INTEROCC case–control study: lack of association between glioma tumors and occupational exposure to selected combustion products, dusts and other chemical agents

Brain tumors constitute a mixed group of intracranial neoplasms which could be benign or malignant. Gliomas account for almost 80% of primary malignant brain tumours [1]. Of the six major histological subtypes of glioma, glioblastoma constitutes the highest grade and the most common type, and is associated with very poor survival [1, 2]. The incidence of glioma differs according to age, gender, race, ethnicity and country; in many regions slight increases have been reported between the 1970s and 2000, though it is unclear whether this ostensible increase simply reflected diagnostic improvement [3].

Except for ionizing radiation, which is a well-established risk factor for glioma [4, 5], the aetiology of this tumor is largely unknown [4, 6, 7]. Some case–control studies have found a lower risk of glioma among subjects reporting allergies or other atopic conditions [4]. There is inconsistent and largely negative evidence regarding possible associations between smoking, diet or alcohol and gliomas [3, 4]. Concerns have been raised about possible effects of exposure to non-ionizing radiation including radiofrequency (RF) fields due to mobile phones or electromagnetic fields (EMF) in the extremely low frequency range (ELF), but the evidence remains inconsistent and uncertain [8–12].

A few studies have reported on glioma risks in relation to occupations or industries, but there has not been any strong consistent pattern across studies. Slight increased risks have been reported among workers in synthetic rubber manufacturing [13], petrochemical refineries [14] and pulp and paper industries [15], as well as among physicians [14, 16, 17], firefighters [14, 17], farmers [16, 18] and legal and social services workers [14]. More studies have reported on occupational risks for all brain cancers combined [18–20]. There have been reports of increased risk of brain tumors for polyvinyl chloride production workers [21, 22], anatomists, pathologists and embalmers [23–25], painters, machinists, industrial mechanics and plumbers [26, 27]. Few studies have delved into the possible associations between specific occupational exposures in those occupations/industries and the risk of brain cancer. There is a need for studies that go beyond the job or industry title to investigate brain cancer risks in relation to chemical exposures.

The INTEROCC study was built on two important databases: the INTERPHONE study, the largest international collaborative case–control study of brain cancer yet conducted [28], and FINJEM, a database permitting the translation of job histories into occupational exposure histories [29]. The INTERPHONE study involved interviews with over 2,400 meningioma cases, 2,700 glioma cases and over 5,600 controls in regions of 13 countries, with a primary focus on use of mobile phones. A subset of INTERPHONE study investigators were interested in investigating occupational risk factors and had agreed to include in the interview questionnaire a module to elicit a detailed lifetime job history. That module was used in seven of the INTERPHONE countries, and the subjects in those countries comprise the INTEROCC study subjects. No further data collection was required from the study subjects. The second pillar for this project, the FINJEM, allowed us to infer exposure to many occupational agents. Following examination of the agents that were available in FINJEM, and a review of previous epidemiologic evidence concerning occupations and brain cancer risk, we established a list of 28 agents that would be worthwhile to evaluate in relation to brain cancer. These 28 substances comprised solvents, metals, combustion products, dusts and other chemical agents. For most of them there is little epidemiologic evidence, and little or no literature on possible mechanisms of brain cancer carcinogenesis. This should not be an impediment to studies such as the present one, since most known carcinogens were discovered by epidemiologic or clinical observation, before there were evidence-based biological plausibility arguments [30]. The present article focuses on risks of glioma in relation to 10 of the agents, namely combustion products (diesel exhaust emissions, gasoline exhaust emissions, benzo(a)pyrene), dusts (asbestos, crystalline silica, wood dust, animal dust) and certain other chemical agents (formaldehyde, oil mist, sulphur dioxide). Subsequent papers will address risks related to other agents and to meningioma.

Of the eligible subjects who completed the interview, 54 cases and 36 controls were removed from the analyses because it was technically impossible to link part of their occupational history to the INTEROCC JEM. An additional set of 9 cases and 6 controls were removed due to a family history of neurofibromatosis and/or tuberous sclerosis. Following these removals, the study sample consisted of 1,800 glioma cases and 5,160 controls. Response rates for cases and controls were 68% and 50%, respectively.

We believe INTEROCC is the largest study ever conducted on occupational chemical exposure and risk of glioma. We examined the association between glioma and 10 different chemical agents using three different exposure indices. A large number of sensitivity analyses were conducted, varying the definition of exposure as derived from the INTEROCC JEM, covariates and other analytic tactics and stratifying on the nature of the tumor, sex of subjects, respondent status (self/proxy). Considering all the sensitivity analyses, we analyzed over 200 different models for each chemical agent and thus more than 2,000 models in total. Despite the large sample size, none of the 10 agents demonstrated a pattern of findings that would be persuasive of a causal association with glioma. The handful that were borderline statistically significant could most likely be attributed to multiple testing and random chance.

Although there have been few previous studies on occupational risk factors for glioma, in particular, there have been many previous studies on possible associations between brain cancer and occupational circumstances. However, most of these studies have suffered from some combination of low statistical power due to few study subjects combined with rare exposures, and inadequate assessment of occupational exposures. While there have been no consistently and persuasively reported associations, there are a few leads that deserve attention and that have motivated the present analyses.

There have been some reports of possible excess risks of brain cancer among anatomists, pathologists and embalmers [23–25] and this has led to hypotheses regarding a possible role for formaldehyde. But other studies of formaldehyde-exposed workers have failed to support an etiologic association between formaldehyde and brain cancer [39–42]. Our results also failed to support such an association.

While there have been some reports linking PAH exposure to brain cancer risk [43, 44], such an association was not observed in all studies [45]. In our study, we focused specifically on benzo(a)pyrene, and our results do not support a possible role of such exposure in the aetiology of gliomas. Still, exposure to PAHs is ubiquitous, occurring in many different occupations and industries, and often in environments where workers are simultaneously exposed to many other agents. Thus, analysing the specific role of PAHs is challenging.

Whereas some previous results suggested a possible increase of brain cancer risk among subjects exposed to wood dust [15, 46–48], two large studies, one a cohort study [49] and the other a large international multi-center case–control study [50], failed to detect any risk. In our study, we did not find any association between glioma and wood dust.

For diesel exhaust emissions and gasoline exhaust emissions, an increased risk of brain cancer has been reported among subjects employed as motor vehicle operators [17, 51], but our results did not support such an association. The borderline significant OR for one subset analysis of diesel exhaust emissions is most likely due to random chance since we did not observe any excess risks in relation to cumulative exposure.

An association between asbestos exposure and brain cancer risk has been suggested by some studies [47, 52], but other investigations, including a recent large multi-center case–control study of brain tumors in adults, did not find any association between gliomas and asbestos or insulation products in both men and women [50]. These findings are in agreement with our results, with ORs near the null.

There has not been previous research on sulphur dioxide in relation to glioma. While we found a borderline statistically significant OR between sulphur dioxide, the least prevalent of the 10 agents, and glioma, this was due to an excess of cases with very short duration, and it is likely to be explained by random variability.

Finally, previous evidence for an association between occupational exposure to oil products and/or animal products and glioma is sparse and inconclusive [43, 50], and our results fail to support any such association.

Participation rates were 68% among glioma cases and 50% among controls. While these rates are not atypical for contemporary population-based case–control studies, they are low enough to raise concerns about a potential participation bias. This issue has been explored in some depth in the parent INTERPHONE study with regard to mobile phone use and brain cancer [53], and it was estimated that differential participation rates could those estimated ORs by 10% to 15%. Such biases, related to the differential cellphone usage of participants vs. non-participants, are unlikely to be applicable to the present analyses of occupational exposures because whereas cellphone usage is a prevalent behavioural characteristic that might conceivably be related to the behaviour of being willing and available to participate in a study, it is less likely that being exposed to chemicals at work is similarly correlated with study participation. Even if there was a bias in the same order of magnitude, it would hardly change the overall inferences of the present analysis, as the ORs are far from being statistically elevated.

The only solidly established risk factors for glioma are rare (ionizing radiation, rare medical conditions), probably rarer than the agents that we are examining. Also, these risk factors are probably not strongly correlated with the agents we are examining. These conditions militate against the likelihood that there was confounding of our results by known risk factors. Further, since our observed ORs were mainly around the null value, if confounding had occurred it would have distorted a true causal association to a null one, and this would require quite strong negative correlation between the known risk factor(s) and the agents that we are examining. Since there has been some debate about the possible association between socioeconomic status and brain cancer risk [13, 26, 45, 54, 55], we adjusted all results for socioeconomic status using two different markers, one based on the Standard International Occupational Prestige Scale and another one based on the maximum level of education attained in the household. In some sensitivity analyses, we also adjusted for marital status (married vs. other) and this additional adjustment did not change results. While it is theoretically possible that our essentially null findings were biased by other unmeasured confounders, this seems unlikely, as such factors would have to be quite strongly negatively associated with the occupational agents we examined, conditional on the socio-economic factors we did adjust for.

The exposure assessment was based on the subjects’ reported job histories, complemented by the data in the INTEROCC JEM. The assessments did not rely on self-reports of exposures, which have been shown to be unreliable [56]. Besides the advantage of providing occupational exposure assessments quickly and with modest cost, especially considering the large number of study subjects that had to be evaluated (approximatively 7,000 subjects), the use of a JEM has the advantage of providing a standardized tool to assign exposure and thereby minimizing the possibility of recall and reporting bias. On the other hand, a JEM allocates the same exposure estimate to each subject in a given occupation in a given era, without taking into account inter-individual variability in performed tasks and job environments. The error involved in applying a JEM is non-differential with respect to case/control status and therefore can lead to some attenuation of true odds ratios [57]. Furthermore, in implementing the INTEROCC JEM, we had to deal with values of probability of exposure ranging from 0% to 100%. It was not self-evident how to use this information in developing an index of exposure for our study. We decided to create an ever/never exposed variable by dichotomising the continuous probability entries at 25% probability of exposure as assessed by the INTEROCC JEM. Such a decision, favouring sensitivity over specificity, is debatable and can lead to further exposure misclassification [58]. In sensitivity analyses, we used 5% and 50% as alternative cut-points, and the results were almost identical. Higher values of the cut-point were impractical since the number of subjects considered exposed would be too low to sustain valid meaningful analyses.

A final source of uncertainty and error relates to the generalizability of the exposure profiles from Finland to other countries. There must surely be some variability between countries in the exposures of workers. Firstly, the industrial profiles differ between different countries. But this in itself is not a source of error in our study. What would cause error is the fact that even within a given occupation, workers in different countries may experience different exposures. All of the countries in INTEROCC have modern industrial economies where, a priori, we would not expect industrial processes to be radically different within occupations as would be the case if we were comparing countries at very different levels of industrialisation. Two comparisons have been carried out between FINJEM estimates and those derived by local experts, one in Australia [59] based on a set of exposure agents that were not part of the present analysis, and one in Canada [60], based on a set of exposures that included those analysed here. Both comparisons demonstrated varying degrees of concordance between exposure estimates derived from FINJEM compared with those provided by local experts. For some agents there was high concordance, while for other agents, the concordance was rather low. An indirect indication that our results were not compromised by inter-country patterns of exposure in relation to occupations is the fact that when we carried out analyses within countries, the results were the same as when we combined countries (data not shown).

Statistical power to detect risks was limited by the low prevalence of exposure for some agents, by the measurement error inherent in our exposure assessment, and by the fact that many of the exposed subjects may have experienced relatively low concentrations of exposure. Thus, the essentially null results presented here should not be interpreted as strong evidence for an absence of any risk. Still, this is the strongest evidence to date on glioma risks from these agents.

This is the largest study to date to investigate associations between selected occupational exposures and glioma. Our findings do not support an association between any of the chemical agents examined and risk of glioma.