MATERIALS AND METHODS

The base population for this historical cohort study was comprised of all Swedish men and women who: were gainfully employed at the time of the 1970 census, were present in the country during the 1960 census, and were still alive and over age 24 as of 1 January 1971. This encompassed 1 779 646 men and 1 101 669 women, aged 25–64 years at the beginning of the study and subsequently followed up for 19 years until the end of 1989. The total of 31 668 842 person-years among men after 19 years of follow up ended with 5620 newly diagnosed pancreatic cancer cases. Among women, these figures were 20 087 479 person-years and 2657 cases. The resulting standardised rates were 18.4 and 13.8 cases per 100 000 person-years, using the standardised European population as reference, respectively among men and women, while the cumulative risk for ages 25–79 were 1.4% among men and 1.1% among women. After excluding histological pancreatic tumours other than adenocarcinoma of the pancreas, we ended with 4420 cases of pancreatic adenocarcinoma in men and 2143 in women, the case population used for this study.

Information was drawn from two data sets. The first source of data was the Swedish cancer environment register, comprising all cancer cases including information on occupation, residence, and different demographic variables from the 1960 and 1970 censuses.^11,12 This register was used to compute specific rate numerators, with pancreatic cancer defined as any case with code 157 of the International Classification of Diseases (7th revision). The second data source comprised all individuals in the 1970 census, with information on occupation and residence in 1970, occupation in 1960, and, if applicable, date of death. This register was used to calculate specific rate denominators.

During the study period, death was the only event defined as end of follow up. Participants not reported as dead were considered to be alive until end of follow up. This led to a slight overestimation of person-years, since those who emigrated were not withdrawn. Nevertheless, the annual emigration rate among Swedish citizens was very low, approximately 1 per 1000.^13–15

In the 1970 census, occupations were coded according to the Nordic Classification of Occupations.¹² Nearly the same codes were used in the 1960 census. Every occupation is represented by a three-digit number. The first digit refers to one of 10 major occupational sectors (0–9), where higher numbers indicate manual occupations, and lower numbers, occupations requiring a longer education and higher socioeconomic status.

The overall person-time that each person contributed to the study was allocated to the corresponding cells of the variables of stratification. These variables were:

1. Occupation

2. County of residence in 1970

3. Size of town of residence in 1970 (under 2000; 2000–20 000; 20 000–100 000; >100 000 inhabitants)

4. Five year age group, from 25–29 to 75–79

5. Calendar time period (1971–75, 1976–80, 1981–85, and 1985–89).

The exact number of person-years was calculated using Clayton’s algorithm.¹⁶ Age standardised incidence rates per occupation for the whole period were computed using the age distribution of the European population as standard. Standardised incidence ratios (SIR) were calculated for men and women separately in each occupation. SIRs were adjusted by age and period using the whole cohort to provide the reference rates. On the assumption that the observed number of cases was distributed in each stratum as a Poisson variable, we fitted log linear Poisson models, comparing occupations against others in the general cohort, adjusting for geographical risk area and town size. Given that the expected number was computed on the basis of the age and period specific reference rates, relative risk (RR) for each occupation was likewise age and period adjusted.

Counties were grouped into three SIR based categories (<90, 90–110, and >110). To take into account the relation between social class, lifestyle (mainly tobacco consumption), and occupation, the same analysis was carried out for the selected occupations, but this time solely taking other occupations in the same major group (that is, those with the same first digit) as reference.

For occupations that still showed an excess risk of at least 10%, relative risks (RRs) were calculated separately: for subjects reporting such occupations in both censuses; and for those reporting such occupations in 1970 but not in 1960. In each case, the RR was calculated taking the other occupations in the same major group as reference. Furthermore, assuming greater exposure (at least in terms of years) for individuals reporting the same occupation in both censuses, the statistical significance of a gradient of risk was obtained. In order to obtain more valid estimates, RR were adjusted for other important confounders (county and town size), and definition of occupation was further refined by means of additionally computing RRs for those participants reporting the same occupation in both censuses.

RESULTS

In our cohort, pancreatic cancer incidence varied markedly among Swedish counties. The highest SIR for men (1.33) and women (1.37) was registered in Västerbotten North Sweden and the lowest (0.63 in men and 0.62 in women) in Kristianstad South Sweden. The same was true for urban and rural areas, with the incidence rate ratio for towns with more than 100 000 inhabitants versus those with less than 2000 being 1.42 for men and 1.22 for women. These factors were potential confounders in our study, since some occupations are more predominant in urban than in rural areas, and vice versa. In general, RRs from Poisson models taking geographical area and town size into account were lower than the corresponding SIRs. The mentioned attenuation might be due to the confounding effect of these two variables.

RRs for the 10 major occupational sectors (first digit of the occupational codes) showed some differences, ranging from 0.89 among men (0.82 among women) for “administrative and managerial personnel” to 1.19 among “services and military workers” (1.17 in “production workers” among women). When comparing risk estimates as a result of intragroup comparisons (taken as reference subjects with job codes having the same first digit) versus risk estimates taking all the population as reference, attenuation of the risks towards unity was observed for intragroup comparisons in occupational sectors with lower education and socioeconomic status in both sexes. This might be due to a higher prevalence of pancreatic cancer risk factors in these main occupational sectors.

From the 279 different occupations reported among men in 1970, 140 had at least five observed cases and their RR were computed using Poisson models. Among these occupations, 69 showed a standardised incidence ratio greater than 110, of which 53 showed a relative risk for pancreatic cancer equal or higher to 1.1 (table 1). Statistically significant RRs were observed among “technical assistants”, “travelling agents”, “other metal processing workers”, “baker and pastry cook”, “docker and freight handler”, and “waiters”. Relative risks over 1.5 were seen also among “road traffic supervisors”, “gold and silver smiths”, “telephone-telegraph install repair”, “grain mill and oil press workers”, “stone cutters and carvers”, “greasers”, and “other service workers” (table 1). In most cases, when restricting the analyses to those aged under 65, risks estimates remained unchanged (data not shown). Some occupations suspected of increasing pancreatic cancer risk showed slightly increased risks: “motor vehicle drivers”, “blacksmiths”, “watchmakers”, “butcher and meet preparer”, “chemical and laboratory”, “shoe workers”, “rubber product workers”, “cleaners”, and “hairdressers”. No excess of risk was observed among male “farmers”, “gardeners”, or “dry cleaners”.

Women reported 270 different occupations in total. About one third (n = 77) of them showed at least four pancreatic cancer cases and were considered in the following analyses. Thirty two of these occupations had standardised incidence ratio greater than 110, and 26 had a relative risk for pancreatic cancer over 1.1 (table 2). In one third of them, the increased RR reached statistical significance, although most of these were low numbers. These occupations included “educational methods advisors”, “librarian, archivist, curator”, “motor vehicle driver”, “typographer, lithographer”, “purser, steward, stewardess”, “other housekeeping and related workers”, “other electric and electronic workers”, and the groups of occupations of “glass, pottery, and tile workers”. When restricting the analyses to those aged under 65, risks estimates tended to increase slightly (data not shown). Some occupations believed to increase the risk of pancreatic cancer showed little risk: “launderer and dry cleaner” (RR = 1.08, 95% CI: 0.68 to 1.73) and “textile worker” (RR = 1.06, 95% CI: 0.82 to 1.37), or null risk (farming) among women.

For job codes shown as having at least a 10% excess risk in the SIR, tables 3 and 4 set out the RRs for men and women respectively reporting occupations in both censuses (1960 and 1970). Only those occupations with an RR over 1.1 and at least four cases are shown. Both sexes showed excess of pancreatic cancer in occupations entailing food processing, “hunt and livestock”, “metal workers”, and “electronic related workers”. Among men (table 3), most significant associations showed higher estimates except for “waiters”, and in general, RRs tended to increase. When restricting the analyses among those aged under 65 years, RRs changed slightly, although they increased notably among “other metal processing workers” (RR = 4.50, 95% CI: 1.45 to 14.0), “sheet metal workers” (RR = 1.81, 95% CI: 1.12 to 2.93), “frame, circular sawyer, and planer” (RR = 2.37, 95% CI: 0.89 to 6.35), and “rubber products workers” (RR = 2.09, 95% CI: 0.94 to 4.67). Among women, with the exception of shoe workers (RR = 1.66, 95% CI: 0.74 to 5.35), after limiting the analyses to those reporting the same occupation in both censuses (table 4) point estimates of RRs only changed slightly.

DISCUSSION

From the results from this study, occupation seems to play a small role in pancreatic cancer aetiology in the Swedish population. A few occupations showed an increased risk of pancreatic cancer in both men and women, and the associations observed were in accordance with previous studies.

“Gold and silver smiths” among men, and workers on electronics among both sexes showed an increased risk of pancreatic cancer. A possible exposure among those workers could be chlorinated solvents while cleaning the material they work with. A recent meta-analysis studying the association between chlorinated solvents and pancreatic cancer, reported that polychlorinated biphenyls, trichloroethylene, methylene chloride, and vinyl chloride (but not carbon tetrachloride) could be associated with pancreatic cancer.³ The authors of the study also point out a possible interaction between these solvents and other pancreatic carcinogens.³ However, due to the high correlation of exposure among those compounds, the specific responsible agent is difficult to identify. A case-control study in Shanghai reported an increased risk among workers in “electrical, electronics, and related”, and attributed the increase of risk to electric magnetic fields.¹⁷ Electromagnetic fields exposure has been suggested to suppress pineal gland production of melatonin, thereby promoting the occurrence of cancer.¹⁸ Also, it has been hypothesised to alter normal removal of damaged cells; to affect hormone release; and to influence cellular growth and differentiation.¹⁹ Other exposures among these workers include cutting oils and PCBs.

Another occupation showing increased risk among men and women is the group of metal workers. This increase in the risk is consistent with findings of many previous epidemiological studies,^17,20–24 but not all.^1,25 PAHs might explain the risk excess among those workers. The carcinogenic properties of benzo[a]pyrene are widely known; considering its presence in tobacco smoke (the only well established risk factor for pancreatic cancer) we find this excess credible. The review of the epidemiological evidence of cancer in aluminium reduction plant workers by Ronneberg and Langmark suggested an association between pancreatic cancer and potroom workers.²⁶

The increase among stone cutters in men, and glass, pottery, and tile workers among women has been also reported in other studies,¹⁷ including a nationwide case-control study in Finland.^6,27 Inorganic dust might be the most prevalent exposure among those workers.

Similarly to our study, an increased pancreatic cancer risk has been observed among workers who hold sedentary jobs,^4,27,28 a factor which may activate a biological pathway involving obesity.^8,29–32 Lack of physical activity could possibly explain the results for librarian/archivist/curators, teachers, cashiers, working proprietors, managing directors, bank employees, and clerical workers. In the absence of any other exposure these associations might indicate a role for lack of physical activity in pancreatic cancer.³¹

The observed increased risk among printing workers in women was mainly due to typographers and lithographers. Other studies have also reported increased risk of pancreatic cancer among printing workers.^33–36 A study published in the early 1990s reported increased mortality from diabetes mellitus and pancreatic cancer among Swedish pulp and paper mill workers.³⁷ However, other studies in northen European countries did not observe any association with pancreatic cancer among those workers.^38,39

We did not observe an excess risk among female hairdressers,⁴⁰ but we found a slight increase in the risk of pancreatic cancer among male hairdressers. A Finnish cohort of hairdressers showed increased risks of pancreatic cancer among both men and women,⁴¹ as did another case-control study, also in Finland.²⁷ Some of the dyes used by hairdressers are aromatic amines; a number of aromatic amines have proved to be pancreatic carcinogens in animal models, and it has been suggested that they may play a role in human pancreatic cancer.⁴² An excess of pancreatic cancer has also been reported in some studies of exposure to dyes and aniline derivatives.^5,25,43–45

Although some recent publications support the association between pesticide use and pancreatic cancer in different countries,^2,5,46 the lack of association between farming and pancreatic cancer in our study is in accordance with other reports from studies developed in Sweden among male farmers,⁴⁷ female farmers,⁴⁸ and pesticide applicators.⁴⁹ Some explanations include the fact that spray seasons are short and control of toxic pesticides is probably the strictest in the world in Sweden; active ingredients are therefore restricted, exposure levels probably relatively low (except maybe in greenhouses), and cumulated exposures low because of both low levels and low time fractions of exposure. Also, while other studies reported increased risk among textile workers,^17,27,28,50 we did not find indications of such an association among Swedish women. The most recent mortality analysis from a Russian cohort of shoe workers⁵¹ was not able to replicate the initially observed excess of pancreatic cancer.⁵² Other recent European studies have not found increased risk of pancreatic cancer among shoe workers.⁵³ “Waiters and waitresses”, “store and warehouse workers”, “working proprietors”, and “travelling agents” showed increased risk of lung and bladder cancer in our cohort. A higher consumption of cigarette smoking in these groups might explain the excess of observed cases of pancreatic cancer. Other associations putatively related to cigarette smoking would include staff officers and food processing work related occupations. A higher tobacco consumption among Swedish brewery workers was suggested as the explanation for the highest incidence of pancreatic cancer in this occupational sector.⁵⁴ The pattern of RR between size of population and cancer observed in pancreatic cancer can be also observed in lung cancer. The fact of controlling for population size, and using as a reference population other workers within occupational sectors might provide some adjustment for tobacco consumption. The point risk estimate for waiters decreased from 2.36 to 2.08 when workers from the same main occupational sector were used as reference to compute the RRs.

Other limitations of this study include the heterogeneity of exposures within job titles (ensuing uncertainties in attributing chemical agents as explanations); the lack of adjustment from well known risk factors for pancreatic cancer, namely cigarette smoking, diabetes, chronic pancreatitis, and body mass index; and the high number of comparisons run. However, most results are consistent with international literature, thus reducing their likelihood of being chance findings. As the referent population only includes working individuals, it is unlikely to be confounding due to differences in confounding factors between the Swedish working population and the Swedish general population employed, although differences in tobacco consumption across the different occupations are probable. In Sweden the prevalence of smoking among males was around 35% during the 1980s, declining linearly to 18% at the start of the twenty first century, while the corresponding figure among women during the 1980s was around 30%, declining slowly since then to 20% in 2000.⁵⁵ A prospective study reported an RR for pancreatic cancer among female current smokers of 1.7 versus never smokers, and the relations with amount smoked daily and age at starting smoking were not pronounced.⁵⁶ The relatively low prevalence of smoking in the Swedish population during the study follow up, the modest magnitude of association between smoking and pancreatic cancer (versus the association between cigarette smoking and cancers from the upper respiratory tract, lung, or stomach), and the fact that the point estimate for most statistically significant associations is higher than the putative RR for cigarette smoking in the Swedish population somehow limit (but do not completely remove) the potential of having confounded results from tobacco consumption in our study.

Diagnostic misclassification has been shown to seriously bias risk estimates for pancreatic cancer.^57–59 While many other studies included other histology diagnoses, we only considered in our study histologically confirmed pancreatic cancer cases with adenocarcinoma of the pancreas (the most frequent pancreatic malignancy). Even though the definition of the exposure is limited in this population study, the availability of additional information about the 1960 census allowed us to increase the specificity of this definition considering the subcohort of subjects exposed in both censuses. SIRs were computed taking only age and period into account, since the reference rates proved to be unstable when stratification for geographical area and town size was attempted. As an alternative, Poisson regression analysis possesses major advantages over standardisation in controlling for confounding.⁶⁰ However, the general theory of goodness of fit analysis in Poisson regression requires the number of expectations per stratum not to be too small.⁶⁰ In our models, the total number of cases for any occupation was split into the corresponding categories, producing strata with very low figures. Nevertheless, the use of observed and expected values afforded the opportunity of collapsing the different strata (that is, different age groups and periods), thereby increasing their stability and rendering regression analysis more efficient.

Although occupation might play a small role in pancreatic cancer in the Swedish population, a few occupations were found to increase the risk of pancreatic cancer in both men and women, and some occupations entailing exposure to chlorinated hydrocarbons solvents and PAHs deserve further research to clarify the accuracy of and reasons for these findings.