This work was a collaboration between researchers from Emory University (Atlanta, GA, USA) and Chiang Mai University (Chiang Mai, Thailand), who are studying neurodevelopmental impacts of prenatal pesticide exposure in the Study of Asian Women and their Offspring’s Development and Environmental Exposures (SAWASDEE) birth cohort. Our target study population included healthy pregnant women, at all stages of pregnancy, using the antenatal care (ANC) clinic at Fang Hospital, Fang District, Chiang Mai Province (Thailand). According to the World Bank, 99% of pregnant women in Thailand received prenatal care in 2009 [35]. Chiang Mai Province was chosen based on previous work showing pesticide exposure in women and children in this area [25,36,37,38,39]. Study nurses enrolled 76 ANC patients during prenatal visits to Fang Hospital in January and February 2011. Participants were volunteers who provided informed written consent. Participation was limited to Thai nationals or foreigners (e.g., Burmese migrant workers) with health insurance cards who had resided in Fang District for at least nine months before enrollment. Guidance and oversight for this work was provided by the human subjects research committees of Emory and Chiang Mai Universities.

We developed our KAP questionnaire based on previous work conducted by others [31,40,41,42] as well as new questions we designed to collect pregnancy-specific pesticide information. Pesticide knowledge questions collected information on pesticide training, as well as knowledge about exposure routes, populations at risk of exposure, acute and chronic health effects, toxicity symptoms, and effective methods for preventing exposure. Pesticide attitudes questions collected information on beliefs about responsibility for safe use, susceptibility to health effects, pesticide effectiveness, and reasons for pesticide use. Safe practices questions addressed occupational pesticide use, home pesticide use, PPE use, and use of other safety precautions during and after pesticide spraying. Questions regarding pesticide practices were asked prior to pesticide knowledge and attitude questions to avoid biased answers that may result from reflection on pesticide hazards and risks.

We translated the questionnaire into Thai (and back-translated it to English), pre-tested it with co-workers at Chiang Mai University, and pilot tested it with seven ANC patients during July 2010. Feedback from the pre- and pilot-testing was incorporated into the final questionnaire design.

The full questionnaire in English is presented in the Supplementary Material. Questionnaires were administered via face-to-face interview in Thai language by two study nurses trained in basic interview techniques. Questionnaire responses were checked by verifying the electronic database against the original paper surveys to ensure accuracy.

We calculated knowledge, attitudes, and practices scores using previously published methods, where available. This resulted in seven scores—one measuring knowledge, four measuring attitudes, and two measuring practices. The scores generally did not follow normal or log-normal distributions, thus were dichotomized at the median for the majority of analyses. We calculated a continuous knowledge score that measured the percent of questions answered correctly. This continuous knowledge score was used in the majority of analyses. We also categorized knowledge scores greater than the median as a high degree of knowledge, and those below the median as a low degree of knowledge, using a method modified from Dasgupta et al. [43]. Following McCormack et al. [44], we considered “don’t know” responses incorrect.

We calculated four separate attitudes scores, including two ‘susceptibility’ attitudes scores: one ranging from 0–4 measuring attitudes about personal susceptibility to the health effects of pesticides, and the other ranging from 0–8 measuring the participant’s attitudes about her child’s susceptibility to the health effects of pesticides. For these scores, the highest values in the ranges indicated the strongest belief in susceptibility to health effects from pesticides, while a score of 0 indicated the weakest belief. Following Sam et al. [40], we tabulated a third attitude score, measuring the extent to which a participant believed she had a personal responsibility for the safe use of pesticides. This score ranged from 0–12, with higher values indicating a stronger acceptance of personal responsibility for safe use. We dichotomized these three attitude scores at the maximum score because approximately 50% of participants scored at the maximum on each. We calculated a fourth ‘pesticide usefulness’ attitude score to indicate the degree of the participant’s belief in the usefulness of pesticides. We calculated this score, based on the number of options participants specified as reasons for using pesticides, only for women who personally applied pesticides at work or at home. We dichotomized this score, which ranged from 0–13, at the mean (4.2) because it was approximately normally distributed.

We measured pesticide practices by tabulating the number of reported “risky behaviors” (defined in Goldman et al., [45]), or behaviors known or believed to be associated with pesticide exposure, that the participant reported engaging in, including: improper handwashing, delayed bathing, lack of protective clothing, improper storage of clothing, low frequency of house cleaning, eating fruits and vegetables directly from the field, wearing work shoes into the house, and wearing work clothes into the house [46,47,48,49,50,51,52]. We added an additional risky behavior to the Goldman et al. list [45]—storing pesticides in or around the home. Since not all of our study participants were agricultural workers, we used two separate measures for risky behaviors—those at work (for agricultural workers) and those at home (for all participants). We dichotomized both pesticide practices scores into ‘no risky behaviors’ or ‘some risky behaviors’. Detailed score calculation procedures are presented in the Supplementary Material.

Demographic and pesticide use characteristics of the 76 participants are presented in Table 1 and Table 2. The mean age was 26.0 (±6.8) years, and participants were relatively evenly distributed across the first (28%), second (33%), and third (39%) trimesters of pregnancy. Approximately half (45%) of the participants reported working in agriculture since becoming pregnant. Twenty-three (30%) women, all agricultural workers, reported that pesticides were applied at their job. Pesticides were applied in the homes of 39 (51%) participants since they became pregnant, with 21 (28%) personally applying those pesticides. There were no missing data for any of the variables included in the descriptive statistics.

Table 3 presents descriptive statistics for the seven KAP scores for agricultural workers, non-agricultural workers, and all participants together. The median overall knowledge score was 0.86 (IQR = 0.10), and knowledge did not differ significantly between agricultural and non-agricultural workers (Wilcoxon p = 0.10). However, agricultural workers scored significantly lower in attitudes about responsibility for the safe use of pesticides and significantly higher in the number of risky behaviors at home (Wilcoxon p < 0.01). Of the 34 agricultural workers, 16 (47%) reported regularly engaging in at least one risky behavior at work. Among all participants, 55 (72%) regularly engaged in at least one risky behavior at home.

Higher knowledge was significantly associated with having at least some formal education (chi-sq p = 0.03), receiving pesticide training (Fisher’s p = 0.03), and believing in personal responsibility for the safe use of pesticides (Wilcoxon p = 0.02). Higher knowledge was also marginally associated with being Thai (chi-sq p = 0.09). Women in their first trimester were more likely to be agricultural workers (chi-sq p = 0.02) and were less educated than women in later trimesters (chi-sq p < 0.01).

Table 4 presents results from the univariate and final multivariable regression models showing associations with risky behaviors at work and home (any versus none). There were no missing data in these analyses, though some questions were not applicable to all participants. In univariate tests of our study hypotheses, engaging in risky behaviors at home was marginally associated with the continuous knowledge score (D; Wald chi-square p = 0.10) and significantly associated with pregnancy trimester (E; Wald chi-square p = 0.04). Neither factor was associated with the number of risky behaviors at work (A, B; Wald chi-square p > 0.20).

The final multivariable model describing the association between the continuous knowledge score and risky behaviors at work contained knowledge as the only predictor. The association between knowledge and risky behaviors at work was positive (A; odds ratio (OR) = 1.1) but not statistically significant (p = 0.21; 95% confidence interval (CI) 0.9–1.4) and the likelihood ratio test showed low explanatory power (p = 0.20). The final model describing the association between the knowledge score and risky behaviors at home also contained knowledge as the sole predictor, with a negative (D; OR = 0.9) and marginally significant association (p = 0.10; 95% CI 0.7–1.0). Explanatory power was also marginally significant (likelihood ratio p = 0.09).

The univariate association between pregnancy trimester and risky behaviors at work was not statistically or marginally significant so the corresponding multivariable model was not constructed. The final model of the association between pregnancy trimester (first versus second/third) and risky behaviors at home included education as a significant confounder. Although pregnancy stage was significantly associated with risky behaviors at home in the univariate model (E; OR = 5.0, 95% CI 1.1–23.9), the association was not statistically significant when education was included (F; OR = 4.1, 95% CI 0.8–20.6).

Knowledge areas with the lowest median scores were pesticide toxicity symptoms and pesticide exposure intake routes (0.78 and 0.80, respectively, out of 1.0). Only 5% of participants indicated that they knew that different pesticides have different health effects. Although nearly all participants (96%) agreed that spraying pesticides in the home could harm their fetuses, 28 (37%) reported using pesticides in the home since becoming pregnant. Similarly, all of the 18 (24%) participants who reported not wearing gloves while using pesticides in the home expressed knowledge that wearing gloves was an effective strategy to prevent pesticide exposure. Women with these inconsistent responses were more likely than those with consistent responses to report using pesticides in the home before becoming pregnant, and more likely to report engaging in other risky behaviors including smoking and allowing agricultural workers to wear shoes into the home after work (chi-sq and Fisher’s p < 0.05).

The regression analyses presented in Table 4 identified populations with increased odds of engaging in risky behaviors and pointed to targets for intervention beyond simple knowledge dissemination. For example, we found that agricultural workers who engaged in more risky behaviors at home were significantly more likely to engage in risky behaviors at work (C; OR = 2.2, 95% CI 1.1–4.5). Having a job involving farm work before becoming pregnant, using pesticides in the home before becoming pregnant, having a previous child, and having a strong belief in the child’s susceptibility to pesticides were significantly associated with engaging in risky behaviors at home (G; Wald chi-sq p < 0.05).

In general, study participants demonstrated relatively high levels of knowledge about pesticides, with most answering over 80% of the knowledge questions correctly. Knowledge was higher among our participants than reported in previous studies of agricultural workers in Thailand [12] and other locations such as Bangladesh, Brazil, and Palestine [31,40,56]. This may be due to our inclusion of non-agricultural workers, who had higher levels of education and higher knowledge scores than agricultural workers.

Consistent with previous findings [19], attitudes on personal susceptibility to the health effects of pesticides were not associated with pesticide practices among the women in our study (Fisher’s p > 0.20). In our small cohort, beliefs in personal responsibility for the safe use of pesticides were lower among agricultural than non-agricultural workers. The agricultural workers were also more likely to engage in risky behaviors in the home. However, certain behaviors were only considered risky when the participant had a household member who worked in agriculture, which was significantly more common among participants who worked in agriculture themselves (chi-sq p < 0.01). Thus, while it is difficult to assess the true association between working in agriculture and engaging in unsafe practices in the home, we preliminarily conclude that the agricultural workers in our study likely have a greater potential for exposure to pesticides in the home due to the significant association with risky behaviors.

Risky behaviors were less common in our study population than in other populations, including pregnant agricultural workers in California [45] as well as other Thai agricultural worker populations [12,18,57]. While the women in our study may truly be less likely to engage in certain risky behaviors, our pilot tests showed that some participants may be unlikely to admit to engaging in risky behaviors due to a desire to please researchers. Other survey-based studies of behavior among populations of Asian women have introduced similar concerns about the potential for reporting or desirability bias [58,59]. Although we attempted to minimize this bias by asking the behavior questions before the knowledge questions, this bias may not be completely removed through a multiple choice survey conducted in a face-to-face interview [60]. Future studies might consider anonymous surveys or open-ended qualitative surveys, interviews, and focus groups to further minimize this bias (see Flocks et al. [29]).

In this study, risky behaviors were used as a measure of potential pesticide exposure during pregnancy to help identify women and children at greater risk of pesticide exposure and reveal potential targets for interventions. Higher knowledge was marginally associated with decreased odds of engaging in risky behaviors at home. On average, the odds of engaging in risky behaviors at home decreased by approximately 10% for every additional knowledge question answered correctly, preliminarily indicating that interventions designed to increase knowledge among pregnant women from all backgrounds could be effective at reducing exposures in the home. However, this association did not meet our criterion for statistical significance, thus a study with a larger sample size would be needed to confirm or reject the relationship. Knowledge was not significantly associated with risky behaviors at work in our study, perhaps also due to small sample size (n = 34 agricultural workers).

As we hypothesized, women in later stages of pregnancy (e.g., second or third trimesters) were significantly less likely to engage in risky behaviors at home. In addition, more women had worked in a job involving potential pesticide exposure, personally applied pesticides, or had pesticides applied in their home before becoming pregnant than after becoming pregnant. These observations may indicate that women alter their pesticide use behaviors when they become pregnant as well as when they advance to later stages of pregnancy. However, when we included education as a covariate in our models, the association between pregnancy trimester and risky behaviors at home was only marginally significant. In our small sample population, first trimester participants were less likely to be educated than participants in their second or third trimesters. Pregnancy trimester was not associated with risky behaviors at work, potentially due to small sample size.

The knowledge gaps we identified in this pilot study could be used to design knowledge-based interventions aimed at pregnant women in Thailand. However, these types of interventions may not be sufficient to reduce or eliminate prenatal pesticide exposures. For example, some of the women we surveyed reported knowledge of risky behaviors but engaged in them nonetheless. These women reported other unsafe practices and their previous pesticide use habits did not appear to be influenced by their becoming pregnant. These important behavioral observations should be considered when planning interventions in this or similar populations.

In addition, regression modeling showed that neither knowledge nor pregnancy trimester alone were significantly associated with risky behaviors, but that other factors were involved. The number of risky behaviors at home was significantly associated with increased odds of engaging in risky behaviors at work. Specifically, the odds of engaging in risky behaviors at work increased two-fold for each risky behavior the participants reported at home. We preliminarily interpret this to mean that interventions designed to decrease the number of risky behaviors at home may be effective in decreasing risky behaviors at work as well. In addition, risky behaviors at home might serve as a proxy for risky behaviors at work in future studies. Behaviors at home can be assessed among all pregnant women, in contrast to behaviors at work, which typically only apply to agricultural workers.

Four other covariates were significantly associated with increased odds of engaging in risky behaviors at home in our study. These observations may help future researchers prioritize sub-groups for educational interventions. Agricultural workers were more likely to report risky behaviors at home, so the additional finding that having an agricultural job before becoming pregnant was significantly associated with risky behaviors in the home was not surprising. However, it helps underscore the importance of identifying agricultural workers as a group at elevated risk of pesticide exposure during pregnancy. Similarly, we observed that participants who reported using pesticides in the home before becoming pregnant were more likely to report unsafe pesticide use behaviors while pregnant. In addition, women with a previous child were significantly more likely to engage in risky behaviors at home during the current pregnancy. This is consistent with previous findings that within certain populations in the United States, women were more likely to engage in potentially harmful behaviors such as use of tobacco, lower utilization of prenatal care, and failure to meet diet quality and nutritional recommendations during later pregnancies than during their first [61,62]. Thus, women who already have children may need reminders that safe pesticide practices during pregnancy are necessary to protect their developing fetus. Although the predictors identified in backward elimination (having a job involving farm work before becoming pregnant, using pesticides in the home before becoming pregnant, having a previous child, and having a strong belief in the child’s susceptibility to pesticides) can help identify populations at risk of exposure, these factors are not preventable through interventions implemented during pregnancy. Thus, our pilot study shows that pesticide knowledge may remain an important target for prevention activities.

Our questionnaire was based on previously published work, pre-tested, pilot tested, and edited to ensure accurate translation, coherence, and relevance. However, we did not conduct a full-scale validation study, such as in Sam et al. [40]. Further, knowledge may be better measured using open-ended questions, where participants are asked to provide information freely, without being influenced by potentially leading questions or a restricted number of choices. Additionally, we did not receive any critical feedback during pilot testing, so some women in our study may have wanted to please the researchers instead of answer questions accurately. Such a reporting bias may partially explain why safe practices were more prevalent in our study than in previous findings.

Our study is also limited due to small sample size (n = 76). A larger number of participants would likely be needed to detect true associations among the variables of interest within smaller subgroups such as agricultural workers or women who personally applied pesticides. Further, our primary ‘outcomes’ of interest were not direct measures of exposure, but proxy questionnaire responses. Although there is evidence that unsafe pesticide practices are associated with increased exposure in Thai populations [12,18], this association has not been studied among pregnant women in Thailand or their children. While it is practical to assume that risky behaviors lead to increased exposures in this population, actual exposure measurements would be needed to validate the assumption.

Finally, behavior change interventions may have only modest impact without changes to national and international pesticide policy. The context of pesticide and agricultural policy in Thailand is unique and challenging, with weak enforcement of existing regulations and the absence of a uniform system for pesticide management [63]. While addressing the issue of prenatal pesticide exposure at the community and societal level is outside the scope of this study, these areas offer another opportunity for intervention and the results of our KAP survey in this understudied population may be useful in such efforts.