Estimates of the proportion of bladder cancers attributable to occupation.

VINEIS P, SIMONATO L. Estimates of the proportion of bladder cancers attributable to occupation. Scand J Work Environ Health 12 (1986) 55-60. In order to estimate the proportion of bladder cancers attributable to occupation in different countries, three criteria were defined , each with a different degree of strictness, for the inclusion of job titles or industrial activities in thc estimates. Such criteria wereapplied to the available case-referent studies, the range for the less severe criterion estimates being 0-19 0J0 and that of the strictest being 1-19 010. The variability was greater among the studies than among the criteria of inclusion of occupations; therefore the study design and time- and place-specificity of exposures play an important role in the magnitude of such estimates.

From several case-referent (control) studies undertaken in industrialized countries, it has been estimated that the proportion of bladder cancer cases attributable to occupational exposure in the general male population ranges from 8-10 % (10,25) to 20 % (1, 19). These studies used different designs and different methods in the collection and classification of occupational data; in addition, the criteria used for the estimation of attributable risks were not always fully reported.
The present paper reevaluates the relevant available literature using predefined and standardized criteria for the estimation of attributable risks. Job titles and /or industrial activities to be included in the estimates have been defined a priori on the basis of different (more or less stringent) criteria, which were subsequently applied to each of the available studies in the literature. A single overall estimate would not be informative, since attributable risks in the population are time-and place-specific (5); in particular, they depend on (i) the proportion of people employed in different jobs and (ii) the extent and intensity of the exposure to carcinogens in the relevant jobs .

Attributable risk
The attributable risk in a population (ARp) is the epidemiologic measure representing the proportion of the cases of a given disease attributable to a given exposure in a population; it represents also the proportion of cases preventable by the elimination of that exposure.
In the estimation of attributable risk in a population we have used the formula suggested by Miettinen for case-referent studies (12): ARp = (RR-l)/RR x Pee' where RR = relative risk and Pee = proportion of exposed cases. It should be stressed that, rigorously speaking, the use of such a formulation is adequate for population-based case-referent (control) studies only , in which the cases are representative of all cases occurring in a definite population and the referents are a representative sample of the same population.

Criteria for the inclusion of occupations
In strict terms, the exposures at issue for our purpose should be represented by occupations causally associated with bladder cancer. According to recent reviews (5,11,21) these are occupations in the rubber, dyestuff, and gas industries. However, an attributable risk computed in this way for a population would probably represent an underestimation due to the exclusion of a number of occupational exposures involving a real but not yet clearly recognized risk for cancer of the bladder. We have therefore defined a priori the following three different criteria for the inclusion of occupational categories reported by available investigations: I. Occupational categories showing a statistically significant association in at least one study, with relative risks greater than 1.0 in all the informative studies (ie, those providing absolute numbers of cases and referents for those occupational categories). II. Occupational categories showing a statistically significant association in at least one study with relative risks greater than one in all the informative studies with at least 5 % of the referents exposed. III. Occupational categories showing a statistically significant association in at least two studies with 55 relative risks greater than 1.0 in all the informative studies with at least 5 070 of the referents exposed. Five percent of the referents exposed was introduced as a threshold because it corresponds to an acceptable statistical power (see table 4 and the Discussion).
Rubber, dyestuff, and gas workers were included in each estimate. Jobs and/or industrial activities falling into the criteria were identified from the case-referent studies included in the survey. All of them provided age-adjusted estimates of the relative risks with the exception of Wynder et al (27) and Tola et al (23). Doseresponse relationships and biological plausibility, which would represent other criteria for causality assessment in addition to the consistency and statistical significance of the associations, could not be considered due to the unavailability of information.

Selection of the studies
Only case-referent studies allowed for the estimation of attributable risks in the population. Cohort studies might be used in particular cases, eg, when a crosssectional survey on the prevalence of the exposure at issue is available or the study population is a sample of the general population; however, these conditions do not apply to any study we reviewed. All published case-referent studies reporting information on occupations and absolute numbers of cases and referents were taken into account. The following were subsequently excluded: (i).two papers giving information on a few chemical exposures but not on job titles or industrial acti,vities (24,26); (ii) those reporting data only for a very.limited list of occupations (less than 10) (3,7,9,16,18); (iii) 'those aiming at hypothesis generation, such as routine mortality statistics by oc- Hospital-based study: sex-and age-matched referents; papillomas excluded ; cancers of the respiratory system and the upper alimentary trac t and myocardial infarction excluded from the referents; interview on the whole occupational history; control of confound ing by age (matching) Hospital-based study: cases recruited in 1958-1967,340 surgical and 312 other cancer referents (exclud ing genitourinary) recruited in 1955-1958; interview on the whole occupational history; matching by age, residence, and smoking habits Populat ion-based study: cases randomly selected among all (491) incident cases occurring in 1967-1968 in 87 cities of the Boston area (20-89 years of age); referents a random sample of 20-to 89-year-old residents, matched for sex and year of birth ; 140/470 case and 781500 referent interviews obtained from spouse or next-of-kin (men and women); interview on the whole occupational history; control of confounding by age and smoking habits Population-based study: elig ible cases were all patients with newly diagnosed bladder cancer in three Canadian provinces (1974-1976)(77 % interviewed); referents matched for sex, age, and neighborhood (refusals were 20, 4, and 0 % in the three provinces; refusing referents were substituted); male cases had higher education and income than the referents; interview on the whole occupational history; control of confounding by age (matching) and smoking Mixed design: originally eligible cases were all those (N = 274) reported to the Finnish Cancer Registry in 1975-1976 for five Finnish provinces; postal questionnaires sent to 269 cases and 271 sex-and age-matched hospital referents or their relatives; responses were 80 % (cases) and 81 % (referents); the whole occupational history collected; control of confounding by age (matching) Mixed design : incident and prevalent cases in West Yorkshire; sex-and age-matched hospital referents (25 % arterial disease; 60 % accident, minor or cold surgery; 10 % chest conditions); interview on the whole occupational history; control of confounding by age and health district (matching) Population-based study: all histologically confirmed incident cases (81 % interviewed); referents randomly selected by digit dialing (age 21-64 years) or from lists of the Health Care Admin istration (age;;,: 65 years) (respondents 84-89 % ); Interview on all jobs lasting at least si x months; control of confounding by age, smoking, and other high-risk indust ries Poputetion-besed study: incident cases with histologically confirmed carcinoma (90 % interviewed); selection of referents as in Silverman et al (20) (87 % interv iewed); interv iew on all jobs lasting at least six months; control of confounding by age, smoking, and five other variables Hospital-based study: incident + prevalent cases; referents were 225 persons with benign urological conditions and 371 with mostly benign surgical diseases; refusals were 11 cases and 12 referents ; interview on all jobs lasting at least six months, control of confounding by age and smoking Table 2. Absolut e n um be rs of c as es/re f erents (ag e-adj u sted relative risks in parenthe s es) f o r t he job t it les or industria l ac tivit ies assoc iated w ith at least one s tat istically s ign ificant i ncre as ed relat ive ri sk for bladder cancer in t h e case-referent studies re vi ewed (men o n l y ' Out of 1 24 1 occ upatio ns in 345 cases and 1 241 occupa ti ons in 345 referent s. , Unspecif ied or oth er than dyestu ff and petro leum. " Painters, metal worke rs, plumbers, and barbers/hairdre ssers were included because of several st udies reporting a relative risk of> 1.0, altho ugh none was stati stica lly signif icant ; they are not considered for inclusi on in t able 3. , Expect ed numbers. i Start of employment before 1940 and duration of employment > 10 years. cupatio nal categories (13,1 7). All other published investigations were considered. One rep ort (22) was not included because of difficulties in ident ifying the absolute numbers of exposed cases and referents. The report by Decoufle et al (8) on the Roswell Park Memorial Institu te experience was not used because the absolute number of bladder cancers was not traced in the publication , and therefore the attributable risk could not be estimated for the popul ation . Ho we et al (10) reported only discord ant pairs for each of the occupations listed with a possible underestimation of the attributa ble risk due to the exclusion of concordant pairs from the computation of the prop ort ion of exposed cases. From Cole et al (4) only " ever employed" cases and referents were extracted. Only men have been considered in the review.

Results
Tabl e I repor ts the stud y designs of the case-referent stud ies included in the review. All were based on personal interviews on the whole occup ational history, Table 3. Estimate of th e popula tio n attribut able risk s (ARp) for occu pational exposure. The age-adjusted relative risk s (RR) for each job wi thi n each study are presented with the percentage of exposed cases in parentheses . Wynd  with one exception [postal questionnaire (23)]. Five studie s mention one or more specific classifications used for coding the occupational categories; these were the Registrar's General 's classification 1966 (1), the classification of the United States Bureau of the Census (4,19,20), the 1971 Canadian census classification (10), and the United Nations and the International Labour Office classificat ions (25). Confounding by age was controlled in all the studies, mainly by matching. Confounding by smoki ng was controlled in six studies, but smoking-adjusted estimates were reported only occasionally, when differing from unadjusted values. One study controlled confounding by other high-risk occup ations.
Job titles and indu strial activities associated with a statistically significant relat ive risk grea ter than 1.0 in at least one study are reported in table 2. On the basis of the occupat ions listed , job titles and /or industrial activities were aggregated according to the three criteria (table 3). We obta ined the estimates of attributable risks by applying Miettinen's formula (12) to each job included according to each of the men-tioned criteria and summing up the job-specific att ributable risks for the population within each study. The relative risks and the proportions of exposed cases on which the estimates were based are given. It is possible, however, that the same subject was engaged in more than one occupation at risk; this occurr ence would result in an overestimation of the att ributable risk.
As one would expect, the interstud y variability was quite large, ranging from 0 to 19 070 according to the first criterion and from 1 to 19 0J0 according to the other two.
The intercriterion variability was lower, being large only in the studies by Silverman et al (20) and Shoenberg et al (19), moderate in the studies by Cole et al (4) and Ho we et al (10), and virtually absent in the other five studies.

Discussion
Aggregating occupations in order to estimate the attributa ble risk in different popu lations, we have used three different criteria , ranging fro m a less selective Table 5. Attributable risk in the population for smoking and bladder cancer. Men and population-based case-referent studies only were considered .  Power (1-{3) of the studies to detect a relative risk of 2.0 with ex =0.05 (level of statistical significance). (Po = proportion of exposed referents) occupations could have been overlooked and /or their role underestimated due to the low power of many studies, and (iv) exposure to bladder carcinogens for workers sharing the same occupation is likely to have been unequal. The last possibility is expressed by variations in the estimates of the relative risks in the same occupation among different studies. The third point can be particularly relevant. Table 4, in fact, shows the limitations of many studies with respect to statistical power, expecially for a low proportion of exposed referents (I % or less).
In spite of the described limitations and the fact that each estimate should always be regarded as place-and time-specific, this exercise suggests that the attributable risk within a population can be used as a good, albeit imprecise, epidemiologic measure of the cancer burden which can be prevented by the removal of occupational exposure to carcinogenic factors. In fact, the application of three different criteria for the inclusion of occupations does not result, for most of the studies, in a wide variation in the attributable risk .
There are at least two possible explanations for this stability: (i) the differences between the criteria applied were not large enough to detect wide variations or (ii) the product of the magnitude of the risk by the size of the occupational population exposed tends to be approximately constant within the same study in different occupational groups .
The inclusion of further studies or a duplication of the same analysis with other site-specific data sets are warranted to clarify the issue.
For comparison, one (relative risk greater than 1.0 in all the studies, with at least one statistically significant association) to a more conservative one (at least two studies showing a statistically significant association and a relative risk greater than 1.0 in all the informative studies with at least 5 070 of the referents exposed). The estimates which were generated (table 3) show variability Doth between and within the studies. While for most of the investigations from countries other than the United States, with the exception of that by Howe et al (10), the three criteria did not produce largely different intrastudy estimates (I, 2, 23, 25), the American investigations showed wider fluctuations.
These fluctuations could be ascribed to at least two sources of variation in the case of criterion I, ie, inclusion of occupations shared by a high proportion of the subjects (19,20) or exclusion of an occupation (leather work) characterized by a high relative risk and a high prevalence in one study (4) but with inconsistencies in the reported relative risks (some of which were lower than 1.0) in other investigations.
The case of leather workers, for whom the overall epidemiologic evidence supports a causative role of the occupational exposure for cancer of the urinary tract, is a good example of the limitations of such an exercise. In fact, criterion I would not be met, even if two previously discarded studies reporting a high and statistically significant relative risk were reincluded [Schmautz & Cole (18), relative risk 4.8 for cancer of the pelvis and ureter, and Decoufle (7),6.3 for bladder cancer] because of the conflicting results of different studies, some with relative risks lower than 1.0. Such conflicts are resolved when only studies with at least 5 % of the referents exposed are considered.
We also computed the attributable risk within a population only for those occupations with evidence of carcinogenic risk coming from cohort studies (ie, rubber, dyestuff, and gas workers). For two studies (2,27) the attributable risk could not be computed , and in all the other studies the absolute frequency of each of such exposures was very low or absent. Where it could be computed, the attributable risk based on this criterion ranged from I to 5 %.
The choice of the study population does not appear to have affected grossly the variability of the attributable risk, as both in population-based and hospitalbased studies there was a wide fluctuation (table 3).
In principle, the lowest estimate of the attributable risk within a population should derive from the use of criterion III, which is the most conservative, while criterion I should generate the highest estimate. However , both estimates (the latter to a greater extent than the former) are likely to be affected by variability in the quality of the assessment of occupational histories. Other possible sources of inaccuracy were the following: (i) the same subject might have been engaged in more than one occupation, (ii) for a number of occupations only a few studies were informative, (iii) rare mated by the authors of the available population-based case-referent studies. The lower variability of the estimates among studies may be attributed to the higher and more stable proportion of smokers in different western populations, in comparison with specific occupational exposures.