbed10 and may include acute poisoning but also less severe symptoms, including headaches, dizziness, abdominal cramps, and blurred vision.10−12 Long-term (lowlevel) exposure has also been linked to a wide range of chronic conditions, including increased cancer risk,8,12 reproductive abnormalities,13 neurological effects,14,15 and immune suppression and hormone disruption.13 In that respect, transfer of © 2014 American Chemical Society

pesticide residues to farmers’ (young) children is of particular concern because continuing exposure from a young age, or even prenatally, into adulthood may result in developmental problems or increased risk of these chronic conditions at a later age.16 Moreover, because of less strict regulations and lesser familiarity with potential health effects, farmers and farmers’ families in the developing world may be at particular risk.17,18 About 47% of the area of Thailand has an agricultural use, and about 37% of its population is engaged in agricultural work,19 with rice, maize sugar cane, cassava, rubber, coffee, and various fruits and vegetables as the main crops in terms of planted area and economic value. Traditionally, farming in Thailand has been self-sufficient in using only natural fertilizers. However, following trends in other countries, pesticide usage, as well as production, in Thailand has also been increasing over the past 20 years.20 In 2010, Thailand imported approximately 118 152 tons of pesticides, which were primarily herbicides (85 821 tons) but also insecticides (19 709 tons) and fungicides (8485 tons). The majority of pesticides, ∼60%, are used for control of disease vectors (especially malaria and dengue), with about 39% used in agriculture and the remaining 1% for Received: Revised: Accepted: Published: 562

August 8, 2014 November 7, 2014 November 19, 2014 November 19, 2014 dx.doi.org/10.1021/es503875t | Environ. Sci. Technol. 2015, 49, 562−569

Environmental Science & Technology

Article

household use.21 The most common types of pesticides by Thai farmers are organophosphates (45%), carbamates (27%), and pyrethroids (18%), with the remaining 10% a variety of pesticides, including organochlorine.22 Farmers can get exposed to pesticides during their work, especially when spraying pesticides, with dermal exposure being the main route.23,24 However, there is also the risk of downstream exposure of their families when they inadvertently take pesticide residues home via their clothing, boots, and skin.5,6,25 Additionally, inadvertent exposure may also occur through drinking water, diet, and residential pest control.26−29 Data from Thailand25,30−32 as well as other countries33−36 have consistently shown that urinary metabolite levels of children from agricultural families were higher than those of nonagricultural, reference families. These studies further indicated that exposure occurred primarily through transfer of pesticide residues rather than direct exposure from spraying, since differences between agricultural and nonagricultural children were also present in the nonspraying season, while also exposure could have occurred from a variety of different environmental and domestic sources and through various pathways.17,33 In this study we aim to specifically investigate the individual contributions of different exposure pathways,37 using structural equation modeling methods, to assess transfer of pesticide residues from farmers to their children. By identification of the individual contributions of the different exposure pathways to total exposure of farmers’ children, interventions can be specifically targeted at reducing or eliminating the most important pathway and thereby significantly reduce protracted, low-level exposure of Thai farmers’ children to pesticides.

Ethical approval was obtained from the University of Manchester Committee on the Ethics of Research on Human Beings (TPCS/ethics/09225) and in Thailand from the Ethics Committee for Research Involving Human Subjects at the Suranaree University of Technology (ref 5621/0734). Urinary Measurements. Families were provided with urine sample collection packages containing polyethylene containers and an instruction leaflet about collection and storage. The families were asked to self-collect three spot morning urine samples in a week that the farmers were spraying pesticides and also provide a fourth sample on a randomly chosen day in the nonspraying season. Only the urine samples from the spraying week were used in these analyses. Samples were collected by the farmers, spouses, and children themselves prior to the farmer going to work and were stored in a refrigerator immediately after collection by the family until they were picked up on the same day by the research team. The samples were then transported on ice and stored in a freezer at −20 °C at the local health center for 3 weeks until they were transported to the laboratory where they were stored at −80 °C until analysis. The three samples for each person were combined into one sample and analyzed by the Toxicology Center and Professional Imagine Laboratory in Thailand for diethyl phosphate (DEP), diethyl thiophosphate (DETP), diethyl dithiophosphate (DEDTP), dimethyl phosphate (DMP), dimethyl thiophosphate (DMTP), and dimethyl dithiophosphate (DMDTP) using gas chromatography−mass spectrometry (GC−MS),38,39 with the limit of detection (LOD) ranging from 1 to 5 μg/L. Urinary creatinine was measured using the Roche Creatinine Plus assay40 using a Roche Hitachi automatic analyzer (model 912; Hitachi Inc., Japan). Urine results were corrected for creatinine, and samples were considered valid if creatinine concentrations were between 3 and 30 mmol/L. Urinary concentrations of all eight individual metabolites were added together to derive a total urinary dialkyl phosphate (DAP) metabolite concentration, expressed in micrograms per gram of creatinine. The total urinary DAP metabolite concentration was used as a proxy for total organophosphate exposure, assuming that urinary concentration would be highly correlated to absorbed dose. Statistical Modeling. The traditional or standard multiple regression methods describe a model in which all potential explanatory variables are regarded as coequal, their interrelations are unanalyzed, and the resulting unstandardized coefficients are expressed in their original units (partial regression coefficients). As a result it is not possible to answer the question regarding the relative importance of a set of causes to an observed phenomenon, because the correlations between predictors are ignored. Causal modeling using structural equation models (SEMs) provides a theory-oriented method that facilitates the inclusion of specifying bivariate hypotheses about interdependencies between variables using a generalized multiequation framework.41 It combines both the analytical element (e.g., statistical estimation) and the research element (e.g., the prior knowledge of the research team) and has the benefit over traditional multiple regression methods that it enhances the interpretation of the results because it enables us to describe the theoretical relations among variables.42 SEMs express the effect of predictors (directed paths) in semipartial coefficients. These can be describes as “the unique ability of a predictor variable to explain variation in a response variable that cannot be explained by any other predictor variable in the model”.42 In essence, the SEM is not just an estimation method



MATERIALS AND METHODS Population. The target population consisted of all farming families who lived in one of the six subdistricts of the Pakchong district in Thailand. In these areas mainly vegetable, fruit, and arable (rice, maize, and cassava) were farmed. Inclusion criteria were that a household included a currently active farmer or farm worker (aged 18−65 years) who would use pesticides during the study period, their spouse (aged 18−65 years), and one child (randomly selected if more children) aged between 2 and 12 years. No further inclusion criteria with respect to health status or occupational history were applied. Eligible families (n = 1120) were identified through local healthcare centers in the Pak-chong district from the list of identified families. After selection, the households were approached by a local healthcare officer between July and September 2010 in the first instance (40% agreed to participate). The households who agreed to participate were categorized on the basis of their main product (vegetable, fruit, and arable), and 25 eligible families were randomly selected from each category. In the final study population only 22 vegetable farming families were included because 3 families left the study area after inclusion in the study but during the data collection phase. The total sample size was 72 farmers, with their spouses and children. As far as we could evaluate, the distribution of nonresponder households with respect to their main product was comparable to that of the responders. Information on demographics, lifestyle, occupation, farming practices, pesticide handling, housework, and the children was obtained through self-administered questionnaires translated from English to Thai and back-translated by a second bilingual researcher for consistency and context. 563

dx.doi.org/10.1021/es503875t | Environ. Sci. Technol. 2015, 49, 562−569

Environmental Science & Technology

Article

were log-transformed prior to analysis to satisfy the assumption of normal distribution underlying the statistical models. Represented by the directed paths (the single-headed arrows) in Figure 1, as a source of exposure, pesticide storage was hypothesized to influence the exposure level of the farmer (farmer DAPs), the spouse (spouse DAPs), and the child (child DAPs). Further hypotheses for the model were that the farmer (farmer DAPs) was a source of exposure contaminating the home environment (home cleanliness) and could also transfer residues to the spouse and child through direct contact. The home environment (home cleanliness) subsequently influenced the exposure level of the spouse and the child, and the exposure of the spouse affected the exposure of the child through direct contact. There is a sizable literature on the appropriate sample size for SEMs.46−57 However, prescription of an appropriate sample size for SEMs (“rules of thumb”) has been challenged on the basis that the sufficiency of a sample size depends on critical factors such as the size of the model, distribution of the variables, amount of missing data, reliability of the variables, strength of the relationships among the variables, and researcher element; the SEM is an a priori method.45,46,48,51,52,58−61 On the basis of this literature, and despite the fact these analyses are only based on 72 measurements, we propose that our sample size will suffice on the basis that (1) the proposed model is developed upon our a priori understanding of different exposure pathways of pesticide residues to the farmer’s child, (2) the model is a noncomplex model with no latent variable (except that the error terms ε1−ε4 are treated as latent variables in the SEM), (3) the outcome variable is continuous, and (4) there are no missing data. The validity of the model will be tested following two approaches: (1) goodness of fit, that is, the ability of the proposed model to reproduce the data (i.e., the variance− covariance matrix), and (2) performance of the proposed model compared with that of a model using the standard multiple regression methods. We will present and discuss the results of such validity tests in the next section. The estimation method is maximum likelihood, with the standard errors calculated using the observed information matrix (OIM) estimator. Analyses were conducted in Stata v 13.0.

for a particular model in the way that multiple regressions are, but a means of investigating the mechanisms of causal action. Focusing on the families’ home environment, we developed a path model43 adopting the SEM methods (represented graphically in Figure 1) to examine the individual contributions

Figure 1. Path model of transfer of pesticide residues from Thai farmers to their children within the home environment.

of different exposure pathways to assess transfer of pesticide residues from Thai farmers to their children in a week they sprayed pesticides. The specification of the recursive (or hierarchical) model43,44 illustrated in Figure 1 reflects our a priori45 hypothesis of different exposure pathways of pesticide residues to the farmer’s child, to some extent based on our previous study in a similar population.25 All of the variables in Figure 1 are measured (observed) with complete data (i.e., no missing data). At the top left of Figure 1 is the variable “farmer DAPs”; this was a biomarker/indicator of the pesticide exposure level for the farmer, expressed by the farmer’s total urinary dialkyl phosphate metabolites in micrograms per gram of creatinine. At the bottom left is the variable “pesticide storage” indicating whether pesticides were stored in the home environment and was obtained from a self-reporting questionnaire; this was considered an exposure source of pesticides within the home environment. In the middle, the variable “home cleanliness” measured whether the farmer’s spouse reported cleaning their house at least weekly and was used as an indication of the cleanliness of the home environment. The variable “spouse DAPs” was a biomarker/indicator of the pesticide exposure level for the farmer’s spouse, expressed by the spouse’s total urinary dialkyl phosphate metabolites in micrograms per gram of creatine. On the right is the final outcome, the outcome of interest, indicating exposure of pesticide residues to the farmer’s child (“child DAPs”), expressed by the child’s total urinary dialkyl phosphate metabolites in micrograms per gram of creatinine. ε1−ε4 represent the error term or residual variance. Exogenous variables do not have the error term. Pesticide storage and home cleanliness were included in the models as dichotomous variables, while all measured urinary DAP metabolite levels



RESULTS Table 1 presents the descriptive statistics of the variables involved in this paper. A total of 86% of the participant families cleaned their house regularly, and 42% of them stored pesticides in their house. After log-transformation, distributions of urinary DAP measurements of the farmers, spouses, and children resembled Gaussian distributions. The average level of Table 1. Summary Statistics variable home cleanliness pesticide storage variable farmer DAPs (μg/g of creatinine) spouse DAPs (μg/g of creatinine) child DAPs (μg/g of creatinine) a

564

N

percentage

72 72 N GMa 72 72 72

4.68 4.04 3.12

GSDa

86 42 min

max

0.78 0.73 0.72

2.21 2.02 1.62

5.79 5.19 4.59

Geometric mean (GM) and geometric standard deviation (GSD). dx.doi.org/10.1021/es503875t | Environ. Sci. Technol. 2015, 49, 562−569

Environmental Science & Technology

Article

with those of the standard multiple regression model. The χ2 statistic indicates the model is not saturated. An overall RMSEA ≤ 0.06 with a 90% confidence interval ranging from 0.00 to 0.08 indicates a good fit.65 Methodology research, however, has shown that RMSEA over-rejects true models for “small” N (