Methods

The selection of cases and controls

The population-based National Registry of Childhood Tumours (NRCT)1 included 1318 cases of retinoblastoma born and diagnosed between 1962 and 1999 for whom a birth record was available. A further 69 children (18 heritable, 51 non-heritable) who were born overseas, adopted or not traced in the birth registers, were excluded from the study. The birth record for one control child per case matched on sex, date of birth (within 6 months) and birth registration sub-district was selected from the Office for National Statistics (ONS) or General Register Office in Scotland (GROS) birth registers.

Oxfordshire Research Ethics Committee (Oxfordshire REC C, Ref 07/Q1606/45) approved the use of the data reported in this study in 2007.

Heritability status of cases

For the purposes of this study, all bilateral cases (n = 508) and those unilateral cases with a family history of retinoblastoma (n = 58) were classified as heritable, plus one case of unknown laterality with a family history. Bilateral heritable cases with no known family history of retinoblastoma (n = 323) were considered to result from a new germ-line mutation.3 Unilateral cases with no family history were classified as non-heritable (n = 737), as were 14 cases of unknown laterality with no family history.

We analysed the data on retinoblastoma cases separately for non-heritable, heritable and heritable cases with no family history of the disease. Parental mutagenic exposure might be the cause of either newly arising germ-line mutations (which are more commonly associated with the paternal line15) or the additional somatic mutation in heritable cases, so we have analysed all heritable cases and a major subset, that is those without a family history of retinoblastoma. We also analysed all cases of retinoblastoma combined.

Coding of occupational data and derivation of the occupational exposure groups

Paternal occupation (part of the public record of birth registrations) was abstracted verbatim from the birth records from ONS/GROS and coded according to the 1980 Classification of Occupations.16

The occupations were coded (“blind” to case/control status) independently by two coders (resulting in 68% agreement). A third coder then recoded occupations where there were discrepancies and in 70% of cases agreed with one of the original coders. Occupations coded differently by all three were resolved by consensus after consultation with a fourth coder.

The codes allocated to the occupations from the 1980 classification16 were converted to the 1970 Classification of Occupations17 using an automated recoding scheme.18 These codes were then allocated to one or more of 33 occupational exposure groups described previously.18 Each occupation could be classified in more than one exposure group.

Occupations classified within the 33 exposure groups were further defined as having either “definite” (contact with the agent everyday or at very high levels) or “possible” (exposed to the agent but not necessarily at a very high level or everyday) exposure in that group.

Some occupations did not appear in any of the exposure groups and were classified as “unexposed” in all exposure groups.

Some birth records did not have a paternal occupation recorded (60 cases, 54 controls) and were excluded from these analyses. In other instances it was not possible to allocate a code from the 1980 classification (n = 6), or to convert the 1980 code to a 1970 code (n = 13); these records were included in the analyses as “unexposed”.

A code for social class16 was allocated to each 1980 occupation code (where possible) and then categorised as manual (social classes: IIIM, IV, V) or non-manual (social classes: I, II, IIINM). The occupational codes which could not be allocated a code for social class or manual/non-manual status included those not stated or inadequately described, armed forces, full time students, and those of independent means.

Analysis

We present in tables 1–3 the results of analyses of “definite” occupational exposures. A further analysis was carried out on exposure groups including occupations with “definite” and “possible” exposure grouped together, or for “possible” exposure only (for some exposure groups this was the only data available for some groups of retinoblastoma cases).

Conditional logistic regression analyses19 were carried out on the paternal exposure data for each of the case groups (non-heritable, heritable, and heritable with no family history) and their matched controls. As some of the exposure groups contained only small numbers of exposed fathers, the analysis was carried out using LogXact 7 (Cytel Software, Cambridge, MA, 2005). Odds ratios (OR) and 95% confidence intervals (95% CI) were calculated for each of the 33 exposure groups. Statistically significant results were defined as those where the p value was <0.05.

Results

Table 4 describes the cases in the study in terms of their heritability status, sex, birth year and the manual/non-manual social class status of the father.

Table 1 includes estimates of risk for the paternal occupational exposure groups for the non-heritable cases and controls. There is only one statistically significant result here for (definite) exposure to oil mists in metal working (OR = 1.85 (95% CI 1.05 to 3.36)). The risk for this occupational exposure group fell to borderline significance when adjustment was made for the manual/non-manual social class status of the father (OR = 1.73 (0.96 to 3.19)). Odds ratios for other exposure groups were not materially affected on adjustment for manual/non-manual social class status.

To explore the apparent increased risk for exposure to oil mists in metal working further, we calculated odds ratios separately by stratifying the individually matched analysis for sex, four birth periods (1962–1969, 1970–1979, 1980–1989 and 1990–1999) and 10 “standard regions”. These regions are defined by (case or control) address at birth and designate 10 geographical areas. The vast majority of the case/control pairs, which were by definition matched on birth registration sub-district and time of birth, therefore fall into the same (larger) standard region and period of birth. Thus, standard region and birth period were relevant factors which could be incorporated in the model as stratifying variables.

The increased risk for non-heritable retinoblastoma appeared to be restricted to female offspring (OR = 3.50 (1.41 to 8.67) for females, OR = 1.14 (0.56 to 2.34) for males). The difference in ORs between the two sexes was of borderline significance (p = 0.054). Likewise, the increased retinoblastoma risk was concentrated in children born in the middle two decades of the study (OR = 3.50 (1.15 to 10.63) for children born 1970–1979 and OR = 3.25 (1.06 to 9.97) for those born 1980–1989). However, the variation in ORs among the four birth periods fell short of significance (p = 0.09). There was no significant variation in risk among the children born in the 10 different standard regions (p = 0.70).

Results for the paternal occupational exposure groups for the totality of heritable cases (table 2) and for the majority subset of heritable cases with no family history (table 3) revealed no statistically significant associations for any of the 33 occupational exposure groups.

We also analysed the data for all cases of retinoblastoma. The results were similar to those shown for the non-heritable cases and their controls, with the risk for exposure group “metal working (oil mists)” being significantly elevated (OR = 1.77 (1.12 to 2.85)). On adjustment for manual/non-manual social class status of the father this was reduced to OR = 1.67 (1.05 to 2.72).

Discussion

Our one statistically significant finding was that for “metal working (oil mists)”. The 40 exposed fathers (of non-heritable cases) in this group were employed in the following occupations: “turners”, “machine tool setters”, “setter operators not elsewhere classified”, “machine tool operators”, “toolmakers/tool room fitters” and “press workers and stampers”. Our finding of an increased risk in only the female offspring of these fathers cannot easily be explained. This was an unexpected finding, there is little difference in the overall incidence of retinoblastoma between the two sexes and we can find no other report of any differential effect of paternal occupational exposure by sex of offspring relating to retinoblastoma. This result could be due to chance, arising from a post-hoc subgroup analysis.

Previous findings

A number of epidemiological studies have examined the possible association between paternal occupation and the occurrence of retinoblastoma in offspring.4 5 6 7 8 9 10 11 12 13 14 These include case–control studies,4 5 6 14 cohort studies and those of other design.7 8 9 10 11 12 13 Many of the previous findings are based on small numbers.

There is major difficulty in comparing the results we obtained with those from the other studies because each study classified paternal occupations in a different way, and collected occupational data from different sources.

Case–control studies

Hicks et al4 reported a case–control study of parental exposure to radiation. A significant result relating to retinoblastoma was reported for two fathers employed as “radio and television repairmen”. Our results for “electromagnetic fields” and “ionising radiation” were unremarkable.

Bunin et al5 reported a case–control study where statistically significant increased risks were reported for sporadic heritable retinoblastoma where paternal occupation (in the preconception period) was in a “metal-manufacturing industry” and the “military”. For non-heritable retinoblastoma, paternal work as “welders and machinists” (in the pre- or post-conception period) and in “metal-related occupations” (in the preconception period) was associated with an increased risk.

A third case–control study included cases of retinoblastoma from the Danish Cancer Registry.6 One statistically significant association was reported for paternal occupation as “physician”. Our results for the exposure group “medical/healthcare” were not significant.

A case–control study examining paternal occupational exposure to pesticides or herbicides was carried out using data from an English regional cancer registry.14 This study reported estimates of risk for retinoblastoma (two sources of controls): one estimate was raised and the other was less than unity, and neither was statistically significant. Our study found no significant associations with “agriculture” or “agrochemicals”.

Cohort and other studies

Bonde et al7 quoted a statistically significant relative risk of 3.5 for four fathers from a cohort of Danish metal workers. A study based on a cohort of Norwegian printing workers8 reported one case of “eye cancer” when 1.1 was expected. Another Norwegian cohort study9 involving farm holders reported two cases of retinoblastoma (the sex of the exposed parent was not specified).

Mutanen and Hemminki10 analysed the risk associated with paternal occupation using cases from the Swedish Family Cancer Database. Significantly raised standardised incidence ratios (SIRs) for “eye cancer” were found for two paternal occupations: “shoe and leather workers” and “electrical workers”. Our results for “electromagnetic fields” were unremarkable.

A cohort study of male pesticide applicators in Sweden11 quoted a reduced, non-significant SIR for childhood “eye cancer” based on two observed cases. Fear et al12 reported a reduced proportional mortality ratio of 0.71 (0.15 to 2.08) based on three observed cases of “eye cancer” whose fathers had potential occupational exposure to pesticides at the time of the child’s death. A report from a study of pesticide applicators in the USA,13 quoted a raised SIR for retinoblastoma, but this estimate does not specifically refer to paternal occupational exposure. Our study found no significant associations with “agriculture” or “agrochemicals”.

Interpreting the results of this study

Study power

Despite the large number of cases and controls the study has limited power. For the largest group of cases (ie, non-heritable, n = 751), if there was a prevalence in the control group of 15% for any exposure group, the study would have 85% power to detect an odds ratio of 1.5. In the analyses of the non-heritable cases (table 1), only four groups (hydrocarbons (inhaled), hydrocarbons (dermal), metal and social contact) have this prevalence of exposure. Despite the low power, these data represent all cases of retinoblastoma in the national population-based registry.

Issues arising from methods for coding occupational data

The occupation was recorded before the diagnosis of retinoblastoma, eliminating the possibility of recall bias.

The occupations were coded and discrepancies resolved “blind” to case–control status. The occupations were allocated to exposure groups within an exposure classification scheme devised with the help of an occupational hygienist, which has been used in other studies.12 18 20 21

Our study examines paternal occupation as recorded at the time of birth of the child. Occupations could change in the time between conception, birth and diagnosis of retinoblastoma. However, retinoblastoma is one of the earliest childhood cancers to be diagnosed, so this is less of a consideration than it might be for some other diagnostic groups.

Study limitations

As we examined risk in a large number of exposure groups, our results may be due to chance, and therefore should be interpreted cautiously and in the light of previously reported associations,22 although this is difficult due to differences in study methodology. Additionally, cases or controls could be non-differentially misclassified with respect to exposure group and this would usually be expected to bias the odds ratio towards one, obscuring a real relationship with the exposure group.

We do not have information about biomarkers of occupational exposure, and cannot adjust for the use of protective equipment or factors such as smoking.

We have no information on the level or frequency of the inferred exposure. Occupational practices and exposures may have changed over the 40-year period covered by this study and our analyses do suggest a possible effect of period of birth.

Our odds ratios for exposure to “metal” were close to one and not statistically significant for any type of retinoblastoma. These results do not support the significant excess risks reported by Bunin et al5 for “metal-related occupations” or “welders and machinists”, or Bonde et al7 for “metal workers”.

We did, however, find a statistically significant increased risk (associated with fathers of non-heritable cases) for a subset of metal workers, those definitely exposed to oil mists. No estimates of risk for a comparable group have been reported elsewhere.

Conclusion

We have carried out the largest case–control study of paternal occupation and retinoblastoma possible in Britain. However, our study has limited power for many of the defined exposure prevalences. Differences in study design also make it difficult to examine how our results relate to previous findings for other occupational exposures.

We have outlined a number of limitations in our study methods, but bearing these in mind our findings do not support the hypothesis that paternal occupational exposure is an important aetiological factor for retinoblastoma.