The incidence of non-Hodgkin's lymphoma (NHL) unrelated to HIV infection has steadily increased over the past several decades and remains substantially unexplained. Limited evidence suggests that increased concentrations of polychlorinated biphenyls (PCB) measured in blood or fat tissue are associated with increased risk of NHL. Although PCB congeners vary in their biological activity, the relation between individual congeners and NHL risk has not been examined previously using prospectively collected biospecimens. We examined congener-specific associations in three prospective cohorts. Prediagnostic serum or plasma concentrations of selected PCB congeners were measured among NHL cases and controls from these cohorts: Janus (190 cases and 190 controls) in Norway and CLUE I (74 cases and 147 controls) and the Nurses' Health Study (30 cases and 78 controls) in the United States. All blood samples were collected in the 1970s or 1980s. We used logistic regression to calculate odds ratios (OR) and 95% confidence intervals (95% CI) for the relations between risk of NHL and lipid-corrected plasma or serum concentrations. Several congeners (i.e., 118, 138, and 153) that were present at higher levels and were moderately to highly correlated with each other showed exposure-response trends with risk of NHL in all three cohorts. These associations were observed primarily among subjects diagnosed closer to the date of blood collection in the two cohorts with sufficient cases to permit stratification by time. Among cases diagnosed within the median years of follow-up (16 years in Janus and 12 years in CLUE I), ORs and 95% CIs for increasing fourths of concentration of congener 118 relative to the lowest fourth were as follows: 2.4 (0.9–6.5), 4.9 (1.6–15.3), and 5.3 (1.5–18.8; Ptrend < 0.005) in Janus and 8.1 (1.0–68.9), 6.6 (0.7–59.0), and 13.0 (1.6–106.8; Ptrend < 0.05) in CLUE I. Similar patterns were seen for congeners 138 and 153 and for total PCBs. Limited evidence of exposure-response trends was also observed for several other congeners. The primary 1,1,1-trichloro-2,2-bis(p-chlorophenyl)ethane metabolite, p,p′-DDE, was not significantly associated with NHL in most analyses but slightly to moderately confounded the PCB associations. The results from these three cohorts suggest that concentrations of certain PCBs in blood are associated with increased risk of NHL. [Cancer Res 2007;67(11):5545–52]
- polychlorinated biphenyl
- non-Hodgkin's lymphoma
The incidence of non-Hodgkin's lymphoma (NHL) dramatically increased in the United States, Europe, and elsewhere over the past several decades ( 1, 2). This increase is only partially explained by changes in diagnostic patterns, increased rates of HIV infection, and use of immunosuppressive drugs ( 2, 3). The rapid increase in incidence followed by an apparent leveling off among some demographic subgroups in some countries in recent years ( 4) suggests a strong and changing environmental component of this disease.
Polychlorinated biphenyls (PCB) comprise a group of persistent, lipophilic compounds ubiquitous in the environment. They were widely used in the United States beginning in the 1930s, until their production was banned in 1977 because of concerns about their environmental and biological persistence and their possible health effects ( 5). The United States Environmental Protection Agency and the IARC have listed PCBs as probable human carcinogens, particularly for liver cancer and melanoma, based largely on animal studies with some supporting epidemiologic evidence ( 5). 11 The different PCB congeners exhibit a variety of biological activities ( 6, 7), warranting examination of their effects individually. Many congeners have long biological half-lives, so their levels in blood reflect cumulative exposure over time ( 6).
Three previously published studies reported positive associations between serum concentrations of total PCBs and risk of NHL ( 8– 10). Only one of these ( 8) was based on sera collected before cancer diagnosis and only one ( 10) examined associations with individual PCB congeners. The timing of diagnosis relative to PCB exposure or measurement either was not or could not be examined in these studies.
In this article, we examine the risk of NHL in relation to blood concentrations of individual PCB congeners using a nested case-control design in each of three prospective cohorts, Janus, CLUE I, and the Nurses' Health Study (NHS). These CLUE I analyses are based on data from the study by Rothman et al. ( 8) but extend those previously published analyses to include risks associated with individual congeners and by time from blood collection. These three nested case-control studies, which are the only completed studies to date using prospectively collected blood to measure organochlorine levels, were included in the present article to highlight the similarities and differences in associations observed across diverse study populations. For space reasons, we provide supplementary study results online.
Materials and Methods
Janus. The Janus cohort includes ∼87,600 Norwegian men and women who provided blood samples during routine county health examinations between 1972 and 1978 ( 11). The cohort includes ∼90% of persons ages 35 to 49 years and a 10% random sample of persons ages 20 to 34 years at the time of their health examination who were residing in the sampled areas. Serum was separated and stored at −25°C.
Cohort members with no prior history of cancer who were diagnosed with NHL [International Classification of Diseases (ICD)-8 code 200 or 202] at least 2 years after the health examination and before 1999 were identified by the Cancer Registry of Norway. All 194 subjects who had sufficient serum (1.0 mL) and were judged via a special review of pathology reports to have NHL were included as cases. One control from the cohort was matched to each case by sex, county, age at examination (±1 year), and date of examination (±3 months). Demographic and other data were obtained from the Cancer Registry, Census, and Janus database from the original health examinations and other surveys.
CLUE I. The subjects and follow-up used in the CLUE I analyses were described previously ( 8). In brief, 23,938 residents of Washington County, Maryland participated in 1974 in the Campaign Against Cancer and Stroke (CLUE I). Subjects provided a blood sample and answered a brief questionnaire at that time. Serum was separated and stored at −70°C.
Cases comprised CLUE I participants with no prior history of cancer who were residents of Washington County and were identified through the Washington County Cancer Registry with a diagnosis of histologically confirmed NHL (ICD-8 code 200 or 202) between January 1975 and May 1994. Of 76 cases who had sufficient serum, 51 had slides available for pathology. Two of the 51 were judged not to be NHL. Exclusion of the 25 NHL cases without pathology slides available for review had little effect on NHL risk estimates and, therefore, were retained. Thus, 74 cases were included in this study. Two controls were matched to each case based on race, sex, date of birth (±1 year), and date of blood sample donation (±15 days).
Nurses' Health Study. The NHS cohort was established in 1976 when 121,700 female registered nurses completed a mailed questionnaire on risk factors for breast cancer and other diseases. Participants were 30 to 55 years old and resided in 11 large states at enrollment. Every 2 years, participants completed follow-up questionnaires to update information and report health status. Between 1989 and 1990, 32,826 of the nurses provided a blood sample. Plasma was separated and stored below −130°C. Follow-up of this subcohort through 1994, as a proportion of potential person-years, was 98% complete.
Cases of NHL (ICD-8 202) were identified via self-report on the biennial follow-up questionnaire. Nonrespondents were contacted by telephone, and deaths were identified through next of kin or the National Death Index. All diagnoses were confirmed by review of medical records and pathology reports.
As part of a pilot study, the 33 women with no prior history of cancer who were diagnosed with NHL between the date they provided the blood sample and May 1994 were included as cases. Controls consisted of 78 cohort members with no history of cancer through May 1994 and who were selected previously as controls for a study of organochlorines and breast cancer and whose plasma samples were analyzed for organochlorines at the same time as the NHL cases ( 12).
Janus. Concentrations of 36 PCB congeners [International Union of Pure and Applied Chemistry (IUPAC) congeners 11, 18, 28, 44, 49, 52, 66, 74, 87, 99, 105, 110, 118, 128, 138, 146, 149, 151, 153, 156, 157, 167, 170, 172, 177, 178, 180, 183, 187, 189, 194, 195, 196, 199, 206, and 209] were measured at the Centers for Disease Control and Prevention National Center for Environmental Health (NCEH) in 0.6 mL of serum, using previously published methods ( 13). In brief, serum samples were first spiked with isotopically labeled internal standards and then purified via automated accelerated solvent extraction and high-resolution gel permeation chromatography on a high-performance liquid chromatograph. Concentrated extracts were analyzed by gas chromatography/high-resolution mass spectrometry. Selected PCB congeners and organochlorine pesticides in each sample were quantified from 13C isotope dilution continuing calibration plots, which automatically corrected for extraction efficiency. Values below the instrumental limits of detection (LOD) were imputed as the instrumental LOD divided by ( 14). The pesticides measured in these samples were the same as those measured in the CLUE I samples ( 8, 15). Results of analyses of these pesticides will be reported separately, except for 1,1,1-trichloro-2,2-bis(p-chlorophenyl)ethane (DDT), which is included in the current analyses. Total lipid concentration was calculated for each subject using measurements of total cholesterol and triglycerides in an additional 0.1 mL serum ( 16). Masked quality control samples, including single samples from a large pool and pairs of replicate samples, were interspersed among study samples to assess interset and intraset variability. The serum samples from one case and three controls were not successfully analyzed, leaving 190 case-control pairs with organochlorine measurements.
CLUE I. The laboratory methods used for the CLUE I samples are described in detail elsewhere ( 8). Briefly, concentrations of 28 PCB congeners (IUPAC 28, 52, 56, 66, 74, 99, 101, 105, 110, 118, 138, 146, 153, 156, 170, 172, 177, 178, 180, 183, 187, 189, 193, 194, 195, 199, 203, and 206) were measured at the NCEH in 1.0 mL of serum, which first underwent solid-phase extraction and then the concentrated extract was analyzed via dual-column gas chromatography with electron capture detection ( 17). Congener concentrations were not adjusted for extraction efficiency, which was 60% on average (range, 40–77%) in serum samples containing general population levels, based on published values ( 17). Values below the instrumental LOD were retained because they likely provide the best estimate of the concentration of a compound ( 17). Several organochlorine pesticides, including DDT, were also examined in these sera and were reported elsewhere ( 8, 15). Total lipids were measured for each subject in an additional 0.5 mL serum ( 16). Quality control samples were used as described above. The serum sample from one control was not successfully analyzed; therefore, our analyses included 73 complete case-control sets (i.e., one case and two controls) and one set with one case and one control.
Nurses' Health Study. The laboratory methods used for the NHS samples have been described in detail previously ( 18). Briefly, a polar extract of plasma lipids from 0.5 mL plasma was further enriched by chromatographic cleanup and then analyzed for 21 PCB congeners (IUPAC 28, 66, 74, 82, 99, 101, 105, 118, 128, 138, 141, 153, 156, 170, 171, 174, 180, 183, 187, 199, and 203) by gas chromatography with electron capture detection. Analyses were done at the Mount Sinai School of Medicine Environmental Health Sciences laboratory. Congener concentrations were not adjusted for extraction efficiency, which was 76% to 82% in serum pools with concentrations near the mean of the samples. Missing values for minor congeners that were not detected were imputed from the available congeners using regression analysis ( 12). Values below the instrumental LOD were retained. Total lipids were measured in an additional 120 μL plasma. Pairs of masked quality control samples from a large pool were interspersed among study samples to assess intraset variability. The plasma samples from three cases were not successfully analyzed.
Laboratory personnel were blinded to the case status of samples in all studies.
Because the periods of blood sample collection, the types of samples analyzed, and the laboratory methods used to analyze those samples differed across the cohorts, exposure-disease associations were examined separately by cohort. Statistical analyses were done for all PCB congeners with concentrations above the instrumental LOD in at least half the subjects. Concentrations of total congeners were calculated by summing the concentrations of all measured congeners (including those below the LOD). Results of grouped analyses based on moles were similar to those based on weight of the congeners, so results based on weight are presented for consistency with the literature on this topic. All PCB measures were corrected for total lipids ( 16), although alternatively adjusting for total lipids in the statistical models produced similar results. Analyses of individual PCBs, total PCBs, and a priori groupings of congeners based on potential immunotoxicity and dioxin-like activity, enzyme induction, and degree of chlorination ( 6, 7, 19) were conducted. Because concentrations of the groupings were highly correlated with the concentrations of one or a few constituent congeners and exposure-disease associations were similar, results of the grouped analyses are not presented.
Our criteria for selecting individual congeners for particular focus were that the congener concentration be above the instrumental LOD in at least 50% of subjects and be ascertained with relatively low measurement error across the three studies. Only congeners 118, 138, and 153 met these criteria, as shown below, and are presented in this article. Results for the other congeners are presented in the Supplementary Tables.
Concentrations were above the instrumental LOD in at least half of the subjects for all of the 36 congeners examined in the Janus sera, 7 of the 28 congeners in the CLUE I sera (28, 74, 105, 118, 138, 153, and 180), and 18 of the 21 congeners in the NHS plasma (28, 66, 74, 82, 99, 101, 105, 118, 138, 141, 153, 156, 170, 180, 183, 187, 199, and 203; Supplementary Tables S1–S3).
Blood concentrations and measurement accuracy varied appreciably among individual congeners within each study but showed the same general pattern (i.e., ranking of individual congeners) across the studies (Supplementary Tables S4–S6). Intraset and interset coefficients of variation (CV) among quality control samples for the different congeners tended to be inversely related to the measured concentration of each congener. For congeners 118, 138, and 153, the median concentrations were relatively high and the CVs were relatively low compared with other congeners across studies (range of intraset CVs, 3.5–20.2; range of interset CVs, 6.6–13.0). Congener 180 had a relatively high median concentration and low CVs in the Janus and NHS samples but showed the opposite pattern in the CLUE I samples. Congener concentrations for 118, 138, and 153 were highly correlated, with most Spearman correlation coefficients ≥0.7 (Supplementary Tables S7–S9). Correlation coefficients between these congeners and 180 were more variable but tended to be somewhat lower.
For the Janus and CLUE I analyses, conditional logistic regression was used to assess risk [odds ratio (OR) and 95% confidence interval (95% CI)] by fourths of congener concentration, with quartiles based on congener-specific concentrations among controls. All risk estimates in the Janus cohort were adjusted for body mass index (BMI; thirds) and smoking status (current, former, and never). Those in the CLUE I cohort were adjusted for years of education (<12 and ≥12 years) and current smoking status (yes and no). For the NHS, unconditional logistic regression, controlling for BMI (thirds), smoking (current, former, and never), and age as a continuous variable, was used to assess risk by thirds of congener concentration, based on congener-specific concentrations among controls. Covariates and matching variables were chosen based on availability of data, potential or observed relationships with the exposures and disease being examined, and consistency with previously published studies in this subject area. Additional analyses were carried out including p,p′-DDE in the model. Tests for trend were assessed using log-transformed values as continuous measures. All P values were two sided. Statistical analyses were done with SAS version 9. The study protocol was approved by the Institutional Review Boards of participating institutions.
The three study populations were diverse, differing in geographic region, age at enrollment, sex distribution, education, and duration of follow-up ( Table 1 ). The Janus and CLUE I participants were enrolled during the 1970s, whereas the NHS members were enrolled in 1989 to 1990. The median years to diagnosis among cases were 16.6 in the Janus cohort, 12.6 in CLUE I, and 1.04 in the NHS.
There was evidence of increased NHL risk associated with congeners 118, 138, and 153 ( Table 2 ). Exposure-response relations were most apparent among the CLUE I subjects, especially for congeners 138 and 153. Risk estimates associated with p,p′-DDE were lower and showed no consistent exposure-response pattern. Adjustment for p,p′-DDE had only a minimal effect on the PCB risk estimates (Supplementary Table S10).
When analyses were stratified by median time from blood draw to disease diagnosis in the Janus and CLUE I studies (16 and 12 years, respectively), exposure-response relations were observed primarily in the earlier follow-up period for both studies ( Table 3 ; Supplementary Tables S12 and S13), with weaker, nonsignificant associations in the later period ( Table 4 ; Supplementary Tables S12 and S13). Exposure-response relations for individual congeners tended to be clearest in the Janus cohort. Risk estimates for the Janus cohort generally were smaller and 95% CIs were narrower than the corresponding estimates for the CLUE I cohort, likely due in part to the small number of subjects in the lowest exposure categories for CLUE I. ORs for concentrations above the highest quartile were 5.3 to 13.0 for 118, 2.5 to 7.8 for 138, and 2.7 to 3.6 for 153. Significant exposure-response relations were similarly observed in the NHS, which had only 5 years of follow-up. Separate analyses excluding cases diagnosed within 2, 3, 4, or 5 years of providing a blood sample and their matched controls did not substantively affect results in either Janus or CLUE I. Similarly, defining the early follow-up period in the Janus cohort by the same 12-year cutpoint used for the CLUE I cohort did not substantively change the pattern of risks for the Janus cohort. Results of analyses stratified by age at blood draw did not show a clear age effect; the sample sizes precluded analyzing disease risk stratified by both time from blood draw to cancer diagnosis and age at blood draw. Several other congeners, including 180, had exposure-response trends in one or two of the studies (Supplementary Tables S12–S14). Of note, congeners 157, 189, 206, and 207, which were measured relatively precisely in the Janus cohort and were at least moderately correlated with congeners 118, 138, and 153, showed no apparent exposure-response relations (Supplementary Table S12).
Risk estimates associated with p,p′-DDE seemed to be stronger in the earlier period but did not show clear exposure-response trends ( Tables 3 and 4). p,p′-DDE modestly confounded the relations between PCBs and NHL risk in the earlier, but not the later, follow-up period (Supplementary Table S11). The extent of attenuation was greater for p,p′-DDE than for the individual PCB congeners when these factors were adjusted for each other, and trends in the congener risk estimates remained significant, whereas those for p,p′-DDE did not (Supplementary Table S11). Associations between PCBs and NHL risk were not confounded by the other pesticides measured in either CLUE I ( 15) or Janus (data not shown). In models that included two PCB congeners simultaneously, risk estimates associated with a given congener generally decreased, but ORs remained elevated, and in Janus and CLUE I, exposure-response trends remained significant, for congeners 118, 138, and 153 (data not shown). In analyses of Janus subjects restricted to persons with normal BMI (18≤BMI<25) at enrollment, exposure-response trends were seen for congeners 118, 138, and 153, although risk estimates were imprecise (data not shown).
We found evidence of associations, including exposure-response trends, between risk of NHL and blood concentrations of PCB congeners 118, 138, and 153 across these three prospective cohorts. We also found limited evidence of exposure-response trends associated with several other congeners. Our finding of an increased risk of NHL associated with blood concentration of PCBs is consistent with two previously reported case-control studies of incident, untreated NHL patients ( 9, 10).
The stronger and more consistent associations that we observed for congeners 118, 138, and 153 compared with other congeners may be due to the higher blood concentrations of these congeners and the consequently greater precision with which they were measured. In addition, most of the well-measured congeners were moderately to highly correlated. However, several of these congeners did not show a relationship with NHL risk, suggesting some degree of congener specificity for the associations.
Congener 180 was related to significantly increased risk in the Janus cohort and to nonsignificantly increased risk in the other cohorts. A recent population-based case-control study in the United States observed positive trends between risk of NHL and levels of this congener in both blood plasma and carpet dust from study participants ( 10, 20).
The main p,p′-DDT metabolite, p,p′-DDE, showed a slightly increased risk of NHL across the three cohorts, which was stronger in the earlier follow-up period. However, there were no apparent exposure-response trends in most analyses, and the p,p′-DDE effect was attenuated more by adjustment for PCBs than vice versa. Although several epidemiologic studies have found modest increases in NHL risk related to measured or self-reported p,p′-DDT/p,p′-DDE exposure ( 20– 24), this effect has tended to decrease substantially in those studies in which adjustment was made for other chemical exposures ( 21, 25). In addition, two case-control studies ( 9, 10) that measured p,p′-DDE in either blood or fat tissue found no association with NHL. The evidence is mixed, but suggestive, in relation to other pesticides ( 26). However, no association was observed between serum levels of organochlorine pesticides and NHL risk among the CLUE I participants included in the present study ( 15) or among subjects in a recent population-based case-control study ( 10). Interestingly, two case-control studies of NHL among farmers ( 27, 28) observed significantly increased risks associated with certain pesticides or pesticide groups, including organochlorines, only among cases with a t(14;18) chromosomal translocation. We lacked tumor tissue with which to examine this issue in the present study.
Histopathologic classification systems for NHL evolved during the follow-up periods of this study. Pathology reports were re-reviewed for all identified NHL cases in each cohort and pathology slides were re-reviewed for most cases in the CLUE I cohort, with those cases judged not to be NHL excluded from this study; however, pathology slides necessary to standardize histopathologic classifications over time were unavailable for the large majority of cases in this study. Future studies with the necessary materials could contribute to our understanding of the relationship between PCBs and NHL risk by investigating this association by disease subtype.
It is possible that other factors confounded our analyses. For example, there may be other unmeasured, fat-soluble, bioaccumulating compounds that have similar exposure patterns as PCBs and that may lead to lymphoma. However, a recent population-based case-control study by De Roos et al. ( 10) that measured levels of PCBs and other organochlorines, including dioxins, furans, coplanar PCBs, and pesticides, observed that the elevated risk of NHL associated with PCBs was not confounded by these other organochlorines. There may also be lifestyle factors, such as diet, BMI, or reproductive history, that can affect the body burden of PCBs ( 29) and that a few studies have associated with risk of NHL ( 30– 34). In fact, reported links between blood levels of PCBs and risk of type 2 diabetes and between overweight/obesity and risk of NHL raise the possibility that blood levels of PCBs may be a surrogate measure of dietary intake of animal fat, a primary source of PCBs in the general population, which can increase weight and consequently contribute to increased risk of diabetes and NHL. However, this is a complicated causation pattern and demographic/lifestyle factors, including BMI, did not appreciably confound the associations we observed between PCB concentrations and NHL risk in the cohorts for which we had the necessary data. In addition, we observed exposure-disease associations even in analyses restricted to persons with normal BMI at enrollment. Finally, the magnitude of risks reported for these factors in other studies of NHL is too low to have substantially confounded our risk estimates. De Roos et al. ( 10) also found little evidence of confounding of the PCB-NHL relationship by BMI or other demographic characteristics.
Several retrospective cohort mortality studies of electric utility or capacitor-manufacturing workers and a study of cancer incidence and mortality among Swedish fishermen consuming PCB-contaminated fish found only weak and inconsistent evidence of an association between PCB exposure and lymphohematopoietic malignancies ( 35– 45); however, none of these studies examined risk in relation to measured biological levels, which are likely to provide a more accurate measure of cumulative exposure to these long-lived bioaccumulating compounds. Moreover, all but two of the studies of capacitor-manufacturing workers examined cancer mortality rather than incidence; there were few expected cases of NHL in all but two of these cohorts; and the relatively young age at which vital status was ascertained in some of the studies prevented assessment of risk of cancers with long latency periods. Nonetheless, substudies showed that some of the workers had serum PCB concentrations in the 1970s that were substantially higher than those observed in the three cohorts described here ( 37, 38, 44). The discrepancy between the health effects observed among workers with occupational exposure to industrial-grade PCB mixtures (i.e., Aroclor) and the association between environmental exposure to PCBs as measured in blood and risk of NHL in the general population needs further exploration before conclusions can be made about the carcinogenicity of PCBs.
It is noteworthy that exposure-response trends in this study were strongest in the period closest to blood draw. The best studied risk factor for NHL, immunosuppressive therapy after organ transplantation, shows a median time from transplantation to NHL diagnosis of 1 to 5 years ( 46). In a prospective study of AIDS-related lymphoma, the median time from HIV infection to NHL diagnosis among NHL cases diagnosed before and after introduction of highly active antiretroviral therapy was 6 to 8 years ( 47). However, the lack of data on temporal patterns of exposure among subjects in our study limits interpretation of the temporal patterns of association.
Many of the congeners associated with increased risk in the present analyses are proposed to be immunotoxic ( 7). PCBs have been shown to alter immune function ( 48). Moreover, limited evidence suggests that infection by the Epstein-Barr virus (EBV) may potentiate the effects of PCB exposure ( 8, 9), supporting the theory that any role of PCBs in the etiology of lymphoma may be mediated through immunotoxic mechanisms ( 49). The high correlations between immunotoxic and other congeners in our study precluded our investigating the specificity of this relationship in more detail.
This analysis had several strengths. We were able to compare results from three prospective cohorts. Disease status and treatment, which can affect blood concentrations of PCBs ( 50), could not bias the results because blood samples were collected from subjects before any cancer diagnosis, although we cannot rule out some influence of early disease on PCB levels in the NHS given the short median time to diagnosis in that study. It is important to note, however, that the results from these prospective studies do not fundamentally differ from those from earlier case-control studies. The serum concentrations of total PCBs among controls in the present studies were similar to those observed during the corresponding time periods among U.S. populations that had no occupational or unusual environmental PCB exposures ( 51). The measured concentrations from the 1970s are likely to represent peak or near-peak levels for these populations because PCBs were banned around this time. In fact, environmental and general population levels of PCBs in industrialized countries have declined in recent decades, although there is evidence that this decline has leveled off. 12 In addition, the analytic methods used to measure PCB concentrations were not affected by many potentially interfering compounds ( 17, 52), and all values were corrected for blood lipids.
In conclusion, blood concentrations of PCB congeners with the greatest relative concentrations and measurement precision were related to an increased risk of NHL in all three cohorts. These findings need to be replicated and potential confounding by factors such as correlated exposures to other unmeasured organochlorines or other bioaccumulating compounds needs to be further evaluated. We are aware of several ongoing studies examining the relationship between PCBs and NHL risk that should shed further light on these findings. Our findings, together with results from other studies, suggest that the association between certain PCB congeners and NHL, whether causal or not, could provide an important clue to help explain the increase in incidence of this tumor over the last half of the 20th century.
Grant support: Intramural Research Program of the NIH National Cancer Institute's Division of Cancer Epidemiology and Genetics; National Cancer Institute grants CA98122 and CA/ES62984; Department of Health and Human Services grants CA60754 and ES03819; and Research Career Award HL21670 (G.W. Comstock).
The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.
We thank Randi Elin Gislefoss (Norwegian Cancer Registry, Oslo, Norway) for help with selection of study participants, data retrieval, and processing of serum samples; Steinar Thoresen (Norwegian Cancer Registry) for systematic review of pathology reports among the Janus participants; Svein Erling Tysvær (Norwegian Cancer Registry), Steve Palladino, and Chris Shaffer (Information Management Services, Inc., Silver Spring, MD) for assistance in data processing and preparation; Hilde Langseth (Norwegian Cancer Registry) for assistance in interpreting certain cancer registry data; and Joanne Colt (National Cancer Institute, Bethesda, MD) for advice on an earlier version of the manuscript.
Note: Supplementary data for this article are available at Cancer Research Online (http://cancerres.aacrjournals.org/).
↵12 WHO. Health risks of persistent organic pollutants from long-range transboundary air pollution. WHO Regional Office for Europe (Copenhagen), 2003. Available from: http://www.euro.who.int/Document/e78963.pdf
- Received October 20, 2006.
- Revision received February 23, 2007.
- Accepted April 5, 2007.
- ©2007 American Association for Cancer Research.