Document gE03DQKOow37Kmd8xKMoZZXEa

AMERICAN JOURNAL OF INDUSTRIAL MEDICINE 31:287295 (1997) Review of Epidemiologic Evidence on Benzene and Lymphatic and Hematopoietic Cancers David A. Savitz, PhD,* and Kurtis W. Andrews, PhD Exposure to benzene is generally accepted as a cause of acute myeloid leukemia (AML), but the association with other cell types of leukemia and other lymphatic and hematopoietic cancers is controversial. We compiled epidemiologic research on benzene and lymphatic and hematopoietic cancers in order to assess the pattern of associations. Eighteen relevant community-based and 16 industry-based studies were located. Four of seven studies of lymphatic and hematopoietic cancer in the aggregate identified relative risks of 1.8 or more, and eight of 14 total leukemia studies yielded relative risks in that range. The few available studies of specific histologic types of leukemia do not indicate larger or more consistent elevations in risk for AML compared to other leukemia cell types. Sporadic reports have linked benzene to non-Hodgkin's lymphoma and multiple myeloma, but most studies do not indicate a positive association. Limitations in study quality, particularly exposure assessment, pervade all of the studies reviewed, and the distinction between studies addressing benzene and those addressing jobs in industries that use benzene is somewhat arbitrary. Nonetheless, the epidemiologic evidence linking benzene to leukemia in the aggregate, as well as for subtypes other than AML, is no less persuasive than that for AML alone. Am. J. Ind. Med. 31:287295, 1997. r 1997 Wiley-Liss, Inc. KEY WORDS: benzene; leukemia; lymphoma INTRODUCTION Benzene has been suspected as an occupational carcinogen since the 1920s [International Agency for Research on Cancer, 1982]. In 1981, the evidence that benzene is carcinogenic in experimental animals was characterized by the International Agency for Research on Cancer [1982] as ``limited,'' and the evidence that benzene is carcinogenic to humans was found to be ``sufficient.'' In part because the cancers of primary concern, lymphatic and hematopoietic neoplasms, are rare, case series documenting ``larger than expected'' numbers of such cancers among shoe workers in Department of Epidemiology, School of Public Health, University of North Carolina, Chapel Hill, North Carolina. Contract Grant sponsor: the National Institute of Environmental Health Sciences Training; Contract Grant number: T32-ES07018. *Correspondence to: David A. Savitz, PhD, Department of Epidemiology, CB #7400, University of North Carolina, Chapel Hill, NC 27599. E-mail: DAVID_SAVITZ@UNC.EDU Accepted 24 June 1996. Italy [Vigliani & Saita, 1964] and Turkey [Aksoy et al., 1974] provided important early information regarding increased risk. Although such observations lend general support to the notion that more cases than expected have occurred, in the absence of information on the population at risk and disease rates in some comparison population, the magnitude of association between benzene exposure and lymphatic and hematopoietic cancers cannot be assessed. Subsequent studies of benzene and cancer have attempted to measure exposure and disease risk directly in order to derive dose-response relations. Although controversy remains about the exact shape of the function relating benzene exposure to acute myeloid leukemia (AML) [Austin et al., 1980; Swaen and Meijers, 1989; Paustenbach et al., 1994], with particular uncertainty regarding risks in the low-dose range (around 10 ppm), it is generally accepted that benzene exposure in human populations is causally related to the development of AML. Most of the epidemiologic studies include information on several types of leukemias and often on other types of cancers as well. In fact, many of the early studies found the r 1997 Wiley-Liss, Inc. 288 Savitz and Andrews strongest evidence linking benzene to total leukemia rather than AML [Aksoy et al., 1974; Girard and Revol, 1970; Thorpe, 1974]. In the rubber industry study, specific leukemia cell types other than AML were more strongly implicated [McMichael et al., 1976]. Lymphatic and hematopoietic cancers other than leukemias have also been associated with benzene exposure, including lymphoreticular sarcoma [Vianna & Polan, 1979] and multiple myeloma [Rinsky et al., 1987]. The possibility of a broader array of lymphatic and hematopoietic cancers resulting from benzene has been suggested by Goldstein, based largely on toxicologic data [Goldstein, 1988; Goldstein, 1990; Hricko, 1994]. Given a substantial number of epidemiologic studies addressing benzene and lymphatic and hematopoietic cancer, most of which include cancers other than AML, we undertook a systematic effort to identify and compile the results in order to evaluate the epidemiologic evidence linking benzene to a range of specific types of lymphatic and hematopoietic cancers. MATERIALS AND METHODS We sought all published studies, primarily in the peer-reviewed literature, that generated a risk estimate relating benzene exposure to individual or grouped lymphatic and hematopoietic cancers. There was no minimum quality criterion for inclusion; instead, study quality was noted in the tabulations and interpretation of results. For both community-based and industry-based studies, an explicit effort to assess the role of benzene was required, omitting results from studies of industries in which benzene is known to be present (e.g., oil refineries, rubber manufacturing, shoe manufacturing) unless the researchers tried to isolate the subset of the population thought to have benzene exposure based on work locations or job titles. Similarly, we did not include studies of organic solvent exposure in the aggregate unless a subset of agents or workers was examined that directly addressed benzene. However, studies were not excluded for evaluating benzene as part of a relatively narrow subset of organic solvents. Although we began with the intention of conducting a quantitative meta-analysis, the disparities among studies in exposure assessment in particular made them of problematic comparability. The evaluation of the studies was therefore not based on formal quantitative criteria but rather on the patterns of results, emphasizing the magnitudes of relative risk estimates across studies. A second criterion for inclusion was an explicit comparison of risk for persons thought to be exposed to benzene compared to those not thought to be exposed, usually generating a measure of relative risk (standardized mortality ratio, odds ratio, observed to expected ratio, etc.). We also sought to identify or calculate the number of exposed cases or precision of the risk estimate. This eliminated several case series reports that lacked an indication of the expected number of such cases [Vigliani and Saita, 1964] and one study that did not provide sufficient information to derive the number of exposed cases [Chinese Epidemiologic Study Group of Leukemia and Aplastic Anemia, 1992]. We attempted to include publications in languages other than English when the information was clearly relevant and not available elsewhere [Girard and Revol, 1970; Vai et al., 1989] but we are less confident that all relevant non-Englishlanguage publications were found. When multiple publications were produced from the same study population, we tabulated results from the most complete or up-to-date version of the study [Rinsky et al., 1987; Paxton et al., 1994; Wong, 1995]. When reports differed by the amount of specific information available on benzene, we selected the most detailed report for inclusion in the tables. In one instance, the more recent study's results are not yet complete, so that we include parts of the earlier results [Yin et al., 1989], which are a subset of the later results [Travis et al., 1994]. The major division in the studies was between those that were conducted in a community population versus those that were based in a specific industry, such as oil refining or rubber manufacturing. Overall quality of exposure assessment was characterized as low, moderate, or high. For community studies, ``low'' meant that it was based on a job title or self-report with no evaluation by an industrial hygienist or similar expert, whereas ``high'' meant that the information was assessed by an expert familiar with the use of chemicals in the workplace. For industry-based studies, ``low'' meant that there was no explicit workplace evaluation, ``moderate'' meant that there was an evaluation to determine that benzene was present, and ``high'' meant that a systematic evaluation of job titles or environmental measurements was undertaken by an expert to determine probability of benzene exposure. Some reports examined multiple groups defined by level of exposure, certainty of exposure, duration of exposure, or latency, which we included in the tabulations. When results were available with and without adjustment for potential confounders, we chose to present those which were adjusted. No distinction was made in the tables between incident and fatal cases. In some instances, we had to calculate odds ratios or observed to expected ratios from data provided in the study; 95% confidence intervals were calculated using appropriate methods [Rothman and Boice, 1982]. We used the codes from the Eighth Revision of International Classification of Diseases [US Department of Health and Human Services, 1980] to group cancers regardless of the original classification scheme used by investigators. We examined results for all lymphatic and hematopoietic cancers (ICD codes 200209), non-Hodgkin's lymphoma (ICD codes 200 and 202), multiple myeloma (ICD code 203), leukemia and aleukemia (ICD codes 204207), and leuke- Benzene and Lymphatic/Hematopoietic Cancer 289 TABLE I. Description of Community-based Studies Reporting on Benzene and Lymphatic and Hematopoietic Cancers First author (year) Setting Exposure data source Exposure assignment Quality of exposure assessment Control for confounders Girard (1970) France Interview Self-report Low Age, sex Ishimaru (1971) Hiroshima & Nagasaki Interview Job title Low Age, sex, city, distance from A-bomb Vianna (1979) New York State Death certificate Job title Low None Linos (1980) Olmsted County, Medical records Job title Low Age, sex, residence, dura- Minnesota tion of medical coverage Flodin (1986) Sweden Interview Self-report High None Linet [CLL] (1987a) Baltimore Interview Self-report High Age, sex, race Linet [MM] (1987b) Baltimore Interview Self-report High Age, sex, year, hospital Schumacher (1988) North Carolina Death certificate Job title High Age, race, year Malone (1989) US (multi-center) Interview Self-report High Age, sex, race, residence, education Vai (1989) Italy Hospital record Clinical evaluation High Age, sex, year Crane (1989) Houston Interview Self-report High None Siemiatycki (1991) Montreal Interview Job description High Age, income, smoking Richardson (1992) Paris Interview Job description High Age, sex, ethnicity, resi- dence, hospital, prior cancer therapy Scherr (1992) Boston Interview Self-report High Age, sex, residence Blair (1992) Iowa & Minnesota Interview Job title High Age, sex, state, year, edu- cation, pesticides, family cancer history, hair dyes, direct/surro- gate respondent Cranea (1992) Houston Interview Job title Low Age, sex, prior cancer therapy Heineman (1992) Denmark Tax records Job title High Age Ciccone (1993) Torino, Italy Interview Job title High Age, sex, smoking aSome overlap with subjects in Crane et al. (1989). mias divided by chronicity and cell type (acute and chronic, lymphocytic and myeloid). RESULTS Eighteen community-based (Table I) and 16 industrybased (Table II) studies provided data on benzene exposure and lymphatic and hematopoietic cancers. Communitybased studies identified exposure from self-report of exposure or job title, with each having advantages and disadvantages (Table I). Only four studies integrated self-report, job title, and expert evaluation [Siemiatycki, 1991; Richardson et al., 1992; Heineman et al., 1992; Ciccone et al., 1993], which is optimal in such studies. The earliest studies generally did the poorest job of identifying exposure. Most, but not all, studies adjusted relative risks for age, and where appropriate for sex and race. Few studies included a broader array of potential confounders related to lifestyle and none attempted to incorporate information on other occupational workplace exposures. Industry-based studies are concentrated in the chemical, oil, rubber, and shoe manufacturing sectors (Table II), all of which classified exposure based on industrial-hygienist judgments, sometimes combined with measurements. Most approaches to exposure assessment were considered to be of high quality. Consideration of confounding was largely limited to age and calendar time. Four of seven studies of lymphatic and hematopoietic cancers in the aggregate showed two-fold or greater elevated risks associated with benzene exposure (Table III), with the Tsai et al. [1983] study largely uninformative due to small 290 Savitz and Andrews TABLE II. Description of Industry-based Studies Reporting on Benzene and Lymphatic and Hematopoietic Cancers First author (year) Setting Industry Quality of exposure assessment Control for confounders Aksoy (1974) Thorpe (1974) Rushton (1981) Tsai (1983) Decoufle (1983) Checkoway (1984) Bond (1986) Austin (1986) Wong (1987a, b) Yin (1989) Ott (1989) Hurley (1991) AIPHSa (1992) Paxton (1994) Wong (1995) Greenland (1994) Travis (1994) Istanbul, Turkey Europe United Kingdom Texas US US Michigan Illinois US China US United Kingdom Australia US Massachusetts China Shoe manufacturing Oil/petrochemical Oil/petrochemical Oil/petrochemical Chemical Rubber Chemical Oil/petrochemical Chemical All industries Chemical Coke and coal Oil/petrochemical Pliofilm (rubber) Transformer assembly Multiple industries Low High High High High High High High High High Moderate Moderate High High High High None Age Age, duration of work, year hired Age, sex, year, race Age, year Age Age, year Age Age, year Age None Age, year Age Age, year Age, year of hire, year of death Age, sex aAbbreviation: AIPHS, Australian Institute of Petroleum Health Surveillance Program. TABLE III. Results of Studies Reporting on Benzene and Total Lymphatic and Hematopoietic Cancers First author (year) Study setting Exposed cases Relative risk 95% CI Vai (1989) Tsai (1983) Decoufle (1983) Bond (1986) Wong (1987a) AIPHS (1992) Travis (1994) Community Industry Industry Industry Industry Industry Industry 15 0 4 7 30a 8b 10c 7d 82 13.3 8.022.2 0.0 0.03.3 3.8 1.110.2 1.3 0.52.8 1.3 0.91.8 1.0 0.52.1 1.9 0.74.6 2.3 0.86.5 3.4 1.96.1 aContinuous benzene exposure. bIntermittent benzene exposure. cMedium versus low benzene exposure. dHigh versus low benzene exposure. numbers. Although the comparison between benzeneexposed workers and the general population in Wong's [1987a] study provided evidence against an association with lymphatic and hematopoietic cancers, internal comparisons to the unexposed cohort and, particularly, analyses of risk in relation to cumulative benzene exposure [Wong, 1987b] TABLE IV. Results of Studies Reporting on Benzene and Total Leukemia First author (year) Study setting Exposed cases Relative risk 95% CI Girard (1970)a Ishimaru (1971)b Linos (1980) Thorpe (1974) Rushton (1981) Tsai (1983) Decoufle (1983) Bond (1986) Austin (1986) Wong (1987a) Yin (1989) Hurley (1991) Paxton (1994) Greenland (1994) Community Community Community Industry Industry Industry Industry Industry Industry Industry Industry Industry Industry Industry 30 24 4 8 18 0 3 4 15 6c 1d 25 5 14 NAf NAg 2.9 1.17.4 1.8 0.93.7 3.3 0.627.6 1.2 0.52.4 2.3 1.05.0 0.0 0.08.7 6.8 1.419.9 1.9 0.54.9 NA NA 1.4 0.53.0 0.7 0.03.7 5.7 4.08.0e 0.4 0.11.0 3.6 2.06.0 0.9 0.33.1 1.4 0.63.2 aAcute, CLL, myeloid combined. bWelders and radiologists excluded to exclude presumed x-ray rather than benzene exposure. cContinuous exposure. dIntermittent exposure. eFrom Wu (1988). fAny versus no exposure. gOver 97.5 percentile versus no exposure. Benzene and Lymphatic/Hematopoietic Cancer 291 TABLE V. Results of Studies Reporting on Benzene and Specific Histologic Types of Leukemia TABLE VI. Results of Studies Reporting on Benzene and Non-Hodgkin's Lymphoma First author (year) Study setting Exposed cases Relative risk First author 95% CI (year) Study setting Disease groupa Exposed Relative cases risk 95% CI Lymphocytic Checkoway (1984) Ott (1989) Chronic Lymphocytic Girard (1970) Linet (1987a) Industry Industry Community Community Malone (1989) Myeloid Ciccone (1993) Bond (1986) Austin (1986) Ott (1989) Acute Girard (1970) Richardson (1992) Community Community Industry Industry Industry Community Community Aksoy (1974) Acute Myeloid Flodin (1986) Crane (1989) Crane (1992) Wong (1995) Industry Community Community Community Industry 4 2 9 60a 43b 31c NA 9 4 15 5 17 22d 15e 26 0 6 4 6 2.5 NA 1.5 NA 3.7 1.210.7 1.4 0.92.2 1.5 0.92.5 0.9 0.51.5 1.1 0.62.0 1.7 0.65.5 4.4 1.211.4 NA NA 1.0 NA 3.0 1.18.1 1.3 0.82.3 2.8 1.35.9 2.2 1.43.2 0.0 -- 1.1 0.33.8 3.3 0.522.3 5.0 1.811.0 aCalculated from Table 4 based on self-reported exposure. bBased on Hoar et al. (1980) job-exposure matrix. cBased on National Occupational Hazards Survey (Sieber et al., 1991) job-exposure matrix. dAny exposure. eMedium or high exposure. support a positive association with relative risks in the range of 25 in the most highly exposed groups. Overall, given the diversity of methods and likely bias toward the null from nondifferential exposure misclassification, the replicated evidence of increased risk is notable across a series of relatively high-quality industry-based studies. Leukemia in the aggregate provides stronger evidence for an increased risk, with relative risks above 2 in two community-based studies [Girard and Revol, 1970; Linos et al., 1980] and four industry-based studies [Rushton and Alderson, 1981; Decoufle et al., 1983; Yin et al., 1989; Paxton et al., 1994] (Table IV). Two studies yielded relative risk estimates of 1.5 to 2.0 [Ishimaru et al., 1971; Bond et al., 1986], but several studies generated relative risk estimates near or below 1.0. The comparison of benzene-exposed workers to the general population yielded the relative risk estimates reported for Wong [1987a] in Table IV, with Girard (1970) Vianna (1979) Schumacher (1988) Community Community Community LS/RS LS LS/RS, Other LC Siemiatycki (1991) Community All NHL Scherr (1992) Blair (1992) Checkoway (1984) Bond (1986) Wong (1987a) Community Community Industry Industry Industry All NHL All NHL LS LS/RS LS/RS Ott (1989) Greenland (1994) Industry Industry All NHL All NHL 1 499 56b 10c 26d 6e NA 153 6 2 3f 1g 5 NA 0.8 0.17.4 2.0 1.82.1 0.8 0.51.1 0.9 0.42.1 0.7 0.51.1 0.8 0.31.9 1.2 0.52.6 1.1 0.91.4 3.0 NA 1.8 0.26.6 1.1 0.23.3 1.2 0.06.4 1.0 NA 1.0 0.24.5 aLS, lymphosarcoma; RS, reticulsarcoma; LC, lymphatic cancer; NHL, non-Hodgkin's lymphoma. bWhites, 95% CI estimated based on 90% CI. cBlacks, 95% CI estimated based on 90% CI. dAny exposure, 95% CI estimated based on 90% CI. eSubstantial exposure, 95% CI estimated based on 90% CI. fContinuous exposure. gIntermittent exposure. internal cohort comparisons not calculable due to an absence of cases in the unexposed cohort, but dose-response analyses showing a concentration of risk in the most highly exposed group [Wong, 1987b]. For example, in the group with over 720 ppm-months of occupational benzene exposure, the SMR was 2.8 (95% confidence interval 0.68.1), based on three observed cases. Analysis of specific cell types of leukemia provides the most direct suggestion of a general rather than specific association of benzene with leukemia (Table V). Studies grouped leukemias in different ways, e.g., aggregating all lymphatic or all acute leukemias, so that the number of reports for individual cell types is limited. Also, imprecision is more problematic for specific histologic types of leukemia than for leukemia in the aggregate. Both studies examining lymphocytic leukemia and two of three studies addressing chronic lymphocytic leukemia generated relative risks of 1.5 or more. Two of four studies of myeloid leukemia and acute myeloid leukemia met this criterion, whereas all three studies generated relative risks of 2 or more for acute 292 Savitz and Andrews TABLE VII. Results of Studies Reporting on Benzene and Multiple Myeloma Reference (year) Study setting Exposed cases Relative risk Linet (1987b) Heineman (1992) Ott (1989) Wong (1995) Community Community Industry Industry NA 137a 52b 5 4 1.2 1.3 0.8 1.4 2.9 aPossible exposure. bProbable exposure. 95% CI 0.43.6 1.01.6 0.61.1 NA 0.87.5 leukemia. The quality of studies is quite mixed, limited particularly for the community studies which contribute over half of the relative risk estimates, perhaps because it is difficult to accrue a sufficient number of cases in industrybased studies to examine specific histologic types of leukemia. The overall pattern does not indicate an association limited to AML, with the evidence for lymphocytic leukemia similar to that for myeloid leukemia. Non-Hodgkin's lymphoma was generally not associated with an overall elevated relative risk (Table VI), with only three studies providing support for such an association [Vianna and Polan, 1979; Checkoway et al., 1984; Bond et al., 1986] and a number of well-designed, large studies failing to report increased risks [Schumacher and Delzell, 1988; Siemiatycki, 1991; Blair et al., 1992]. Multiple myeloma, first linked to benzene in the report of two cases by DeCoufle et al. [1983] was strongly associated with benzene exposure in the study of Pliofilm workers [Rinsky et al., 1987; Wong, 1995], in which four cases were observed versus 1.37 expected (Table VII). Other studies [Linet et al., 1987; Ott et al., 1989; Heineman et al., 1992] found only very weak associations (relative risks under 1.5). DISCUSSION The pattern of epidemiologic study results across broad and narrow groups of lymphatic and hematopoietic cancers does not lead to a conclusion that benzene is a cause of AML alone. Considering the strength and precision of associations, the data are most supportive of a rather broad effect on lymphatic and hematopoietic cancers or leukemia in general. Since these diseases are subsets of one another, extremely strong associations for AML would have to influence the total leukemia results, which would, in turn, influence the results for all lymphatic and hematopoietic cancers. However, the diseases are not quantitatively dominant subsets, with AML making up approximately 24% of all leukemias, and leukemias making up 41% of all lymphatic and hemato- poietic cancers [Young et al., 1981]. Thus, although these findings reflect overlapping groups, they do not simply indicate an association with a single subset of cases. Wong [1995] recently highlighted the strong association between benzene exposure and AML in the Pliofilm cohort (six observed, 1.19 expected, relative risk of 5.03), yet by subtraction from the observed and expected numbers of total leukemias [Paxton et al., 1994], the residual of leukemias other than AML are also associated with benzene exposure (eight observed, 2.70 expected, relative risk of 2.96) [Savitz and Andrews, 1996]. The perception that the association between benzene and AML dominates the overall pattern of epidemiologic study results is not supported by our review. Restricting our analyses to epidemiologic studies excludes a number of lines of evidence. Extensive laboratory research [Goldstein, 1988] was not considered, even though such studies provide an important component of the overall assessment of the health effects of benzene. Case series on benzene and leukemia provide evidence both for specific associations with AML as well as other cell types of leukemia [International Agency for Research on Cancer, 1982], but were not included in the tabulations unless explicit comparison was made of risk among exposed and unexposed individuals. Even epidemiologic studies of industries in which benzene is known to have been used were excluded unless benzene was examined explicitly. The extensive data from the petroleum industry [Delzell et al., 1988; Savitz and Moure, 1984] and rubber industry [McMichael et al., 1976] are potentially relevant to assessing the health risks associated with benzene, but without some restriction to the subsets of workers with benzene exposure the data are difficult to apply. We did not exclude any studies for failure to meet minimum quality criteria. Exposure assessment, particularly in community-based studies, is subject to substantial error and may well reach a point at which the authors' assertions that benzene is under investigation can be challenged. Inferences based on aggregations of job titles, without careful review by industrial hygienists [Vianna & Polan, 1979] or other corroboration of inferences regarding benzene exposure, are subject to a great deal of uncertainty. Self-reported exposure of cases and controls [Girard and Revol, 1970; Malone et al., 1989] may be superior to assessment by job title if the study subjects have sufficient understanding of their work environment. The optimal approach in community-based studies is the synthesis of self-report with expert evaluation [Siemiatycki, 1991] or specialized questionnaires designed to elicit essential information for assessment of exposure [Stewart and Stewart, 1994]. Few community-based studies attained that level of quality. Where cases and controls contribute subjective information, there is the potential for biased reporting (over-reporting of exposure by cases) and falsely elevated associations. Otherwise, for inferences based on job titles or Benzene and Lymphatic/Hematopoietic Cancer 293 other more objective data, the errors are likely to be similar for cases and controls and the resulting bias toward failing to identify associations that are truly present. Industry-based studies have used more objective exposure assessment methods, relying on blinded coding by experts or inferring exposure based on job characteristics. Some involvement with exposure assessment experts was almost always involved, ensuring a higher minimum quality. Nonetheless, even with certainty that benzene is present in the work environment and that some individuals are exposed, there is often substantial uncertainty regarding which individual jobs involved exposure to benzene and the magnitude of that exposure. Quantitative inferences relating levels of benzene exposure to risk of cancer is impossible in all but a few of the epidemiologic studies reviewed [Paustenbach et al., 1994]. Given the diversity of approaches to exposure assessment and their varying quality, the potential for the intensity and accuracy of exposure assessment to produce notably different risk estimates should be noted. Failure to isolate benzene from closely associated agents that may themselves cause lymphatic and hematopoietic cancers could introduce confounding. In the rubber industry, for example, associations between benzene and lymphocytic leukemia were reported [Checkoway et al., 1984], yet other solvents were as or more strongly associated with lymphocytic leukemia, making isolation of the etiologic agent difficult. Environments associated with the petrochemical industry, shoe manufacturing, and painting contain a wide array of organic solvents and other potentially carcinogenic chemicals which were not fully addressed in the published reports. In fact, no matter how carefully job titles and work settings are examined, the ability to isolate benzene from concomitant exposures will be limited. Diagnostic quality of lymphatic and hematopoietic cancers can be problematic, especially in past time periods and for deaths rather than incident cases. Diagnoses would generally be superior in community-based studies in which incident cases are identified through hospital records or tumor registries. Although death certificates, typically used in industry-based studies, are accurate for broad groupings like leukemia, they are subject to substantial error in making inferences about specific histologic types of leukemia [Percy et al., 1981]. Such misclassification across histologic types would tend to dilute associations that occur only within a specific subtype of leukemia, and reduce precision if many cases are relegated to the ``unspecified'' category. The methods for identifying all relevant research to be included in this review are vulnerable to bias as well. The decision of the authors to publish articles and include selected results may be affected by the pattern of those results. Evaluations of benzene and cancer that did not produce positive associations may have been less likely to be published given the general knowledge that benzene is carcinogenic. Even in studies that were published, the decision to provide results specifically for benzene among an array of agents of interest may have been influenced by whether an association was found. One might expect such bias to be strongest for AML, in which failure to find associations is seen as aberrant. In light of the research findings reviewed here, sufficient uncertainty remains to encourage additional study with refined exposure assessment. Clarification of the pattern of benzene associations with specific leukemia and lymphoma cell types requires extremely large studies, so that cohort studies with sufficient statistical precision are required. If based on mortality as the endpoint, such studies may be lacking in sufficient diagnostic precision to address this issue. Community-based case-control studies, which have the potential for attaining adequate study size and accurate diagnoses, are severely limited in their ability to assess exposure across diverse industries. Although a number of agents have been effectively reconstructed by experts [Gerin et al., 1984], benzene poses particular challenges: It was rarely used in pure form, but as one among several solvents; it is no longer used as a commercial solvent, so that present-day measurements provide no information relevant to past exposures; and the benzene levels of concern are relatively low. In spite of limitations, mortality follow-up of historically exposed cohorts should continue, with attention to identifying diagnostic subtypes of these cancers, as well as refined exposure assessment in community-based studies. Even where exposure has terminated, continued follow-up is warranted to fully understand the long-term consequences of benzene exposure. A recently completed study in China may provide the clearest evidence to date regarding benzene and lymphatic and hematopoietic cancers [Yin et al., 1994; Dosemeci et al., 1994]. Methods, but not detailed results, from this large cohort and nested case-control study of multiple factories have been described, and it promises to provide refined exposure and cancer data on a cohort of nearly 80,000 exposed men and women, as well as 36,000 unexposed workers [Yin et al., 1994]. Differences in leukemia cell type distributions between Caucasian and Chinese populations, however, may affect the generalizability of their results, at least in regard to the types of leukemia affected by benzene exposure. CONCLUSION The epidemiologic evidence linking benzene to leukemia in the aggregate, as well as acute and chronic lymphocytic and myeloid leukemia, is no less persuasive than that for AML alone. Specificity of association was once cited as a hallmark of causality [Hill, 1965], yet with expansion of knowledge regarding agents such as tobacco, ionizing radiation, and asbestos, the criterion of specificity has little 294 Savitz and Andrews value. By analogy with other known carcinogens, it would be surprising if benzene's only carcinogenic effect was on AML. Limitations in the literature, particularly in regard to exposure assessment and the ability to isolate benzene from concomitant workplace exposures, preclude more definitive conclusions. X-C, Lin X-F, Meng J-F, Zho J-S, Wacholder S, Kneller R, Blot WJ (1994): Cohort study among workers exposed to benzene in China: II. Exposure assessment. Am J Ind Med 26:401411. Flodin U, Fredriksson M, Axelson O, Persson B, Hardell L (1986): Background radiation, electrical work, and some other exposures associated with acute myeloid leukemia in a case-referent study. Arch Environ Health 41:7784. ACKNOWLEDGMENTS We would like to acknowledge the support provided by Dr. Herschel Hobson to initiate this project and the helpful review of the manuscript by Dr. Andrew Olshan. Kurtis Andrews was supported in part by National Institute of Environmental Health Sciences Training Grant T32ES07018. REFERENCES Aksoy M, Erdem S, DinCol G (1974): Leukemia in shoe-workers exposed chronically to benzene. Blood 44:837841. Austin H, Cole P, McCraw DS (1986): A case-control study of leukemia at an oil refinery. J Occup Med 28:11691173. Austin H, Delzell E, Cole P (1980): Benzene and leukemia: A review of the literature and risk assessment. Am J Epidemiol 127:419439. Australian Institute of Petroleum Health Surveillance Program (1992): Health Watch, Ninth Report. Department of Public Health and Community Medicine, University of Melbourne, Melbourne, Australia. Blair A, Linos A, Stewart PA, Burmeister LF, Gibson R, Everett G, Schuman L, Cantor KP (1992): Evaluation of risks for non-Hodgkin's lymphoma by occupation and industry exposures from a case-control study. Am J Ind Med 23:301312. Bond CG, McLaren EA, Baldwin CL, Cook RR (1986): An update of mortality among chemical workers exposed to benzene. Br J Ind Med 43:685691. Checkoway H, Wilcosky T, Wolf P, Tyroler H (1984): An evaluation of the associations of leukemia and rubber industry solvent exposures. Am J Ind Med 5:239249. Chinese Epidemiologic Study Group of Leukemia and Aplastic Anemia (1992): Risk factors analysis of leukemia and aplastic anemia in China (in Chinese). Acta Academiae Medicinae Sinicae 14:185189. Ciccone G, Mirabelli D, Levis A, Gavarotti P, Rege-Cambrin G, Davico L, Vineis P (1993): Myeloid leukemias and myelodysplastic syndromes: Chemical exposure, histologic subtype and cytogenetics in a case-control study. Cancer Genet Cytogenet 68:135139. Crane MM, Keating MJ, Trujillo JM, Labarthe DR, Frankowski RF (1989): Environmental exposures in cytogenetically defined subsets of acute nonlymphocytic leukemia. JAMA 262:634639. Crane MM, Godwin JE, Annegers JF, Keating MJ (1992): Is histological subtype a marker for environmental exposures in acute myelogenous leukemia? Cancer Epidemiol Biomarkers Prev 1:183188. Decoufle P, Blattner WA, Blair A (1983): Mortality among chemical workers exposed to benzene and other agents. Environ Res 30:1625. Delzell E, Austin H, Cole P (1988): Epidemiologic studies of the petroleum industry. Occup Med: State of the Art Reviews 3:455474. Dosemeci M, Hayes, RB, Yin S-N, Linet M, Chow W-H, Wang Y-Z, Jiang Z-L, Dai T-R, Zhang W-U, Chao Z-J, Ye P-Z, Kou Q-R, Fan Y-H, Zhang Gerin M, Siemiatycki J, Kemper H, Begin D (1985): Obtaining occupational exposure histories in epidemiologic case-control studies. J Occup Med 27:420426. Girard R, Revol L (1970): La frequence d'une exposition benzenique au cours des hemopathies graves. Nouvelle Revue Francaise d'Hematologie 10:477484. Goldstein BD (1988): Benzene toxicity. Occup Med: State of the Art Reviews 3:541554. Goldstein BD (1990): Is exposure to benzene a cause of human multiple myeloma? Ann NY Acad Sci 609:225230. Greenland S, Salvan A, Wegman DH, Hallock MF, Smith TJ (1994): A case-control study of cancer mortality at a transformer assembly facility. Int Arch Occup Environ Health 66:4954. Heineman EF, Olsen JH, Pottern LM, Gomez M, Raffn E, Blair A (1992): Occupational risk factors for multiple myeloma among Danish men. Cancer Causes Control 3:555568. Hill AB (1965): The environment and disease: Association or causation? Proc Roy Soc Med 58:295300. Hoar SK, Morrison AS, Cole P, Silverman DT (1980): An occupation and exposure linkage system for the study of occupational carcinogenesis. J Occup Med 22:722726. Hricko A (1994): Rings of controversy around benzene. Environ Health Perspect 102:276281. Hurley JF, Cherrie JW, MacLaren W (1991): Exposure to benzene and mortality from leukaemia: Results from coke oven and other coal product workers. (Letter). Br J Ind Med 48:502504. Ishimaru T, Okada H, Tomiyasu T, Tsuchimoto T, Hoshino T, Ichimaru M (1971): Occupational factors in the epidemiology of leukemia in Hiroshima and Nagasaki. Am J Epidemiol 93:157165. International Agency for Research Cancer (1982): ``IARC Monographs on the Evaluation of the Carcinogenic Risk of Chemicals to Humans. Some Industrial Chemicals and Dyestuffs.'' World Health Organization International Agency for Research on Cancer, Lyon, France, 94148. Linet MS, Stewart WF, Van Natta ML, McCaffrey LD, Szklo M (1987a): Comparison of methods for determining occupational exposure in a case-control interview study of chronic lymphocytic leukemia. J Occup Med 29:136141. Linet MS, Harlow SD, McLaughlin JK (1987b): A case-control study of multiple myeloma in whites: Chronic antigenic stimulation, occupation, and drug use. Cancer Res 47:29782981. Linos A, Kyle RA, O'Fallon WM, Kurland LT (1980): A case-control study of occupational exposures and leukaemia. Int J Epidemiol 9:131135. Malone KE, Koepsell TD, Daling JR, Weiss NS, Morris PD, Taylor JW, Swanson GM, Lyon JL (1989): Chronic lymphocytic leukemia in relation to chemical exposures. Am J Epidemiol 130:11521158. McMichael AJ, Spirtas R, Gamble JF, Tousey PM (1976): Mortality among rubber workers: Relationship to specific jobs. J Occup Med 18:178185. Ott MG, Teta MJ, Greenberg HL (1989): Lymphatic and hematopoietic tissue cancer in a chemical manufacturing environment. Am J Ind Med 16:631643. Benzene and Lymphatic/Hematopoietic Cancer 295 Ott, MG, Townsend JC, Fishbeck WA, Langner RA (1978): Mortality among individuals occupationally exposed to benzene. Arch Environ Health 33:310. Paustenbach DJ, Bass RD, Price P (1994): Benzene toxicity and risk assessment, 19721992: Implications for future regulation. Environ Health Perspect Suppl 101(Suppl6):177200. Paxton MB, Chinchill VM, Brett SM, Rodricks JV (1994): Leukemia risk associated with benzene exposure in the Pliofilm cohort: I. Mortality update and exposure distribution. Risk Anal 14:147154. Percy C, Stanek E, Gloeckler L (1981): Accuracy of cancer death certificates and its effect on cancer mortality statistics. Am J Public Health 71:242250. Richardson S, Zittoun R, Bastuji-Garin S, Lasserre V, Guihenneuc C, Cadiou M, Viguie F, Laffont-Faust I (1992): Occupational risk factors for acute leukaemia: A case-control study. Int J Epidemiol 21:10631073. Rinsky RA, Smith AB, Horning R, Filloon TG, Young RJ, Okun AH, Landrigan PJ (1987): Benzene and leukemia: An epidemiologic risk assessment. N Engl J Med 316:10441050. Rothman KJ, Boice JD Jr (1982): ``Epidemiologic Analysis with a Programmable Calculator.'' Epidemiology Resources, Boston, Massachusetts. Rushton L, Alderson MR (1981): A case-control study to investigate the association between exposure to benzene and deaths from leukaemia in oil refinery workers. Br J Cancer 43:7784. Savitz DA, Andrews KW (1996) Risk of myelogenous leukaemia and multiple myeloma in workers exposed to benzene (Letter). Occup Envirn Med 1996;53:357. Savitz DA, Moure R (1984): Cancer risk among oil refinery workers. J Occup Med 26:662670. Scherr PA, Hutchison GB, Neiman RS (1992): Non-Hodgkin's lymphoma and occupational exposure. Cancer Res (Suppl) 52:5503s5509s. Schumacher MC, Delzell E (1988): A death-certificate case-control study of non-Hodgkin's lymphoma and occupation in men in North Carolina. Am J Ind Med 13:317330. Sieber WK Jr, Sundin DS, Frazier TM, Robinson CF (1991): Development, use, and availability of a job exposure matrix based on National Occupational Hazard Survey data. Am J Ind Med 20:163174. Siemiatycki J (1991): ``Risk Factors for Cancer in the Workplace.'' CRC Press, Boca Raton, Florida. Stewart WF, Stewart PA (1994): Occupational case-control studies: I. Collecting information on work histories and work-related exposures. Am J Ind Med 26:297312. Swaen GMH, Meijers JMM (1989): Risk assessment of leukaemia and occupational exposure to benzene. Br J Ind Med 46:826830. Thorpe JJ (1974): Epidemiologic survey of leukemia in persons potentially exposed to benzene. J Occup Med 16:375382. Travis LB, Li C-Y, Zhang Z-N, Li D-G, Yin S-N, Chow W-H, Li G-L, Dosemeci M, Blot W, Fraumeni JF Jr, Hayes RB, Linet MS (1994): Hematopoietic malignancies and related disorders among benzene-exposed workers in China. Leukemia and Lymphoma 14:91102. Tsai SP, Wen CP, Weiss NS, Wong O, McClellan WA, Gibson RL (1983): Retrospective mortality and medical surveillance studies of workers in benzene areas of refineries. J Occup Med 25:685692. US Department of Health and Human Services (1980): ``The International Classification of Diseases,'' Ninth Revision, Clinical Modification, Volume 1. DHHS Publ. No. (PHS) 801260. Vai T, Radice L, Catenacci G, Biscaldi GP, Guercilena S, Pesatori AC, Bertazzi PA (1989): Studio a distanza di 304 casi di sospetta patologia da benzene osservati negli anni 19501971 (in Italian). Med Lav 80:397404. Vianna NJ, Polan A (1979): Lymphomas and occupational benzene exposure. Lancet 1:13941395. Vigliani EC, Saita G (1964): Benzene and leukemia. N Engl J Med 271:872876. Wong O (1987a): An industry wide mortality study of chemical workers occupationally exposed to benzene. I. General Results. Br J Ind Med 44:365381. Wong O (1987b): An industry wide mortality study of chemical workers occupationally exposed to benzene. II. Dose-response analyses. Br J Ind Med 44:382395. Wong O (1995):Risk of acute myeloid leukemia and multiple myeloma in workers exposed to benzene. Occup Environ Med 52:380384. Wu W (1988): Occupational cancer epidemiology in the People's Republic of China. J Occup Med 30:968974. Yin S-N, Li G-L, Tain F-D, Fu Z-I, Jin C, Chen Y-J, Luo S-J, Ye P-Z, Zhang J-Z, Wang G-C, Zhang Z-C, Wu H-N, Zhong Q-C (1989): A retrospective cohort study of leukemia and other cancers in benzene workers. Environ Health Perspect 82:207213. Yin S-N, Linet M, Hayes RB, Li G-L, Dosemeci M, Wang Y-Z, Chow W-H, Jiang Z-L, Wacholder S, Zhang W-U, Dai T-R, Chao X-J, Zhang X-C, Ye P-Z, Kou Q-R, Meng J-F, Zho J-S, Lin X-F, Ding C-Y, Kneller R, Blot WJ (1994): Cohort study among workers exposed to benzene in China: I. General methods and resources. Am J Ind Med 26:383400. Young JL Jr, Percy CL, Asire AJ (eds) (1981): ``Surveillance, Epidemiology, and End Results: Incidence and Mortality Data, 197377.'' US Department of Health and Human Services, NIH Publication No. 81-2330, National Cancer Institute Monograph 57. National Cancer Institute, Bethesda, Maryland.