Polychlorinated biphenyls (PCB) have been suspected as possible contributors to increasing non-Hodgkin's lymphoma incidence during the latter half of the 20th century based on their toxicologic properties and provocative epidemiologic reports. We investigated PCBs and other organochlorines and risk of non-Hodgkin's lymphoma in a population-based case-control study in the United States. Congeners of PCBs (including coplanar congeners), dioxins, furans and pesticides or pesticide metabolites were measured in plasma of 100 untreated cases and 100 control subjects. We used a multiple imputation procedure to fill in missing values of levels determined to be below the detection limits. Risks of non-Hodgkin's lymphoma associated with each analyte were estimated using conditional logistic regression for the continuous measure, exposure quartiles, trend across quartile categories, and exposures above the 95th percentile. Certain PCB congeners were associated with increased risk of non-Hodgkin's lymphoma, including coplanar PCBs 156, 180, and 194, with odds ratios for the highest versus lowest quartile ranging from 2.7 to 3.5, and significant trends. Each of the furan congeners was associated with risk of non-Hodgkin's lymphoma, as were total furans, with 3.5-fold increased risk for the highest versus lowest quartile and a significant trend across quartiles (P = 0.006). The toxic equivalency quotient (TEQ), a summed metric that weights congeners by their dioxin-like potency, was associated with non-Hodgkin's lymphoma, with 35% increased risk per 10 TEQ pg/g lipid (95% confidence interval, 1.02-1.79). Our results add to existing literature, which suggests that exposure to organochlorines contributes to non-Hodgkin's lymphoma risk; these risks were most apparent for certain PCBs and furans.
- non-Hodgkin's lymphoma
- polychlorinated biphenyl
Several studies have suggested a role of the polychlorinated biphenyls (PCB) in development of non-Hodgkin's lymphoma ( 1– 3), a provocative finding in light of rising incidence of non-Hodgkin's lymphoma during the latter half of the 20th century ( 4, 5). During this period, use of and environmental contamination by PCBs increased before cessation of production in the United States in 1977 ( 6). PCBs are not thought to be genotoxic but may promote lymphomagenesis through alteration of gene transcription and immunosuppression ( 7– 9). To identify potential environmental causes of non-Hodgkin's lymphoma, we conducted a case-control study in four geographic areas of the United States covered by Surveillance, Epidemiology, and End Results (SEER) registries, collecting interview and questionnaire data on exposures as well as environmental samples and biospecimens.
We were interested in confirming previous reports of associations with PCBs. We were concerned that previous associations may have been confounded by other organochlorine chemicals, including dioxins, furans, coplanar PCBs, and pesticides. Although the sources of production and use of different organochlorine classes vary considerably, the lipid-soluble nature of these compounds results in similar bioaccumulation patterns in the food chain. Of these other classes of organochlorines, there was the most evidence of an association between dioxins and non-Hodgkin's lymphoma. The most potent dioxin congener, 2,3,7,8-tetrachlorodibenzo-p-dioxin (TCDD), has been declared a human carcinogen by the IARC for all cancer sites based on sufficient evidence in animals, a convincing mechanistic model, and limited evidence in humans ( 10). For non-Hodgkin's lymphoma specifically, there is compelling evidence of increased risk resulting from exceptionally high exposures in occupational settings ( 11– 13) and following an accidental industrial release in Seveso, Italy ( 14); there have been few studies of the effects of dioxins within the general population. Furans and coplanar PCBs exhibit similar toxicologic properties as dioxin through their potential to bind to the arylhydrocarbon receptor, whereas other (noncoplanar or standard) PCBs elicit biological responses that are primarily mediated through other pathways ( 15).
The prior evidence of PCB and dioxins as carcinogens from animal studies and suggestive epidemiologic data make the study of organochlorines as a potential cause of non-Hodgkin's lymphoma timely. In our case-control study of non-Hodgkin's lymphoma ( 16), we collected adequate volume of blood to measure a wide spectrum of organochlorines, including noncoplanar PCBs, dioxins, furans, coplanar PCBs, pesticides, and pesticide metabolites. Simultaneous measurement of several classes of organochlorine compounds makes possible analyses accounting for correlation among organochlorine classes and analyses in which organochlorines are grouped by structure or toxicity.
Materials and Methods
Study population. The study population included a subset of participants from a case-control study of non-Hodgkin's lymphoma conducted by the National Cancer Institute (NCI; “the parent study”). The parent study has been described previously ( 16– 18). Between July 1998 and June 2000, subjects were enrolled from four U.S. SEER registry areas: the state of Iowa, Los Angeles County, and the metropolitan areas of Detroit and Seattle. Cases included 1,321 patients with newly diagnosed non-Hodgkin's lymphoma ages 20 to 74 years without evidence of HIV infection. Data on non-Hodgkin's lymphoma subtypes were obtained from each local SEER registry and were based on abstracted reports of the diagnosing pathologist. Histologic diagnoses were coded using the International Classification of Diseases for Oncology, Second Edition ( 19) and subsequently translated to categories of the WHO classification ( 20). Population controls (n = 1,057) were identified by random digit dialing (ages <65 years) and from Medicare eligibility files (ages ≥65 years) and were frequency matched to cases by age, sex, and race. Overall response percentages were 59% and 44%, for cases and controls, respectively. Among eligible subjects we attempted to contact, 76% of cases and 52% of controls participated in the study.
Written informed consent was obtained from each participant before interview. A computer-assisted personal interview was administered that contained questions about demographic characteristics, hair coloring, occupational history, and pesticide use history. Some environmental samples were collected from the subject's residence, including carpet dust and tap water. All study subjects were asked to provide a blood or buccal cell sample. Blood samples were collected from 62% of cases and 66% of controls.
The study of plasma organochlorines and non-Hodgkin's lymphoma included untreated cases only to avoid misclassification of prediagnostic exposure resulting from possible changes in organochlorine levels during chemotherapy ( 21). Because a greater sample volume is critical in plasma analysis for dioxins, only subjects with at least 7.1 mL banked plasma were included [the total banked plasma for each subject included a combination of vials containing heparin plasma and acid, citrate, dextrose (ACD) plasma]. Of the untreated cases with adequate plasma volume, 100 were randomly selected for the study. Controls were matched 1:1 to cases by birth date (±4 years), date of blood draw (±1 year and 3 months), sex, and study site. Each control was selected from a list of potential matches for each case, starting with the closest match according to birth date first and date of blood draw second. If a potential control had already been assigned to another case, then the next closest match was chosen.
Laboratory methods. Organochlorines in plasma were measured by the Dioxin and Persistent Organic Pollutants Laboratory of the Centers for Disease Control and Prevention (CDC) in Atlanta, GA. An aliquot of 7.1 mL plasma was prepared for each of the 100 cases and 100 controls. These samples were organized in pairs, with each case next to its matched control in the same batch. Quality-control (QC) samples were prepared using stored heparin and ACD plasma from participants of two unrelated NCI studies, which handled and processed the samples in a similar fashion and during an overlapping period as the current study. The 40 QC samples representing duplicate samples from 20 pools (20% of the total case group size) were included as pairs within two of every three laboratory batches. QC of laboratory measures was assessed using the average intrabatch coefficient of variation (CV) and intraclass correlation coefficient (ICC) among duplicates for each analyte.
Measurements were conducted for 36 noncoplanar PCB congeners (18, 28, 44, 49, 52, 66, 74, 87, 99, 101, 105, 110, 118, 128, 146, 149, 151, 153, 156, 157, 167, 170, 172, 177, 178, 180, 183, 187, 189, 194, 195, 201, 206, 209, 138-158, and 196-203), 7 dioxins (1234678D, 123478D, 123678D, 123789D, 12378D, 2378D, and OCDD), 10 furans (1234678F, 1234789F, 123478F, 123678F, 123789F, 12378F, 234678F, 23478F, 2378F, and OCDF), 4 coplanar PCBs (77, 81, 126, and 169), and 13 organochlorine pesticides or pesticide metabolites [p,p′-DDE, o,p′-DDT, p,p′-DDT, aldrin, β-hexachlorocyclohexane (β-HCCH), γ-hexachlorocyclohexane (γ-HCCH), dieldrin, endrin, hexachlorobenzene (HCB), heptachlor epoxide, mirex, oxychlordane, and transnonachlor (tNONA)]. These substances were measured in plasma by high-resolution gas chromatography/isotope-dilution high-resolution mass spectrometry ( 22). Each batched analytic run was conducted blind to case-control status and consisted of eight unknown plasma samples, a method blank, and two internal laboratory QC samples. Subquantities of the entire sample volume were dedicated to measurement of noncoplanar PCBs and pesticides (average, 1.0 g; range, 0.7-1.2 g), dioxins, furans, and coplanar PCBs (average, 5.8 g; range, 5.0-6.2 g) and plasma lipids (∼0.1 g). The analytic results were reported on both whole-weight basis and lipid-corrected basis. Detection limits (DL), the minimum level that could be detected by the gas chromatography/mass spectrometry technique, were reported for each sample on a whole-weight basis and lipid-adjusted basis, corrected for sample weight and analyte recovery.
Exposure coding and statistical analyses. All statistical analyses were conducted using SAS version 9.1 (SAS Institute, Cary, NC). We excluded measurements when the sample contained compounds that coeluted with the target analyte and the result was nonreportable due to “interferences” (see Table 1 ) and considered laboratory measurements “missing” when concentrations were below DLs. Because replacement of a measurement below its DL with a single value, such as 0, DL/2, or DL/√2, distorts exposure assessment, we accounted for measurements below DLs using an unbiased multiple imputation approach ( 23). The approach was based on our preliminary analyses indicating that lipid-adjusted measured concentrations were consistent with log-normal distributions. Using all data from a target analyte, we used maximum likelihood methods to estimate parameters for the log-normal distribution adjusting for age (in years), sex, study site, and year of blood draw. For each measurement below its DL, we randomly sampled a value from the appropriate log-normal distribution as the imputed value below the DL. This imputation procedure is related to the “fill-in” approach described by Helsel ( 24) and applied by Moschandreas et al. ( 25). Thus, analyte means, percentiles, and categorizations are unbiased under the log-normal assumption. Because fill-in data are artificial, we appropriately accounted for the random variation by conducting multiple imputation creating five replicate data sets as discussed in detail in Lubin et al. ( 23) and Colt et al. ( 26). In the current study, imputation parameters were based on the congener distributions among control subjects only to reflect characteristics of the general population ( 27).
We included in our main analysis all analytes that had been measured in at least 30% of samples (≤70% nondetects). A simulation study by Lubin et al. indicated that for 100 cases and 100 controls the approach is unbiased, with nominal α levels for hypothesis testing with missing data in ≤50% measurements and with testing only slightly anticonservative (i.e., confidence intervals too narrow) with 70% missing data ( 23).
In addition to continuous values for each analyte, we constructed several additional exposure metrics. Four categories of exposure were coded based on quartiles of the analyte distribution among controls in the five combined data sets. Extreme exposures were characterized for each analyte as levels above the 95th percentile of the control distribution. Summary exposure variables were constructed for groups based on chemical structure as total summed molar concentrations (mol/g lipid) of dioxin, furan, and noncoplanar PCB congeners. The groups only included congeners that were included in our main analyses (those with ≥30% measured values). Sums of PCB congeners were grouped by the degree of chlorination for lower (di-, tri-, and tetra-biphenyls), moderate (penta-, hexa-, and hepta-biphenyls), and highly (octa-, nona-, and deca-biphenyls) substituted PCBs ( 28). The toxic equivalency quotient (TEQ), a toxicity-weighted summary metric indicating biological activity through the arylhydrocarbon receptor pathway, was calculated as the sum of the lipid-adjusted levels of dioxin, furan, coplanar PCB, and mono-ortho-chlorinated PCB congeners, each weighted by its congener-specific toxic equivalency factor (TEF; congener potency relative to TCDD; refs. 10, 29). We included in our TEQ calculation 11 of the 29 congeners for which WHO TEFs have been assigned ( 29), representing those with at least 30% measured values in our data. In addition to an overall TEQ, we calculated TEQ concentrations specific to dioxins, furans, and PCBs. The TEQ is heavily weighted by dioxins (e.g., TCDD TEF = 1.0 and 12378D TEF = 1.0), moderately weighted by furans (e.g., 23478F TEF = 0.5 and 123678F TEF = 0.1) and coplanar PCBs (e.g., PCB 126 TEF = 0.1), and weakly weighted by mono-ortho-substituted PCBs (e.g., PCB 118 TEF = 0.0001 and PCB 156 TEF = 0.0005). Eighty percent to 90% of the total TEQ can be estimated from the heavily weighted congeners, TCDD and 12378D, in combination with five of the congeners included in our analysis ( 30). To check the sensitivity of our TEQ risk estimates to misclassification from noninclusion of the heavily weighted congeners (TCDD and 12378D), we conducted additional TEQ analyses when adding those two congeners to the metric despite high proportions of imputed values.
Odds ratio (OR) and 95% CI for the risk of non-Hodgkin's lymphoma associated with each exposure metric were estimated using conditional logistic regression for each of five imputed data sets, and results were combined as per Rubin ( 31) using the Proc MIANALYZE in SAS. This approach adjusts the variance for the imputation, resulting in wider CIs than would be obtained from a single imputation approach. Relative risk associated with each congener was estimated for a 10-unit increase in the continuous analyte level based on a log-linear model for the OR. Risks associated with quartiles and with levels above the 95th percentile were each estimated in separate models using values below the lowest quartile as the reference. Tests for trend across quartiles were conducted by creating a continuous variable with assigned values equal to the median level among controls within each category; the P for the trend test was based on the logistic model coefficient for this continuous variable divided by its SE.
The conditional logistic modeling procedure provided adjustment for the matching variables age, sex, study site, and date of blood draw. We considered other potential confounders of plasma organochlorine-non-Hodgkin's lymphoma associations, including education (indicator variables: less than high school graduate as reference, college, graduate school), race (indicator variables: White as reference, Black, unknown), usual body mass index [BMI; kg/m2; normal as reference (20-25), underweight (<20), overweight (25-30), obese (>30), unknown], and family history of non-Hodgkin's lymphoma (no as reference, yes, unknown), but no confounding was apparent. An extensive amount of missing data on smoking (61% of the 200 subjects) precluded adjustment for this factor. We also examined the relation between analyte levels and employment with potential exposure to one or more classes of organochlorines, including farming, gardening/groundskeeping, electrical repair/fabrication, and military occupations or industries. For each analyte that was associated with non-Hodgkin's lymphoma risk (considered as analytes with elevated ORs of ≥1.5 for a 10-unit increase in the continuous measure or ≥2.0 for the highest versus the lowest exposure category), we explored potential confounding by other organochlorine classes (total PCBs, total dioxins, and total furans); for example, we explored confounding of the association between PCB 180 and risk of non-Hodgkin's lymphoma by adjustment for total dioxins and total furans.
Separate analyses were conducted for men and women to evaluate differential effects of organochlorines by sex. We also explored whether observed effects differed by histology [i.e., diffuse (n = 14) versus follicular (n = 25) non-Hodgkin's lymphoma]. Models examining the effect of organochlorines on certain non-Hodgkin's lymphoma histologic subtypes included all controls to provide a consistent comparison group using unconditional logistic regression with adjustment for the matching factors.
In addition to analyses using imputed values, we conducted analyses using only measured values. Measured values above the DLs were categorized according to control quartiles, and risk estimates were generated using conditional logistic regression, with the lowest quartile as the reference. We compared results of our two modeling strategies for assurance that where elevated risk estimates and trends were observed in analyses of imputed data the general patterns were also observed in analyses of measured values only.
Of 70 analytes measured, 28 had adequate proportions of detects (≥30%) to include in all statistical analyses ( Table 1). The average intrabatch reliability among duplicate samples was good, with CVs ranging from 0.8% to 14.5%. ICCs representing agreement between organochlorine measurements among duplicate samples were high for pesticides [ICCs = 0.92-0.99, except mirex (ICC = 0.63) for which only three duplicate pairs had measured values] and PCBs (ICCs = 0.81-0.96) but were generally lower for dioxins and furans (ICCs = 0.49-0.99), likely reflecting higher assay sample volume requirements for accurate measurement of these congeners.
Characteristics of the study population are shown in Table 2 for the study of plasma organochlorines and for the parent non-Hodgkin's lymphoma study. Organochlorine study cases who were untreated differed from the parent non-Hodgkin's lymphoma study cases in that they were less likely to be of diffuse histology. They were also of older ages and were more likely to be from the Iowa study site. By the matched study design, organochlorine study cases were similar to their controls with regard to age, sex, study site, and date of blood draw. Organochlorine study cases were more likely than controls to have higher education (≥16 years), to be White, to be underweight (BMI <20 kg/m2), and to have a family history of non-Hodgkin's lymphoma, although not significantly so. Organochlorine study cases had significantly lower measured plasma total lipids than controls (625.7 versus 683.0 mg/dL; data not shown in Table 2). Although smoking data were available for only a subgroup of the study population, there was no apparent difference in smoking between organochlorine study cases and controls.
Several organochlorine analytes were, on average, higher among those who had ever been employed in farming occupations or industries (n = 23), significantly so for each of the dioxins (124678D, 1235678D, and OCDD) and furans (1234678F, 123478F, 123678F, and 23478F), and the pesticide metabolite oxychlordane; these differences were stronger among cases than controls. In addition, farming employment was slightly more frequent among cases than controls (14% versus 9%; P = 0.27). Plasma organochlorine levels did not differ by the other employment groups examined. Neither farming nor the other employment groups confounded associations between plasma organochlorines and non-Hodgkin's lymphoma.
Several noncoplanar PCB congeners showed modest associations with non-Hodgkin's lymphoma risk ( Table 3 ), including PCBs 156, 180, and 194, with ORs for the highest versus lowest quartile ranging from 2.7 to 3.5, and significant trends across categories (Ps < 0.05). These estimates were unconfounded by other organochlorines or demographic factors; for example, the 3.5-fold increased risk associated with the highest quartile of PCB 180 persisted after adjustment for BMI (OR, 4.2; 95% CI, 1.4-12.9), total dioxins (OR, 4.2; 95% CI, 1.5-11.6), or total furans (OR, 2.9; 95% CI, 1.0-8.0). Analyses of levels above the 95th percentile showed slightly stronger associations than levels in the highest quartile for PCB 170 (OR, 4.1; 95% CI, 0.9-17.7), 180 (OR, 4.5; 95% CI, 1.1-19.0), and 194 (OR, 3.8; 95% CI, 0.9-16.7) but not 156 (OR, 1.4; 95% CI, 0.3-7.5). Results of continuous analyses generally reflected the trends observed across exposure categories, with the exception of the continuous PCB 180 measure, for which a small, borderline significant 8% increase was associated with each 10 ng/g lipid increase. Qualitatively similar results for all congeners were obtained from models including only measured (not imputed) values.
Results shown in Table 4 suggest little evidence of an association between dioxins and the risk of non-Hodgkin's lymphoma, but several furan congeners were associated. TCDD was detected in only seven samples (two cases and five controls), hampering risk inference. ORs for dioxin congeners (those in Table 4) were somewhat elevated in analyses of levels above the 95th percentile (data not shown ORs ranged from 1.3 to 3.8) but were not statistically significant. Each of the furan congeners we examined was associated with increased non-Hodgkin's lymphoma risk, with ORs ranging from 1.8 to 2.6 per 10 pg/g lipid increase. Estimates for the highest quartiles were also elevated, ranging from 1.9 to 4.2, although category-specific ORs were not always monotone and some were not statistically significant. Of the furans, 1234678F was most strongly associated with non-Hodgkin's lymphoma, with >10-fold increased risk observed for exposure above the 95th percentile (OR, 10.5; 95% CI, 1.6-68.6). A monotonic increase across exposure categories was more apparent for 123478F in models with exposure categories based on measured values only than in models of imputed data (with lowest quartile as reference: ORsecond quartile, 2.0; 95% CI, 0.5-8.5; ORthird quartile, 3.9; 95% CI, 0.8-18.3; ORhighest quartile, 5.7; 95% CI, 1.2-26.1). On adjustment for total dioxins or PCBs, associations remained elevated for all of the furans in continuous and categorical analyses (data not shown). The coplanar PCB 169 was associated with 50% increased risk of non-Hodgkin's lymphoma per 10 pg/g lipid increase, and elevated but imprecise increased risk was observed for levels above the 95th percentile (OR, 24.4; 95% CI, 2.6-226); these associations were unconfounded by other classes of organochlorines.
None of the organochlorine pesticide or pesticide metabolites was associated with the risk of non-Hodgkin's lymphoma in analyses of quartiles ( Table 5 ). In analyses of extreme levels of exposure, we found elevated ORs for p,p′-DDT (OR, 3.3; 95% CI, 0.7-15.9), oxychlordane (OR, 2.9; 95% CI, 0.6-13.2), tNONA (OR, 2.0; 95% CI, 0.5-7.5), and β-HCCH (OR, 2.0; 95% CI, 0.5-7.6); indeed, the continuous measures of these four analytes indicated slight increases in risk (8-17%) per 10 ng/g lipid increases in organochlorine level.
There was some indication that the total summed noncoplanar PCBs was associated with increased risk of non-Hodgkin's lymphoma ( Table 6 ) and that PCB congeners with higher degrees of chlorination were more strongly associated than lower- or moderate-chlorinated congeners. Total PCBs were associated with a nonsignificant 85% increased non-Hodgkin's lymphoma risk among the highest quartile, and similar risk was observed for levels above the 95th percentile (OR, 1.9). Highly chlorinated PCBs showed a 2.7-fold increased risk of non-Hodgkin's lymphoma for the highest versus lowest quartile and 3.8-fold risk above the upper 95th percentile (95% CI, 0.9-16.7); however, this group contained data on only one congener, PCB 194, because other nona- and deca-chlorinated biphenyls were not detected in our samples. Total summed dioxins were not associated with the risk of non-Hodgkin's lymphoma, in contrast to total furans, which were associated with a 3.5-fold increased risk of non-Hodgkin's lymphoma for the highest quartile, a significant trend across quartiles, and >7-fold increased risk above the 95th percentile (OR, 7.4; 95% CI, 1.7-32.9). The TEQ was associated with non-Hodgkin's lymphoma ( Table 7 ), with 35% increased risk per 10 pg/g lipid (95% CI, 1.0-1.8) and a nonsignificant 2-fold increased risk for the highest compared with the lowest TEQ quartile (OR, 2.3; 95% CI, 0.8-6.2; P for trend = 0.06). There were suggestions of increasing risk by TEQ concentrations for dioxins, furans, and PCBs separately, with the strongest associations observed for furans. The TEQ results were unchanged or only slightly attenuated by inclusion in the summed metric of heavily weighted congeners that had high proportions of imputed values.
Increased non-Hodgkin's lymphoma risks associated with organochlorine exposures were more apparent among women than men (data not shown), particularly for the furan congeners 1234678F (highest versus lowest quartile: women OR, 21.3; 95% CI, 1.8-247; P for trend = 0.02; men OR, 2.0; 95% CI, 0.5-8.2; P for trend = 0.37; P for interaction with sex = 0.09) and 123478F (women OR, 7.1; 95% CI, 1.6-31.6; P for trend = 0.009; men OR, 0.8; 95% CI, 0.3-2.4; P for trend = 0.82; P for interaction with sex = 0.03). Increased non-Hodgkin's lymphoma risk associated with PCB 180 was present among both sexes, with >3-fold risk associated with the highest quartile for each group. Some associations were more prominent among the 14 diffuse non-Hodgkin's lymphoma cases than the 25 follicular cases (data not shown), and significant trends of increasing risk across exposure quartiles were observed for diffuse but not follicular non-Hodgkin's lymphoma for PCBs 180 and 187, furans 1234678F, 123478F, and 123678F, dioxin 123678D, and coplanar PCB 169.
Our results contribute to the evidence that exposure to certain PCB congeners at levels experienced by the general population may increase the risk of non-Hodgkin's lymphoma and add new data about a potential association between furans and non-Hodgkin's lymphoma. We also found a positive association of the TEQ with non-Hodgkin's lymphoma, indicating possible importance of a carcinogenic mechanism operating through congener binding with the arylhydrocarbon receptor and subsequent disruption of the transcription of numerous genes, including cytochrome P4501A1 ( 10). A strength of our study is the measurement of a wide spectrum of organochlorine chemicals to adjust for potential confounding between different classes of organochlorines. In addition, our reliable laboratory methods, as evidenced by high repeatability between blind, duplicate samples, provide assurance that the positive and null findings are not driven by differences in measurement precision among compounds. Although we conducted many comparisons, and some associations are expected due to chance, analyses were based on a priori hypotheses and our results can be interpreted in the context of prior data.
Total noncoplanar PCBs measured in serum, plasma, or adipose tissue have been associated with increased non-Hodgkin's lymphoma incidence in several studies ( 1– 3), and our data provide additional evidence that at least certain congeners, including 156, 180, and 194, are associated with the risk of non-Hodgkin's lymphoma. Furthermore, ours and two other studies ( 2, 32) suggest that the risk of non-Hodgkin's lymphoma may be particularly evident for PCB congeners with higher degrees of chlorination. Hardell et al. reported from a small, hospital-based study, conducted in Sweden, that several moderately and highly chlorinated congeners measured in adipose tissue, including PCBs 156, 180, and 194, were higher in cases than controls, whereas most lower-chlorinated PCBs did not differ ( 2). In a congener-specific analysis using data from the earlier report by Rothman et al. ( 1), Engel et al. ( 32) has recently reported risks with several PCBs, including 28, 74, 118, and 138, measured in serum among subjects from Maryland, and relative risks were most elevated for summed highly chlorinated congeners (octa-, nona-, and deca-biphenyls). The levels of specific PCB congeners in our plasma samples were generally lower than those in previous studies ( 2, 32), likely reflecting decreasing levels since bans on many of these compounds ( 33).
The current study provides needed evidence that associations between noncoplanar PCBs and non-Hodgkin's lymphoma are not likely to be confounded by other classes of organochlorines, BMI, or demographic factors. PCB congener-non-Hodgkin's lymphoma associations were not strongly confounded by other classes of organochlorines; however, moderately high correlations between PCB congeners (Pearson correlation coefficients ranged from 0.14 to 0.95) created uncertainties in precisely identifying differential risks among the specific congeners. Our observation of increased risk associated with plasma levels of PCB 180 is internally consistent, corresponding to a finding from the same parent study in which carpet dust levels of PCB 180 were associated with increased non-Hodgkin's lymphoma risk ( 26).
Dioxins have been classified as carcinogenic by IARC ( 10) although not based specifically on evidence for causation of non-Hodgkin's lymphoma in humans. Our study does not provide strong support on association between specific dioxin congeners or summed dioxins and risk of non-Hodgkin's lymphoma. We observed nonsignificantly elevated ORs at extreme levels of dioxin exposure, indicating that dioxins may pose increased non-Hodgkin's lymphoma risk for higher levels than those measured in our study population. We also observed increased but not statistically significant increases in non-Hodgkin's lymphoma risk according to the dioxin-specific TEQ, which underscores the potential importance of a toxicologic pathway involving the arylhydrocarbon receptor. A previous study of nonoccupationally exposed persons found no differences in adipose tissue levels of dioxins congeners between 33 non-Hodgkin's lymphoma cases and 29 controls ( 34). Although we aimed to study the risk associated with dioxin levels experienced by the general population, several dioxin congeners were higher among subjects who were ever employed in farming and may actually have been indicative of occupational exposures. Farmers may have been exposed to dioxins and/or furans in previous formulations of phenoxy herbicides or by releases from agricultural burning ( 30).
A limitation of our dioxin analysis was the detection rate for laboratory measurement of low-level dioxins. Our sample volume for measurement of dioxins of ∼6 mL allowed detection of dioxin congeners present in relatively high concentrations only. Missing data due to levels determined to be below the DLs for various dioxin congeners may have obscured associations. However, among subjects with detected levels of dioxins, there was little evidence of case-control differences. Our summed metrics of total dioxins and the TEQ, which included only those dioxin congeners with measured levels above the DL for at least 30% of samples, may not have adequately ranked subjects in terms of dioxin exposure that would be represented had all congeners been detected. Nevertheless, we tested the sensitivity of the TEQ results by inclusion of two heavily weighted dioxin congeners (2378D and 12378D) despite high proportions of nondetected values, and the results were only slightly attenuated for the dioxin TEQ concentration (per 10-unit increase, OR, 1.57; 95% CI, 0.99-2.50), indicating that our ranking using a limited number of analytes may have provided a fair representation of exposure for the risk analysis.
The association of non-Hodgkin's lymphoma with dioxins, PCBs, pesticides, or furans has appeared stronger in the presence of EBV markers in previous studies ( 1, 34, 35). It seems unlikely that we missed associations due to our inability to stratify on this factor, because the previous studies showed associations even without stratification by EBV markers. In our study, one of the dioxin congeners (123678D) was more strongly associated (albeit nonsignificantly) with diffuse than follicular non-Hodgkin's lymphoma, and associations limited to high-grade lymphomas may not have been apparent in our study, including multiple case types, due to limited power.
Although there was more prior evidence for an association between non-Hodgkin's lymphoma and dioxins, we instead observed associations with the furans. Each of the furan congeners we examined was associated with increased risk of non-Hodgkin's lymphoma, as were summed furans and the furan TEQ. Like dioxins, furan levels were generally higher among subjects with a history of farming employment; however, associations with non-Hodgkin's lymphoma persisted when farmers were excluded from analyses. Furans are formed in the same processes as dioxins: as byproducts during the manufacturing of certain pesticides, such as phenoxy herbicides and chlorophenols; during the bleaching of pulp and paper; and during thermal reactions, such as waste incineration and wood combustion ( 10). Furans have not been associated previously with non-Hodgkin's lymphoma, although observations of increased non-Hodgkin's lymphoma incidence in occupational cohorts with high exposures to phenoxy herbicides and chlorophenols cannot rule out possible furan effects ( 11, 36). Further evidence is gleaned from toxicologic effects observed in Taiwanese and Japanese patients who were poisoned by food supplies contaminated with PCBs and furans; animal studies have since shown that immunosuppression and other biological effects observed in these patients can be largely attributed to the furans ( 7, 37, 38). Inconsistent with our finding, Hardell et al. ( 34) found no significant differences in furan levels measured in adipose tissue in his small hospital-based study. However, our highest exposure categories for estimating risks were set at higher cut points than in the previous study ( 34) for all of the congeners examined, and it is possible that increased risks are associated only with higher exposures.
Effects of organochlorine pesticides or metabolites in plasma were not completely clear in our data. In the parent non-Hodgkin's lymphoma study within which our organochlorine study was conducted, chlordane level in residential dust was associated with increased risk of non-Hodgkin's lymphoma, as was self-reported residential termite treatment before the ban on chlordane in 1988. 8 In addition, dust level of DDE was associated ( 26). When we looked at the extremes of our exposure distributions for pesticides in plasma, we found elevated risks for certain pesticides, including chlordane and p,p′-DDT, but wide 95% CIs because of the small numbers of subjects with these levels of exposure. There is evidence that exposure to (and corresponding biological levels of) organochlorine pesticides, including DDT and chlordane, have been decreasing in the United States over the past few decades ( 30), and it is possible that any increased risk associated with historical exposure levels is not apparent at the lower levels measured in our study population.
There have been few previous studies of biological measures of organochlorine pesticides in relation to non-Hodgkin's lymphoma. A prospective cohort study did not find any associations with DDT or its metabolites ( 1) or other pesticides ( 39) measured in serum. However, a small case-control study nested within the Nurses' Health Study cohort found nonsignificant increased risks associated with DDE measured in plasma ( 3), and two case-control studies, in which organochlorines were measured in adipose tissue ( 40, 41), found increased risks for several pesticides, including DDE ( 40), HCB ( 40, 41), and chlordane ( 41). Different biological media for organochlorine measurement could partially account for discrepant results between studies; however, a high correlation between organochlorines detected in serum, breast adipose, and gluteal adipose tissue has been documented ( 42).
Engel et al. ( 32) found stronger associations with non-Hodgkin's lymphoma incidence for PCBs measured in prediagnostic serum samples collected 1 to 12 years before diagnosis versus >12 years, suggesting that organochlorine exposure proximate to disease may be more important for lymphomagenesis than distant exposure. If there is a short latency for non-Hodgkin's lymphoma with regard to organochlorine effects, then the relevant exposures would be missed if blood was collected much earlier than diagnosis. Following this argument, case-control studies with blood samples collected around the time of diagnosis may capture a more relevant cumulative organochlorine concentration compared with cohorts in which blood samples would have been collected at a time point many years before incident disease. Despite this possible advantage of case-control studies, concerns about biological changes after cancer diagnosis may limit inference. We designed our study to only include cases who had not undergone chemotherapy or radiation treatment following non-Hodgkin's lymphoma diagnosis, because chemotherapy has been observed to alter serum organochlorine levels ( 21). Nevertheless, our untreated cases differed from controls in that their plasma lipid levels were lower. This biological difference could be due to changes resulting from disease symptoms but unrelated to treatment, such as changes in diet and weight. However, the magnitude of difference in lipids we observed (<10%) could also be plausibly due to transient changes resulting from feeding patterns during the day before blood draw ( 43). We used lipid-corrected organochlorine values for our analysis to account for differing organochlorine levels that may be solely due to different lipid levels. This type of adjustment would provide a reasonable correction only for the situation in which differences in blood lipids are not related to changes in the rate of release of organochlorines into the bloodstream. Without further knowledge about the nature of biological changes postdiagnosis, it is impossible to determine whether our results are affected by bias. Nevertheless, we would expect any such bias to act equally on all analytes, and because we observed only certain organochlorines to be associated with non-Hodgkin's lymphoma risk, there does not seem to be a global bias occurring for all associations.
The total body of available evidence now suggests that certain PCB congeners, particularly higher chlorinated PCBs, may contribute to non-Hodgkin's lymphoma risk. Our data also provide new evidence that furans may be associated with increased risk of non-Hodgkin's lymphoma. Our next step will be to examine the potential effect of persistent organochlorine chemicals in the presence of reactivated EBV or susceptibility genotypes to further elucidate biological mechanisms of associations observed in this study. Future research to clarify possible health effects of furans and other organochlorines should be conducted in studies with substantial amounts of blood to provide adequate sample volume for detecting low levels of organochlorines common in the general population or perhaps by pooling blood samples for different strata of participants to minimize the quantity of sample needed per individual ( 44). Additional information from several recently completed or ongoing prospective cohort studies of PCBs and non-Hodgkin's lymphoma will help with an overall assessment of the roles of various organochlorine compounds in the development of non-Hodgkin's lymphoma and would additionally advance knowledge in this field through assurance that associations are not due to possible bias introduced by measuring organochlorines in samples collected after disease diagnosis.
Grant support: Intramural Research Program of the Division of Cancer Epidemiology and Genetics of the NCI, NIH; NCI-Iowa SEER contract N01-CN-67008, NCI-Los Angeles County SEER contract N01-CN-67010, NCI-Seattle SEER contract N01-PC-67009, and NCI-Detroit SEER contract N01-PC-65064; and NCI-CDC interagency agreement Y1-CP-8028-02 for the measurement of organochlorines.
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 Carol Haines (Westat, Rockville, MD) for study coordination and management, Laura Capece and Irish Lonn (Information Management Services, Inc., Silver Spring, MD) for assistance in data processing and preparation, Jackie King (BioReliance, Rockville, MD) for all aspects of blood sample handling and shipping, and Wayman Turner (Dioxin and Persistent Organic Pollutants Laboratory, Centers for Disease Control) for management of and assistance in interpretation of the organochlorine measurement data.
↵8 J.S. Colt et al. Residential insecticide use and risk of non-Hodgkin lymphoma, submitted to Cancer Epidemiology, Biomarkers & Prevention.
- Received May 23, 2005.
- Revision received August 9, 2005.
- Accepted September 14, 2005.
- ©2005 American Association for Cancer Research.