Plasma organochlorine levels and the risk of breast cancer: An extended follow-up in the Nurses' Health Study
The environmental organochlorines 2,2-bis(p-chlorophenyl)1,1,1,trichloroethane (DDT) and polychlorinated biphenyls (PCBs) have been implicated as potential causes of female breast cancer. We continued follow-up of our 1997 case-control study nested in the Nurses' Health Study cohort, adding 143 postmenopausal cases and controls to the original 238 pairs, and examining specific PCB congeners for the first time. We measured plasma levels of 2,2-bis(p-chlorophenyl)ethylene (DDE), the major metabolite of DDT, and PCBs prospectively, comparing women who were diagnosed with breast cancer between 1 month and 4 years after blood collection with control women in whom breast cancer did not develop. Median concentrations of lipid-adjusted DDE, total PCBs, and PCB numbers 118, 138, 153 and 180, assessed individually, were similar among the cases and controls. The multivariate relative risk of breast cancer for women in the highest quintile of exposure as compared with women in the lowest quintile was 0.82 for DDE (95% confidence interval [CI]: 0.49–1.37) and 0.84 for total PCBs (95% CI: 0.47–1.52), 0.69 for PCB 118 (95% CI: 0.39–1.22), 0.87 for PCB 138 (95% CI: 0.50–1.50), 0.83 for PCB 153 (95% CI: 0.47–1.48), and 0.98 for PCB 180 (95% CI: 0.55–1.75). Sub-group analyses were also performed. Overall, our results do not support the hypothesis that exposure to DDT and PCBs increases the risk of breast cancer. © 2001 Wiley-Liss, Inc.
The environmental organochlorines 2,2-bis(p-chlorophenyl)1,1,1,trichloroethane (DDT) and polychlorinated biphenyls (PCBs) have been implicated as potential causes of female breast cancer. Some of these compounds have demonstrated weak estrogenic properties in vitro1, 2 and have been shown to adversely affect reproduction and sex determination in wildlife.3, 4 Others have dioxin-like activity and are considered anti-estrogenic.5 Some PCBs have been observed to induce P450 metabolizing enzymes6 or affect the immune system7 in animal experiments. Although use and manufacture of DDT and PCBs were banned in the United States in 1972 and 1977, respectively; the compounds persist in the environment. Furthermore, PCBs and 1,1-dichloro-2,2-bis(p-chlorophenyl)ethylene (DDE), the main metabolite of DDT, are lipid soluble and stored in adipose tissue, in the lipid components of blood and in breast milk. Because they are resistant to metabolism and have long half-lives, measurements of these organochlorines in these biological media represent cumulative exposures over time.8, 9
Thirteen large epidemiological studies (each with at least 50 cases) have been published that assessed exposure to DDE and PCBs in either blood (plasma or serum) or in adipose tissue in relation to risk of breast cancer, including one from our group.10–22 A prospective study of breast cancer cases in New York found a significant increase in the risk of breast cancer with higher serum levels of DDE and a nonsignificant positive association with PCBs collected 1–6 months prior to diagnosis.20 In a San Francisco Bay area study of incident breast cancer cases diagnosed from 6 months to 20 years after blood collection, there was a nonsignificant elevated risk associated with DDE among African Americans and whites in ethnicity-specific analyses, but the risk in Asians and all participants combined was not elevated.15
Findings from subsequent studies have been inconsistent. Results from a retrospective case-control study in western New York state suggested an association between highly chlorinated PCBs and postmenopausal breast cancer risk among parous women who had never breast-fed.17 A Canadian study comparing levels of organochlorines in adipose tissue from breast cancer cases and benign breast disease controls observed a similar association in the same subgroup for Mirex, an organochlorine insecticide, but not for PCBs.10 These authors did observe elevated breast cancer risk with higher levels of PCB congeners 105 and 118 among premenopausal women (71 cases) and with higher levels of congeners 170 and 180 among postmenopausal women (143 cases).10 A case-control study of serum levels in a population in Colombia suggested an increased risk associated with elevated levels of DDE.18 However, most studies have not supported the hypothesis that exposure to DDT and PCBs increases the risk of breast cancer overall.11–14, 16, 17, 19, 21, 22 In the largest prospective study published to date, Helzlsouer et al.12 did not observe an elevated risk of breast cancer associated with PCBs and DDE, regardless of lactation status or decade of blood collection.
In 1997, we published an analysis of DDE and PCBs and breast cancer risk in a case-control study of 238 incident cases, nested in the Nurses' Health Study. We did not observe any evidence of an increased risk of breast cancer among women with relatively high levels of plasma DDE or PCBs.14 In the current analysis we expand the period of follow-up to include an additional 143 cases of invasive postmenopausal breast cancer. In this larger sample, we have assessed the association of breast cancer risk with specific PCB congeners, as well as total PCBs and DDE. We also evaluated risk in potentially susceptible sub-groups, including those defined by lactation, body mass index and other breast cancer risk factors. Our previous publication was criticized because we had not adjusted plasma organochlorines for triglycerides in addition to plasma cholesterol;23 in the present analysis we obtained triglyceride measurements and performed this additional adjustment.
The Nurses' Health Study is an ongoing prospective cohort study established in 1976 when 121,700 female registered nurses completed a mailed questionnaire on risk factors for breast cancer and other diseases. At enrollment, the participants were between the ages of 30 and 55 years and resided in 11 large states (California, Connecticut, Florida, Maryland, Massachusetts, Michigan, New Jersey, New York, Ohio, Pennsylvania and Texas). Every 2 years participants completed follow-up questionnaires to update information and to report the occurrence of breast cancer and other illnesses.
Diagnoses of breast cancer were reported on the biennial follow-up questionnaire. Nonrespondents were contacted by telephone, and deaths were identified through next of kin or searches of the National Death Index. Each diagnosis was confirmed by review of medical records and pathology reports, and estrogen receptor status and nodal involvement was extracted from these records.
From 1989 to 1990, 32,826 of these women provided us with a blood sample using blood collection kits supplied by the Nurses' Health Study. The participants sent the samples to our laboratory via overnight courier; 97% of the samples arrived within 26 hours of being drawn. Upon arrival, the blood samples were centrifuged and aliquotted into plasma, white blood cell and red blood cell components. The cryotubes were then stored in liquid nitrogen freezers at temperatures of no higher than −130°C.
Women who sent a blood sample were similar to other women in the cohort with respect to reproductive risk factors for breast cancer such as age at menarche, parity and age at the birth of their first child. Women who gave a blood specimen were slightly more likely to have a history of benign breast disease or a family history of breast cancer. These differences should not influence the internal validity of comparisons between case patients and controls in the sub-cohort of women who gave a blood specimen. The completeness of follow-up as a proportion of potential person-years through 1994 is 98% for this sub-cohort.
We defined cases as women who did not have a diagnosis of cancer (other than non-melanoma skin cancer) when they sent in the blood specimen and in whom breast cancer was subsequently diagnosed before June 1, 1994. Although we included premenopausal breast cancer (n = 46) and carcinoma in situ in the first period of follow-up (1989–1992), we restricted the 1992–1994 follow-up to postmenopausal invasive cancers due to budgetary limitations and the limited number of premenopausal cases. There were 381 eligible cases: 199 invasive and 39 carcinoma in situ cases diagnosed prior to June 1, 1992 and 143 invasive cancers diagnosed between June 1, 1992 and June 1, 1994. For each case, we selected a control woman who had not reported a diagnosis of cancer, matched on year of birth, menopausal status, month and time of blood collection, fasting status at blood draw and postmenopausal hormone use.
Assessment of covariates
Information on known and suspected breast cancer risk factors was obtained from the biennial questionnaires and the questionnaire completed by each participant at the time of blood collection. We included the following risk factors in the multivariate models: age at menopause, age at menarche, parity, age at first full-term pregnancy, body mass index (weight [kg]/height [m]2) at blood collection, history of benign breast disease and history of breast cancer in a mother or sister. Menopausal status, one of the matching factors, was defined based on a woman's response to the question whether her periods had ceased permanently. Women who had had a hysterectomy with one or both ovaries left intact were classified as premenopausal until the age at which 10% of the cohort had undergone natural menopause (46 years for smokers and 48 years for nonsmokers) and as postmenopausal at the age at which 90% of the cohort had undergone natural menopause (54 years for smokers and 56 years for nonsmokers). In the intervening years these women were classified as being of uncertain menopausal status and excluded from menopause-specific analyses.
The laboratory methods have been described in detail elsewhere.20, 24 Briefly, a polar extract of plasma lipids was enriched by chromatographic cleanup and then analyzed by gas chromatography with electron-capture detection. All steps were scaled appropriately for 0.50-ml aliquot volumes. Previously, using Nurses' Health Study specimens, we had demonstrated that the precision with the use of this volume and an optimized analytic procedure is similar to that with previous procedures using 1- and 2-ml aliquots.25 Results were obtained in parts per billion (ppb) of p,p′-DDE (which is equivalent to nanograms of DDE per milliliter plasma), the principal metabolite of DDT, and of 21 individual PCB congeners. For this analysis we calculated a sum of the 16 higher PCB congeners—compounds with retention times longer than that of DDE (pentachlorobiphenyls, hexachlorobiphenyls and heptachlorobiphenyls). This total is the equivalent of the sum of PCBs assessed in our previous publication14 and by others.15, 20 International Union of Pure and Applied Chemistry (IUPAC) congeners 118 (a pentachlorinated isomer), 138, 153 (hexachlorinated isomers) and 180 (a heptachlorinated isomer) accounted for an average of 64% of the total sum. Therefore, if one of these congeners was not detected then the sample was considered to be unreliable and was excluded from statistical analyses of total PCBs. Otherwise, when a minor congener was not detected, its predicted value was calculated from the available congeners using regression analysis. These prediction equations worked well; the correlations between the predicted values and the observed values ranged from 0.53 to 0.83 for all but one congener for which the correlation coefficients were between 0.24 and 0.36 depending on the equation used. In general, the predicted values for the undetected congeners were close to 0. In the one case in which the predicted value was negative, we set it to 0. We also performed sensitivity analyses evaluating the effect of setting all of the not detected congeners to 0, the limit of detection, half the limit of detection and the square root of the limit of detection. Additionally, we evaluated congeners 118, 138, 153 and 180 individually. The limits of detection were less than 1 ppb for both DDE and PCBs, based on a value that was 3 times the standard deviation26 of 24 determinations over the course of sample analyses of a quality-control plasma pool with approximately 1 ppb each of DDE and PCBs. The limit of detection for the individual congeners was 0.07 ppb. Both DDE and PCBs are stable in frozen blood; organochlorine levels in serum frozen at −20°C were unchanged over a period of 1 year (Wolff's unpublished data).27
Plasma sample sets were constructed to contain matched pairs of cases and controls (with the order of samples randomized) and were sent to the laboratory in batches of 12 pairs. In addition, each batch included 2 blinded split samples from pooled plasma from premenopausal or postmenopausal women. For each batch we calculated the coefficient of variation. Based on 35 possible batches, the median coefficient of variation was 5.0% for DDE and 12.0% for the sum of 16 PCB congeners. The median coefficient of variation for the 4 main PCB congeners ranged from 8.1% for congener 153 to 12.4% for congener 138.
Samples were assayed separately for total cholesterol and triglyceride levels. We calculated lipid-adjusted values using the formula described by Phillips et al.28 and present the data in units of micrograms of organochlorine per gram of lipid (μg/g). Both PCBs and DDE values were missing for one member of 5 case-control pairs and lipids were missing for one member of 4 pairs because the samples were lost during analysis. Total PCB values were missing for one member of an additional 2 pairs in which at least one of the 4 main congeners was not measured due to chromatographic interference.
We examined the distribution of risk factors for breast cancer within tertiles of levels of DDE, the sum of PCBs and the individual congeners 118, 138, 153 and 180 among the controls. For the continuous variables, the medians in the tertiles were compared using the Kruskal–Wallis test. The chi-square test was used to compare the categorical variables. Because the distributions of the organochlorines were not normally distributed, we used the Wilcoxon signed-rank test for paired data to compare plasma levels of the different organochlorines between cases and controls. We divided the control distribution into quintiles of each organochlorine and calculated the relative risk (RR) and 95% confidence interval (CI) for each quintile relative to the lowest quintile using conditional logistic regression, controlling for established risk factors for breast cancer in addition to the matching factors. Tests for trend were assessed by using the log-transformed organochlorines as continuous variables. To examine whether the associations with organochlorines were modified by conventional risk factors for breast cancer, we conducted unconditional analyses of tertiles of each organochlorine within strata of the risk factors for breast cancer, controlling for the matching variables. If there was an obvious order to the stratification variable then we used the Wald test to assess the statistical significance of the interaction term. Otherwise, we evaluated the interaction using the likelihood ratio test, comparing the model with indicator variables for the tertiles of the organochlorine and for the levels of the breast cancer risk factor with the model including these main effect variables and terms for each level of the risk factor cross-classified with the organochlorine tertiles. All p values are 2-sided. We also repeated these analyses, restricting the data set to invasive cases and their matched controls and to postmenopausal women.
The 238 women diagnosed with breast cancer before June 1992, their controls and the distributions of risk factors for breast cancer in the 2 groups were described previously.14 The addition of the 143 postmenopausal invasive breast cancer cases diagnosed between June 1992 and June 1994 and their controls did not substantially change these relationships. In the whole study sample, the median age of the subjects was 60 years (range, 43–69 years). Among the 298 matched pairs in which both the case and the control subjects were postmenopausal at the time of the diagnosis of the case, the median age at menopause was 49.5 years for the cases and 49 years for the controls. Maternal history of breast cancer, history of breast cancer in a sister and history of benign breast disease were more common in the cases than in the controls (p < 0.05); differences for other breast cancer risk factors were not statistically significant. Furthermore, levels of plasma triglycerides and total lipids as calculated by the Phillips equation28 were not independently associated with breast cancer risk (multivariate RR for fifth quintile of lipids compared with the first: 0.99 [95% CI: 0.59–1.65], test for trend: p = 0.73).
Among the controls, the relationships between established and suspected risk factors for breast cancer and plasma levels of DDE and the sum of PCBs were similar to those presented in our original publication.14 Across tertiles of DDE and the measurements of total and individual congeners of PCBs, plasma levels increased with age. There was a positive association of body mass index with lipid-adjusted levels of DDE (median body mass index 23.6, 24.4 and 24.9 kg/m2 in tertiles 1, 2 and 3, respectively, p = 0.04). Total PCBs and congeners 118 and 138 were not linearly associated with body mass index. However, body mass index decreased across increasing tertiles of congeners 153 and 180 (median body mass index for congener 153 tertiles = 25.0, 24.9, and 23.6 kg/m,2p = 0.004; and for congener 180 tertiles = 26.3, 24.1, and 23.3 kg/m,2p < 0.0001). Among parous women, more women in the lowest tertile than in the highest tertile of DDE had breast-fed their children for more than 6 months; however, this association was not statistically significant. Inverse associations of breast-feeding with levels of total PCBs or individual congeners were not evident. There was no evidence of a linear association with the other breast cancer risk factors.
Adjusting for plasma lipids did not substantially affect the relative ranking of the unadjusted organochlorine levels; the Spearman correlation coefficients comparing lipid-adjusted (μg/g) to non-adjusted levels (ppb) were 0.97 for DDE and 0.88 for the sum of PCBs. Consequently, results were similar, and therefore we present only the results from lipid-adjusted analyses.
Exclusion of 55 pairs in which the case was diagnosed within 6 months of blood sampling had little impact on these findings. The multivariate RR for the fifth quintile of DDE compared with the first was 0.76 (95% CI: 0.44–1.32), p for trend = 0.09, and the equivalent RR for PCBs was 1.00 (95% CI: 0.53–1.89), p for trend = 0.26. No consistent evidence was found for an association between DDE and PCBs and breast cancer risk when we extended this exclusion to include cases diagnosed within the first year after blood draw (N = 99).
There was no significant association of lipid-adjusted levels of DDE, sum of PCBs or the individual congeners of PCBs with breast cancer. Median levels of DDE were slightly higher in the controls than in the cases and levels of the different measurements of PCBs were essentially the same in the 2 groups (Table I). Restricting the analysis to the 333 cases with invasive breast cancer and their controls did not materially alter the results. There were also no associations when we limited the cases to estrogen-receptor-positive or estrogen- and progesterone-receptor-positive breast cancer. Results from analyses excluding the women who were premenopausal at diagnosis and their controls were also similar. Because of the potential for greater relative measurement error in quantifying the low concentration PCB congeners, we also calculated an exposure measure summing only the 4 most common congeners (118, 138, 153 and 180). This sum was highly correlated with the original sum of 16 congeners (Spearman r = 0.97), and the association with breast cancer risk was the same (results not shown). Fifty-four samples were missing values for at least one of the minor congeners. Compared with the analyses using the predicted values for these samples, the results were not affected by setting the missing congeners to 0 or to the limit of detection.
Table I. Lipid-Adjusted Plasma Levels of Organochlorines Among Breast Cancer Cases and Controls in The Nurses' Health Study
We found no evidence of any increase in risk in the highest categories of lipid-adjusted DDE, total PCBs or the individual congeners and risk of breast cancer (Table II). Comparing the fifth quintile to the first quintile, the multivariate RR for DDE was 0.82 (95% CI: 0.49–1.37) and for PCBs the RR was 0.84 (95% CI: 0.47–1.52). The equivalent RRs for the individual PCB congeners ranged from 0.69 for congener 118 to 0.98 for congener 180, and were not statistically significant. The multivariate RRs for the highest decile of each congener as compared with the lowest decile were all less than 1 and not statistically significant: congener 118: RR = 0.61, 95% CI: 0.27–1.38; congener 138: RR = 0.55, 95% CI: 0.25–1.22; congener 153: RR = 0.71, 95% CI: 0.32–1.54; congener 180: RR = 0.75, 95% CI: 0.34–1.68. Again, restricting the case definition to invasive or estrogen-receptor-positive breast cancers and restricting the study population as described above did not alter the results (results not shown).
Table II. Relative Risk of Breast Cancer According to Quintile of Lipid-Adjusted Levels of Plasma DDE, Sum of PCBs, and PCB Congeners 118, 138, 153 and 180 at Baseline in the Nurses' Health Study, 1989 to 1994
| μg/g lipid||0.007–0.427||0.428–0.703||0.708–0.955||0.955–1.441||1.466–6.054|
| Cases (n)||88||85||48||83||68|
| Controls (n)||74||75||74||75||74|
| RR matched (95% CI)||1.00||0.96 (0.62–1.50)||0.56 (0.35–0.89)||0.93 (0.60–1.44)||0.75 (0.47–1.21)||0.13|
| RR multivariate (95% CI)2||1.00||0.95 (0.59–1.53)||0.51 (0.31–0.86)||0.91 (0.57–1.47)||0.82 (0.49–1.37)||0.15|
| μg/g lipid||0.131–0.404||0.406–0.491||0.491–0.596||0.602–0.763||0.766–1.986|
| Cases (n)||86||65||65||80||74|
| Controls (n)||74||74||74||74||74|
| RR matched (95% CI)||1.00||0.74 (0.47–1.18)||0.74 (0.45–1.20)||0.90 (0.54–1.49)||0.83 (0.49–1.42)||0.67|
| RR multivariate (95% CI)2||1.00||0.73 (0.44–1.21)||0.75 (0.44–1.28)||0.85 (0.49–1.47)||0.84 (0.47–1.52)||0.56|
| μg/g lipid||0.014–0.045||0.045–0.060||0.061–0.074||0.074–0.101||0.101–0.313|
| Cases (n)||88||62||61||90||69|
| Controls (n)||74||74||74||74||74|
| RR matched (95% CI)||1.00||0.66 (0.40–1.09)||0.68 (0.41–1.11)||0.99 (0.60–1.66)||0.75 (0.44–1.28||0.77|
| RR multivariate (95% CI)2||1.00||0.68 (0.39–1.17)||0.62 (0.36–1.06)||1.02 (0.59–1.77)||0.69 (0.39–1.22)||0.67|
| μg/g lipid||0.004–0.066||0.066–0.087||0.087–0.108||0.109–0.142||0.143–0.402|
| Cases (n)||83||69||75||65||78|
| Controls (n)||74||74||74||74||74|
| RR matched (95% CI)||1.00||0.82 (0.52–1.31)||0.89 (0.56–1.43)||0.76 (0.47–1.25)||0.91 (0.55–1.49)||0.34|
| RR multivariate (95% CI)2||1.00||0.82 (0.49–1.37)||0.90 (0.53–1.50)||0.71 (0.41–1.20)||0.87 (0.50–1.50)||0.21|
| μg/g lipid||0.009–0.078||0.078–0.094||0.095–0.121||0.121–0.159||0.159–0.447|
| Cases (n)||89||58||75||69||79|
| Controls (n)||74||74||74||74||74|
| RR matched (95% CI)||1.00||0.63 (0.38–1.02)||0.81 (0.51–1.30)||0.77 (0.47–1.26)||0.87 (0.52–1.45)||0.39|
| RR multivariate (95% CI)||1.00||0.67 (0.39–1.14)||0.69 (0.41–1.15)||0.77 (0.45–1.31)||0.83 (0.47–1.48)||0.26|
| μg/g lipid||0.000–0.055||0.055–0.068||0.069–0.082||0.082–0.103||0.103–0.467|
| Cases (n)||89||65||62||63||91|
| Controls (n)||74||74||74||74||74|
| RR matched (95% CI)||1.00||0.70 (0.43–1.15)||0.65 (0.39–1.08)||0.67 (0.41–1.10)||1.01 (0.60–1.68)||0.50|
| RR multivariate (95% CI)2||1.00||0.70 (0.41–1.20)||0.65 (0.37–1.11)||0.70 (0.41–1.19)||0.98 (0.55–1.75)||0.67|
Results from analyses stratified on body mass index, lactation and region are presented in Tables III and IV for DDE and total PCBs, respectively. For the 4 individual PCB congeners, the RRs for the third tertiles compared with the first tertiles, and tests for trends and interaction are summarized in Table V. There was no evidence of an interaction of body mass index or of lactation with DDE. For the sum of PCBs, RRs in tertiles 2 and 3 were modestly elevated among women with a body mass index of less than 25 kg/m.2 However, the 95% CI included 1 and the test for trend was not significant. There was a statistically significant inverse trend with increasing PCBs among the obese women, and the overall test for interaction of PCBs with body mass index was statistically significant. This pattern of a positive but not statistically significant association with breast cancer risk among the lightest women and of an inverse association among the heaviest women was evident for congener 118, but not for the other individual PCB congeners. The statistically significant interaction with body mass index was not present in analyses in which PCBs were not adjusted for plasma lipids.
Table III. Relative Risks of Breast Cancer According to Tertile of Lipid-Adjusted DDE, Stratified on Body Mass Index, Lactation and Region of Residence
|Body mass index (kg/m2)|
| <25||70||73||1.00||51||70||0.73 (0.44–1.20)||74||64||1.19 (0.73–1.94)||0.51|
| 25–29.9||54||36||1.00||26||35||0.48 (0.25–0.95)||38||38||0.64 (0.34–1.21)||0.10|
| ≥30||23||15||1.00||11||19||0.32 (0.11–0.95)||25||22||0.75 (0.30–1.90)||0.59||0.223|
| Nulliparous||11||13||1.00||6||11||0.52 (0.13–2.07)||5||9||0.58 (0.13–2.65)||0.03|
| Parous, never lactated||40||34||1.00||25||38||0.57 (0.28–1.13)||37||46||0.70 (0.37–1.32)||0.29|
| Parous, ever lactated||96||77||1.00||57||75||0.60 (0.38–0.95)||95||69||1.13 (0.72–1.77)||0.65||0.464|
| Northeast||73||69||1.00||42||62||0.58 (0.34–0.99)||71||51||1.25 (0.75–2.06)||0.82|
| Midwest||37||27||1.00||17||25||0.44 (0.19–1.00)||13||21||0.44 (0.18–1.09)||0.02|
| South||21||18||1.00||16||19||0.72 (0.28–1.83)||25||21||1.01 (0.41–2.48)||0.33|
| West||16||10||1.00||13||18||0.52 (0.16–1.68)||28||31||0.61 (0.21–1.81)||0.44||0.124|
Table IV. Relative Risks of Breast Cancer According to Tertile of Lipid-Adjusted Total PCBs, Stratified on Body Mass Index, Lactation and Region of Residence
|Body mass index (kg/m2)|
| <25||48||63||1.00||66||66||1.39 (0.82–2.36)||80||77||1.41 (0.84–2.37)||0.81|
| 25–29.9||49||45||1.00||33||37||0.80 (0.42–1.49)||36||27||1.21 (0.63–2.31)||0.32|
| ≥30||28||15||1.00||19||21||0.40 (0.15–1.05)||11||19||0.26 (0.09–0.76)||0.01||0.023|
| Nulliparous||5||12||1.00||5||14||0.81 (0.18–3.68)||12||6||5.30 (1.06–26.57)||0.11|
| Parous, never lactated||43||37||1.00||27||40||0.60 (0.31–1.17)||31||40||0.68 (0.35–1.31)||0.54|
| Parous, ever lactated||77||74||1.00||86||70||1.19 (0.75–1.88)||84||77||1.04 (0.66–1.65)||0.61||0.024|
| Northeast||45||57||1.00||62||55||1.43 (0.82–2.49)||78||69||1.42 (0.83–2.42)||0.19|
| Midwest||32||17||1.00||19||26||0.35 (0.14–0.84)||16||29||0.25 (0.10–0.64)||0.003|
| South||22||26||1.00||20||18||1.34 (0.57–3.18)||19||14||1.73 (0.68–4.36)||0.80|
| West||26||23||1.00||17||25||0.67 (0.28–1.59)||14||11||1.47 (0.50–4.30)||0.95||0.014|
Table V. Relative Risks of Breast Cancer Comparing The Third Tertile to The First Tertile For The Individual PCB Congeners 118, 138, 153 and 180, Stratified on Body Mass Index, Lactation and Region of Residence
|Body mass index (kg/m2)|
| <25||1.28 (0.78–2.10)||0.74||1.18 (0.71–1.95)||0.83||1.21 (0.72–2.04)||0.95||0.91 (0.54–1.56)||0.56|
| 25–29.9||1.29 (0.66–2.52)||0.32||0.88 (0.46–1.67)||0.92||1.06 (0.54–2.07)||0.81||1.19 (0.62–2.29)||0.22|
| ≥30||0.28 (0.09–0.85)||0.005||0.06||0.50 (0.19–1.34)||0.06||0.13||0.47 (0.16–1.40)||0.07||0.19||0.76 (0.24–2.42)||0.10||0.88|
| Nulliparous||7.10 (1.42–35.6)||0.06||9.47 (1.80–49.9)||0.02||5.56 (1.08–28.6)||0.12||3.17 (0.70–14.4)||0.13|
| Par, no lactated||0.66 (0.34–1.28)||0.51||0.56 (0.29–1.09)||0.34||0.76 (0.38–1.50)||0.56||0.87 (0.46–1.65)||0.94|
| Par, ever lactated||1.06 (0.67–1.68)||0.59||0.01||0.92 (0.59–1.45)||0.36||0.003||0.92 (0.58–1.46)||0.37||0.06||0.87 (0.55–1.38)||0.32||0.33|
| Northeast||1.22 (0.72–2.08)||0.62||1.24 (0.73–2.11)||0.66||1.48 (0.85–2.59)||0.46||1.61 (0.96–2.70)||0.20|
| Midwest||0.31 (0.12–0.80)||0.003||0.30 (0.12–0.79)||0.01||0.27 (0.10–0.69)||0.01||0.42 (0.17–1.01)||0.004|
| South||1.99 (0.71–5.59)||0.11||0.86 (0.35–2.13)||0.57||0.79 (0.31–2.02)||0.64||0.53 (0.20–1.38)||0.53|
| West||1.41 (0.52–3.86)||0.77||0.02||1.29 (0.42–3.93)||1.00||0.11||1.74 (0.61–4.97)||0.55||0.02||1.04 (0.37–2.93)||0.75||0.01|
There was a statistically significantly elevated risk of breast cancer among nulliparous women, specifically those with the highest levels of total PCBs and congeners 118, 138 and 153. There was also some evidence of a linear trend. However, there were few nulliparous women; these results were based on 17 observations (5 cases) in the first tertile and 18 (12 cases) in the third, and should therefore be interpreted with caution.
In this cohort, plasma levels of DDE were greater for women residing in the western United States, and levels of PCBs were greater in the Northeast and Midwest than in the other regions of the country.29 We have included a discussion of the interaction with location because of this observed variation, and also because most U.S. studies published to date have focused on residents in the Northeast. Risks of breast cancer were non-significantly elevated in the upper 2 tertiles of total PCBs compared with the lowest tertile in the Northeast and South (Table IV). Again these observations are based on small numbers and confidence intervals are wide and not statistically significant. Breast cancer risk decreased with increasing levels of both DDE and PCBs in the Midwest (Tables III and IV). Similar results were observed in the Northeast and Midwest for the individual congeners and also in the South for congener 118 (Table V). The inverse association in the Midwest was independent of body mass index. There was no evidence of an association between DDE, PCBs and breast cancer within strata of age, age at menarche, age at birth of first child, number of children, history of benign breast disease or family history of breast cancer.
In this expanded follow-up with more than 50% more cases than our original report, we did not observe any convincing evidence of an increased risk of breast cancer among women with relatively high levels of plasma DDE (fifth quintile compared with the first: multivariate RR = 0.82; 95% CI: 0.49–1.37) or PCBs (fifth quintile compared with the first: multivariate RR = 0.84; 95% CI: 0.47–1.52). These results are more compatible with a null association than those we reported in our previous analysis, in which there was some suggestion of an inverse association with PCBs (fifth quintile compared with the first: multivariate RR = 0.66; 95% CI: 0.32–1.37). With a larger sample size we were able to examine plasma levels of the individual PCB congeners 118, 138, 153 and 180; none were associated with breast cancer risk.
In sub-group analyses, we observed a suggestion of a decreased risk of breast cancer associated with total lipid-adjusted PCBs and congener 118 among obese women. This association was not apparent for non-lipid-adjusted PCBs, suggesting that, although lipids were not associated with breast cancer risk overall, they may be a confounder among the heavier women. In contrast to the observation of Moysich et al.17 of an increased risk with total PCBs among parous women who had never breast fed, we observed a non-significant inverse association in a similarly defined sub-group. Interestingly, Aronson et al.10 found an elevated risk in this sub-group with higher levels of Mirex, an organochlorine insecticide, but not with PCBs. We did observe a suggestion of an increased risk of breast cancer with increasing levels of total PCBs and congener 118, 138, and 153 among the 62 nulliparous women. This observation is based on few cases, and therefore the statistical analyses are not stable. Furthermore, because this is only one positive association among many comparisons and because it has not been observed in other studies, conclusions about its importance should be made with caution. In this analysis, we present our data individually adjusted in units of μg/g plasma lipid, using the equation developed by Phillips et al.28 to derive total lipids from total cholesterol and triglycerides. This method could create negative confounding by blood lipid levels if they are directly associated with risk of breast cancer. However, in this population, there was no independent association. In the 1997 analysis we accounted for blood lipid content by regressing the log-transformed DDE and PCB values on plasma cholesterol concentration and adding the residuals to the predicted DDE or PCB level at the median cholesterol value. The organochlorine levels from the 2 methods are highly correlated (Spearman r: DDE 0.99, PCBs 0.95) and yield similar results with respect to breast cancer risk. Lipid adjustment serves to minimize intra-individual variability and to better approximate the concentrations in adipose tissue.28
The larger population size in this study compared with our first study gave us more power to perform sub-group analyses and to evaluate individual PCB congeners, which are measured with less relative precision than the total sum. Much discussion in the literature has revolved around the appropriate grouping of PCBs.17, 30–32 Different congeners have different biologic activities, including both estrogenicity and anti-estrogenicity;5 therefore, summing them together might obscure an association with breast cancer if one exists. We chose to evaluate congeners 118, 138, 153 and 180 individually because they were the congeners with the largest concentrations in the blood samples and therefore their measurements were the most stable. These 4 congeners are reported consistently in environmental samples, in other human populations, and they were major components of technical PCB formulations.33 Congener 118 is dioxin-like, and therefore potentially antiestrogenic and immunotoxic.32 Congener 138 might have limited dioxin-like activity.32 There is some evidence that congener 153 or its metabolite has estrogenic properties,34 and this congener and congener 180 have been demonstrated to induce CYP1A1.32 We did not observe evidence that any of these congeners were associated with either an increased or decreased risk of breast cancer. Dorgan et al.11 also prospectively evaluated the relationship of congeners 118 and 138 with breast cancer risk among 105 cases and controls in Columbia, Missouri. They reported a modestly elevated risk associated with higher levels of these congeners among the 54 women diagnosed within 2.7 years after blood donation; however, this association was not evident when the interval between diagnosis and blood donation was longer,11 and was not evident in our study.
The scope of our findings is limited. Even with the relatively large overall sample size, we still had small numbers in some categories in the sub-group analyses; therefore, we are limited in our ability to evaluate sub-groups or high-risk populations. We measured blood levels of DDE and PCBs in adult women more than 10 years after use and manufacture of these compounds were banned in the United States. Current levels are likely to be indicative of past exposures because of the long half-lives of the compounds and their resistance to metabolism, but the evaluations of risks associated with low levels from a general population sample, years after exposure, could be biased to the null due to non-differential measurement error. Furthermore, our conclusions may not be generalizable to exposures experienced decades ago in utero nor during adolescence. DDE and the higher chlorinated PCBs are the most persistent organochlorines and are detected consistently in biological specimens. Thus, they are the most logical candidates for a possible role in the etiology of breast cancer, even though they are not necessarily the most toxic. The absence of an association with DDE and PCBs does not rule out the possibility that other pesticides and environmental contaminants may be associated with breast cancer.
Since our original analysis was published in 1997, the body of literature on this topic has grown. Our results from this more detailed evaluation in the Nurses' Health Study, together with most of the evidence from other publications are reassuring that adult exposure to DDT and PCBs, as measured in blood plasma, is unlikely to be an important cause of breast cancer in the general U.S. population.
We are indebted to the nurse participants in this study. We also thank Rachel Meyer, Michelle Lachance, Kathryn Starzyk, Jeanne Sparrow, Karen Ireland and Nancy Niguidula for expert technical assistance.