Associations between several sites of cancer and twelve petroleum-derived liquids. Results from a case-referent study

SIEMIATYCKI Associa tions betweenseveralsites of cancer and twelve petroleum-derived liquids: Results from a case-referent study in Montreal. Scand J Work Environ Health 13(1987) 493-504. A population-based case-referent study providedinformation on the associationsbetweenseveraltypesof cancerand 12petroleum-derived liquids. Allsite-exposure combinationswereinvestigated. The most interestingresultsconcernedthe fol lowingcombinations: leaded gasoline-stomach cancer, aviation gasoline-kidney cancer (and the possible implications of this association for a similar effect of unleaded automoti.ve gasoline), mineral spirits squamous-cell cancerof the lung, dieselfuel-nonadenocarcinoma lungcancer, lubricatingoils-squamous celllungcancer, fluids-bladdercancer, cancer, and lubricating oils-prostate

A large population-based case-referent monitoring study was carried out in Montreal. It focused on occupational exposures as potential risk factors (34,35). About 20 sites of cancer were included in the study. For each patient, information was obtained concerning past exposure to about 300 substances. The overall analytical strategy was to analyze subsets of substances at a time to determine whether there seemed to be any remarkable cancer-exposure associations.
This report examines the associations between the cancers in our study and the following 12 petroleumderived liquids: automotive gasoline, aviation gasoline, mineral spirits, kerosene, jet fuel, diesel fuel, heating oil, cutting fluids, hydraulic fluids, lubricating oils and greases, other mineral oils, and crude oil. These substances include some of the most common exposures in the 20th century industrial environment. Over the years, they have been used in a fairly well circumscribed set of occupations and industries, although their compositions have evolved over time. None of them is a pure compound; rather they are complex mix-tures of tens or hundreds of hydrocarbons of various molecular weight, including both saturated and unsaturated aliphatics and ring compounds.

Subjects and methods
A full description of the fieldwork and analytical methods can be found in another article (35). A brief outline follows.
Each of these groups constituted a case series which was investigated in relation to each of the petroleumderived liquids. For each case series, a reference group was selected from among the other cancer patients interviewed. Thus each subject could serve as a case in one analysis and as a "referent" in others. The criteria for selecting "referents" among the other cancers and the numbers of referents thereby selected for each site have been presented in another article (35). (For example, lung cancer patients were not included as referents in any of the analyses).
The in-depth interview elicited a detailed job history of the subjects and information on potentially confounding covariables. A team of chemists and hygienists examined each completed questionnaire and translated each job into a list of potential exposures (15). They did this on a checklist which explicitly listed some 300 of the most common occupational exposures in Montreal. For each product thought to be present in each job, the chemists noted their confidence that the exposure actually occurred (possible, probable, definite), frequency of exposure during a normal workweek « 5,5-30, and >30 070), and the level of concentration of the agent in the work environment (low, medium, high).
For each subject, the data set comprised semiquantitative information on the degree of exposure and the number of years of exposure to each of several hundred occupational substances. For the purpose of the analyses, two indices of exposure to each substance were computed: one comprising the concentration, frequency, and confidence measures cumulated over the working lifetime (cumulative exposure) and the other dividing the cumulative exposure by duration to derive an average level of exposure.
The analysis was carried out in stages. First a screening analysis based on the Mantel-Haenszel (26) approach estimated the odds ratio (OR) between each petroleum-derived liquid and each type of cancer, stratifying on age, ethnic group, socioeconomic status, smoking, and an index of the overall dirtiness of the subject's jobs (ie, blue collar/white collar).
In fact this screening analysis was repeated four times with two different definitions of "exposed" and with two different definitions of the study populations. The cumulative index of exposure was cut at the median to provide a level of exposure that we call "substantial." The basic analysis was carried out once with exposed status defined by any versus none and then by substantial versus none. One set of basic analyses was carried out among all the subjects interviewed, and another set was carried out only among French Canadians, a relatively homogeneous social and genetic grouping that constituted about 60 0J0 of the study population. Any association that appeared to have an elevated odds ratio in any of the four screening runs was earmarked for in-depth analysis.
In-depth logistic regression analyses First, each association thus selected underwent an analysis to determine which of the hundreds of available covariables might be confounders, and then an analysis to estimate the odds ratio was performed that took these confounders into account. The search for 494 confounders was based on the empirical principle of finding those covariates which, when included as stratification variables, changed the estimate of the diseaseexposure odds ratio by more than 10 0J0. Some established risk factors were included as confounders whether or not they satisfied this criterion. For instance, asbestos, nickel, and chromium were included for any association involving lung cancer.
Using logistic regression methods (2) and including the potential confounders identified, we estimated the disease-exposure odds ratio associated with any level or duration of exposure to the substance, the odds ratio associated with different levels of exposure to the substance, and the odds ratios of subgroups who received their exposure to the substance in different occupations.

Defining the petroleum-derived liquids
The substances selected for analysis in this report are, for the most part, derivatives of petroleum crude oil. They have certain chemical and physical properties in common, and there is considerable overlap in their use patterns. They are all liquid at room temperature. Each substance is a complex mixture with a composition that has varied considerably over time and in different uses. In fact, even within a given era and a given use, the composition of each of these substances can have varied considerably according to such factors as the geographic source of the crude oil, the particular refining process used, and the blending formulation (3). It is conceivable that, within the substance categories we examined, some formulations are dangerous, whereas others are not. Our study could detect risks due to one of the substances only if the most prevalent formulations of the substance were carcinogenic.
Some of the nomenclature used to define these substances is vague and requires clarification. For example, it can be assumed that all those exposed to automotive gasoline in our study population were exposed to leaded gasoline. Those exposed recently may also have had exposure to unleaded gasoline. We made no attempt to identify this subset.
The term "mineral spirits" is used broadly to encompass petroleum-derived solvent mixtures known at various times or in various countries as white spirit, Stoddard solvent, varnish makers' and painters' (VM and P) naphtha, rubber solvent, benzine, and ligroin. Mineral spirits are composed of organic compounds with chain lengths that range from C, to C n (25). These hydrocarbons consist mainly of aliphatics (30-90 0,70), cyclic aliphatics (10-55 0J0), and aromatics (1-20 0J0). In the 1970s, concern over the carcinogenicity of benzene led the major petroleum refineries to offer solvent mixtures containing lower concentrations of aromatics of low molecular weight. The aromatic content of these mixtures is now comprised mainly of alkylbenzenes with higher molecular weights.
The compositions of cutting fluids, hydraulic fluids, and lubricating oils have also evolved over time from Table 1. Percentage of all the 3 726 subjects exposed to each of 12 petroleum-derived liquids according to degree of exposure. being mainly mineral oil-based to having a variety of formulations, some of which contain little or no mineral oil (eg, emulsified cutting fluids). The term "other mineral oil" is used to cover such diverse oils as rolling oils, heat treating oils, textile oil, penetrating oil, die lubricants, protective coating oils, and oils used in printing ink.
Kerosene is a mixture of petroleum hydrocarbons with carbon chain lengths ranging from 9 to 16 atoms per molecule (30). Its most common uses by our study subjects were as a metal-cleaning solvent, as, for example, among forestry workers who used kerosene to clean and lubricate sawing equipment, and as stove oil, eg, among construction workers in winter. Jet fuels, as coded in this study, include both kerosene and "wide cut" fuels.
Exposure to the combustion products of these substances has not been included in this analysis, but it will be covered in a subsequent report. Table I describes the exposure patterns of our entire study population (3 726 subjects) for each of the 12 petroleum liquids. Lubricating oils was by far the most common exposure; 31.9 % of all subjects were considered to have had potential exposure to lubricating oils in at least one of their jobs. Most of these subjects were considered definitely exposed (23.8 1110 of the entire sample). However, only 7.0 % of the entire sample had been exposed at a high frequency (more than 30 % of the day) and only 3.1 % were exposed at a high concentration level (on a relative scale). A large number (14.0 %) had over 20 years' exposure to lubricating oils at one level or another of frequency, concentration, and confidence. In contrast, crude oil was the least common, with a lifetime work prevalence of 0.8 % for any level or length of exposure. Table  2 shows the main occupational groups in which exposure occurred for each substance. Most of these substances occurred in many job classes apart from those shown in the table. In addition the indication that a largest occupational category exposed to kerosene was "forestry workers," the exposure level was much lower among forestry workers than among mechanics and repairmen. b This is the number of persons exposed at any level; N is the denominator for each percentage corresponding t<1 the substance in question.

Results
substance was attributed to some workers bearing a given job title does not imply that all workers with that job title were attributed that exposure. For instance, while many of those exposed to hydraulic fluids were " mechan ics and repairmen " only a frac tion of " mecha nics and repai rmen" were conside red to have been exposed to hyd raulic fluids. As expected, severa l of these substances occ urred in many of the same occupa tions. For each substance coded by our team of chemists, the pr esumed route of exposure was indica ted . For mo st of the substances in this repo rt , expos ure occurre d via bot h respiratory and cutaneous contact. For some of the heavier o nes (lub rica ting oils, hydrau lic fluids, other mineral oils) exposure was more often cuta neous tha n respiratory .

Screening results
Mantel-Hae nszel screen ing analyses were carried out among all the subjects and among the subset of French Canadians. Associations which were elevated among the French Canadians were also elevate d in the whole gro up . We have the refore presented only th e results for all the subjects. We estimated the od ds rat ios between th e 12 petro leum-derived liquids and 20 types of cancer. For the rare types of cancer there was a very low power to detect risks; the confidence intervals around th e odd s ratio estimates were so wide that the findi ngs were not very informative. Tables 3-5 show   Table 3. Odds ratios (OR) between 12 sites of cancer -and exposur e" to automotive gaso line , aviation gasoline, mi neral spirits, and kerosene , on th e basis of Mant el -Haenszel ana lyses wi th five strat ifying varlab tes .s (N = numb er of exposed cases, , OR was sig nifican tly great er than 1.0 at P = 0.05 (one-sided). Not e that a lower lim it of 1.0 fo r the 90 % CI does not necessarily im ply a significantly elevated OR. The lower limit may have been rounded to 1.0. In addi tion the tes t and the con fiden ce in terval computat io ns have been based on separate alogorithms (35). , OR was sig nif ic antly greater than 1.0 at P = 0.05 (one-sided). No te t hat a lower limit of 1.0 .fo r t he 90 % CI does not necessari ly im ply a sig nifi ca ntly elevat ed OR. The lower lim it may have been rounded to 1.0. In add ition the te st and t he conf idence int erval co mput at ion s have been based on sepa rate alogorit hms (35). -OR was significantl y greater than 1.0 at P = 0.05 (one-sided). Not e th at a low er limit of 1.0 fo r the 90 % CI does not necessaril y imply a signific antly elevated OR. The lower limit may have been round ed to 1.0. In addition the test and the co nf idence int erval co mputatio ns have been based on separate alogo rithms (35).
the screening results for the 12 substa nces by the 12 cancer types which had over 150 cases and thus reasonable statistical power. These tabl es are based on any exposure versus no exposure . Twelve associations were significan tly elevated (P < 0.05, one -sided), and all of the 12 were selected for in-depth analysis. We also carried out the corresponding set of screening ana lyses with expos ure defined by substa ntial exposure versus none. Altho ugh we have not shown these results because of space limitations, we did select any association which was significant in these results for in-depth anal ysis. Six more associatio ns were thereby selected [diesel fuel-rectu m cancer, diesel fuel-lung (squamous cell) cancer, heating oil-rect um cancer, cutting f1uidslung (oat cell) cancer, cutti ng flui ds-non-Hodgkin 's lymphoma , ot her mineral oil-bladder cancer]. In addition , among the ra rer cancer types omitted from tables 3-5, there was only one statistically significant association based on five or more exposed cases, that between Hodgkin ' s lymphoma and mineral spirit s exposure . Finally, we also selected for in-dep th analysis three associa tions which were of borderline stati stical significance, which hinted at higher risk among the substantially exposed , and which were of interest be-, cause of other work . Th e cutti ng fluids-bladd er cancer association had been reported in the literature (39). The lung (squam ous cell) cancer-lubricating oils association largely involved an occupation (mechanics and repa irmen) which had been repor ted to be at risk for lung cancer (12). Th e lung (squa mo us cell) cancermineral spirits association was of inte rest because one of the main occupationa l gro ups exposed to mineral spir its (na mely, pai nters) had been reported to be at risk fo r lung cancer (12). A to tal of 22 associations were thereby selected, three with stomac h cancer, th ree with rectal cancer, seven with var ious histological types of lung cancer, three with prostate ca ncer, two each with bladder and kidney cancer , and one each with Hodgkin's and non-Hodgkin's lymphoma.

In-dep th analyses
Each selected association was analyzed with the purpose of obtaini ng odds ratio estimates for various exposure subgro ups , adj usted for all potential confound ers. First a series of analyses was carri ed out to identify a sho rt list of potential con founders. Then, using logistic regression, we estima ted odds ratios for vario us exposure subgro ups . Th e results for all 22 associations are shown in ta ble 6.
We estimated risk associated with any level or duration of exposure, as well as that associated with subgro ups at different levels of expo sure. When an association was based on 20 or more exposed cases, we categorized them into four exposure subgroups, based on the level and duration of exposure. We have called the groups short-low, short-high, long-low, long-high. When ther e were fewer than 20 exposed cases, we categorized them into two exposure subgro ups, based on cumulative exposure (which combines level and duration). These we have called non substantial and substantial.
Co ncerni ng the occupation-specific odds rati os, it should be not ed that "exposed" is defined as exposed to the substa nce of interest in the occupation in question and " unexposed" is defined as not exposed to the substance regardless of whether or not the subject worked in the occupation of interest. For each associat ion, we examined the risk in up to six of the main occupations in which the exposure occurred . However in tab le 6 we present only tho se occupations in which the risk was particularly high. The data-based potential confounders were divided into the following three categories: nonoccupational covariates (eg, beverage consumption, marital status), other occupational exposures than other petroleumderived liquid s, and other petroleum-derived liquid s. Each model was built up gradually in five cumulative steps. First, we estimated the crude od ds ratio . Second, we included the same a priori confounders that were included in the Mantel-Haenszel analyses, though the co ntinuo us variables among them were included as cont inuous variables. Then we included in sequence each of the three aforementioned categories of the a These are the associations which were significant in at least one of the two screening runs and which had at le ast five exposed cases. A few associati ons which were of borderline significance were also selected .
b N = number of ex posed cases.
c The odds ratio (OR,) est imates for each associat ion are based on a logistic regre ssion model including the following five a priori co var-1at85: age, ethn ic group. socioeconom ic stat us, smokin g, job "dirtiness." • The odds ratio (OR, ) estimal es for each as soc iati on are based on a logi stic regression model including all potenti al confound ers identified in th e con founder search ing proc edu re descri bed in the Meth ods section , exce pt for ot her petroleum-derived liquids examined in this paper.
For each subs tance, it is implici tly assumed that unexposed men have an OR of 1.0. Fo r each asso ciat ion , t he ref erents cons is ted of ot her types of can c er as indic ated in refe rence 35. e 90 % confidence interval for OA 2 _ I Exposure level was divided into two or four categ orie s depending on whet her Iher e were fewer or mo re th an 30 ex posed c ases . When Ih e tot al number wa s less than 30, the level of exposure was defined by the cumulative expo sure index used in th e Mantel -Haenszel analys es and dichotomized at t he same point along the scale of that inde x to provide two categories which have been called nonsubstanti al and subst ant ial. When the total number exceeded 30, fo ur categories were def ined by dichotom izing the duration of exposure and di chotomi zing the degree of exposure. Short -and long-term exposure was defin ed by a 10-yearcut point. Low and high exposure was defined by the medi an of the distribut ion of values of the index comprisi ng the foll owing characferistics of the exposure : conc entration, frequency , and the chemists ' confidence that tne exposure occurred. See reference 35 for a descriptio n of index E, used in the two-ca tego ry def inition , and index A, used in the fou r-category defin ition . • Occupati on refe rs to the occ upatio n in which t he subject was expose d to t he substance in quest ion . Seiected for presentat ion were one or two occupati ons in Which the odd s ratio diff ered f rom t he overall odds ratio . If the od ds ratios we re consta nt ac ros s the main occu pati on s in which the substa nce was found , no occ upation s were lis ted. Each man was cla ss ified into only one occupation cafegory; if he was expose d 10 the substance in two different occupation s, the job of long est dur ation was used. Thu s the sum of the numbers across occupations is the same as th at across exposu re levels , and it equals the tot al number of men with thi s site of cancer and exposure to this substa nce. Note th at the odds ratios correspondin g to each occupati on refer to the risk for thos e men in the occupation who have been exposed to the substance of interest , not for all men in the occupati on . The level of exposure was ignored in these occupat io n-substance analyses . data-based confounders. It was not clear-cut which of these steps provided the most "valid" odd s rat io estimate (5). Space limitation mitigates again st presenting all five, and in any event the variation in the odds ratio estimates across steps was generally minor, especially acro ss the last three steps. We decided to present the estimates from two models, ie, that based on a prio ri confounders only and that based on all variables except other petroleum-derived liquids. If the results from the full model differed from the latter, it is mentioned in the text and its meaning discussed. For each association there was a distin ct regression model containing from 5 to 25 covariates, depending on which covariables were earmarked in the data-based search for confounders. While there may be so me interest in showing which variables went into the respective models, in fact it is not importa nt for the interpretation of the disease-exposure odd s ratios because we have " controlled " for all variables in our data set, either by con firming in the initial step that their inclusion in the model did not affect the odds ratio estimate or by including them in the regression model. Because it would take co nsiderable space to present them all, we have chosen not to present the covariables included in each model.

Discussion
There were many significant finding s, some undoubtedly by cha nce and some possibly because of real cause-and-effect relations. While acknowledging the possibility of false positive results, we must also note the possibility of false negat ives. As implied by the width of the confidence inter vals in table s 3-5 , the power to dete ct risks was onl y moderate fo r most of the associations ana lyzed . The power may have been further compromised becau se of a misclassification error in the exposure assessment. Furthermore the strategy of employing other cancer patients as referent s for each case series was a "conservative" strategy, possibly leading to some attenuation of risk estimates . Finally the inclusion of data-based confounders in the models may also be a conservative strategy . On the one hand , including more variables than is strictly necessary increases the variability of estimates. On the other, it may also introduce some overadjustment.
We recognize the arbitrary and limited nature of statistical significa nce as a criterion for selecting associatio ns for in-depth anal ysis. Neverthele ss, for our purpose, we believe it is as useful an arb itrary criterion as any other. The main objective of the in-depth anal ysis was to try to separate the false positives from the true positives . In this process we used criteria such as the sta bility of statistical significance once th e confounders were included in the model, the strength of associat ion, dose-response, and coherenc e with experimental or other epidemiologic information. Unfortunatel y, there have been very few oth er epidemiologic studies bearing directly on the carcinogenicity of any of these sub stances. The available evidence, such as it is, derives indirectly from stud ies of occupational or indu strial groups who may have been exposed to the substance in question . Most of these studies were based on the occupations mentioned on the death certificates or tumor registers. Such evidence suffers from several deficiencies -notably, the poor validity of the occupational information, its que stionable appropriateness as a surrogate for specific sub stances, the que st ion able valid ity of the attributed cause of death as an indi cator of cancer incidence, and the lack of information on potential confounding factors (34).
When odd s ratios are discussed without qualification , it is assumed that we are referring to the more fully adj usted odds rat io in table 6, namel y, that designated as OR 2 •

Automotive and aviation gasoline
We found an association between automotive gasoline expo sure and sto mach cancer. The odd s ratio was 2.0 among exposed mech anic s and repairmen, compared to 1.5 among other exposed workers. Although only a minority of mechanics were attributed exposure to automoti ve gasoline , th ey were often expo sed at relatively high levels becau se they used it regularly as a degreasing agent. We carried out an analysis of selected job titl es in our data set and found that all mechanics had an odds ratio of 1.3 for stomach cancer, in contrast with the odds ratio of 2.0 for the subgro up exposed to automotive gasoline. The stom ach cancerautomotive gasoline association was significant with eviden ce of a dose-respon se relationship . Even when other petroleum liquid s (kerosene , hydraul ic fluid s) were added to the model, the odds ratio s for automoti ve gasoline did not decrease.
There ha ve been no other epidemiologic studies directly providing eviden ce on stomach ca ncer risks from automotive gasoline exposure. Leaded automotive gasoline has usually been formulated with two lead-scavenger additives , dichloroethane (ethylene dichloride) and dibromoethane (ethylene dibromide). Both compounds are mutagens and induce tumors of the for estomach in rat s a nd mice aft er oral administration (31). Ethylene d ichloride has produced tumors in a spectru m of other organ s, while topic al administration of ethylene dibromide to mice ha s resulted in skin tumors (31). Both compounds can be absorbed percutaneously. The stomach cancer-automotive gasoline association is compatible with the carcinogenic action of one or both of these additives, po ssibly aided by percutaneous absorption.
Insofar as there has been suspicion of human carcinogenicity from exposure to automotive gasoline, the evidence has pointed to a risk for kidney cancer. While the epidemiologic evidence for kidney cancer risk has been inconclusive (11, 14,27 ,32,42) , expo sure to unleaded automotive gasoline ha s been associated with renal cancer in rats (24). Although the slight excess risk of kidney cancer among workers exposed to automotive gasoline (OR = 1.2) was not statistically significant, we did find a significant cluster of seven kidney cancer patients with exposure to aviation gasoline. Six of them were also exposed to jet fuel, which was also associated with kidney cancer. Because of the high correlation between aviation gasoline and jet fuel, it was difficult to disentangle the effe cts on kidney cancer. When both were included in a regression model, the odds ratio for jet fuel decrea sed somewhat, while that for aviation gasoline was una ffected. Thi s result hint s at a greater role for aviation gasoline than for jet fuel. Aviation gasoline differs in composition from leaded automotive gasoline by its high content of alk ylate naphthas, constituted mainly of branched alkanes (9). These compounds arc strongly suspected of being responsible for the animal nephrotoxicity of various petroleum products (17). It has furthermore been hypothesized th at there may be a causal relation between nephrotoxic changes and the appearance of renal neoplasms in rats exposed to unleaded automotive gasoline (24) . It is thus tempting to hypothesize a link between exposure to highly branched alkanes and renal cancer. This link could explain the animal carcinogenicity of unleaded automotive gasoline, which is known to contain about 20 % alkylate stream (24), a nd the present finding of excess kidney canc er due to aviation gasoline, since aviation gasoline also contains a high proportion of alkylate stream (50-70 llJo) (9). Thus our finding of a kidney cancer-aviation gasoline association is important because it suggests a possible association between kidney cancer and unleaded automotive gaso line.

Mineral sp irits
Exposure to mineral spirits was significantly associated with squamous-cell lung cancer, especially amon g tho se with long-hi gh expo sure. The risks were particularly elevated in small clusters of metal machini sts and janitors. Although not shown in table 6, there was also some excess lung cancer among co nstru ction workers exposed to miner al spirits, many of whom were painters (OR = 1.4). In fact the janitors received their expo sur e to mineral spirits through painting act ivities. In the anal yses of job titles, the entire groups of metal machini sts and co nstruction workers had odds ratios of less than 1.0 for squamo us-cell lung cancer . Thi s finding implie s that exposure to mineral spirits is an important factor. Several studies have reported excess lung cancer risk among painters (12,13,28,29,37). In Dubrow & Wegman's (12) synthesis of several large studies involving standardized mortality ratios and proportionate mortality ratios, the aggregate standardized mortality rat io for lung cancer among painters was 140 (12). At least one case-referent study also found a large excess of lung cancer among painters, especially among tho se who did not wear protective equipment (37). The possible etiologic role of mineral spirits ha s not previously been addressed, but seems to us to be reali stic.
showed signs of association with mineral spirits . The evidence in our data was stronger in the case of prostate cancer than for Hodgkin's lymphoma, although the biological plausibility of such an association is not self-evident.

Kerosene and jet fuels
Although the screening analyses turned up an association between kerosene and stomach cancer, the evidence from the in-depth analyses was not persuasive. The association was entirely attributable to a stomach cancer risk among forestry workers. The level of exposure to kerosene among forestry workers was generally low. The odds ratio for stomach cancer among the entire group of forestry workers was as high as that among the subset exposed to kerosene. A significant excess risk for stomach cancer among forestry workers has also been reported in a large proportionate mortality ratio (PMR) study carried out in the state of Washington in the United States (PMR = 115) (29).
The jet fuel-kidney cancer association has already been mentioned. It is interesting to note further that nephrotoxic effects have been reported in male rats exposed to various kinds of military jet fuels, including wide cut and kerosene base types similar to the ones coded in our study (25). Exposure to synthetic jet fuels, associated with the development of kidney cancers in male rats in the same study, was not, however, judged to be present in our study population.
Based on small numbers, a nonsignificant excess of colorectal cancers was found among the workers exposed to jet fuel (table 4). In Washington State, which contains a large aircraft industry, there was a slight excess of colorectal cancer among aircraft mechanics (22 observed, 18 expected) (29).

Diesel fuel and heating oil
Diesel fuel and heating oil are similar in composition. In the screening analyses and in the in-depth logistic regressions, there was a significant association between diesel fuel and squamous-cell lung cancer. Furthermore, although they were not significant, there were excesses both for oat-cell and for "other" lung cancer cell types in relation to diesel fuel. We therefore carried out a separate set of analyses combining all types of lung cancer except adenocarcinomas. The odds ratio corresponding to OR 2 in table 6 for "any" exposure was 1.6 (90 alo confidence interval 1.1-2.4, N = 39). The odds ratios corresponding to the four subcategories short-low, short-high, long-low, and long-high were 1.9, 2.0,1.1, and 2.0, respectively. These results support the hypothesis of a risk for nonadenocarcinoma lung cancer due to diesel fuel. There was also a significant odds ratio between heating fuel and oatcell lung cancer. Becauseof the similarity between heating and diesel fuel, this finding may also be thought to support the lung cancer-diesel fuel hypothesis. AI-though there has been some previous evidence of an association between diesel exhaust and lung cancer (19), there has been little evidence of direct relevance to the liquid itself. Nevertheless exposure to the liquid and to the exhaust products undoubtedly has some elements in common.
The association between diesel fuel and prostate cancer was also significant in the logistic regression, but there was, if anything, an inverse dose-response relationship. Finally both of these substances were associated with rectal cancer. After logistic regression, both associations were reduced, though both maintained a hint of dose-response. The numbers were small and the confidence intervals wide.

Hydraulic fluids
Hydraulic fluids comprise a chemically heterogeneous class of substances with very little epidemiologic or experimental evidence. Automobile mechanics who were exposed to transmission and brake fluids constituted the main group in this category. The level of exposure of most mechanics to these substances was rather low. Only stomach cancer was significantly associated with hydraulic fluids in the Mantel-Haenszel screening runs. While the odds ratio estimate shown in table 6 was significant, it was not persuasive for two reasons . First, the risk -was elevated only for those workers with the lowest exposure. Second, when gasoline exposure was added as a confounder to the regression model, the odds ratios decreased and were no longer significant.

Lubricating oils
Lubricating oils is one of the most commonly used classes of substances in the industrial environment. They are used in all kinds of occupations and industries and would include both used and unused oils. The overall association with prostate cancer which turned up in the screening analyses virtually disappeared in the indepth analysis. There remained excesses of prostate cancer among farmers and mechanics exposed to lubricating oils, who together comprised less than onethird of all workers with exposure to lubricating oils. We carried out analyses within each of these two subgroups to determine if in either case there was a doseresponse relation between level of lubricating oil exposure and prostate cancer risk. There was none. The excess of prostate cancer among farmers is consistent with the results presented in several reports (6,12,33). The reasons remain obscure. It is not clear whether the apparent excess among mechanics and repairmen exposed to lubricating oils reflects a risk from these substances or from some other factor. On the basis of small numbers, Wen et al (43) found a "nonsignificant" standardized mortality ratio of 182 for prostate cancer among oil refinery workers involved in the processing of lubricating oil. Vena et al (38) found no excess prostate cancer in an auto engine and parts manufacturing complex, where again there was presumed exposure to lubricating oils, though this finding too was based on small numbers.
In our data, lubricating oils were also associated with squamous-cell lung cancer. Although the OR z was only 1.3, it was based on large numbers and was of borderline statistical significance. There was no clear dose-response relation. Dubrow & Wegman (12), in their synthesis of national studies involving standardized mortality ratios (SMR) and proportionate mortality ratios, concluded that lung cancer risks are high for mechanics (aggregate SMR = 122). Most workers with this job title are exposed to lubricating oils. There have been a number of historic cohort studies among workers with probable exposure to lubricating oils. Two found proportionate mortality ratios in excess of 100 for lung cancer (16,38), while others did not (4,43). However, none of these studies was large or convincing. In sum, there is some suggestion of an increased risk of both lung and prostate cancer, though the evidence is not persuasive for either site. If there is a risk for lung cancer, it may be limited to squamouscell tumors.

Cutting fluids and other mineral oils
The categories of cutting fluids and other mineral oils have had varying formulations over the years and also of excess risk, -= no evidence of excess risk, --= evidence against the hypothesis of excess risk (eg, inverse dose-response). b Based on the results of the logistic regression for "any" exposure. It takes into account the magnitude of the odds ratio, its statistical significance, and the number on which it is based. c Refers to the trend among subgroups at different levels andlor durations of exposure and to the odds ratio in the highest exposure sub-group_ d Some experimental evidence which supports this association is presented in the text. e Some previous epidemiologic evidence which supports this association is presented in the text. For the bladder-cutting oil association that evidence was considerable; for the other associations, the evidence was indirect and weak.
502 different uses. Both were associated with bladder cancer. Though these associations were of borderline significance, the odds ratios were greatest in the respective long-high exposure categories. The bladder cancer risk was somewhat higher among machinists and plumbers exposed to cutting fluids than among other workers with the same exposure. There were associations between cutting fluids and oat-cell lung cancer and non-Hodgkin's lymphomas which were no longer significant after the adjustment for confounders. An association between "other mineral oils" and "other lung cancer" was barely significant and was based on small numbers.
Aside from the well-documented association of cutting fluids with cancer of the scrotum (41) and skin cancer (21), there has been less conclusive evidence concerning the effects of these substances on the sites of cancer included in our study. Several studies have reported excess bladder cancer risks in occupational groups with presumed exposure to cutting fluids (1, 7,8,18,36,38,40), while only one investigation found no such excess (10). Some studies have reported excess lung cancer (8,38), while others have not (10,22). Some have reported excess stomach cancer risk (10,22), while others have not (38). The studies providing data on stomach and lung cancer were based on small numbers and cannot be considered persuasive. Our findings are clearly negative for stomach cancer and fairly negative for lung cancer, but suggestive for bladder cancer. In the past, some cutting oils have been formulated with aromatic amines, a class of compounds which includes known human bladder carcinogens (20,39). Also, N-nitrosamines, a class of animal carcinogens, have been detected in cutting fluids (23). Given the prior evidence from other epidemiologic studies and the fact that cutting oils have in the past been formulated with known carcinogens, the fact that both cutting fluids and other mineral oils should have turned up in our study as more strongly associated with bladder cancer than with any other cancer lends credibility to these associations.

Crude oil
Crude oil was apparently associated with rectal and squamous-cell lung cancers, but these associations were based on very small numbers. One of the main groups in which this exposure occurred, namely, seamen, would likely have had very different life-styles than the rest of our study population.

General comments
While we presented and discussed the associations under the headings of the various substances, some readers may be interested to see them grouped by cancer site. In addition it is useful to summarize briefly the evidence presented. Table 7 presents an admittedly rough summary of the evidence from our study on each association that was examined in-depth.
Our purpose was to generate hypotheses. In our view the most promising leads to follow-up from our results are the following: (i) the effects of exposure to leaded automotive gasolin e on the occurrence of stomach cancer , (ii) the effect s of exposure to aviation gasoline on the occurrence of kidney ca ncer and the possible implications of this finding for a similar association for unleaded automotive gasoline, (iii) the effects of exposure to mineral spirits on the occurrence of squamouscell cancer of the lung, (iv) the effects of exposure to diesel fuel on the occurrence of nonadenocarcinoma lung cancer , (v) the effects of exposure to lubricating oil on the oc currence of squamous-cell lung ca ncer , (vi) the effects of exposure to cutting oils and other m ineral oils on the occurrence of bladder cancer , (vii) the effects of exposure to mineral spirits and diesel fuel on the occurrence of prostate cancer.
Some of the hypotheses suggested will be followed up in our own data set with additional analyses regarding latency, interactions with smoking and other factors, effect modification, and more complex regression models . Such analyses were beyond the sco pe of thi s initial paper. ment Program, and the National Cancer Institute of Canada. Ms