Joanne S. Colt

Epidemiologic Evaluation of Measurement Data in the Presence of Detection Limits

Quantitative measurements of environmental factors greatly improve the quality of epidemiologic studies but can pose challenges because of the presence of upper or lower detection limits or interfering compounds, which do not allow for precise measured values. We consider the regression of an environmental measurement (dependent variable) on several covariates (independent variables). Various strategies are commonly employed to impute values for interval-measured data, including assignment of one-half the detection limit to nondetected values or of “fill-in” values randomly selected from an appropriate distribution. On the basis of a limited simulation study, we found that the former approach can be biased unless the percentage of measurements below detection limits is small (5–10%). The fill-in approach generally produces unbiased parameter estimates but may produce biased variance estimates and thereby distort inference when 30% or more of the data are below detection limits. Truncated data methods (e.g., Tobit regression) and multiple imputation offer two unbiased approaches for analyzing measurement data with detection limits. If interest resides solely on regression parameters, then Tobit regression can be used. If individualized values for measurements below detection limits are needed for additional analysis, such as relative risk regression or graphical display, then multiple imputation produces unbiased estimates and nominal confidence intervals unless the proportion of missing data is extreme. We illustrate various approaches using measurements of pesticide residues in carpet dust in control subjects from a case–control study of non-Hodgkin lymphoma.

Etiologic heterogeneity among non-Hodgkin lymphoma subtypes.

Understanding patterns of etiologic commonality and heterogeneity for non-Hodgkin lymphomas may illuminate lymphomagenesis. We present the first systematic comparison of risks by lymphoma subtype for a broad range of putative risk factors in a population-based case-control study, including diffuse large B-cell (DLBCL; N = 416), follicular (N = 318), and marginal zone lymphomas (N = 106), and chronic lymphocytic leukemia/small lymphocytic lymphoma (CLL/SLL; N = 133). We required at least 2 of 3 analyses to support differences in risk: (1) polytomous logistic regression, (2) homogeneity tests, or (3) dichotomous logistic regression, analyzing all 7 possible pairwise comparisons among the subtypes, corresponding to various groupings by clinical behavior, genetic features, and differentiation. Late birth order and high body mass index (>/= 35) kg/m(2)) increased risk for DLBCL alone. Autoimmune conditions increased risk for marginal zone lymphoma alone. The tumor necrosis factor G-308A polymorphism (rs1800629) increased risks for both DLBCL and marginal zone lymphoma. Exposure to certain dietary heterocyclic amines from meat consumption increased risk for CLL/SLL alone. We observed no significant risk factors for follicular lymphoma alone. These data clearly support both etiologic commonality and heterogeneity for lymphoma subtypes, suggesting that immune dysfunction is of greater etiologic importance for DLBCL and marginal zone lymphoma than for CLL/SLL and follicular lymphoma.

Positional accuracy of two methods of geocoding.

Background: Geocoding is often used in epidemiologic studies to map residences with geographic information systems (GIS). The accuracy of the method is usually not determined. Methods: We collected global positioning system (GPS) measurements at homes in a case–control study of non-Hodgkin lymphoma in Iowa. We geocoded the addresses by 2 methods: (1) in-house, using ArcView 3.2 software and the U.S. Census Bureau TIGER2000 street database; and (2) automated geocoding by a commercial firm. We calculated the distance between the geocoded and GPS location (positional error) overall and separately for homes within towns and outside (rural). We evaluated the error in classifying homes with respect to their proximity to crop fields. Results: Overall, the majority of homes were geocoded with positional errors of less than 100 m by both methods (ArcView/TIGER 2000, median = 62 m [interquartile range = 39–103]; commercial firm, median = 61 m [interquartile range = 35–137]). For town residences, the percent geocoded with errors of ≤100 m was 81% for ArcView/TIGER 2000 and 84% for the commercial firm. For rural residences, a smaller percent of addresses were geocoded with this level of accuracy, especially by the commercial firm (ArcView/TIGER 2000, 56%; commercial firm, 28%). Geocoding errors affected our classification of homes according to their proximity to agricultural fields at 100 m, but not at greater distances (250–500 m). Conclusions: Our results indicate greater positional errors for rural addresses compared with town addresses. Using a commercial firm did not improve accuracy compared with our in-house method. The effect of geocoding errors on exposure classification will depend on the spatial variation of the exposure being studied.

A Case–Control Study of Smoking and Bladder Cancer Risk: Emergent Patterns Over Time

BACKGROUND Cigarette smoking is a well-established risk factor for bladder cancer. The effects of smoking duration, intensity (cigarettes per day), and total exposure (pack-years); smoking cessation; exposure to environmental tobacco smoke; and changes in the composition of tobacco and cigarette design over time on risk of bladder cancer are unclear. METHODS We examined bladder cancer risk in relation to smoking practices based on interview data from a large, population-based case-control study conducted in Maine, New Hampshire, and Vermont from 2001 to 2004 (N = 1170 urothelial carcinoma case patients and 1413 control subjects). We calculated odds ratios (ORs) and 95% confidence intervals (CIs) using unconditional logistic regression. To examine changes in smoking-induced bladder cancer risk over time, we compared odds ratios from New Hampshire residents in this study (305 case patients and 335 control subjects) with those from two case-control studies conducted in New Hampshire in 1994-1998 and in 1998-2001 (843 case patients and 1183 control subjects). RESULTS Regular and current cigarette smokers had higher risks of bladder cancer than never-smokers (for regular smokers, OR = 3.0, 95% CI = 2.4 to 3.6; for current smokers, OR = 5.2, 95% CI = 4.0 to 6.6). In New Hampshire, there was a statistically significant increasing trend in smoking-related bladder cancer risk over three consecutive periods (1994-1998, 1998-2001, and 2002-2004) among former smokers (OR = 1.4, 95% CI = 1.0 to 2.0; OR = 2.0, 95% CI = 1.4 to 2.9; and OR = 2.6, 95% CI = 1.7 to 4.0, respectively) and current smokers (OR = 2.9, 95% CI = 2.0 to 4.2; OR = 4.2, 95% CI = 2.8 to 6.3; OR = 5.5, 95% CI = 3.5 to 8.9, respectively) (P for homogeneity of trends over time periods = .04). We also observed that within categories of intensity, odds ratios increased approximately linearly with increasing pack-years smoked, but the slope of the increasing trend declined with increasing intensity. CONCLUSIONS Smoking-related risks of bladder cancer appear to have increased in New Hampshire since the mid-1990s. Based on our modeling of pack-years and intensity, smoking fewer cigarettes over a long time appears more harmful than smoking more cigarettes over a shorter time, for equal total pack-years of cigarettes smoked.

Persistent organochlorine chemicals in plasma and risk of non-Hodgkin's lymphoma.

Polychlorinated biphenyls (PCB) have been suspected as possible contributors to increasing non-Hodgkins 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-Hodgkins 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-Hodgkins 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-Hodgkins 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-Hodgkins 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-Hodgkins 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-Hodgkins lymphoma risk; these risks were most apparent for certain PCBs and furans.

Comparison of pesticide levels in carpet dust and self-reported pest treatment practices in four US sites

Epidemiologic studies have used both questionnaires and carpet dust sampling to assess residential exposure to pesticides. The consistency of the information provided by these two approaches has not been explored. In a population-based case–control study of non-Hodgkins lymphoma, carpet dust samples were collected from the homes of 513 control subjects in Detroit, Iowa, Los Angeles, and Seattle. The samples were taken from used vacuum cleaner bags and analyzed for 30 pesticides. Interviewers queried subjects about the types of pests treated in their home using a detailed questionnaire accompanied by visual aids. Geographic variations in pesticide levels were generally consistent with geographic differences in pest treatment practices. Los Angeles residents reported the most treatment for crawling insects, fleas/ticks, and termites, and Los Angeles dust samples had the highest levels of propoxur, chlorpyrifos, diazinon, permethrin, and chlordane. Iowa had the most treatment for lawn/garden weeds, and also the highest levels of 2,4-dichlorophenoxyacetic acid and dicamba. Although Seattle had the highest proportion of subjects treating for lawn/garden insects, the lawn/garden insecticides were higher in other sites. Multivariate linear regression revealed several significant associations between the type of pest treated and dust levels of specific pesticides. The strongest associations were between termite treatment and chlordane, and flea/tick treatment and permethrin. Most of the significant associations were consistent with known uses of the pesticides; few expected associations were absent. The consistency between the questionnaire data and pesticide residues measured in dust lends credibility to both methods for assessing residential exposure to pesticides. The combined techniques appear promising for epidemiologic studies. Interviewing is the only way to assess pesticide exposures before current carpets were in place. Dust sampling provides an objective measure of specific compounds to which a person may have been exposed through personal use of a pesticide or by drift-in or track-in from outside, and avoids recall bias.

Etiologic Heterogeneity Among Non-Hodgkin Lymphoma Subtypes: The InterLymph Non-Hodgkin Lymphoma Subtypes Project

BACKGROUND Non-Hodgkin lymphoma (NHL) comprises biologically and clinically heterogeneous subtypes. Previously, study size has limited the ability to compare and contrast the risk factor profiles among these heterogeneous subtypes. METHODS We pooled individual-level data from 17 471 NHL cases and 23 096 controls in 20 case-control studies from the International Lymphoma Epidemiology Consortium (InterLymph). We estimated the associations, measured as odds ratios, between each of 11 NHL subtypes and self-reported medical history, family history of hematologic malignancy, lifestyle factors, and occupation. We then assessed the heterogeneity of associations by evaluating the variability (Q value) of the estimated odds ratios for a given exposure among subtypes. Finally, we organized the subtypes into a hierarchical tree to identify groups that had similar risk factor profiles. Statistical significance of tree partitions was estimated by permutation-based P values (P NODE). RESULTS Risks differed statistically significantly among NHL subtypes for medical history factors (autoimmune diseases, hepatitis C virus seropositivity, eczema, and blood transfusion), family history of leukemia and multiple myeloma, alcohol consumption, cigarette smoking, and certain occupations, whereas generally homogeneous risks among subtypes were observed for family history of NHL, recreational sun exposure, hay fever, allergy, and socioeconomic status. Overall, the greatest difference in risk factors occurred between T-cell and B-cell lymphomas (P NODE < 1.0×10(-4)), with increased risks generally restricted to T-cell lymphomas for eczema, T-cell-activating autoimmune diseases, family history of multiple myeloma, and occupation as a painter. We further observed substantial heterogeneity among B-cell lymphomas (P NODE < 1.0×10(-4)). Increased risks for B-cell-activating autoimmune disease and hepatitis C virus seropositivity and decreased risks for alcohol consumption and occupation as a teacher generally were restricted to marginal zone lymphoma, Burkitt/Burkitt-like lymphoma/leukemia, diffuse large B-cell lymphoma, and/or lymphoplasmacytic lymphoma/Waldenström macroglobulinemia. CONCLUSIONS Using a novel approach to investigate etiologic heterogeneity among NHL subtypes, we identified risk factors that were common among subtypes as well as risk factors that appeared to be distinct among individual or a few subtypes, suggesting both subtype-specific and shared underlying mechanisms. Further research is needed to test putative mechanisms, investigate other risk factors (eg, other infections, environmental exposures, and diet), and evaluate potential joint effects with genetic susceptibility.

Hepatitis C virus infection and non-Hodgkin lymphoma: Results of the NCI-SEER multi-center case-control study

Several studies have noted elevated hepatitis C virus (HCV) prevalence among patients with non‐Hodgkin lymphoma (NHL), suggesting that HCV infection increases NHL risk through chronic immune stimulation. Population‐based data from the U.S. are lacking. In a population‐based case‐control study of NHL in the United States, we identified HCV infection using an enzyme immunoassay, confirmed by recombinant immunoblot assay or HCV RNA detection. The association between HCV and NHL was assessed using logistic regression, adjusting for demographic factors, illicit drug use or medical history. Thirty‐two of 813 (3.9%) NHL cases and 14 of 684 (2.1%) controls were HCV‐infected [odds ratio (OR) 1.96, 95%CI 1.07–4.03]. For separate NHL subtypes, numbers were limited. Nonetheless, positive associations were noted for follicular (OR 2.46, 95%CI 1.01–5.81), marginal zone (3.99, 0–13.6) and mucosa‐associated lymphoid tissue (2.04, 0–7.20) NHLs. For all NHLs combined, the HCV‐NHL association changed little after adjustment for sex, age, race and study center (OR 1.89, 95%CI 1.00–4.00). HCV was common in controls who had injected drugs (40%) or used other illicit drugs (6.5%), but adjustment for drug use did not affect the HCV‐NHL association (OR 1.87, 95%CI 0.95–4.10). Transfusion history was unrelated to HCV status, and adjustment for this exposure did not attenuate the HCV‐NHL association (OR 2.15, 95%CI 1.12–4.76). Excluding 4 subjects with a history of hemodialysis or 3 subjects with organ transplants also did not affect the results. Our study demonstrates an association between HCV infection and NHL in the United States. HCV infection may be a cause of NHL.

Proximity to Crops and Residential Exposure to Agricultural Herbicides in Iowa

Rural residents can be exposed to agricultural pesticides through the proximity of their homes to crop fields. Previously, we developed a method to create historical crop maps using a geographic information system. The aim of the present study was to determine whether crop maps are useful for predicting levels of crop herbicides in carpet dust samples from residences. From homes of participants in a case–control study of non-Hodgkin lymphoma in Iowa (1998–2000), we collected vacuum cleaner dust and measured 14 herbicides with high use on corn and soybeans in Iowa. Of 112 homes, 58% of residences had crops within 500 m of their home, an intermediate distance for primary drift from aerial and ground applications. Detection rates for herbicides ranged from 0% for metribuzin and cyanazine to 95% for 2,4-dichlorophenoxyacetic acid. Six herbicides used almost exclusively in agriculture were detected in 28% of homes. Detections and concentrations were highest in homes with an active farmer. Increasing acreage of corn and soybean fields within 750 m of homes was associated with significantly elevated odds of detecting agricultural herbicides compared with homes with no crops within 750 m (adjusted odds ratio per 10 acres = 1.06; 95% confidence interval, 1.02–1.11). Herbicide concentrations also increased significantly with increasing acreage within 750 m. We evaluated the distance of crop fields from the home at < 100, 101–250, 251–500, and 501–750 m. Including the crop buffer distance parameters in the model did not significantly improve the fit compared with a model with total acres within 750 m. Our results indicate that crop maps may be a useful method for estimating levels of herbicides in homes from nearby crop fields.

Residential exposure to polychlorinated biphenyls and organochlorine pesticides and risk of childhood leukemia.

Background Incidence of childhood leukemia in industrialized countries rose significantly during 1975–2004, and the reasons for the increase are not understood. Objectives We used carpet dust as an exposure indicator to examine the risk of childhood leukemia in relation to residential exposure to persistent organochlorine chemicals: six polychlorinated biphenyl (PCB) congeners and the pesticides α- and γ-chlordane, p,p′-DDT (dichlorodiphenyltrichloroethane), p,p′-DDE (dichlorodiphenyldichloroethylene), methoxychlor, and pentachlorophenol. Methods We conducted a population-based case–control study in 35 counties in northern and central California in 2001–2006. The study included 184 acute lymphocytic leukemia (ALL) cases 0–7 years of age and 212 birth certificate controls matched to cases by birth date, sex, race, and Hispanic ethnicity. We collected carpet dust samples from the room where the child spent the most time before diagnosis (similar date for controls) using a specialized vacuum. Results Detection of any PCB congener in the dust conferred a 2-fold increased risk of ALL [odds ratio (OR) = 1.97; 95% confidence interval (CI), 1.22–3.17]. Compared with those in the lowest quartile of total PCBs, the highest quartile was associated with about a 3-fold risk (OR = 2.78; 95% CI, 1.41–5.48), and the positive trend was significant (p = 0.017). Significant positive trends in ALL risk were apparent with increasing concentrations of PCB congeners 118, 138, and 153. We observed no significant positive associations for chlordane, DDT, DDE, methoxychlor, or pentachlorophenol. The associations with PCBs were stronger among non-Hispanic whites than among Hispanics despite similar distributions of PCB levels among controls in each racial/ethnic group. Conclusions Our findings suggest that PCBs, which are considered probable human carcinogens and cause perturbations of the immune system, may represent a previously unrecognized risk factor for childhood leukemia.