Risk factors for acute myeloid leukemia and multiple myeloma: a combination of GIS and case-control studies. (Features).

Abstract

Risk factors for acute myeloid leukemia (AML) and multiple myeloma (MM) include exposure to toxic chemicals present in tobacco smoke, as well as to emissions from industrial operations and petroleum refinery waste dumps. The study reported here identified these risk Factors among case patients and control patients in Orange County, California, from 1984 to 1993 and determined the significance of the risk factors in the study population.

A case-control study was performed for 604 cases of AML and 643 cases of MM; there were 7,112 control subjects who had colon cancer. The model included the variables smoking history, occupational history, and residence in a census tract with a petroleum refinery waste dump. A geographic information system (GIS) analysis also was performed to correlate the incidence of AML and MM with proximity to the six dump sites that received large amounts of petroleum refinery waste.

Current smokers were found to be at an increased risk of AML with an odds ratio of 2.0. Laborer/equipment cleaners and transportation workers/movers were at risk of AML with odds ratios of 3.5 and 2.4, respectively. Construction/resource extraction workers were at risk of MM with an odds ratio of 2.8. GIS analysis determined that the risk for MM was 1.6 cases per mile for 10 or more years of residence near a large chemical dump.

The authors were able to identify census tracts with a high incidence of AML and MM, and to perform distance analysis using a statistical measure of spatial randomness. The case-control study identified occupational and lifestyle risk factors for AML and MM that were not apparent from census-tract-level data.

Introduction

Environmental epidemiology attempts to characterize the health consequences of exposure to elements that occur in the environment. The fact that exposures can occur at low levels and over long time periods makes it very difficult to establish an association between exposure and disease, let alone to prove causality. Yet these relatively small exposures may be responsible for large attributable risks in the general population, and the health consequences may be chronic diseases with long latency.

The purpose of this study was to determine risk factors for acute myeloid leukemia (AML) and multiple myeloma (MM) among a population of affected patients and control patients in Orange County California. It has been found that exposure to petroleum hydrocarbons over extended periods of time may result in hematopoietic cancers such as AML and MM (Infante, Schwartz, & Cahill, 1990). Low-level exposures to these toxic chemicals can come from residential or occupational sources (Hunting et al., 1995; Jakobsen, Ahlbom, Bellander, & Lundberg, 1993; Lindquist, Nilsson, Eklund, & Gahrton, 1991; Nilsson, Norlinder, Horte, & Jarvholm, 1998; Rushton, 1993) as well as from lifestyle factors such as smoking (Austin & Cole, 1986; Brownson, Novotny, & Perry, 1993; Mills, Newell, Beesin, Fraser, & Phillips, 1990; Paffenbarger, Wing, & Hyde, 1978; Severson, 1987; Williams & Horm, 1977).

This study relied on two databases available in Orange County, California. The population-based Cancer Surveillance Program of Orange County (CSPOC) maintains medical histories of cancer patients, occupational histories, records of patient tobacco and alcohol use, and family cancer histories. The Orange County Environmental Health Division maintains records of the fate and transport of petroleum hydrocarbons from leaking underground storage tanks, former refinery waste landfills, and contaminated industrial sites.

The authors combined case-control study design with geographic information system (GIS) mapping to focus on both individual and census-tract levels of analysis. The study was able to identify areas of high incidence with dump sites; to combine distance analysis with a statistical analysis of spatial randomness; and to identify occupational and lifestyle risk factors not apparent from census-tract data.

Methods

Cancer Surveillance Program Database

CSPOC maintains a registry of all cancer cases diagnosed in Orange County from 1984 to the present (Anton-Culver, Culver, Kurosaki, Osann, & Lee, 1988). This population-based registry is mandated by law to receive the case histories of all patients diagnosed with any type of cancer from hospitals, medical groups, and private practitioners. CSPOC receives the case histories in monthly data transmittals from the State of California Department of Health Services in Sacramento. After a random sampling is made of the case histories for quality control inspection, they are entered into the Cancer Registry Information System (CRIS) database. For every cancer diagnosis, a case history is entered into the database approximately six months after the diagnosis is made.

Selection of Cases and Controls for the Case-Control Study

The authors designed a case-control study to provide information at the individual level of analysis. Cases and controls were indexed in the CRIS database by their Surveillance Epidemiology End Results (SEER) codes. The SEER code for AML (35021) included International Classification of Diseases-Oncology (ICD-O) disease classification 205.0 and histology classifications 9861 and 9867. MM cases were indexed as SEER code 34000, which corresponded to ICD-O code 203 and included histology classifications 9730 and 9731 (all sites). The control patients with colon cancer were indexed as SEER codes 21041 through 21049, which corresponded to ICD-O codes 530 through 539 and 590. Case histories were available for the period 1984 to 1993. All of the case patients were residents of Orange County at the time of diagnosis. Patients of all ages were included in the study.

The control patients with colon cancer were selected from the CRIS database because occupational and lifestyle data, comparable to the data for AML and MM cases, were available and because petroleum hydrocarbon exposure has not been shown to be a risk factor for colon cancer. Exposure status was assessed from medical charts and by residential proximity to known point sources of exposure at an aggregate level of analysis.

The risk factors from case histories were coded by one trained, professional nosologist to ensure uniformity. Patient demographics such as age, sex, ethnicity, and census tract of residence were automatically entered into the registry when the record was transmitted from the state. Because of CSPOC contractual requirements, exact addresses of the case patients and the control patients were not transmitted with the patient records and were not, therefore, available for this study. The nosologist was required, however, to read each case history abstract in order to code family health history, smoking habits, alcohol habits, occupation, and industry. Abstracts with ambiguous phrasing such as "none," "not to excess," or "not heavy" were omitted from the study to prevent misclassification of exposure. Patient occupation and industry histories were coded according to the 1980 census coding scheme. The authors had complete occupational histories for 58 percent of AML cases, 40 percent of MM patients, and 59 percent of patients with colon cancer. Complete histories regarding industries in which the patients were employed were available for 56 percent of the AML patients, 30 percent of the MM patients, and 58 percent of the control patients.

Statistical Analysis in the Case-Control Study

The case-control study used both univariate and multivariate unconditional logistic regression. A variable for age was included in the model, coded from 1 to 18 to represent five-year age blocks from age zero to age 85 and over. Ethnicity was incorporated into three variables: Hispanic, non-Hispanic white, and non-Hispanic black. Regression analysis was performed for the ethnicity variables against the reference group "other" ethnicities. Smoking status was addressed in a variable for current smokers and a variable for former smokers, with nonsmokers as the reference group. Nine categories of hazardous occupations were selected from the 1980 census coding scheme and included as variables in the model. Likewise, six industrial categories were included in the model on the basis of the census occupation and industry cross-reference procedure. The logistic regression was performed on a VAX/VMS minicomputer with the SASS System for Elementary Statistical Analysis (SASS software, version 6.12).

Selection of Parameters for the GIS Study

The study used data that were available to the general public on the location of possible large point sources of exposure. The Orange County Environmental Health Division of the Health Care Agency had maintained a database of all closed, inactive, or abandoned dump sites in the county. The database consisted of a spreadsheet listing 144 sites throughout the county that had been reported to accept waste or trash. Of these 144 sites, only six locations were reported to have accepted petroleum refinery waste. The other 138 sites had not received petroleum refinery waste at any time in the past. The study included only the six dump sites that were reported to have accepted petroleum refinery waste at some time in the past. The amount of waste received by any of these sites is unknown. The authors, using Atlas Select GIS software (Environmental Systems Research Institute; Redlands, California), determined the distance from every Orange County census tract centroid to each dump site.

Additional demographic data for the census tracts were available in the Atlas database. This database contained the entire 1990 census long form sorted into spreadsheet format by census tract. The spreadsheet summarized, in five-year groups, the numbers of individuals who reported various initial years of residence in the census tract. Thus, the total number of people residing in the census tract was stratified into five-year time blocks beginning in 1959 and ending in 1990. These data were used to weight the distances from the census tract centroids to the dump sites. The dependent variable was assigned as the number of AML or MM cases in the census tract and was obtained directly from the General Statistical File of the CSPOC database. Regression analysis was performed for the number of cases of AML and MM per census tract against the distance to each dump; the authors used the Poisson regression module of the Stata statistical analysis system (Manchester Information and Associated Services, Manchester, En gland). The likelihood ratio test was computed against a Poisson-model Chi squared. The regression model was modified to accept a negative binomial error term to detect deviation from a purely random spatial distribution of cases.

Results

Case-Control Study

A total of 604 AML patients and 643 MM patients were included in this study, along with 7,112 colon cancer control patients who had colon cancer. The authors were able to abstract smoking history for 29 percent and 24 percent of AML and MM patients, respectively, and, by contrast, from 46 percent of the control patients. The demographic characteristics of case patients and control patients are presented in Table 1. Control subjects were more likely to be nonsmokers. In many cases, however, it was not possible to reconstruct a smoking history because of inaccurate descriptions in the case abstracts. The term "none" was used to describe the smoking history of many subjects. This term did not differentiate between subjects who had no case history of smoking and subjects who had never smoked.

Employment in hazardous occupations (e.g., laboratory science, laboratory-technician jobs, agriculture/forestry/fishing, mechanics/repair, construction/resource extraction, precision production, machine operation, transportation work, and laboring) and industries (e.g., resource extraction, construction, manufacturing, transportation, and repair) was found to be higher among the AML and MM cases than among the controls. Only 11.5 percent of the control patients were found in hazardous occupations, compared with 15.5 percent of AML patients and 22.4 percent of MM patients. The largest percentages of both case patients and control patients were employed in the manufacturing industry. The percentage of AML patients employed in manufacturing (13.4 percent) was slightly higher than that of control patients (11.2 percent). The next largest percentage employment by industry for case patients and control patients was in transportation. AML and MM patients had higher levels of employment in the transportation industr y (5.3 percent and 8.3 percent, respectively) than did controls (3.4 percent). The authors tested a total of 21 potential risk factors for AML and MM using logistic regression.

Current smoking status was found to be a significant risk factor for AML in the logistic regression model (Table 2), with an odds ratio of 2.0 (95 percent confidence interval [CI]: 1.4-3.1). The amount and duration of smoking were not found to be significant. Employment as a transportation worker was significant, with an odds ratio of 2.4 (CI: 1.1-5.2). In addition, laborers and equipment cleaners were found to be at increased risk for AML, with an odds ratio of 3.5 (CI: 1.2-10.8). The authors also found construction workers and people employed in natural resource extraction (including petroleum) to be at an increased risk, with an odds ratio of 2.8 (CI: 1.0-7.4).

The logistic regression model determined that Hispanics were at significantly more risk for both AML and MM than for colon cancer. While 9.6 percent of the AML cases and 9.3 percent of the MM cases were reported to be of Hispanic ethnicity only 5.5 percent of the colon cancer controls were labeled as Hispanic. A significant odds ratio of 1.8 (CI: 1.4-2.4) was found for Hispanics and AML, while a significant odds ratio of 1.8 (CI: 1.3-2.4) was found for Hispanics and MM. This finding has been observed in other studies, in which Hispanics in the Los Angeles area were found to have the highest rates of leukemia in the world (Groves, Linet, & Devessa, 1995).

GIS Analysis

To expand the analysis beyond personal risk factors, the authors included geographic location as a potential risk factor. Hazardous waste sites were mapped in relation to case patients and control patients with GIS, and it was found that the distance from the Large Chemical Dump to the census tract of residence was spatially correlated with the number of cases of MM in the census tract. Furthermore, the number of cases in each census tract was weighted by person-years of residence in the at-risk population. This weighting produced a correlation coefficient that can be interpreted in terms of length of residence in the census tract. Residence in an Orange County census tract, therefore, is associated with a risk of 1.6 cases per mile per 10 years of residence in proximity to the Large Chemical Dump.

Figures 1 through 3 show maps of the age-adjusted incidence rates for AML, MM, and colon cancer in Orange County by census tract. There does not appear to be any obvious geographic clustering of high AML incidence rates within Orange County. Census tracts with high AML incidence were, however, seen around the Large Chemical Dump and the Boucher Site. Likewise, Figure 2 does not indicate a cluster of census tracts with high MM incidence rates. The incidence rates of MM do, however, seem higher in the census tracts around the Davis Dump, the Large Chemical Dump, and the Aminoil Site. As can be seen from Figure 3, the incidence rate of colon cancer is elevated in the census tract adjacent to the Aminoil Site. A census tract in the south-central portion of the county (524.04) is noticeable because of an exceptionally high rate of colon cancer (1,548 per 100,000). This census tract is occupied by a large military airbase (exclusive of base housing) and a sparsely populated adjacent community. Six cases of colon c ancer were found among the total population of 14 persons in the five age groups comprising ages 15 to 39. The rate of colon cancer in this census tract is, therefore, inflated because of the age demographics of the residents.

The ratio of observed AML and MM cases to expected cases was plotted according to Poisson regression (Figure 4). The vertical axis represents the ratio of observed cases to the number of cases predicted from the Poisson regression. The horizontal axis represents the number of cases found in an individual census tract. The graph on the left depicts the relationship for AML, the graph on the right the relationship for MM. The peak in the AML graph represents a census tract with nine cases, exceeding the number of cases randomly predicted by 70 times; nine AML patients reside in census tract 995.10. Likewise, the MM graph shows that a census tract with 12 cases, exceeding the number of cases expected in a tract by a factor of 10; census tract 626.23 has 12 patients with MM.

Discussion

Summary of Results

This investigation identified several established risk factors for AML and MM in addition to geographic proximity to toxic waste dumps. At the same time, the spatial analysis determined that residence in a census tract in proximity to a petroleum refinery waste dump constituted a significant risk for MM. It should be noted that this risk was not detected in the case-control study at the individual level of analysis. The sensitivities of the case-control studies and GIS analysis to different measures of risk provide improved reliability over either type of study alone. Thus, the combination of case-control study with GIS analysis should be considered for environmental epidemiology studies in general to improve exposure assessment.

Limitations of the Study

Despite the overall effectiveness of the case-control study, the methodology placed some limitations on the results. Complete smoking and occupational histories for many of the case patients and control patients were not available or obtainable; as can be seen from Table 1, smoking histories were available for only 29 percent of the AML cases and 24 percent of the MM patients. Similarly, in the AML case-control study, occupational histories were available for only 58 percent of patients. In the MM case-control study, only 40 percent of patients had complete occupational histories. Despite this potential source of selection bias, the study was nevertheless able to identify smoking as a risk factor for AML. This result was consistent with published data (Mills et al., 1990).

Exposure misclassification and biased selection of cases and controls are common limitations of case-control studies. These limitations are best addressed by a proper study design. In the present study it was important to maintain the exposure status of case patients and control patients so that each group had an equal probability of exposure. The nosologist who coded exposure variables for statistical analysis worked blinded as to the case or control status of the patient abstract at the time of coding. As a result, exposure misclassification was minimized, since exposure and occupation/industry variables were assigned irrespective of AML, MM, or colon-cancer status.

Recall bias is another common concern in case-control studies. CSPOC expends considerable effort to control recall bias, through educational workshops and ongoing communication with the oncology staff of the local reporting hospitals. Obviously, recall bias can best be minimized through use of trained occupational health nurses to perform the interviews. It was not, however, practical to use occupational health nurses to conduct the interviews in the present study. Nevertheless, the abstracting methods of the individual hospitals that report to CSPOC were carefully scrutinized by in-house CSPOC occupational health nurses and quality control staff. The accuracy of the interview data also can be compared against data obtained from death certificates. For the purpose of case-control studies, however, interview data are more detailed than death certificate data.

To preserve confidentiality, the State of California Department of Health Services places strict limitations on the use of street addresses in cancer research. CSPOC does not, therefore, code the street addresses of cancer patients for statistical use. The addition of geographic longitudes and latitudes to the CRIS database would have been useful in the study reported here. For example, there are several hundred leaking underground gasoline storage tanks in Orange County, as well as dozens of underground storage tanks that contain possibly carcinogenic chemicals. A Poisson regression that used all of these potential point sources would have required high-resolution mapping of their geographic locations in relation to the residences of case and control subjects.

Similarly, the GIS analysis could have been made more robust if the authors had been able to identify distances from the dump sites to the addresses of the AML or MM patients. The use of census-tract centroids as surrogates for the individual case and control addresses imparted a certain amount of variability to the GIS analysis. The variability was larger for census tracts that had greater overall area.

Distances were calculated by Atlas to a resolution of 0.0001 mile. This value represents a distance of about 6 inches. It is doubtful that the accuracy of the digital plots of the census tracts is on this same order of magnitude. The survey and digitizing accuracy of the Atlas census boundaries is not known. Satellite global positioning systems, however, could be used to obtain more accurate boundary and distance measurements. Such an effort was beyond the scope of the current project.

Occupation and Industry Risk Factors

This analysis indicates that employment in a construction or resource extraction occupation is a risk factor for MM, and that employment as a machine operator, laborer, equipment cleaner, or transportation worker is a risk factor for AML. To the knowledge of the authors, these findings have not been previously reported in the literature. An occupational link between AML and MM and reported exposure to gasoline is, however, well established (Hunting et al., 1995; Jakobsen et al., 1993; Lindquist et al., 1991; Nilsson et al., 1998; Rushton, 1993). The occupational category of resource extraction used in this study included mining and petroleum drilling/recovery. Similarly, construction workers, transportation workers, machine operators, laborers, and equipment cleaners may have been required to operate gasoline-fueled equipment as well as to handle petroleum-based cleaning solvents with or without personal protective equipment. The occupations that this study found to be at risk, therefore, shared many potentia l exposure scenarios with individuals directly employed in the distribution of gasoline.

Several cohort follow-up studies involving gasoline-distribution workers, drivers, or mechanics have reported elevated standardized mortality ratios or odds ratios for myeloid leukemias (Hunting et al., 1995; Jakobsen et al., 1993; Lindquist et al., 1991; Nilsson et al., 1998; Rushton, 1993). Lindquist et al. showed that the risk to individuals with five or more years of exposure or more than one year of exposure in the five to 20 years prior to diagnosis was increased fivefold. For truck drivers and rally drivers who had been exposed for less than five years or less than one year in the five to 20 years prior to diagnosis, the risk of myeloid leukemias still was found to be increased threefold (Lindquist et al.). The present study reports similar findings, with the risk of AML increased 2.4-fold for transportation workers, although the transportation industry category in this study did not include service station attendants (the 1980 census occupation coding scheme defines the latter group as workers in the "retail trade"). It would seem possible, therefore, that this study has underreported the incidence of AML and MM for transportation workers in the CRIS database.

Smoking History as a Risk Factor

The authors found an association between smoking and AML. This finding is affirmed by other studies (Austin & Cole, 1986; Browuson et al., 1993; Mills et al., 1990; Severson, 1987; Paffenbarger et al., 1978; Williams & Horm, 1977). Mills et al. determined a relative risk for smokers of 2.9 for myeloid leukemia. In addition, a weak dose-response correlation based on cigarettes consumed per day was noted. A 1978 case-control study conducted in Pennsylvania and Massachusetts reported an odds ratio of 2.4 for myelocytic leukemia, which is comparable to the twofold risk for current smokers and AML found in this study (Brownson et al.). Several studies have reported that cigarette smoking may be a possible source of exposure to inhaled carcinogens, including benzene.

A healthy-worker effect may have been exhibited in an observed lack of smokers and drinkers in occupations that require safety or physical-fitness standards. In a study of petroleum refinery workers, Wong, Harris, & Smith (1993) determined that occupational ,prohibitions against smoking may explain the lack of reported smokers in the potentially exposed occupational groups. The same study concluded that certain hazardous occupations may require a healthier lifestyle because of periodic routine physical examinations. In the study reported here, the healthy-worker effect may account for a lower odds ratio for AML and MM in otherwise potentially exposed occupational groups.

GIS Analysis

The results of the GIS analysis indicate an association between the Large Chemical Dump and the incidence of MM in the surrounding community. It is important to note that, first, the effect represents a deviation from a purely random geographic distribution of cases; second, the effect is positive, because the number of cases increases as one moves closer to the dump site; and, finally, the magnitude of the effect can be interpreted in terms of person-years of residence.

The Poisson distribution has been used to describe the probability of rare events. In cancer epidemiology, the mean of the Poisson distribution often describes the expected number of cases in a population sample. The expected number of cases is equal to the true incidence of cases in the total population times the size of the population sample. To satisfy a Poisson distribution, the dependent variable (number of expected cases) must be a small number in comparison with the population size. Mathematically, if p is the incidence of cases in a total population and n is the number of persons in a random sample of the population, up generally is 5 or less. The Poisson distribution also can be described as a special approximation of the binomial distribution, which, as p becomes larger and it becomes smaller, assumes the shape of a normal distribution. The Poisson distribution is usually however, skewed to the left, indicating a large number of random samples with a low number of cases.

Poisson regression was used in the current study to generate the number of cases expected in each census tract. A model was then constructed based on person-years of residence to place the distances from the census tract centroids into individual cells of a contingency table. Regression of the ratio of observed to expected cases against the distances to the dump sites, weighted by the person-years of residence, produced a multivariate model. The independent variables were the distances to the individual dump sites. The coefficients of the multivariate model represented the risk, in number of cases per mile, associated with residence in proximity to the dump sites.

An absolute measure of risk for MM is given by the Poisson regression, in contrast to the relative-risk measures produced in logistic regression analysis. Because the risk decreases as the distance from the dump site increases, the distance to the dump site can be interpreted as a surrogate for exposure to the toxic petroleum refinery waste that was deposited in the dump. Poisson regression was used for the spatial analysis in this study because the probability of an AML or MM case falling within a specific census tract constitutes that event as a rare one.

The GIS analysis detected an effect that was not apparent in the case-control study The logistic regression employed independent variables for each of the six petroleum refinery waste dumps that were included in the GIS study Table 1 lists the number of case patients and control patients who lived in the same census tracts as the dumps. Similar numbers of case and control patients lived in census tracts where a petroleum refinery waste dump was located (4.9 percent of AML patients, 5.6 percent of MM patients, and 5.0 percent of control patients). The effect of residence in the same census tract as a dump site was not, however, detected in the case-control study It is apparent that the GIS study provided greater power to detect the geographic effect than did the case-control study.

Corresponding Author: Hoda Anton-Culver, Epidemiology Division, Department of Medicine, University of California, Irvine, 224 Irvine Hall, Irvine, CA 92697-7550. Email: <[e-mail]>.

Did you know...

Up to 60,000 deaths each year in the u.s. are caused by outdoor air pollution.

Source: Environmental Health Foundation citing the American Lung Association and the Harvard University Medical School.

REFERENCES

Anton-Culver, H., Culver, D., Kurosaki, T., Osann, K., & Lee, J.B. (1988). Incidence of lung cancer by histological type from a population-based registry Cancer Research, 48(22), 6580-6583.

Austin, H., & Cole, P (1986). Cigarette smoking and leukemia. Journal of Chronic Diseases, 39(6), 417-421.

Brownson, R., Novotny, T., & Perry, M. (1993). Cigarette smoking and adult leukemia. Archives of Internal Medicine, 153(4), 469-475.

Groves, ED., Linet, M.S., & Devessa, 5.5. (1995). Patterns of occurrence of the leukaemias. European Journal of Cancer, 31A(6), 941-949.

Hunting, K., Longbottom, H., Kalavar, S., Stern, F., Schwartz, E., & Welch, L. (1995). Haematopoietic cancer mortality among vehicle mechanics. Occupational and Environmental Medicine, 52(10), 673-678.

Infante, P, Schwartz, E., & Cahill, R. (1990). Benzene in petrol: A continuing hazard (letter). The Lancet, 336(8718), 814-815.

Jakobsen, R., Ahlbom, A., Bellander, T., & Lundberg, I. (1993). Acute myeloid leukemia among petrol station attendants. Archives of Environmental Health, 48(4), 255-259.

Kinlen, L., & Rogot, E. (1988). Leukaemia and smoking habits among United States veterans. British Medical Journal, 297(6649), 657-659.

Lindquist, R., Nilsson, B., Ekiund, G., & Gahrton, G. (1991). Acute leukemia in professional drivers exposed to gasoline and diesel. European Journal of Haematology, 47(2), 98-103.

Mills, P, Newell, G., Beesin, W, Fraser, G., & Phillips, G. (1990). History of cigarette smoking and risk of leukemia and myeloma: Results from the Adventist Health Study Journal of the National Cancer Institute, 82(23), 1832-1836.

Nilsson, R., Nordlinder, R., Horte, L., & Jarvholm, B. (1998). Leukaemia, lymphoma, and multiple myeloma in seamen on tankers. Occupational and Environmental Medicine, 55(8), 517-521.

Paffenbarger, R., Wing, A., & Hyde, R. (1978). Characteristics in youth predictive of adult-onset malignant lymphomas, melanomas, and leukemias. Brief communication. Journal of the National Cancer Institute, 60(1), 89-92.

Rushton, L. (1993). A 39-year follow-up of the U.K. oil refinery and distribution center studies: Results for kidney cancer and leukemia. Environmental Health Perspectives, 101 (Suppl. 6), 77-84.

Severson, R. (1987). Cigarette smoking and leukemia. Cancer; 60(2), 141-144.

Williams, R., & Horm, J. (1977). Association of cancer sites with tobacco and alcohol consumption and socioeconomic status of patients: Interview study from the Third National Cancer Survey. Journal of the National Cancer Institute, 58(3), 525-547.

Wong, O., Harris, F., & Smith, T. (1993). Health effects of gasoline exposure II. Mortality patterns of distribution workers in the United States. Environmental Health Perspectives, 101 (Supp. 6) 63-76.

[Figure 1 omitted]

[Figure 2 omitted]

[Figure 3 omitted]

[Figure 4 omitted]

TABLE 1

Characteristics of AML and MM Cases and Controls: Smoking Status,
Occupation, and Proximity to Dump Sites, 1984-1993

Variable                                          AML Included
                                                  (%)

Total                                             604

Demographics

 Female                                           284 (47.0)
 Median age                                       65 (range: 0--100)
 Hispanic                                         58 (9.6)
 Non-Hispanic white                               500 (82.8)
 Non-Hispanic black                               3 (0.5)
 Other                                            43 (7.1)

Smoking status

 Current smokers                                  63 (35.6)
 Former smokers                                   35 (19.8)
 Nonsmokers                                       79 (44.6))

Hazardous occupations
(1980 Census code) (*)

 Lab scientists (069--083)                        1 (0.3)
 Lab technicians (223-225)                        0 (0)
 Agricultural, forestry, and fishing (473--499)   2 (0.6)
 Mechanics and repairmen (503--549)               6 (1.7)
 Construction and resource extraction (553--617)  6 (1.7)
 Precision production and craftsmen (628--699)    12 (3.4)
 Machine operators (703--799)                     15 (4.3)
 Transportation and movers (803--859)             3 (0.9)
 Laborers and equipment deaners (864--889)        9 (2.6)
 Nonhazardous occupations                         296 (84.6)

Hazardous industries
(1980 Census code) (**)

 Agriculture, forestry, and fishing (010-032)     2 (0.6)
 Resource extraction (040-050)                    0 (0)
 Construction (060)                               10 (3.0)
 Manufacturing (100-392)                          45 (13.4)
 Transportation (400-472)                         18 (5.3)
 Repair services (751-760)                        1 (0.3)
 Nonhazardous industries                          261 (74.6)

Petroleum refinery dump sites (***)

 Aminoil                                          3 (0.5)
 Ascon Landfill                                   5 (0.8)
 Boucher                                          1 (0.2)
 Davis Dump                                       12 (2.0)
 Large Chemical Dump                              7 (1.2)
 McColl                                           1 (0.2)

Variable                                          MM Included
                                                  (%)

Total                                             643

Demographics

 Female                                           298 (46.3)
 Median age                                       70 (range: 16--95)
 Hispanic                                         60 (9.3)
 Non-Hispanic white                               545 (84.8)
 Non-Hispanic black                               6 (0.9)
 Other                                            32 (5.0)

Smoking status

 Current smokers                                  37 (23.6)
 Former smokers                                   43 (27.4)
 Nonsmokers                                       77 (49.0)

Hazardous occupations
(1980 Census code) (*)

 Lab scientists (069--083)                        1 (0.4)
 Lab technicians (223-225)                        0 (0)
 Agricultural, forestry, and fishing (473--499)   5 (2.0)
 Mechanics and repairmen (503--549)               5 (2.0)
 Construction and resource extraction (553--617)  13 (5.1)
 Precision production and craftsmen (628--699)    9 (3.5)
 Machine operators (703--799)                     8 (3.1)
 Transportation and movers (803--859)             10 (3.9)
 Laborers and equipment deaners (864--889)        6 (2.4)
 Nonhazardous occupations                         198 (77.6)

Hazardous industries
(1980 Census code) (**)

 Agriculture, forestry, and fishing (010-032)     4 (2.1)
 Resource extraction (040-050)                    2 (1.0)
 Construction (060)                               16 (8.3)
 Manufacturing (100-392)                          36 (18.7)
 Transportation (400-472)                         16 (8.3)
 Repair services (751-760)                        2 (1.0)
 Nonhazardous industries                          117 (60.6)

Petroleum refinery dump sites (***)

 Aminoil                                          4 (0.6)
 Ascon Landfill                                   6 (0.9)
 Boucher                                          0 (0)
 Davis Dump                                       19 (3.0)
 Large Chemical Dump                              2 (0.3)
 McColl                                           5 (0.8)

Variable                                          Controls
                                                  Included (%)

Total                                             7,107

Demographics

 Female                                           3623 (51.0)
 Median age                                       (72 range: 14--101)
 Hispanic                                         390 (5.5)
 Non-Hispanic white                               6,347 (89.3)
 Non-Hispanic black                               35 (0.5)
 Other                                            335 (4.7)

Smoking status

 Current smokers                                  670 (20.4)
 Former smokers                                   835 (25.4)
 Nonsmokers                                       1,777 (54.2)

Hazardous occupations
(1980 Census code) (*)

 Lab scientists (069--083)                        9 (0.2)
 Lab technicians (223-225)                        5 (0.1)
 Agricultural, forestry, and fishing (473--499)   18 (0.4)
 Mechanics and repairmen (503--549)               43 (1.0)
 Construction and resource extraction (553--617)  95 (2.3)
 Precision production and craftsmen (628--699)    111 (2.6)
 Machine operators (703--799)                     92 (2.2)
 Transportation and movers (803--859)             70 (1.7)
 Laborers and equipment deaners (864--889)        42 (1.0)
 Nonhazardous occupations                         3,731 (88.5)

Hazardous industries
(1980 Census code) (**)

 Agriculture, forestry, and fishing (010-032)     17 (0.4)
 Resource extraction (040-050)                    10 (0.2)
 Construction (060)                               108 (2.6)
 Manufacturing (100-392)                          459 (11.2)
 Transportation (400-472)                         138 (3.4)
 Repair services (751-760)                        24 (0.6)
 Nonhazardous industries                          3341 (81.5)

Petroleum refinery dump sites (***)

 Aminoil                                          54 (0.8)
 Ascon Landfill                                   73 (1.0)
 Boucher                                          11 (0.2)
 Davis Dump                                       154 (2.2)
 Large Chemical Dump                              13 (0.2)
 McColl                                           41 (0.6)

(*)Defined as the Census occupational code developed from the 1980
Standard Occupational Classification (SOC) Manual.

(**)Defined as the Census industrial code developed from the 1972
Standard Industrial Classification (SIC) Manual.

(***)Defined as residence in a Census tract that is the location of the
indicated dump site.
TABLE 2

Risk Factors for Acute Myeloid Leukemia and Multiple Myeloma, Cases and
Controls 1984-1993

AML Risk Factors

Univariate                           Odds Ratio
Analysis                             (95% CI)
Age                                  0.8 (0.8-0.8)
Hispanic(a)                          1.8 (1.4-2.4)
Non-Hispanic white (a)               0.5 (0.4-0.7)
Current smoker(b)                    2.2 (1.6-3.0)
Machine operator (c)                 2.0 (1.2-3.5)
Laborers and equipment cleaners (c)  2.6 (1.3-5.4)

Multivariate
Analysis

Age                                  0.8 (0.8-0.9)
Current smoker (b)                   2.0 (1.3-3.1)
Laborers and equipment cleaners (c)  3.5 (1.2-10.9)
Transportation workers (c)           2.4 (1.1-5.2)

AML Risk Factors                          MM Risk Factors

Univariate                           Univariate
Analysis                             Analysis
Age                                  Age
Hispanic(a)                          Sex
Non-Hispanic white (a)               Hispanic (a)
Current smoker(b)                    Non-Hispanic white (a)
Machine operator (c)
Laborers and equipment cleaners (c)

Multivariate                         Multivariate
Analysis                             Analysis

Age                                  Construction and
Current smoker (b)                   resource extraction (c)
Laborers and equipment cleaners (c)
Transportation workers (c)

AML Risk Factors                     MM Risk Factors

Univariate                           Odds Ratio
Analysis                             (95% CI)
Age                                  0.9 (0.9-1.0)
Hispanic(a)                          0.8 (0.7-1.0)
Non-Hispanic white (a)               1.8 (1.3-2.4)
Current smoker(b)                    0.7 (0.5-0.8)
Machine operator (c)
Laborers and equipment cleaners (c)

Multivariate
Analysis

Age                                  2.8 (1.0-7.4)
Current smoker (b)
Laborers and equipment cleaners (c)
Transportation workers (c)

CI = confidence interval

(a)Reference group is non-Hispanic blacks.

(b)Reference group is nonsmokers.

(c)Reference group is all other occupation