The Collaborative Perinatal Project (CPP) was a prospective study of neurologic disorders and other conditions in children (Niswander and Gordon 1972
). From 1959 to 1965, the study enrolled 48,197 women upon presenting for prenatal care at 12 medical centers in the United States. Centers were located in Baltimore, Maryland; Boston, Massachusetts; Buffalo, New York; Memphis, Tennessee; Minneapolis, Minnesota; New Orleans, Louisiana; New York, New York (two centers); Philadelphia, Pennsylvania; Portland, Oregon; Providence, Rhode Island; and Richmond, Virginia. Eleven centers recruited participants from the prenatal clinics of a university hospital, and one (Buffalo) recruited from 13 private obstetric practices. The study was not intended to be representative of the United States, and each clinical site had its own sampling approach (varying from 10 to 100% of eligible women, either by enrolling a random or systematic sample or all women).
Women were ineligible if they were incarcerated, planned to leave the area after delivery, planned to place their child for adoption, or if they delivered on the day they were recruited for the study. Four percent of the participants were lost to follow-up before delivery. As part of data collection, the mothers were asked to donate nonfasting blood samples at approximately 8-week intervals throughout their pregnancies. Serum samples were stored in glass vials at –20°C with no recorded thaws.
There were 142,130 pregnancies among the 48,197 women, including 54,390 pregnancies prospectively (observed) captured by the CPP. The children were systematically assessed for the presence of birth defects and other outcomes at birth and through 7 years of age, with follow-up completed for approximately 75% of the children born into the study. All mothers provided verbal consent to participate (Hardy 2003
). The present study was approved by an institutional review board at the National Institutes of Health.
To evaluate our hypothesis, we used data from a nested case–control study of cryptorchidism and hypospadias among sons within the CPP. Details of the nested case–control study have been described previously (Longnecker et al. 2002
). Inclusion criteria were based on the characteristics of the mother and infant. The sole maternal inclusion criterion was the availability of a 3-mL aliquot of third-trimester serum. Inclusion criteria based on the characteristics of the infant included alive at birth, male sex, and singleton birth. Of the 28,444 boys born to mothers enrolled prospectively in the CPP cohort, 6,097 were not eligible for inclusion in the current analysis because there was no maternal 3-mL aliquot of third-trimester serum available (n
= 5,389), the son was not a singleton (n
= 441), or the son was not live born (n
= 267). Among the eligible 22,347 boys, there were 241 cases of cryptorchidism and 214 cases of hypospadias. Five boys had both cryptorchidism and hypospadias and were included in each group for analysis. For comparison, we randomly selected a group of boys (n
= 599) from the remaining eligible boys without a diagnosis of cryptorchidism and/or hypospadias. Controls were selected so that the control:case ratio would be > 2:1 for each case group.
Hypospadias (urethral opening on the ventral side of the penis) was defined as having a diagnosis any time within the first 7 years of life. The degree of hypospadias was not captured in the medical records. The diagnosis of cryptorchidism (failure of one or both testicles to descend into the scrotum) was made by pediatricians based on serial examinations that included inspection and palpation of the genitalia. We defined cryptorchidism as having had a diagnosis of undescended testis(es) at any time during the first year of life. Boys with undescended testis(es) after the first year of life were not considered cryptorchid because they may have had retractile testes. Among the boys with undescended testicle(s) at birth (n
= 138), all but one had documented orchiopexy or a subsequent observation of cryptorchidism in at least one of the three subsequent examinations (ages 4 months, 1 year, 7 years). For the remaining boys diagnosed as cryptorchid during the first year of life (n
= 103), study records indicated that the testicles were descended at birth, suggesting that these boys may have had acquired, rather than congenital, undescended testis (Barthold and Gonzalez 2003
). To evaluate whether a relationship with oxychlordane or trans
-nonachlor varied by type of cryptorchidism, we considered boys with testis(es) descended at birth separately in an additional analysis.
For analyses in CPP, the socioeconomic index was calculated as the mean of three percentile scores: education of head of the household, occupation of head of the household or chief wage earner, and family income. The score used to calculate the percentile for an occupation was based on the percentiles of education and income among persons with the same occupation (Myrianthopoulos and French 1968
Serum levels of chlordane (trans
-nonachlor and oxychlordane) were measured at the Centers for Disease Control and Prevention (CDC) after solid-phase extraction cleanup and dual-column gas chromatography using electron capture (Brock et al. 1996
). Serum levels of p,p
´-DDE) and 11 polychlorinated biphenyls (PCBs) were measured in the same laboratory, and the laboratory methods have been described previously (Longnecker et al. 2002
; McGlynn et al. 2009a
). Serum cholesterol and triglycerides were measured using standard enzymatic assays.
The between-assay coefficient of variation was 25% for trans
-nonachlor and 20% for oxychlordane. These were determined at concentrations of 0.52 and 0.57 μg/L (301 batches), respectively. Limits of detection (LODs) were 0.28 μg/L for trans
-nonachlor and 0.20 μg/L for oxychlordane; 29% of the values for trans
-nonachlor were below that value, whereas 31% of the values for oxychlordane were below the LOD. For values below the LOD, we used the signal recorded by the instrument, when available, because it is thought that the signals below the instrument’s LOD yield better estimates of true concentration than imputed values (Chevrier et al. 2010
). Undetected values were set to missing.
Statistical analysis. Of the 1,054 total subjects (241 cryptorchid, 214 hypospadias, 599 controls), 3% were missing data on trans-nonachlor. Of those with data on trans-nonachlor, 6% were missing data on other covariates, and 5% were also missing data on oxychlordane. A total of 971 (217 cryptorchid, 197 hypospadias, 557 controls) subjects were included in the analysis of trans-nonachlor, and a total of 919 (206 cryptorchid, 181 hypospadias, 532 controls) were included in the analysis of oxychlordane.
We categorized trans
-nonachlor and oxychlordane concentrations according to the quartile distributions in controls, with the lowest quartile serving as the reference category. Odds ratios (ORs) and 95% confidence intervals (CIs) for the association between trans
-nonachlor or oxychlordane and cryptorchidism or hypospadias were estimated using conditional logistic regression conditioned on study center (12 strata). We assessed a linear trend across quartile categories by including a single independent variable taking the value of the corresponding median of the category. Models were adjusted for serum p,p
´-DDE as a five-stratum categorical variable, total PCBs as a four-stratum categorical variable, and serum triglycerides and cholesterol as continuous variables. We included serum triglycerides and cholesterol as independent variables in all statistical analyses to account for interindividual variations in serum lipid concentration. The results of analyses that modeled lipid-adjusted trans
-nonachlor and oxychlordane produced results similar to those presented. We included serum lipids as a covariate in our model rather than using lipid-standardized chlordane concentrations because the latter may be prone to bias, depending on the underlying mechanism of the chlordane–lipid disease association (Gaskins and Schisterman 2009
All models included trans
-nonachlor or oxychlordane as the main exposure and were adjusted for total PCBs, p,p
´-DDE, triglycerides, and cholesterol as a priori
selected variables. Additional variables were assessed as potential confounders using the change in estimate method (Maldonado and Greenland 1993
), starting with all variables in the models with deletion of one by one in a stepwise manner. If, on deletion, the OR for the contrast of the highest-to-lowest chlordane strata or the OR from the trend test changed by ≥ 15%, the factor was considered a confounder and was included in the adjusted analyses. Potential confounding factors included race, maternal age, maternal history of previous live birth, season of birth, socioeconomic index, smoking during pregnancy, and gestational hypertension, as defined in . Additional potential confounding factors included the categorical variables (yes/no) hyperemesis gravidarum, history of infertility, menstrual cycle irregularity, estrogen use during pregnancy, and progesterone use during pregnancy, as well as the continuous variables age at menarche and weight gain during pregnancy. Socioeconomic index was the only variable that changed the OR by ≥ 15%. We also considered the effects of adjustment for the continuous variables birth weight and placental weight and the categorical variables preterm birth and small-for-gestational-age, even though these were potentially intermediate variables.
Characteristics of mothers and sons according to case–control status of the son, CPP, 1959–1965.
We evaluated effect modification by maternal age, race, smoking, prepregnancy body mass index (BMI), previous live births, triglycerides, cholesterol, serum p,p´-DDE, total PCBs, gestational hypertension, and socioeconomic index, using the cross-product terms. Variables were coded as defined in , with the exception of prepregnancy BMI, which was coded as < 25, 25–29.9, ≥ 30 kg/m2. We supplemented evaluation of effect modification by categorical variables with more than two categories by comparing the model fit statistics for models with and without the cross-product terms. If the p-value associated with the interaction term based on the likelihood ratio test had a value < 0.10, the degree of potential effect modification was further evaluated by examining tables stratified by the potentially modifying factor(s). Statistical significance was set at p < 0.05 for main effects based on two-sided tests. Statistical analyses were conducted using SAS statistical software package, version 9.2 (SAS Institute Inc., Cary, NC, USA).