Application of two job indices for general occupational demands in a pooled analysis of case–control studies on lung cancer

Objectives: We investigated general job demands as a risk factor for lung cancer as well as their role in the association between occupational prestige and lung cancer. Methods: In 13 case–control studies on lung cancer, as part of the international SYNERGY project, we applied indices for physical (PHI) and psychosocial (PSI) job demands – each with four categories (high to low). We estimated odds ratios (OR) and 95% confidence intervals (CI) for lung cancer by unconditional logistic regression, separately for men and women and adjusted for study centre, age, smoking behavior, and former employment in occupations with potential exposure to carcinogens. Further, we investigated, whether higher risks among men with low occupational prestige (Treiman’s Standard International Occupational Prestige Scale) were affected by adjustment for the job indices. Results: In 30 355 men and 7371 women, we found increased risks (OR) for lung cancer with high relative to low job demands in both men [PHI 1.74 (95% CI 1.56–1.93), PSI 1.33 (95% CI 1.17–1.51)] and women [PHI 1.62 (95% CI 1.24–2.11), PSI 1.31 (95% CI 1.09–1.56)]. OR for lung cancer among men with low occupational prestige were slightly reduced when adjusting for PHI [low versus high prestige OR from 1.44 (95% CI 1.32–1.58) to 1.30 (95% CI 1.17–1.45)], but not PSI. Conclusions: Higher physical job demands were associated with increased risks of lung cancer, while associations for higher psychosocial demands were less strong. In contrast to physical demands, psychosocial demands did not contribute to clarify the association of occupational prestige and lung cancer.

Lung cancer risks are largely attributed to tobacco smoking, and occupational exposures to lung carcinogens (1,2). Occupational social prestige and socioeconomic status are also identified as important risk factors, but -apart from supposed residual effects of smoking and exposure to occupational carcinogens -the pathways from occupational social determinants to lung cancer remain uncertain (3)(4)(5)(6). Occupational conditions including psychosocial strain have been associated with elevated lung cancer risk (7) and may help to understand increased risks for occupations with a lower societal standing. Occupational prestige assigns a position in a perceived, hierarchical order of occupations that particularly captures work-and rank-related psychosocial demands. In addition, as an occupational indicator, it reflects material aspects of subject's socioeconomic position (via income) and is directly linked with health outcomes by physical occupational hazards (8).
We extended analyses of the association between occupational prestige and lung cancer, previously identified in the international SYNERGY project (3), to investigate the role of further occupational exposures in this association. To cover a broad range of exposures and with regard to available job histories in SYNERGY, we applied two job-title based indices for general occupational demands (9) that have not yet been applied in the context of lung cancer. One was an index for environmental/physical demands, potentially also indicating effects of occupational carcinogens, and the other an index for psychosocial occupational demands. To our knowledge, to date, psychosocial demands have not been analyzed together with occupational prestige and lung cancer.
Before extending analysis of occupational prestige, we examined if the two occupational indices themselves were associated with lung cancer and thus appropriate for further analysis. This could additionally show if the job-title based indices are suitable for facilitated assessment of work environment risks when detailed occupational exposure information is not available.
Thus, in the first step, we analyzed the association of the two indices for general job demands and lung cancer and, in the second step, the role of these demands in the association of occupational prestige and lung cancer.

Methods
The detailed methodology employed in SYNERGY has been published elsewhere (10). For this analysis of lung cancer and job indices, we included 13 European and Canadian case-control studies with 19 study centres of the SYNERGY dataset. Details and distribution of cases and controls are included in the supplementary material (www.sjweh.fi/article/3967), table S1. After exclusion Hovanec et al of subjects with largely (>50%) missing or invalid occupational histories (N=1236) and missing smoking information (N=25), the dataset included 37 726 men and women (16 909 cases, 20 817 controls). To extend the previous social prestige analysis (3), we adapted inclusion criteria so that prestige analyses were restricted to 12 studies (18 study centres) and male gender (11 420 cases, 14 130 controls).
Job demands were assigned by two indices for general job demands (9). These indices were constructed and validated using German survey data for men and women and contain two/three dimensions of occupational demands: (i) a physical index (PHI) for ergonomic demands and environmental exposures (including acid, dust, fumes, climatic conditions, radiation, environmental tobacco smoke (ETS), dirt, noise, vibrations, low/ glaring light, or need for protective clothing) and (ii) a psychosocial index (PSI) for mental (eg, overload, disruptions, low error tolerance), social (eg, lacking work control, conflicts, lacking support), and temporal (eg, on-call service, excessive working hours, shift work) demands. Originally, both indices may be summarized to an overall index, which we did not apply due to its high correlation with the PHI (Spearman correlation coefficient 0.95). We assigned both indices (range of 1-10 from low to high demands) to the subjects' entire occupational histories and calculated time-weighted average (TWA) scores. TWA-scores were categorized into four categories: low (1, 2), lower middle (3)(4)(5), upper middle (6)(7)(8), and high (9, 10) demands (9). In sensitivity analyses, we recalculated scores disregarding the last ten years before diagnosis/interview to consider cancer latency. In the opposite direction, we used the last job to rather consider job demand effects on tumor promotion or progression.
To estimate lung cancer risks for job-demand indices (PHI, PSI), we calculated odds ratios (OR) with 95% confidence intervals (CI) by unconditional multiple logistic regression in a pooled analysis of all studies. We first adjusted for age (ln(age)) and study centre, then added smoking habits (smoking status [never (<1 packyear in lifetime), former, current (including quitting smoking before <2 years), and other type of tobacco, including subdivision of former smokers by time since quitting smoking (2-7, 8-15, 16-25, >25 years)] and cigarette pack-years [ln(pack-years + 1)], and finally added ever employment in occupations and industries known to be associated with lung cancer with potential exposure to carcinogens ('list A' occupations) (12, 13) (final model). OR were estimated separately for main histological lung cancer subtypes [squamous cell carcinoma (SQCC), small cell lung cancer (SCLC), adenocarcinoma (ADC)]. In addition, job-demand indices were included as continuous variables to test for linear trends. To consider effects of individual studies, we compared results from the pooled analyses with metaanalyses (random-effects model) using the Paule-Mandel heterogeneity variance estimator (14) and displayed heterogeneity by I 2 .
For the prestige analysis, we adopted TWA prestige scores of Treiman's Standard International Occupational Prestige Scale (SIOPS) (15), based on subject's occupational history, and categorized it into low, medium, and high TWA prestige (3). We repeated models according to the original publication, adjusting for factors mentioned above (final model), education (<6, 6-9, 10-13, >13 years), and additionally the respective job index.
All calculations were performed with SAS, version 9.4 (SAS Institute Inc, Cary, NC, USA).

Results
Descriptive information on the study population is shown in table 1. Both indices revealed higher job demands for cases than controls, with less pronounced differences for women and psychosocial exposures. TWA prestige was lower among cases.
In regression analysis (

Discussion
In our analysis of lung cancer and job-demand indices in men and women, we found elevated lung cancer risks in particular for high physical job demands and less strong associations for psychosocial job demands. Adjustment for PHI reduced lung cancer risks of men with low occupational prestige but adjustment for PSI did not influence results.
We made use of the large SYNERGY database with its detailed smoking information and occupational histories. Previous SYNERGY analyses have identified possible residual effects of smoking due to potential information bias, lacking data on ETS, and possibly the inclusion of occasional smokers among non-smokers (defined by <1 cigarette pack-year) (3,4). Similarly, we confirmed higher risks for higher job demands in the subtypes of lung cancer that are particularly related to smoking (SQCC, SCLC) and decreased risks for ADC (3,4,16). A potential limitation lies in the German database of the job indices, which we applied to international data. However, these data were all from (post-) industrial countries (Europe and Canada), and results of the random-effects model, considering study-specific variances, were similar to pooled estimates.
The applied job indices were constructed to allow assignment of general occupational demands on the basis of occupational job codes in the absence of more detailed information (9), which are included in SYN-ERGY for selected occupational carcinogens. We considered occupational lung carcinogens in general by ever exposure in 'list A' industries and occupations, a simplified exposure assessment. Occupational carcinogens therefore may also mainly account for the elevated risks for higher physical job index, ie, manual jobs, which may also include exposure to occupational fumes, dusts, and ETS. The reduction of risks of lower prestige occupations by adjustment for PHI might account for these previously uncaptured exposures to occupational carcinogens. Therefore, the physical index appears as crude but easily applicable proxy for occupational lung cancer hazards when only job titles were solicited.
Associations with lung cancer were lower for psychosocial compared to physical job demands. However, the PSI includes indicators for potential (lung) cancer

Hovanec et al
risk factors, in particular chronic stress. Our results were similar to one study on lung cancer and workrelated stress among men (7), while other studies did not find significantly increased risks (17,18). We found an overall pattern of higher lung cancer risks for men, increasing with job demands, but no increase of risks for women with moderate psychosocial demands. The reasons for this finding remain unclear also because the job indices were constructed for men and women.
Generally, methodological issues in the assignment of job demands are critical in occupational cancer risk estimation as shown for two analyses of oesophageal cancer and psychosocial exposures (19,20): one of which used personal questionnaires on job strain exposure and did not find an association for higher job strain (19), whereas in contrast, increased risks were detected when deducing job strain from job titles (20). However, in comparison to physical demands, derivation of psychosocial dimensions by objective job titles may be limited and dependent more on individual characteristics  1.66-1.99) 1.47 (1.32-1.63) 1.43 (1.29-1.59) 1358 1611 1.30 (1.08-1.58) 1.34 (1.08-1.66) 1.32 (1.06-1.64 -years+1)). c Adjusted for ln(age), study centre, smoking status including time since quitting (non-smoker, quitted 2-7, 8-15, 16-25, >26 years before interview/diagnosis, current smoker, other types of tobacco only) and cigarette pack-years (ln(pack-years+1)) and ever employment in occupations and industries with potential exposure to carcinogens. d P for linear trend <0.001. e P for linear trend <0.05. (9). This could explain why the observed associations were lower compared to the physical demands. This limitation has to be considered particularly for our analysis of occupational prestige and lung cancer, ie, we could have missed possible effects by adjusting for psychosocial job demands due to insufficient capture of these demands by job titles.

Concluding remarks
The job-title based indices suggested a role of occupational demands for lung cancer, beyond exposure to known occupational carcinogens, and their application in understanding work environment risks in the absence of detailed quantitative occupational exposure information. Lung cancer risks were particularly increased for higher physical job demands, likely due to capturing undetermined effects of occupational lung carcinogens. The index for psychosocial demands was less clearly associated with lung cancer, and -in contrast to physical demands -did not contribute to clarify the association of occupational prestige and lung cancer.