METHODS

This was a matched case-control study nested in a cohort of cabin attendants (n=1532, all female) which served as a study base and was collected from the union of Icelandic cabin crew and two airline companies, as described previously.⁵ Record linkage, using the personal identification number, with the Icelandic Cancer Registry provided information on breast cancer in the cohort. Similarly, record linkage with the registry of childbirth at the Genetical Committee of the University of Iceland provided the information on reproductive factors (age at first childbirth and number of children). The cohort was then subsequently record linked to the Icelandic Cancer Registry in order to extend the follow up and include breast cancer cases accumulated up to the end of year 2000.

Altogether 35 histologically confirmed cases of breast cancer were found, including four cancers in situ and two women who each had two breast cancers. In the previous cohort study⁵ involving 26 of the cases, all but one of the breast cancers were of ductal type where the pathology reports decided whether the histology showed lobular or ductal types. All 35 cases were included in the study, assuming that they shared aetiological factors; this has been the procedure in other studies dealing with aetiology of breast cancer.⁸ Each of the cases was individually matched by year of birth with four controls from the cohort of cabin attendants. The controls had to have entered the cohort at least in the year the respective case was diagnosed.

Employment time in the cohort study was recorded without taking into account any part time job, or sick or maternity leave. The length of employment was calculated for the cases from the first year of employment as a cabin attendant to the last year of employment or the year of diagnosis of breast cancer, whichever occurred first. The length of employment was calculated for the controls from the first year of employment to the last year of employment or the year of pseudo-diagnosis—that is, the year of diagnosis for the respective case. The Icelandic airline companies introduced jet aircraft on all international routes in the year 1971⁵; jets are flying at higher altitude and faster than non-jets. Therefore, the length of employment was divided by whether it occurred before 1971 and in the year 1971 or later. The cruising altitude difference between non-jet aircraft and jet is about threefold, hence greater exposure to cosmic radiation.

A multivariate case-control analysis was performed by a conditional logistic regression.⁹ The adjusted odds ratio and exact computation of 95% confidence intervals (CI) were calculated using STATA software. Case-control status was the dependent variable. Length of employment was treated as a continuous variable expressed in years. In a different analysis the length of employment was treated as a dichotomous variable, classified either as five or more years or less than five years. Five years were chosen as the cut off for the duration of employment as the annual exposure to ionising radiation sustained by aircrew is low,¹⁰ and five year employment was considered a minimum exposure time. Length of employment, both as continuous and dichotomous variables, was also calculated according to the cut off in year 1971. Age at first childbirth was classified in five categories, including one category for nulliparous. The number of children was classified in six categories (0, 1, 2, 3, 4, and 5 or more). In yet another separate analysis we counted the remaining length of employment 20 and 30 years before the date of diagnosis (or pseudo-diagnosis) (subtracting 20 and 30 years from the date of diagnosis) and in that way implemented 20 and 30 years lag time.

The National Bioethics Committee and the Data Protection Commission approved the study.

RESULTS

The odds ratio for breast cancer associated with continuous length of employment, adjusted for age at first birth, was 1.00 (95% CI 1.00 to 1.05). The odds ratio for those who had five or more years of employment compared with those with less than five years of employment, adjusted for age at first childbirth, was 2.10 (95% CI 0.93 to 4.73).

Table 1 shows the adjusted odds ratio, taking into account two periods of length of employment according to the division in year 1971. The odds ratio was 5.24 (95% CI 1.58 to 17.38) for those who had five or more years of employment before 1971 compared to those with less than five years of employment before 1971, adjusted for age at first childbirth and length of employment 1971 or later. The corresponding odds ratio was 0.82 (95% CI 0.34 to 1.97) for those who had five or more years of employment after 1971 compared to those with less than five years of employment after 1971, adjusted for age at first childbirth and length of employment 1970 or earlier. When adjusted for number of children instead of age at first childbirth, the odds ratio was 4.31 (95% CI 1.33 to 14.00) for those who had five or more years of employment before 1971 compared with those with less than five years of employment before 1971, adjusted for number of children and length of employment 1971 or later.

When excluding the two second breast cancers (and their controls) from the analysis the odds ratio was 4.30 (95% CI 1.26 to 14.64) for those who had five or more years of employment before 1971 compared to those with less than five years of employment before 1971, adjusted for age at first childbirth and length of employment at 1971 or later. Similarly when the four breast cancers in situ (and their controls) were excluded from the analysis the odds ratio was 4.42 (95% CI 1.09 to 18.02) for those who had five or more years of employment before 1971 compared to those with less than five years of employment before 1971, adjusted for age at first childbirth and length of employment at 1971 or later.

When taking 20 years lag time into consideration, the odds ratio was 3.42 (95% CI 1.05 to 11.20) for those who had five or more years of employment compared to those with less than five years employment, adjusted for age at first childbirth. When applying 30 years lag time the odds ratio was 3.51 (95% CI 0.75 to 16.50) for those who had five or more years of employment compared to those with less than five years of employment, adjusted for age at first childbirth.

The mean age of the breast cancer cases at year of diagnosis was 47 years (range 28–72).

DISCUSSION

Our calculations show a significant association between length of employment, measured as five or more employment years before 1971, and risk of breast cancer adjusted for reproductive factors among cabin attendants. In spite of the small number of cases, the odds ratio was high and the 95% confidence intervals did not include unity. Our observed association is supported by the evidence from the cohort studies,^1–7 suggesting an association of the breast cancer risk with the occupation of cabin attendants.

The results of the present case-control study are not directly comparable to the results of the previous cohort study.⁵ In the cohort study an external comparison group was used, but in the present study a type of internal comparison was applied. There are nine more cases in the present study than in the cohort study. The cut off for duration of employment in pre and post 1971 was made differently in the two studies, and in the present case-control study an adjustment was made for reproductive factors.

The possible causal factor or factors in the occupation of cabin attendants have been reviewed recently.¹⁰ They are exposed to cosmic radiation including neutrons and gamma rays, engine exhaust gases, ozone, electromagnetic fields, and radiofrequency radiation.¹⁰ Jetlag and disturbances in the circadian rhythm have also been suggested as possibly playing a part, as well as the previous use of pesticides in aircraft.^4,5,11 The cabin attendants in our study base have all flown on international routes. Keflavik airport (Reykjavik) in Iceland was the hub, cities in Europe and North America were the main destinations; all cabin attendants have therefore sustained disturbances of their circadian rhythms and been exposed to cosmic radiation during northerly flight profiles.^5,12,13 The estimated annual doses received by aircrew are 4–5 mSv¹² for routes between Europe and North America, which may be twofold higher than doses for intracontinental routes.

Exclusion of the second cancers and the cancer in situ did not materially affect the estimates of the odds ratios, which still did not include unity, suggesting that our assumption of shared aetiology was valid.

As far as we are aware, our study is the first published case-control study of breast cancer among cabin attendants adjusting for individual reproductive factors. In previous cohort studies showing an excess of breast cancer among cabin attendants,^1–5,7 the possible confounding of reproductive factors were not directly controlled for. In the Norwegian cohort no association was found between length of employment and the risk of breast cancer after adjustment were made for reproductive factors.⁶ In the present study the picture of the relation between breast cancer risk and age at childbirth and nulliparity is the opposite of the findings from most studies on breast cancer and reproductive factors. In the present study the highest risk was found for age at first childbirth ⩽24 years. It is notable that all confidence intervals for odds ratios of the reproductive factors included unity. However, this relation between breast cancer risk and reproductive factors was also seen in the Norwegian cohort study on cabin attendants.⁶ In previous cohort studies of cabin attendants, using the general population as reference, social class differences were considered to be a possible confounder for the breast cancer risk found.¹ Social class is unlikely to be confounding the result of the present study, as the study base is a cohort of cabin attendants belonging to the same social class.

Use of the comprehensive population registers in Iceland strengthens our study. Misclassification of case and control status is unlikely because the recording of cases in the cancer registry is based on histological examination and is independent of occupation. The reproductive factors were also obtained from an ongoing systematic registration of childbirths covering the whole population. The primary data on employment time and reproductive factors were collected prior to follow up and diagnosis of the breast cancer cases, which eliminates the possibility of recall bias. The small size of the study did not allow us to separate out the possible effects of induction period and calendar time.

In our study the increased risk of breast cancer among cabin attendants is related to length of employment before 1971, the period before jet aircraft were taken into operation. We cannot conclude that the exposure related to the increased risk of breast cancer is solely confined to the period before 1971, because long lag time may be required for inducing breast cancer. Our analysis with 20 and 30 year lag time is compatible with this view and corresponds to a long induction period between ionising radiation exposure and development of breast cancer.¹⁴

More detailed information on the different components of the occupational exposure is needed in further studies in order to understand their importance or their possible combined effects.