Birth malformations and other adverse perinatal outcomes in four U.S. Wheat-producing states.

Chlorophenoxy herbicides are widely used in the United States and Western Europe for broadleaf weed control in grain farming and park maintenance. Most of the spring and durum wheat produced in the United States is grown in Minnesota, Montana, North Dakota, and South Dakota, with more than 85% of the acreage treated with chlorophenoxy herbicides such as 2,4-dichlorophenoxyacetic acid (2,4-D) and 4-chloro-2-methylphenoxyacetic acid (MCPA). Rates of adverse birth outcomes in rural, agricultural counties of these states during 1995-1997 were studied by comparing counties with a high proportion of wheat acreage and those with a lower proportion. Information routinely collected and made available by federal agencies was used for this ecologic study. Significant increases in birth malformations were observed for the circulatory/respiratory category for combined sexes [odds ratio (OR) = 1.65; 95% confidence interval (CI), 1.07-2.55]. A stronger effect was observed for the subcategory, which excluded heart malformations (OR = 2.03; 95% CI, 1.14-3.59). In addition, infants conceived during April-June--the time of herbicide application--had an increased chance of being diagnosed with circulatory/respiratory (excluding heart) malformations compared with births conceived during other months of the year (OR = 1.75; 95% CI, 1.09-2.80). Musculoskeletal/integumental anomalies increased for combined sexes in the high-wheat counties (OR = 1.50; 95% CI, 1.06-2.12). Infant death from congenital anomalies significantly increased in high-wheat counties for males (OR = 2.66; 95% CI, 1.52-4.65) but not for females (OR = 0.48; 95% CI, 0.20-1.15). These results are especially of concern because of widespread use of chlorophenoxy herbicides.

). Increased numbers of soft-tissue sarcoma were reported for exposed Italian female rice weeders (Vineis et al. 1986). Studies in the United States and Canada confirmed the association with non-Hodgkin lymphoma (Hoar et al. 1986;McDuffie et al. 2002;Woods et al. 1987;Zahm et al. 1990). Dose-response relationships with herbicide-sprayed acreage (mostly chlorophenoxy herbicides) were reported for mortality from non-Hodgkin lymphoma (Wigle et al. 1990) and prostate cancer (Morrison et al. 1993). Short-term immunosuppression, possibly related to cancer etiology, was observed in Italian farmers after application of chlorophenoxy herbicides (Faustini et al. 1996). The lymphocyte replicative index, a measure of cell division kinetics, was increased among applicators exposed solely to 2,4-D over 3 months .
A recent ecologic study investigated associations between cancer mortality and wheat acreage in agricultural counties of Minnesota, Montana, North Dakota, and South Dakota, where most of the spring and durum wheat produced by the United States is grown (Schreinemachers 2000). Because chlorophenoxy herbicides were used predominantly on wheat, wheat acreage per county was used as a surrogate for chlorophenoxy herbicide exposure. With increasing wheat acreage per county, increasing cancer mortality rates were observed for the stomach, rectum, pancreas, larynx, cervix, ovary, prostate, thyroid, bone, brain, leukemia, eye, nasal cavity, and oral cavity. Another ecologic study, investigating 1989-1992 Minnesota live births, observed increased rates of birth malformations in western Minnesota, where chlorophenoxy herbicides are applied to wheat (Garry et al. 1996). The comparison region was the urban/forested region of the state. This increase was observed not only among private pesticide applicators but also among the general population, suggesting nonoccupational exposure. Rates were highest for births conceived in the spring, peak time of herbicide application.
Both occupational and nonoccupational exposure to chlorophenoxy herbicides can occur through inhalation, ingestion, or dermal contact. 2,4-D has been measured in the urine of children living near a herbicide manufacturing plant and in the urine of adults exposed to recently sprayed turf (Harris and Solomon 1992;Hill et al. 1989). Detectable levels of 2,4-D were measured in semen of recently exposed farmers (Arbuckle et al. 1999b). 2,4-D was detected in indoor air and on surfaces inside homes after lawn application, with indoor activities, children's play, and pets resuspending indoor dust containing 2,4-D (Nishioka et al. 2001). Indoor airborne 2,4-D was associated with inspirable particles. Based on an experimental model, low levels of 2,4-D were expected to be found in residential carpet dust 1 year after turf application (Nishioka et al. 1996). A recent study on pesticide exposure to agricultural workers and their children observed pesticide metabolites in the children's urine (Curl et al. 2002). Although that study did not include 2,4-D, the results pointed to the possibility of a "take-home" exposure pathway. Atmospheric measurements of 2,4-D indicate drift beyond the target area and presence beyond time of application, depending on meteorologic conditions (Larney et al. 1999;Renne and Wolf 1979;Waite et al. 2002). Concentration of 2,4-D in rain has been known to exceed maximum contaminant levels occasionally [U.S. Geological Survey (USGS) 1997]. 2,4-D has been observed in urban and agricultural streams and rivers (USGS 2001).
The ecologic study presented here is based on results of the two previous ecologic studies, which indicated increased rates for two distinct health end points-birth malformations I thank J.P. Creason, S.E. Fenton, and V.F. Garry for their review of the manuscript and valuable comments, A. Herring for statistical advice, and the anonymous EHP reviewers for their useful suggestions. I am grateful to O. Axelson for his support and encouragement.
The research described in this article has been reviewed in accordance with U.S. Environmental Protection Agency policy and approved for publication. Approval does not signify that the contents necessarily reflect the views and policies of the agency.
The author declares she has no conflict of interest. and cancer mortality-in counties where wheat is the major crop and exposure of residents to chlorophenoxy herbicides is likely (Garry et al. 1996;Schreinemachers 2000). Birth malformations among 1995-1997 births in agricultural, rural counties of Minnesota, Montana, North Dakota, and South Dakota were investigated, comparing high-wheat to low-wheat counties. The purpose of this exploratory study is to identify a potential, regional health hazard. Results should be viewed in this light.

Materials and Methods
Information on newborns and infants for 1995-1997 births was obtained from the Linked Birth and Infant Death files, National Center for Health Statistics (NCHS 1995(NCHS , 1996(NCHS , 1997. Agricultural information on crop acreage by state and county and herbicide use by state was obtained from the U.S. Department of Agriculture 1992 census (USDA 1992a(USDA , 1992b.
To reflect populations more likely exposed to agricultural pesticides than urban populations, counties in Minnesota, Montana, North Dakota, and South Dakota were selected if at least 50% of the county's population was rural and if at least 20% of the county's land was dedicated to cropland. By selecting counties from these four states, a wide range for the percentage of land dedicated to wheat was obtained. A county was assigned to either the low-wheat or high-wheat group depending on its percentage of wheat acreage with respect to the median of all selected counties. Wheat acreage was used as a surrogate measure for exposure to chlorophenoxy herbicides.
White, singleton births to mothers 18 or more years old were selected if the birth's county of residence was included in the study.
The assumption was that county of birth would be the same as the county where the newborn was conceived and the parents lived during pregnancy. Only white, singleton births were included, thereby preventing unreliable results for race (only 18% of all births were to nonwhite parents) and excluding malformed or preterm births due to multiple gestations. The proportion of the following outcome variables in combined high-wheat counties was studied with low-wheat counties as the referent group: a) malformations diagnosed at birth; b) preterm birth, defined as gestational age < 37 weeks; c) small for gestational age (SGA), defined as birth weight below the sex-specific 10th percentile for California white, non-Spanish singleton births for given gestational age (Williams et al. 1982); and d) infant death from birth malformations. Malformations were investigated as presented on the birth certificates and in combination with other malformations based on organ system classifications.
Generalized estimating equations methodology using logistic regression was used to estimate the effects on perinatal outcomes in high-wheat counties compared with lowwheat counties (Proc Genmod, SAS/STAT software; SAS Institute 2001). This method accounted for the potential correlation between observations within a county. Odds ratios (ORs; the odds of the effect among the exposed, divided by the odds of the effect among the unexposed), and 95% confidence intervals (CIs) were used to compare frequencies of perinatal outcomes between high-wheat and low-wheat counties. Models were run for combined and single categories of birth outcomes, and for combined and single sexes, provided at least five observations were available for the anomaly category. If more than five observations per exposure group were available for combined sexes but not for one of the sexes, the analysis was conducted for combined males and females only, to prevent small-sample problems. An exception was made for urogenital malformations, which were presented for boys only, because they contributed most of the observations.
Covariate adjustment was performed only if an abnormal birth event was significantly increased in high-wheat counties compared with low-wheat counties. The following covariates were used: maternal age (≥ 35 vs. < 35 years); maternal education (less than high school vs. at least high school); marital status (not married vs. married); parity (first birth vs. second or higher birth); prenatal care (prenatal care in second or third trimester or no prenatal care vs. prenatal care in first trimester); previous preterm or SGA birth (yes vs. no); tobacco use during pregnancy (yes vs. no); alcohol use during pregnancy (yes vs. no); sex of child (male vs. female); time of conception [conception during April through June (time of herbicide application) vs. conception during other months]. Models were run only if at least five observations per covariate were available in each exposure group. A covariate with a p-value ≤ 0.1 for either the male, female, or combined male-female analyses was retained for the final set of covariates for all three analyses, male, female, and combined male-female.

Results
From among the 262 counties in Minnesota, Montana, North Dakota, and South Dakota, 147 agricultural counties with a mostly rural population were selected for the low-wheat (n = 73) and high-wheat (n = 74) groups (Table 1). The major field crops spring and durum wheat, corn, and soybeans were heavily treated with herbicides (USDA 1992b). Chlorophenoxy herbicides (2,4-D and MCPA) were applied predominantly to spring and durum wheat (88% of acreage) but also to some of the corn acreage (13%) and soybean acreage (4%).
Dicamba (3,6-dichloro-2-methoxybenzoic acid), with similar chemical structure and functional groups, was applied to spring and durum wheat (30% of acreage) and corn (48% of acreage), with corn receiving nearly a five-fold higher concentration. The other major herbicide used on corn was atrazine (33% of acreage). Soybeans were mostly treated with imazethapyr (70% of acreage) and trifluralin (47% of acreage). Some of the durum wheat, representing 3% of the total spring and durum wheat acreage, was also treated with trifluralin. Because among herbicides applied to spring and durum wheat, chlorophenoxy herbicides were the most predominant, and other crops were mostly treated with other herbicides, wheat acreage per county was used as a surrogate measure for environmental exposure to chlorophenoxy herbicides. Wheat acreage in 1992 was highly correlated with 1997 acreage (Spearman rank correlation = 0.99). The 1992 acreage was chosen as an estimate for the 1995-1997 time frame of this study.
The number of births during 1995-1997 in the selected counties was 43,634, including 51.4% males, 7.0% premature births, 4.7% SGA births, and 1.9% births diagnosed with malformations. Most births took place in a hospital (99.2%), and most were attended by a physician (95.4%). Absence or presence of malformations at birth was not confirmed on 2% of the birth certificates. Missing information for preterm and SGA births was less than 1%. The covariates sex, maternal age, and marital status were available for all births. Missing percentages for the other covariates were as follows: maternal education, 0.6%; Children's Health | Schreinemachers 1260 VOLUME 111 | NUMBER 9 | July 2003 • Environmental Health Perspectives parity, 0.2%; prenatal care, 2.0%; previous preterm or SGA birth, 1.9%; smoking or alcohol use during pregnancy, 15%. Covariate information specific to the low-wheat and high-wheat counties is presented in Table 2.
ORs comparing birth malformations and other prenatal outcomes in high-wheat counties with those in low-wheat counties are presented in Table 3. Anomalies available from birth records were analyzed as single categories, provided enough data were available, and as aggregate categories. Based on the 1989 revision of the U.S. Standard Certificate of Live Birth classification scheme (NCHS 1998), the following categories were included: all central nervous system anomalies-anencephalus, spina bifida/meningocele, hydrocephalus, microcephalus, other central nervous system anomalies; all circulatory/respiratory anomalies-heart malformations, other circulatory/respiratory anomalies; all digestive system anomalies-cleft lip/palate, rectal atresia/stenosis, tracheo-esophageal fistula, omphalocele, other gastrointestinal anomalies; all urogenital anomalies-genital malformations, renal agenesis, other urogenital anomalies; all musculoskeletal/integumental anomaliespoly-/syn-/adactyly, clubfoot, diaphragmatic hernia, other musculoskeletal/integumental anomalies; all chromosomal anomalies-Down syndrome, other chromosomal anomalies; all other congenital anomalies.
Significant increases were observed for circulatory/respiratory and musculoskeletal/ integumental anomalies among combined male and female births, and for infant death from congenital anomalies among boys. Among births with circulatory/respiratory anomalies [International Classification of Diseases, 9th revision (ICD-9 1989) 745-748], adjustment for the significant covariates (maternal age and conception during April-June, peak time of herbicide application) did not result in a change for the effect of high-wheat counties, as shown by the following adjusted ORs: combined male-female, OR = 1.64 (95% CI, 1.06-2.53); males, OR = 1.81 (95% CI, 1.05-3.11); females, OR = 1.65 (95% CI, 0.89-3.04). Further analysis of the circulatory/respiratory category showed that the strongest effects were observed for the "other" subcategory (ICD-9 747-748), which excludes heart malformations (ICD-9 745-746) but includes anomalies of the aorta, patent ductus arteriosus, other anomalies of the circulatory system, and all anomalies of the respiratory system. Among births with these "other" circulatory/respiratory malformations, the only significant covariate was conception during April-June. Adjustment for this covariate did not change the high-wheat effects presented in Table 3, as shown by the following adjusted ORs: combined male-female, OR = 1.99 (95% CI, 1.12-3.53); males, OR = 2.02 (95% CI, 1.01-4.02); females, OR = 2.06 (95% CI, 0.92-4.61). ORs for the effect of conception during April-June were as follows: combined male-female, OR = 1.75 (95% CI, 1.09-2.80); males, OR = 2.42 (95% CI, 1.42-4.15); females, OR = 1.02 (95% CI, 0.45-2.34). Given both wheat production and month of conception, boys conceived during April-June and born in high-wheat counties were almost five time more likely to be diagnosed with a birth anomaly coded as ICD-9 747-748 than were boys in low-wheat counties conceived during other months of the year. The difference of the seasonality effect between boys and girls was confirmed by a statistical test for interaction between sex and conception during April-June (p = 0.05). Lowwheat and high-wheat counties in separate analyses showed similar increases for circulatory/respiratory anomalies, excluding heart malformations. among boys conceived during April-June: low-wheat counties, OR = 2.04 (95% CI, 1.08-3.84); high-wheat counties, OR = 2.13 (95% CI, 1.26-3.64). In other words, the more than 2-fold increase for these anomalies among April-June conceptions was not specifically tied to chlorophenoxy herbicides. Other herbicide use needs to be considered, for example, dicamba.
Infant death from congenital anomalies was significantly increased among boys but not among girls, with most deaths caused by heart and musculoskeletal anomalies. This difference between boys and girls was confirmed by a statistical test for interaction between the high-wheat county group and sex (p = 0.002).
The male:female ratios of births with any congenital anomaly were 1.67 and 1.60 in the low-and high-wheat counties, respectively, whereas these ratios for all births were 1.07 and 1.03 for low-and high-wheat counties, respectively, suggesting that males may be more susceptible to congenital anomalies than are girls. Similar observations about the sex ratios have been made previously (Francannet et al. 1993;Garry et al. 1996Garry et al. , 2002Imaizumi et al. 1991).
Additional congenital anomalies, not diagnosed at birth, were identified from death certificates for 24 infants, including 20 infants with circulatory/respiratory anomalies. Combining these additional cases with those obtained from birth records, the OR for births with any anomaly (combined boys and girls) did not change (OR = 1.07; 95% CI, 0.87-1.30), whereas the OR for births with circulatory/respiratory anomalies decreased slightly (OR = 1.55; 95% CI, 1.04-2.32).

Discussion
Results from this study indicate that in rural, agricultural counties where wheat acreage occupies a larger percentage of the land and where use of chlorophenoxy herbicides is higher, anomalies of the circulatory/respiratory and musculoskeletal/integumental system significantly increased. To interpret these results, one should bear in mind the choice of reference group. The advantage of selecting rural, agricultural, low-wheat counties as Children's Health | Birth malformations in wheat-producing states Environmental Health Perspectives • VOLUME 111 | NUMBER 9 | July 2003 referent was that counties included in the study would be more alike, except for wheat acreage, the factor under investigation. The disadvantage was that the reference group was not a null-data referent. If, for example, a specific anomaly had been associated with both chlorophenoxy herbicides in high-wheat counties and other herbicides applied to corn and soy beans in low-wheat counties, no effect might have been observed for this anomaly in high-wheat counties. In other words, use of low-wheat counties as the referent may have produced an underestimate of effects in association with high-wheat counties. Although selection of urban counties as referent would have had the advantage of little or no exposure to agricultural herbicides, the disadvantage would have been that other, nonagricultural factors might also be involved in causing a lower level of birth malformations in urban counties. For example, easier access to prenatal care may be associated with elective abortion after prenatal diagnosis of an anomaly (Cragan and Khoury 2000). Underreporting may be more frequent in urban than in rural counties, because in large hospitals information on birth malformations is provided by obstetricians, in contrast to small hospitals, where pediatricians are the source of information (Hexter and Harris 1991). Therefore, selection of urban counties as referent would likely have overestimated the effects in high-wheat counties. The contribution of this study is that by selecting rural, agricultural, low-wheat counties as referent, effect estimates, although conservative, could be more specifically tied to chlorophenoxy herbicides and/or contaminants. Several limitations of this study need to be considered. Data for this study were based on birth certificates and therefore subject to significant underreporting (Snell et al. 1992). For example, only 28% of malformations recognizable at birth among 1989-1990 Georgia births, reported by the Metropolitan Atlanta Congenital Defects Program, were also reported by Georgia birth certificates (Watkins et al. 1996). In the present study, fewer than 2% of birth certificates were not marked for presence or absence of malformations, in contrast to at least 8% in urban counties. This supported the notion that reporting by small hospitals probably was more complete than reporting by larger hospitals in urban counties.
Malformations not recognizable at birth, especially the occurrence of heart malformations among newborns discharged earlier from the hospital, could have contributed significantly to underreporting (Gadow et al. 1996;Watkins et al. 1996). Combining birth malformations based on organ system classification may have resulted in risk estimates of malformations that do not share the same causal agents (Kogevinas and Sala 1998). Wheat acreage per county, an indirect measure of use of chlorophenoxy herbicides, would be appropriate if the amount applied was in direct proportion to wheat acreage, which may or may not have been the case in the selected counties. Results from this ecologic study were designed to estimate regional differences. Conclusions drawn at the population level are not necessarily valid at the individual level (Morgenstern 1995).
It is remarkable that given the data limitations, the results of the present hazard-identification study are consistent with a previous birth malformation study (Garry et al. 1996). A Norwegian agricultural study reported increased rates for central nervous system anomalies, cryptorchism, hypospadias, urinary system anomalies, and limb reduction in association with grain agriculture and pesticide purchase or grain agriculture and use of spray equipment (Kristensen et al. 1997). Conceptions in spring showed an increase in birth malformations, in association with grain farming. A New Zealand study compared incidence of congenital anomalies in specific regions during years of spraying with 2,4,5trichlorophenoxyacetic acid (2,4,5-T) (1972)(1973)(1974)(1975)(1976) with that in years of no spraying (1959)(1960)(1961)(1962)(1963)(1964)(1965). 2,4,5-T is chemically related to 2,4-D and MCPA. Births with any malformation, heart malformations, hypospadias, or clubfoot were significantly increased in births during the years of 2,4,5-T application (Hanify et al. 1981).
Other studies have reported increased levels of birth malformations in association with less specific pesticide exposures. An increase in transposition of the great arteries was reported among newborns, especially males, whose mother had been exposed to any pesticides during the first trimester (Loffredo et al. 2001). An increased risk for nervous system malformations, oral clefts, and multiple anomalies was Children's Health | Schreinemachers 1262 VOLUME 111 | NUMBER 9 | July 2003 • Environmental Health Perspectives reported among offspring of mothers involved in agricultural activities during 1 month before and 3 months after conception, with presumably low levels of pesticide exposure (García et al. 1999). An association was observed between orchidopexy rates and level of pesticide use in agricultural regions in Spain (García-Rodríguez et al. 1996).
Other abnormalities reported in association with chlorophenoxy herbicide exposure may contribute to adverse developmental or reproductive effects: for example, increased risk of abortion at less than 12 weeks of gestation for preconception exposure (Arbuckle et al. 1999a(Arbuckle et al. , 2001; or increased levels of asthenospermia, necrospermia, and teratospermia in farm sprayers who applied 2,4-D (Lerda and Rizzi 1991).
Toxicologic studies have reported adverse developmental outcomes in rodent models. 2,4-D is teratogenic (Schardein 1993). Pure 2,4-D has been shown to be maternally toxic, embryolethal, and a potential inducer of kidney and urogenital malformations in rats (Fofana et al. 2000). Supernumerary ribs were observed in rat litters treated with 2,4-D (Chernoff et al. 1990). A review article on 2,4-D safety concluded that reproductive and developmental effects occur in toxicologic studies, but mostly at maternally toxic doses, and that no effects were expected at the low levels humans were exposed to (Munro et al. 1992). However, recent studies indicate that chlorophenoxy herbicides at low doses do have biologic effects, although not necessarily teratogenic effects. Reduced litter size was observed in pregnant mice exposed to low and environmentally relevant doses (0.01 mg/kg/day) in their drinking water of a commercial formulation of herbicides consisting of 2,4-D, mecoprop, dicamba, and inactive ingredients (Cavieres et al. 2002). An increase of the lymphocyte replicative index was observed for in vitro tests at a low dose (0.005 mM) of commercial 2,4-D . Exposure to 2,4-D may involve endocrine disruption due to interference of 2,4-D with thyroid hormone transport carriers (Van den Berg et al. 1991).
Toxicity of chlorophenoxy herbicides is usually tested on pure or reagent-grade compounds. Biologic responses to these herbicides in presence of contaminants, adjuvants, and fertilizer may be higher. Contaminants present in technical grade 2,4-D depend on the purity of the chemicals used to produce 2,4-D, and on the production process (IARC 1986). Occasionally 2,3,7,8-tetrachlorodibenzo-pdioxin may be present in technical grade 2,4-D (Johnson et al. 1992). Adjuvants may contribute to adverse health effects (Garry et al. 1999).
In vitro assays showed that commercial grade 2,4-D at concentrations of 1 and 10 µg/mL caused cell proliferation, whereas assays with reagent grade 2,4-D did not. In a review on health effects of chlorophenoxy herbicides, Sterling and Arundel (1986) suggested that not only the contaminant dioxins and furans may have carcinogenic and teratogenic properties, but also the uncontaminated phenoxy herbicides themselves. Two toxicologic studies listed above seem to support this notion: Fofana et al. (2000) used pure 2,4-D, and Van den Berg et al. (1991) used chemicals "of the highest purity commercially available." It is not known whether the chlorophenoxy herbicides themselves, their contaminants and adjuvants, or a combination of these chemicals with other agents are involved with the observed excess of birth malformations.
To deal with the complexity of the current chemicalization of our environment, several investigators have suggested that individual risk-factor epidemiology may not be ideal because many of the exposures may be common throughout the study population in a given geographic region (Kogevinas and Sala 1998;McMichael 2002;Pekkanen and Pearce 2001;Shy 1997;Susser 1998). They suggested further that population studies are fundamental in identifying public health problems and that interregion comparison may offer comparison data for the population at risk. Enhanced sensitivity and resolution of environmental toxicant effects within one region can then be garnered by use of more sophisticated molecular toxicant studies in the affected population (Pekkanen and Pearce 2001).
In conclusion, although results from the present study should be viewed with caution, the consistency of these results with other studies points to a potentially hazardous scenario in terms of specific excess birth defects. Future, more targeted studies investigating developmental effects from chlorophenoxy herbicides and/or contaminants should focus on the toxicology of these chemicals at levels as they occur in the environment and on the routes of exposure for pregnant women living in high-wheat regions.