Trichloroethylene Cancer Epidemiology: A Consideration of Select Issues

A large body of epidemiologic evidence exists for exploring causal associations between cancer and trichloroethylene (TCE) exposure. The U.S. Environmental Protection Agency 2001 draft TCE health risk assessment concluded that epidemiologic studies, on the whole, support associations between TCE exposure and excess risk of kidney cancer, liver cancer, and lymphomas, and, to a lesser extent, cervical cancer and prostate cancer. As part of a mini-monograph on key issues in the health risk assessment of TCE, this article reviews recently published scientific literature examining cancer and TCE exposure and identifies four issues that are key to interpreting the larger body of epidemiologic evidence: a) relative sensitivity of cancer incidence and mortality data; b) different classifications of lymphomas, including non-Hodgkin lymphoma; c) differences in data and methods for assigning TCE exposure status; and d) different methods employed for causal inferences, including statistical or meta-analysis approaches. The recent epidemiologic studies substantially expand the epidemiologic database, with seven new studies available on kidney cancer and somewhat fewer studies available that examine possible associations at other sites. Overall, recently published studies appear to provide further support for the kidney, liver, and lymphatic systems as targets of TCE toxicity, suggesting, as do previous studies, modestly elevated (typically 1.5–2.0) site-specific relative risks, given exposure conditions in these studies. However, a number of challenging issues need to be considered before drawing causal conclusions about TCE exposure and cancer from these data.

Despite numerous reviews [Brüning and Bolt 2000; International Agency for Research on Cancer (IARC) 1995; Institute of Medicine 2002; Lynge et al. 1997;McLaughlin and Blot 1997;National Toxicology Program (NTP) 2002;Wartenberg et al. 2000;Weiss 1996;Wong 2004], including those of two multidisciplinary expert panels that concluded that trichloroethylene (TCE) is "probably" (IARC 1995) or "reasonably anticipated to be" (NTP 2002) carcinogenic in humans, the interpretation of the epidemiologic studies on cancer and TCE exposure remains an area of considerable debate. The strongest epidemiologic evidence for associations between TCE exposure and cancer is for liver cancer, kidney cancer, and lymphomas, but perspectives have differed about the causal inferences regarding the human carcinogenicity of TCE that can be drawn from the epidemiologic database as a whole (e.g., Mandel and Kelsh 2001;Wartenberg et al. 2000). Some of the key issues underlying different interpretations are the use of different qualitative and quantitative (e.g., meta-analysis) methods to synthesize the body of evidence and the weight given to studies on the basis of different measures of cancer risk (e.g., incidence versus mortality) and different methods of exposure assessment. In addition, interpretation of data on lymphomas poses unique challenges because of the use of different classification systems and an evolving understanding of their etiology. As discussed in the overview article on this mini-monograph (Chiu et al. 2006a), these are all issues on which the National Academy of Sciences (NAS) has been asked to provide advice.
In this review we first summarize the recent epidemiologic literature on TCE exposure and cancer occurrence and then discuss the issues identified above as key to interpreting the larger body of epidemiologic evidence. Although some scientific conclusions can be drawn from this updated body of data, speculation about the impact of these data on the final TCE risk assessment would be premature at this point, given the ongoing NAS consultation discussed in the overview article by Chiu et al. (2006a) and the planned revision of the U.S. Environmental Protection Agency (EPA) TCE risk assessment. Therefore, the purpose here and throughout this mini-monograph is to review recently published scientific literature in the context of how it informs the key scientific issues believed to be most critical in developing a revised risk assessment.

Epidemiologic Studies on Cancer and TCE Exposure
The epidemiologic analysis in the U.S. EPA draft TCE risk assessment (U.S. EPA 2001) was supported in large part by the review by Wartenberg et al. (2000). This review identified more than 80 studies that evaluated cancer and TCE exposure, concluding that the evidence more firmly supported associations of TCE exposure with kidney and liver cancer while providing some support for associations with non-Hodgkin lymphoma (NHL). Wartenberg et al. (2000) also noted possible associations between TCE exposure and multiple myeloma and prostate, laryngeal, and colon cancer as well as cervical cancer and TCE or perchloroethylene exposure.
A number of studies and literature reviews have been published since 2000. Tables 1-3 provide short descriptions of these studies, which include historical or retrospective cohort studies (Table 1), case-control studies (Table 2), and ecologic or community studies ( Table 3). Most of the TCE cohort and case-control studies involve occupational exposure to TCE, primarily by inhalation, whereas community studies usually involve contaminated groundwater where potential TCE exposure may be through both ingestion of drinking water and inhalation from TCE vapor intrusion into subsurface residential areas or from showering. Many of these studies employed more sophisticated exposure assessment approaches, allowing better identification of likely TCE-exposed subjects (Brüning et al. 2003;Charbotel et al. 2006;DeRoos et al. 2001;Dumas et al. 2000;Hansen et al. 2001;Pesch et al. 2000aPesch et al. , 2000bRaaschou-Nielsen et al. 2003;Zhao et al. 2005). Tables 4-7 show corresponding study results for cancers that either are newly reported to have associations (Table 4, total cancers and cancers of the bladder, breast, and esophagus) or have drawn the most attention in previous reviews [ Table 5, kidney cancer or renal cell carcinoma (RCC); Table 6, cancer of the liver or liver and biliary passages; Table 7, lymphomas]. These recent studies substantially expand the epidemiologic database, providing additional insights on potential causal associations between TCE exposure and cancer occurrence. The following discussion focuses on the three groups of end points-kidney cancer and RCC, liver and biliary cancer, and lymphomas-previously identified as having the strongest evidence for potential causal association with TCE exposure (IARC 1995;NTP 2002;Wartenberg et al. 2000).
The studies available since 2000 report consistent associations between kidney cancer or RCC and TCE exposure (Table 5). Two cohort studies with large numbers of exposed cases (Raaschou-Nielsen et al. 2003;Zhao et al. 2005) observed statistically significant associations with greater exposure level or duration of employment. These findings were supported by three recent case-control studies assessing TCE exposure in the metal industry in Germany (Brüning et al. 2003;Pesch et al. 2000a) and in France . The studies by Brüning et al. (2003) and Charbotel et al. (2006) were designed specifically to examine the a priori hypothesis of an association between RCC and TCE exposure. Charbotel et al. (2006) suggested that exposure intensity may contribute to the risk associated with cumulative exposure because risks were higher for subjects in the highest cumulative exposure category with peak TCE exposure [odds ratio (OR) = 2.7; 95% confidence interval (CI), 1.1-7.1] than for subjects with only high cumulative exposure (OR = 2.2; 95% CI, 1.0-4.6), compared with unexposed subjects.
Most of the recent cohort studies also provide information as to possible association between TCE and liver and/or biliary tract cancer, although many examined only the combined category (Table 6). Grouping the adjacent, but anatomically distinct, end points of primary liver cancer and biliary cancer, which includes cancer of the gallbladder, limits application of mode-of-action data and may introduce misclassification bias. The recent Nordic cohort studies (Hansen 2004;Raaschou-Nielsen et al. 2003) disaggregate these cancers, and the addition of these two studies doubles the total number of epidemiologic studies providing information for primary liver cancer. The study by Raaschou-Nielsen et al. (2003), having greater statistical power because of its larger cohort size, suggested that both sites are possible targets of TCE toxicity, reporting a standardized incidence ratio (SIR) for primary liver cancer similar to that for gallbladder and biliary tract cancer. Risks for the larger category of liver and biliary tract cancers are presented in both the Nordic studies and the two recent community studies (Lee et al. 2003;Morgan and Cassady 2002). These studies together suggest a modest association (risks between 1.1 and 2.8), with no clear pattern with duration of exposure. Furthermore, none of the studies have sufficient power to identify sex differences in susceptibility.
New information on lymphomas, including NHL and leukemia, and TCE exposure comes from cohort and community studies ( Table 7). Both Nordic studies (Hansen et al. 2001;Raaschou-Nielsen et al. 2003) reported statistically significant associations with NHL, with increasing SIRs with increasing duration of employment. The risk of NHL mortality in Zhao et al. (2005) was more consistent than the NHL incidence with risks observed in Nordic cohorts. Except in the and 1997 for at least 3 months and alive as of higher level exposure to TCE) white-collar workers and companies with ≤ 200 workers were 1 January 1968 at 347 Danish TCE-using (339,486 P-Y). Cancer incidence variables identified as increasing the likelihood for TCE companies. Follow-up for cancer incidence rates of the Danish population. exposure. Subjects were identified from the following from 1 April 1968 or date of first industries: iron and metal, electronics, painting, printing, employment through 31 December 1997. chemical, and dry cleaning.
Abbreviations: JEM, job exposure matrix; P-Y; person-years; U-TCA, urinary trichloroacetic acid.  case of Raaschou-Nielsen et al. (2003), numbers of exposed NHL cases are small, limiting statistical power. The one available case-control study observed a strong but imprecise association between maternal exposure to TCE-contaminated drinking water during pregnancy and childhood leukemia (Costas et al. 2002). Aickin (2004) provides further evidence for an association between TCE in drinking water and childhood leukemia. Analyses using Bayesian statistical methods confirmed an elevated mortality in children from leukemia. Examining childhood leukemia incidence, Aickin (2004) reported that a rate ratio ≤ 1.0 was not credible, and risk > 2.0 could not be ruled out.
To illustrate the potential impact of these new studies, Figures 1-4 show relative risks, SIRs, and standardized mortality ratios (SMRs) from cohort studies and ORs from case-control studies for four cancer sites discussed above (liver, liver and biliary passages, kidney, and NHL, respectively). These figures include studies published before 2000 [reviewed in, e.g., Wartenberg et al. (2000)] and those discussed above. The integration of this new information will contribute substantially to the hazard characterization of a TCE health evaluation and become an integral part of the U.S. EPA revised TCE risk assessment. However, this integration requires consideration of a number of key issues related to interpretation and synthesis, as discussed below.

Studies of cancer incidence or cancer mortality.
Both cancer incidence and cancer mortality rates are potentially useful in risk assessment for identifying hazards and assessing dose-response relationships. Incidence rates, generally considered to provide an accurate indication of risk of a   disease in a population, are rarely available. In the absence of incidence data, epidemiologic studies have commonly relied on mortality data to assess exposure-disease associations. An understanding of the accuracy of death certificate information as a surrogate for incidence data is important for evaluating observations in the mortality studies. Known inaccuracies exist between cancer incidence and death certificate recordings for some cancer sites important to evaluating TCE exposure, for example, cancer of liver (primary) and liver and biliary passages (Percy et al. 1990). In their study of death certificate accuracy, Percy et al. (1990) showed that only 53% of 2,388 incident cases of primary liver cancer were actually attributed on the death certificate to this disease. Zhao et al. (2005) were able to examine both incidence and mortality among TCE-exposed workers and observed underreporting on death certificates for several site-specific cancers, including NHL, leukemia, and kidney and bladder cancers. Death certificate inaccuracies would obscure exposure-disease associations toward the null by reducing statistical power and may explain apparent inconsistencies between epidemiologic studies using incidence data versus those based on death certifications. For example, apparent inconsistencies in some observations from cohort studies of American workers, which were primarily based on mortality, and cohort studies of Nordic workers, which were largely based on incidence, may reflect misclassification of death certificates compared with incidence data.
Non-Hodgkin lymphoma. Lymphoma, including NHL, is a disease composed of numerous, etiologically distinct neoplasms (Fisher 2003;Herrinton 1998). Several issues may affect interpretation of NHL associations in the TCE epidemiologic studies and may be important to evaluating the consistency, or lack there of, across studies. First, epidemiologic studies evaluating NHL and TCE exposure have used a number of different International Classification of Diseases (ICD) revisions. All four Nordic studies (Anttila et al. 1995;Axelson et al. 1994;Hansen et al. 2001;Raaschou-Nielsen et al. 2003) classified NHL according to the seventh revision of the ICD [ICD-7; World Health Organization (WHO) 1957], and all reported consistent findings. Other revisions of the ICD were used in the more recent studies by Blair et al. (1998) [ICD Adapted (ICDA)-8, National Center for Health Statistics 1967], Boice et al. (1999) (ICD-9, WHO 1977, Garabrant et al. (1988) (ICD-9 in effect at date of death: ICD-7, ICDA-8, or ICD-9), Morgan et al. (1998Morgan et al. ( , 2000 (ICD in effect at date of death: ICD-7, ICDA-8, or ICD-9), and Ritz (1999) (ICD-9). Few case-control studies on lymphoma are available. NHL cases in Hardell et al. (1994) were histologically verified and were classified using the Rappaport system. Persson et al. (1989) do not identify the system used to classify NHL cases in their study. Classification of lymphomas has changed with each revision.
Second, understanding of histopathologic and immunologic characteristics of lymphoma has grown since 1977, the publication date of ICD-9. Past classifications of lymphomas do not reflect the current biologic understanding of NHL and do not make distinctions between different cell types. From this perspective, lymphomas are defined broadly as B-cell and T-cell lymphomas, with further divisions into precursor neoplasms and mature neoplasms (Cogliatti and Schmid 2002). This implies that lymphomas classified in the past into distinct categories may share common biological properties and differentiation pathways. For example, a lymphoma of B-cell origin may be classified under older schemes as NHL, multiple myeloma, or leukemia. Emerging data on molecular markers of lymphoma suggest stage of cell differentiation at time of exposure as an important factor in NHL development (Staudt and Dave 2005).

Exposure assessment issues in TCE epidemiologic studies. The methods by which exposure
is assessed in epidemiologic studies of TCE are diverse, ranging from use of broad job or industry categories to analysis of biomonitoring data. Generally, greater weight is assigned to studies with more precise and specific exposure estimates. Careful evaluation of a study's exposure assessment method is important in the evaluation of a body of epidemiologic data, particularly if divergent observations may be due to exposure misclassification bias reflecting incorrect assignment of study subjects to exposure groups. Many of the TCE studies lack actual exposure measurements for individual subjects, and surrogates such as available current or historical monitoring data are often used to reconstruct exposure parameters.
The three Nordic cohorts of Axelson et al. (1994), Anttila et al. (1995), and Hansen et al. (2001) identified study subjects using the TCE biological marker of urinary trichloroacetic acid (U-TCA), which provides some evidence of past TCE exposure, although usually not a full exposure history. These studies carry weight in the overall analysis because of their greater precision of exposure assessment compared with methods discussed below for other cohorts; however, a consideration of statistical 1.3 (0.6-2.5) 1-4.9 years 9 1.0 (0.5-1.9) ≥ 5 years 9 1.1 (0.5-2.1) Duration of employment, female ≤ 1 year 2 2.8 (0.3-10) 1-4.9 years 4 4.  Zhao et al. (2005) did not present relative risks for liver or liver and bile duct cancer in their article. b Total liver cancer deaths in the two villages.
power is also important because of fewer subjects compared with cohorts identified using other methods.
Other cohort and case-control studies have adopted a number of approaches for exposure assessment. TCE exposure has been assigned to subjects using surrogate information based on patterns of TCE use by job title obtained from historical job descriptions, from historical industrial hygiene surveys, or from personal interviews to develop job exposure matrices (JEMs). For several cohorts, industrial hygiene measurements either were absent before the 1970s (Boice et al. 1999;Marano et al. 2000;Morgan et al. 1998Morgan et al. , 2000 or were quite limited (Blair et al. 1998;Stewart et al. 1991). Furthermore, some cohort (Ritz 1999) and case-control (Greenland et al. 1994) studies classified study subjects as TCE exposed using information obtained from personal interviews or generic JEMs or job-task exposure matrices (JTEMs) in the absence of historical monitoring. Two issues associated with the use of generic JEMs are sensitivity (i.e., the ability to identify study subjects as exposed) and specificity (i.e., the ability to identify study subjects as not exposed).
Still other cohort studies (Chang et al. 2003(Chang et al. , 2005Costa et al. 1989;Garabrant et al. 1988) have defined exposure using occupation and industry. TCE is identified as one of a number of potential exposures, but no information is provided on individual subjects with TCE exposure. The main shortcoming of this type of study is that the lack of an association with a particular job or industry may mask the effect of exposure to a specific chemical to which only some individuals in the job are exposed (Teschke et al. 2002). For this reason, a consideration of potential exposure misclassification bias is important in weighting these studies in an overall weight of evidence.
In addition, multiple solvents and chemical agents are common in the TCE studies, adding to the complexity of exposure assessment and inferences about causality. Some studies of TCE also identify exposures to other chlorinated solvents such as perchloroethylene and 1,1,1-trichloroethane (Blair et al. 1998;Boice et al. 1999;Marano et al. 2000;Morgan et al. 1998Morgan et al. , 2000Stewart et al. 1991;Zhao et al. 2005). The potential for exposure to multiple chlorinated solvents is an important consideration in the TCE epidemiologic studies for two reasons. First, these chemicals can share similar metabolic profiles or modes of action as TCE (U.S. EPA 2001), and second, some epidemiologic studies have also reported independent associations between exposure to these other

Cohort studies
Case-control study solvents and cancer (Blair et al. 1998;Zhao et al. 2005). Physiologically based pharmacokinetic models such as those discussed by Chiu et al. (2006b) may be useful for better understanding cumulative exposure in these epidemiologic studies.

Approaches for Causal Inference
The practice of causal inference in environmental epidemiology relies on three approaches: narrative reviews, criteria-based inference methods, and, increasingly, meta-analysis (Weed 2002). All three have been employed in various analyses of the epidemiologic literature on cancer and TCE exposure. Narrative reviews of a body of epidemiologic evidence generally do not fully consider potential biases and confounding factors. By contrast, criteriabased approaches for assessing causality evaluate evidence according to a set of criteria or standards applied to the evidence (Weed 2002). For instance, the aspects proposed by Sir Bradford Hill (1965) are widely cited for framing the factors to consider in determining whether statistical associations are likely to be causal. Similar criteria are also presented in the U.S. EPA Guidelines for Carcinogen Risk Assessment (U.S. EPA 2005). Criteria-based approaches have increasingly been supplemented with formal statistical methods such as meta-analysis for reviewing and summing a body of evidence (Weed 2002). Common meta-analytic methods can include fitting of fixed-effects or random-effects models, linear regression analysis to assess doseresponse, or pooled analyses. Pooled analysis of the Nordic studies may be more feasible because of their similar design and similar follow-up period for documenting cancer incidence than for other TCE cohorts. As discussed in the overview article of this mini-monograph by Chiu et al. (2006a), the NAS has been asked to provide advice on appropriate meta-analysis methods, including the classification and weighting of individual studies.

Discussion and Summary
The U.S. EPA draft TCE assessment (U.S. EPA 2001) noted that epidemiologic studies, when considered as a whole, have associated TCE exposure with excess risk of kidney, liver, lymphohematopoietic, cervical, and prostate cancer. Recently published studies appear to provide further support for several of those conclusions, suggesting, as do previous studies, modestly elevated site-specific risk (typically between 1.5 and 2.0), given exposure conditions in the epidemiologic studies.
The recent epidemiologic studies strengthen the evidence that the kidney is a target of TCE toxicity. It should be noted that kidney toxicity besides cancer has been found by Radican et al. (2006), who reported a statistically significant association with end-stage renal disease mortality and exposure to solvents, including TCE. Understanding the mechanism by which TCE may act in kidney toxicity, including cancer, can inform cause-effect evaluations. The glutathione S-transferase (GST) metabolic pathway has been hypothesized as important to mode-ofaction considerations (Caldwell and Keshava 2006), and GST polymorphisms are reported to influence RCC risk associated with TCE exposure (Brüning et al. 1997). Brauch et al. (2004) examined somatic mutation to the von Hippel-Lindau tumor suppressor gene in renal cell tumors of non-TCE-exposed cases, comparing the prevalence of mutation to that found in renal tumors of TCE-exposed subjects reported in an earlier publication (Brauch et al. 1999). A higher prevalence of somatic mutations was found in renal cell tumors of TCE-exposed cases than in tumors of non-TCE-exposed cases. Moreover, the C > T transition at nucleotide 454, detected in some RCCs from TCEexposed subjects, was not found among the non-TCE-exposed RCC cases.
The recent studies also support the liver and immune system as being targets for TCE toxicity, with most of these studies showing elevated (and in some cases statistically significant) cancer risks from TCE exposure. However, although the number of studies assessing primary liver cancer separately from biliary tract cancers has doubled, the total number is still only 4, compared to 11 examining the combined category. With lymphomas, there are also a number of classification issues, including the use of different ICD revisions, and the fact that these groupings may lump together etiologically distinct neoplasms. Moreover, studies evaluating these end points include both incidence and mortality studies, which may have different sensitivity and biases. Thus, the reduced specificity in most studies, in combination with the relatively small number of total cases due to low background incidence, complicates interpretation of these findings.
Of particular importance for assessment of epidemiologic evidence on TCE exposure is characterizing the totality of the evidence in light of factors that may contribute to false positive findings or to false negative observations. The evidence presented on issues regarding data sources, exposure assessment, and disease classification can influence the statistical power of the epidemiologic study to detect whether there is an underlying risk. The challenge is to consider these issues, along with well-articulated approaches when evaluating the body of evidence, including the application of meta-analysis methods and rationale for grouping individual studies, in identifying hazards and drawing causal conclusions.