Main

Central nervous system (CNS) tumours are the second most common group of cancers diagnosed in children. They account for around 25% of all new cases of cancer in children aged <15 years in Great Britain. About 360 cases of CNS tumours are registered each year at the National Registry of Childhood Tumours (NRCT), with an annual age-standardised incidence of 34 per million children (Stiller, 2007).

The most common CNS tumour subtypes in childhood are: astrocytomas (around 40% of all CNS tumours), intracranial and intraspinal embryonal tumours (19%), and other specified intracranial and intraspinal neoplasms (around 12%). CNS tumours overall are distributed evenly across the childhood age groups, although ependymomas are more common before age 4, and the incidence of embryonal tumours decreases with increasing age (Stiller, 2007).

Known risk factors for CNS tumours in children are few. Genetic susceptibility is suggested because medulloblastoma is more common in boys, and brain cancers are more common in individuals with certain genetic conditions such as neurofibromatosis (risk increased 50-fold) and tuberous sclerosis (risk increased 70-fold). Therapeutic and diagnostic doses of ionising radiation are also a known risk factor for childhood brain tumours (Gurney et al, 1999; Pearce et al, 2012). Studies of risk of CNS tumours from a range of other environmental and genetic risk factors have provided only suggestive or inconsistent evidence (Gurney et al, 1999), although some childhood subtypes are associated with high birth weight (Harder et al, 2008).

The aetiology of childhood CNS cancers is not well understood, but prenatal exposures may be important. There has also been concern that paternal occupational exposures may be risk factors for cancer in the children of workers (Colt and Blair, 1998). Fathers may bring harmful substances home on their clothes to which a child or pregnant woman may be exposed. There is also the possibility that exposure to certain chemicals or radiation could cause genetic changes in sperm, which could predispose a child to cancer (Cordier, 2008).

For CNS tumours, a number of paternal occupational risk factors have been reported, including exposure to agriculture, aircraft industry paint, solvents and electromagnetic fields (Cordier, 2008). However, these and other associations have not been consistently reported in the ∼30 studies carried out since the late 1970s. However, many of these studies have suffered from small numbers and imprecise exposure assessments (Savitz and Chen, 1990; Colt and Blair, 1998; McKinney, 2005; Cordier, 2008). We addressed some of the shortcomings of previous studies by drawing the study population from the NRCT, which holds an almost complete record of all childhood cancers registered in the United Kingdom between 1962 and 2006 (Stiller et al, 1998), and which provides sufficient cases to allow analysis of risk by tumour subtype.

The main objective of this study was to investigate possible associations between paternal occupational exposure and CNS tumours overall and tumour subtypes in children in Great Britain, using a matched case-control design. We focus on paternal occupation because it is more completely recorded on birth registrations during our study period than maternal occupation (Fear et al, 1999a). Additionally, we use job title to derive an approximate measure of paternal social class and use this to investigate possible associations between paternal occupational social class (OSC) and childhood CNS tumours.

Materials and methods

Cases and controls

The NRCT contained 12 706 registered cases of CNS tumours in children aged <15 years born and diagnosed between 1962 and 2006 in Britain. A total of 465 cases were excluded because they were born overseas or adopted. In addition, 367 case for whom no birth registration could be found were excluded, leaving 11 874 eligible cases for whom a birth record was available.

Control children (n=11 874) were selected from all birth registrations for Britain, held by the Office for National Statistics (ONS) or the General Register Office for Scotland (GROS). One cancer-free control for each case was selected, matched on sex, date of birth (±6 months) and birth registration sub-district.

The completeness of ascertainment of childhood cancer cases in the NRCT has varied over time, but it contains an almost (>97%) complete record of all registered cases of childhood cancer in Britain from the early 1970s (Stiller, 2007; Kroll et al, 2011a).

Oxfordshire Research Ethics Committee (Oxfordshire REC C, Reference 07/Q1606/45) approved the use of these data in 2007.

Coding of occupational groups

In the United Kingdom, paternal occupation is routinely recorded on the publically available section of birth registrations where the father is named. Paternal occupation was abstracted verbatim from the case and control birth records as supplied by ONS and GROS.

Occupations were coded according to the 1980 Office of Population Censuses and Surveys (OPCS) Standard Occupational Classifications (SOC; Office of Population Censuses and Surveys, 1980). Coding was carried out independently by two coders, using the OPCS (now the ONS coding manuals. Where the two coders disagreed, a third coded the occupation. Where the third coder agreed with one of the original coders that agreed code was assigned. Where all three coders disagreed, the occupation was regarded as ‘uncodable’. At all stages occupations were coded blind to the case control status of the individuals. The 1980 classifications were converted to the codes used in the 1970 SOC (Office of Population Censuses and Surveys, 1970), using a computer program.

The 1970 codes were subsequently allocated to one or more of 33 occupational exposure groups, which have been described elsewhere (Fear et al, 1999a, 1999b). Briefly, the occupational exposure groups were derived by one of the authors (NTF) in conjunction with an occupational hygienist and an occupational researcher. Occupations not appearing in any of the 33 groups were classified ‘unexposed’ in all groups. For the occupational exposure group, ‘social contact’, jobs were classified according to a previously used scheme in which occupations were classified by an occupational hygienist according to whether they had higher than average social contact (Fear et al, 1999b, 2005). Occupations classified to one or more of the exposure groups were further defined as having either ‘definite’ (daily contact with the agent, or contact at a high intensity, for example, carpenters and wood dust) or ‘possible’ (exposure to the agent but not necessarily daily nor necessarily at high intensity, for example, builders and wood dust) exposure in that group (Fear et al, 1999a). Job titles could be coded to more than one occupational exposure group for example, bus drivers appear as exposed in ‘exhaust fumes’, ‘inhaled hydrocarbons’ and ‘social contact’.

Each 1980 occupation code was then assigned to one of six social class codes based on occupation (V unskilled, IV partly skilled, IIIM-skilled non-manual, IIINM-skilled manual, II managerial and technical, I professional) from the 1980 OPCS Classification of Occupations.

For 622 cases and 703 controls, paternal occupation was missing, and these subjects were excluded from the analysis (Figure 1). For some (82 cases and 95 controls), it was not possible to assign a 1980 occupation code, or it was not possible to convert the 1980 code to a 1970 code (51 cases and 37 controls). In these circumstances, the paternal occupation was coded as if missing. For 1 020 cases and 1 172 controls, social class was classified as ‘missing’ because no occupation was given or the occupation falls outside the ONS social classifications (for example, armed forces, student, independent means or sick). The 51 cases and 37 controls excluded from the occupation analysis, because their 1980 code could not be translated to a 1970 code, were included in the social class analysis and appear in the results shown in Table 5

Figure 1
figure 1

Flow chart showing the numbers of eligible CNS tumour cases and controls included in the analysis.

Full size image

A total of 1024 case/control fathers were classified as ‘forces’, comprising the armed forces, police force, fire service, and guards and related workers not elsewhere classified. Within the ‘forces’ group, social class code was unavailable for members of the armed forces, approximately half the group, so OSC was not included in the analysis for this exposure group.

Outcomes

All cancers registered in the NRCT are coded to the International Classification of Childhood Cancer, third edition (ICCC-3; Steliarova-Foucher et al, 2005). Outcomes of interest in this study were: ependymomas (ICCC-3 31 division 1), astrocytomas (ICCC-3 32) and intracranial and intraspinal embryonal tumours (ICCC-3 33). Astrocytomas and other gliomas together (ICCC-3 32+34) were also considered as an outcome group. The group ‘total CNS tumours’ include these and the following additional categories: choroid plexus tumours (ICCC-3 31 division 2), other specified intracranial and intraspinal tumours (ICCC-3 35), unspecified intracranial and intraspinal neoplasms (ICCC-3 36; Table 2).

Analysis

Odds ratios (ORs) and 95% confidence intervals (95% CIs) for our matched analysis were calculated using conditional logistic regression (Breslow and Day, 1980). Matching factors were: sex, period of birth and birth registration sub-district. ORs and 95% CIs adjusted for social class (I, II, IIINM, IIIM, IV, V) were also generated. Our exposed population for occupational exposure consisted of individuals classified as ‘definitely’ exposed. The same analyses were repeated, taking the exposed population as those with either ‘definite’ or ‘possible’ exposures, although results are not shown because of the potential inclusion of misclassified or non-exposed individuals. Statistically significant results were defined as those where the P value was <0.05.

To assess the impact of multiple statistical testing on the likelihood of any of 33 P values for total CNS tumours being statistically significantly different from those expected by chance should the null hypothesis for each test be true, we plotted the empirical cumulative distribution of P values whose null sampling distribution is assessed as uniform on (0,1). We used a Kolmogorov-Smirnov test to test for significant deviation from linearity.

All analyses were carried out using STATA v. 11 (StataCorp LP 2009).

Results

After exclusions, a total 11 119 cases and 11 039 controls were included in analyses of occupation and CNS cancer risk. There was no significant difference between the birth regions of cases and controls, nor was there a difference in social class or occupational status between the two groups (Table 1). Of the cases, 5 722 (51%) were astrocytomas or other gliomas, 2 286 (21%) were embryonal and 985 (9%) were ependymomas (Table 2). There were no marked differences between the results for definite exposure and definite or possible exposure, so results and discussion that follow are based on definite exposures alone.

Table 1 Childhood CNS cancer cases born and diagnosed in Great Britain between 1962 and 2006 for whom a birth record was available, and their matched controls
Full size table
Table 2 Number of cases and their controls by CNS tumour type included in the occupational exposure analysis (see Figure 1 for details of exclusions)
Full size table

Table 3 shows that there was an increased risk for CNS tumours as a whole with paternal occupational exposure to animals, both before and after adjustment for OSC. Paternal occupational exposure to lead showed an increased risk after adjustment for OSC. However, risk of CNS tumours was reduced in children whose fathers were occupationally exposed to metal-working oil mists, both before and after adjustment for OSC.

Table 3 Paternal occupational exposures and ORs with 95% CIs for total CNS tumours
Full size table

Analysis by tumour subtype revealed further associations (Table 4). For ependymomas (ICCC-3 31 division 1), there was a significantly raised OR for exposure to solvents before adjustment for social class (falling to borderline significance after adjustment). For astrocytomas (ICCC3 32), there was a borderline significantly reduced OR for exposure to paints, both before and after adjustment for social class. ORs for high social contact were significantly raised for both astrocytoma and astrocytomas and other gliomas taken together, but these risk associations became nonsignificant after adjustment for OSC (Table 4). For embryonal tumours (ICCC-3 33) the OR for exposure to metals was significantly reduced, both with and without adjustment for OSC.

Table 4 Significant associations between paternal occupational exposures for CNS subtypes
Full size table

There was a significant effect of OSC on risk of both astrocytoma and astrocytomas and other gliomas taken together (Table 5). For both these groupings there was a clear and significant trend of increased risk in the higher social classes and decreased risk in the lower social classes. For embryonal tumours no clear relationship between social class and risk of disease was discernible. For ependymomas, there was some indication of an inverse relationship between social class and risk of disease, in that the risk was lower in the higher social classes (significantly so for social class II), although the overall test for trend was nonsignificant. For CNS tumours overall, the risk was significantly reduced in social class V, and a weaker but still significant trend of increasing risk with higher social class was visible.

Table 5 Childhood CNS tumour risk by paternal occupationally derived social class
Full size table

We assessed the results for the effect of multiple statistical testing, using a Kolmogorov-Smirnov test on the 33 P values generated in the analysis of paternal occupational exposure groups and childhood CNS tumours. Results showed that the distribution was not different from normal that is, the 33 P values were compatible with a distribution that has arisen by chance (results not shown).

Discussion

Summary

In our analysis of risk of all childhood CNS tumours with paternal occupational exposures, the findings for most exposure groups were unremarkable, with most ORs around 1, both before and after adjustment for OSC. Risk was significantly >1 in two exposure groups; animals and to lead, although this latter group only after adjustment for social class. It remained significantly <1 for exposure to metal oils. In the CNS tumour subgroup analyses, the findings for most exposure groups were also unremarkable, although five associations were significant before adjustment for social class. In particular, social contact was a risk factor for both astrocytoma and astrocytomas and other gliomas taken together, and exposure to solvents was a risk for ependymomas. However, these relationships did not persist after adjustment for social class. Two risk estimates were significantly <1; exposure to paints for astrocytomas and to metals for embryonal tumours. Of the risk associations observed for each CNS tumour subtype, the only one that remained significant after adjustment for social class was that between exposure to metal and embryonal tumours.

We found a significant trend in the association between childhood CNS tumours and social class. There was an increased risk of childhood CNS tumours in the higher social classes and a decreased risk in the lower social classes. This effect was driven principally by the association between risk of astrocytomas and social class, where a clear and highly significant trend was observed. For the other major CNS subgroup, ependymomas, a nonsignificant inverse relationship was observed.

Comparison with previous studies

Previous epidemiological studies have identified a number of possible paternal occupational risk factors for childhood CNS tumours. These include working in the aircraft industry; electronics, petroleum industry, paper mill worker, printer, metal-related occupations, paint, solvents, ionising radiation and electromagnetic fields (Gurney et al, 1999). Subsequent epidemiological research has failed to replicate most of these consistently excepting paternal exposure to paints, and to hydrocarbons or exhaust fumes, which have shown consistently raised risks (Colt and Blair, 1998; Cordier, 2008) as, to a lesser extent, have pesticides (Infante-Rivard and Weichenthal, 2007).

Our findings for all CNS tumours are not in keeping with these results; our ORs for paints, solvents, hydrocarbons and exhaust fumes were all around 1. We did, however, see a raised risk for fathers who work with animals. It is possible that this work is agricultural and was thereby associated with exposure to pesticides, which would align this result with those that have shown an association between paternal exposure to pesticides and risk of childhood CNS tumours (Cordier et al, 2001; Infante-Rivard and Weichenthal, 2007). However, we observed no association with exposure to agriculture or agrichemicals, which may argue against an involvement of pesticide exposure. For exposure to EMFs, our ORs were also around 1, in keeping with a recent case-control study of paternal occupational exposure and childhood cancer (Hug et al, 2010).

Although we see an apparently protective effect of exposure to paints on risk of astrocytomas, our results for risk of ependymoma from paternal occupational exposure are in accordance with studies that have showed an association between exposure to solvents and risk of CNS tumours (Colt and Blair, 1998). However, loss of significance of results upon adjustment for OSC suggests that social class may be responsible for these associations. The risk for social contact and astrocytomas or astrocytomas and other gliomas was raised, but nonsignificant after adjustment for OSC. This relationship is consistent with results from another UK-based study, which saw no relationship between paternal social contact and CNS tumours in children (Fear et al, 2005).

Socioeconomic status

When we examined the risk of CNS tumours in children by the OSC of the father, we found that for all CNS tumours, astrocytomas alone and astrocytomas and other gliomas taken together, there was an increased risk in the higher social classes, with the relationship driven by astrocytomas. In contrast, risk of ependymoma was greater in social class IV and least in social class II, but the trend was not significant.

The relationship between risk of astrocytomas and paternal social class is consistent with evidence from studies of paternal SES and other types of childhood cancer for example, childhood leukaemia (Borugian et al, 2005; Poole et al, 2006; Kroll et al, 2011b). There is also some evidence that higher social class maybe associated with increased risk of brain cancer in adults (Preston-Martin et al, 1993). However, there is little direct evidence in the literature to suggest that an increased risk of childhood CNS tumours with higher social class of the father is likely. There are some limited indications that exposure of the mother and to the child to common viral infections may result in a higher risk of childhood brain tumours (Fear et al, 2001; McNally et al, 2002b; McKinney, 2005); if these exposures themselves are associated with paternal social class, the link might be plausible. Either way, the association would benefit from further investigation.

Strengths and limitations

The strengths of this study are that the analysis is based on the case data drawn from the NRCT, which has, over the period studied here, consistently high levels of case ascertainment (Kroll et al, 2011a), and thus can be considered almost complete. Of the previous epidemiological studies that have investigated paternal occupational exposure and CNS cancer risk in children, few have been able to examine risks by CNS subtypes (McKinney, 2005). We have data spanning 40 years and over 10 000 cases of CNS tumours, with nearly 1 000 cases of the smallest subtype considered separately.

Frequent problems with interview-based case-control studies are recall and participation bias. This is not a consideration in the present study as we used routinely collected data, and occupation was documented before diagnosis. The exposure assessment used a well-established occupational and exposure classification (Fear et al, 1999a) to which father’s occupation was coded blind to case-control status. However, our method used occupation recorded at time of birth, and this might differ from the occupation held during a more aetiologically important time period. We also recognise that the use of occupational title to capture occupational exposure is controversial (Schuz et al, 2003). However, using this method does allow comparison with a number of previous studies in this field that have used this method of classifying occupational exposures (McKinney et al, 2003; Fear et al, 2005).

In our study, we have examined exposures from paternal occupations only. Although maternal exposures of potential carcinogens to their children may have an important role in the development of childhood cancer, fathers’ occupations are more completely recorded on birth registrations than they are for mothers. In our study, father’s occupation was available for 94% of registrations and for mothers it was 23%. This may result in some misclassification with respect to unmeasured exposure to carcinogens via the mothers’ work, but we expect that in the analysis of social class and brain tumour risk, fathers’ OSC classification is as appropriate an exposure measure as would be one based on mothers’ OSC.

Interpretation

In our analysis we carried out multiple comparisons, which may have resulted in a number of associations having arisen by chance. Our analysis of the likelihood of these arising by chance showed that the significant P values, both raised and lowered, are unlikely to be real. Consequently, our results should be interpreted with caution. In addition, other, unmeasured risk factors may also explain any apparent association between OSC and childhood CNS tumours.

The aetiology of childhood cancers, including CNS tumours, is not well understood, and there is continuing debate about the role of occupational and environmental exposures (Belson et al, 2007). The importance of the exposure route and timing is also relevant (McKinney et al, 1991; Roman et al, 2005). Exposures to the mother may be relevant during the intrauterine period and to the father pre-conceptually, when germ cells may be affected and, for both parents, postnatally when residues from work may be brought into the home. As we have no information on the frequency or duration of exposure and occupational practices, and exposures may have changed during the long study period, we cannot exclude the possibility of exposure misclassification. It is also possible that during our long study period some cases of childhood cancer were not diagnosed or registered. If those cases were more likely to be from the lower social classes, there could be fewer cases of CNS cancers in the lower-social class groups relative to the higher-class groups, and this might possibly explain the social class effect we detected, although this is an unlikely explanation, given the size of our data sets.

In this study we have analysed risk of CNS tumours by paternal occupational exposure for each major diagnostic subgroup. This is important as each may have a different aetiology (Ross et al, 1994; McKinney et al, 2003). However, once paternal social class was accounted for we did not see any notable relationships between paternal occupation and increased risk of CNS tumours. The finding that astrocytomas are apparently most common in social class II and least common in social class V, however, may warrant further investigation. The relationship between paternal social class and risk of leukaemia is thought to reflect the likelihood of encountering common infections. Whether this hypothesis may also apply to the relationship with astrocytomas is unclear. Our results for risk of astrocytoma, and astrocytoma and other gliomas with exposure to ‘social contact’ showed a (nonsignificant) excess. Although there are well-developed arguments for a role of infection contact in the aetiology of childhood leukaemia (Kinlen et al, 2002; McNally and Eden, 2004), the role of infection in the aetiology of childhood CNS tumours is uncertain. There is limited evidence that maternal viral infection during pregnancy might be a risk for tumours of the brain or nervous system (Fear et al, 2001). Analysis of space-time clustering has indicated that unspecified infection in the child (or mother) might be relevant (McNally et al, 2002a). It is possible that high paternal occupational social contact leads to more frequent exposure to viruses to pregnant women and children at home.

In conclusion, this paper does not add to evidence for paternal occupation as a risk factor for childhood CNS tumours, either overall or for specific subtypes, although it suggests that OSC of the father may be associated with risk of childhood CNS cancers, in particular astrocytomas and other gliomas.