Dear Editor,

Firstly, latent period always refers to the period between the point of the time when disease occurs and point of the time when the disease is detected, while tumour induction time refers to the period between the point of the time when the component cause (can be an exposure) is satisfied and the point of the time when the disease is occurred.[1] Thus only under the extreme condition that one second of mobile use may cause brain tumours, while once the tumour is ‘born’, it can be detected, it would be possible to defined the latency period as the period between ‘first use of a cellular or cordless telephone until tumour diagnosis’.

Secondly, in the research exposure to ‘microwaves’ is classified as a chronic exposure (such as smoking) rather than a point exposure (such as exposure to toxic gas for one minute). While the exposure is measured by the total hours use of different types mobile, and they compared the effect of different exposure dose (lower or above 85 hours) under different length of ‘latency period’. However the effect of certain dose of exposure will only appear after the subjects already exposed to that dose of exposure. For example, the effect of exposure to 85 hours of digital mobile use on the incident of brain tumours, might only possibly appear after the subjects has already been exposed to 85 hours of digital mobile use. Similarly, it is unrealistic to measure the effect of exposure to 10 pack- year smoking on risk of lung cancer, since the first day when the subjects pick up their first cigarettes.

Furthermore, there may be subjects who use more than one kind of the mobile that have been listed in the article, effect of multiple exposures have not been measured in the research.

In addition people with ‘Zero’ exposure may not be a good comparison. People who never use mobile may be rare in today’s society, as to say even control for gender, age, socio-economic status, they may still have special characters different from people who use mobile such as smoking, alcohol use, diet habit, physical activity etc. Above all, exposure is always extremely hard to measure validly, and sometimes it may even hard to define, however these are always essential for the accuracy of a study.

Reference

1. Rothman, J. & Greenland, S. Modern Epidemiology, 1998, Lippincott-Raven.

Dear Editor,

Kyle Steenland raises some interesting points in his commentary on silica [1] both on our papers reporting exposure assessment and mortality in the UK silica sand industry [2,3] and on the adverse effects of silica in general.

With the exception of one quarry, where other exposures such as polycyclic aromatic hydrocarbons could have occurred, no relationship was found with cumulative silica exposure in the UK silica sand study. Steenland points out that the estimated exposure levels were relatively low with only a few workers having a cumulative exposure over 1 mg/m3.years (in fact only 8% of the study population overall and 5 lung cancer cases had a cumulative exposure over 2 mg/m3.years). We agree that this may be why only 2 silicosis deaths were observed, although, as we point out in the paper, fibroses, including silicosis, are poorly recorded on death certificates and therefore could not be accurately assessed.

Steenland draws attention to the current controversy over the relationships between silica, silicosis and lung cancer. The epidemiological literature is indeed inconsistent. OEM readers might be interested to know about a two day workshop organised by the European Association of Industrial Silica Producers (EUROSIL) that was held in New York in August 2004, which brought together leading scientists from Europe, the USA, Canada, China, South Africa and Australia involved with the major studies in the industrial sand, diatomaceous earth, mining, heavy clay, granite, stone, pottery and brick industries [4] (Kyle Steenland was invited but was unfortunately unable to attend). The aim of the workshop was to gain a clear understanding of the epidemiological work to date, and to prioritise future research needs. Following brief presentations summarising the results from these studies, break out groups evaluated the variations between studies in the design, definition and derivation of health outcomes, assessment of exposure, collection of confounding data and statistical methodology. The groups identified the knowledge gaps and discussed the feasibility and desirability of filling these.

The overwhelming conclusion from the workshop was that heterogeneity occurs across all aspects of both the actual nature of the industries in which silica exposure occurs and the design, conduct, analysis and interpretation of the studies that have been carried out.

Heterogeneity in one or more of the following might contribute towards the differing results:

• The physico-chemical features of the silica, including geological species, chemical composition, percentage crystalline silica, particle size, freshness of the fracture.
• Methods of measuring silica exposure, including sampler design, analytical technique, sampling strategy.
• Exposure assessment methodology, including taking account of changes in technology, use of protective equipment such as respirators, retrospective extrapolation.
• Derivation and definition of health outcomes, including accuracy and completeness of death, cancer and other registers, diagnostic procedures such as X rays, CT scans, use of pathological samples.
• Methodology for collection of data on confounding variables such as smoking, which, in many studies, is limited.
• Statistical methodology, including study size, follow-up time, adjustment for confounders and other relevant exposures, appropriate models.

Many of the above are, of course, inherent difficulties in occupational epidemiology in general. However, in the area of silica some of these differences may be so fundamental that it may be necessary to rethink the concept of the silica industry being a uniform entity. As the IARC Working Group pointed out, given the wide range of populations and exposure circumstances some non-uniformity of results would be expected [5] and even within the same industry the results from different studies may vary. Some of these are highlighted below.

At the workshop the group discussing studies in the diatomaceous earth industry highlighted uncertainties in exposure assessment methodologies including the conversion factors used to convert total dust counts to respirable silica, extrapolation methods used for exposures prior to 1950 and lack of adjustment for calcining for exposure before 1930. They were also concerned about co-exposures to asbestos, lack of smoking data, and the difficulty of separating small round opacities from small irregular opacities when reading chest x-rays. Conversion factors and extrapolation in exposure assessment were also of concern regarding the studies in the sand industry as was the origin and composition of the sand. For example, in the North American studies some sands were almost pure quartz, whilst others elsewhere were dune sand or feldspar sands. There were also differences in the aluminium content. Workforces in mining were considered to have the advantages of large and stable workforces, good exposure measurement data and special health programmes and surveillance. Co-exposures, for example, to radon, asbestos and PAHs were highlighted as particular problems, however, as was the variation in the quartz content of dusts from mines extracting tin, gold, or coal.

The group discussing the mining studies also drew attention to the need to investigate exposure metrics such as intensity and peaks of exposure as an alternative to cumulative exposure. Many of the above considerations were also identified as potential problems in the pottery, brick and granite industries. The researchers involved in the Vermont granite studies expressed interest in collaborating to explore the reasons why their studies had produced different results. This group drew attention to the issues of survivor populations and susceptibility. For example, in the studies in China, there is a high incidence of non-malignant respiratory disease at an early age that could impact on the results for lung cancer occurring at older ages.

One of the aims of the workshop was to try and target any future research so that key issues concerning adverse health effects of silica and uncertainties around current knowledge can be addressed. These include:

• In what industrial settings, if any, does silica exposure at current compliance levels, essentially at established ‘Western (US, EU)’ limits, cause cancer?
• What role does silicosis (fibrosis) play?
• Does the lung cancer risk increase for radiographic severity of silicosis?
• Are there pathological differences between fibrosis from silica and asbestos that influence lung cancer?
• How does non-malignant respiratory disease affect lung cancer risks?

From the various presentations and discussions, eight priorities were identified:

1. Effectively control workers’ dust exposure and implement proper evaluation and prevention measures.

2. Harmonise sampling and analytical methods for future collection of dust measurements and develop a standardised job/task industry wide Job Exposure Matrix (JEM).

3. In parallel, collect information on the type and use of personal protective equipment and develop the methodology for incorporating this into the JEM for future exposure assessment.

4. Investigate the toxicological potency of different types of silica using industry samples

5. Focus on industries with similar exposures and review the differences that may have given rise to different estimates of risk.

6. Consider whether pooling of the data might be useful and investigate what this might entail, e.g. development of a harmonised JEM and exposure assessment methodology, bearing in mind that indiscriminate pooling might give misleading and imprecise results.

7. Consider whether current cohorts might be able to re-analyse their data to address the priority areas of concern and/or whether they can collect supplementary data to assist with this.

8. Alternatively carry out a new study(ies) but ensure that there is an agreed protocol and a design that ensures knowledge gaps will be filled.

Some of the above are already being developed. For example, the Industrial Minerals Association in Europe is developing a harmonised dust monitoring strategy and encouraging all its members to implement this. IARC have classified silica as a Group 1 carcinogen, but rather than focusing on hazard identification, workshop participants felt that the focus should be on reducing exposure to a level that would protect against silicosis, and thus probably lung cancer, and be technically achievable across the whole industry. The need for enforcement of current national standards is illustrated by findings of high percentages of samples taken by the US Occupational Safety and Health Administration (OSHA) still exceeding the OSHA Permissible Exposure Limit (PEL) of 0.1 mg/m3 [6]. For example in 1999, 47% and 38% of samples in construction and manufacturing groups, respectively, exceeded the OSHA PEL. The fact that silicosis (and other potentially related diseases) still occurs is likely to be due to the continuation of overexposure that is perhaps not reflected in some of the industries investigated in much of the current epidemiological literature, which tends to focus on producers rather than manufacturers or users.

References

1. Steenland K Silica: déjà vu all over again? Occup Env Med 2005; 62: 430-432

2. Brown TP, Rushton L Mortality in the UK Industrial Silica Sand Industry: 1. Assessment of exposure to respirable crystalline silica. Occup Env Med 2005; 62: 442-445

3. Brown TP, Rushton L Mortality in the UK Industrial Silica Sand Industry: 2 a retrospective cohort study. Occup Env Med 2005; 62: 446-452

4. Eurosil Proceedings of expert workshop: Epidemiological perspectives on silica and health. Brussels, Belgium, European Association of Industrial Silica Producers, 2005

5. IARC Monographs on the evaluation of carcinogenic risk to human, vol. 68: silica, some silicates, coal dust and para-aramid fibrils. Lyon, International Agency for Research on Cancer, 1997

6. NIOSH Work-related lung disease surveillance report 2002 (DHHS (NIOSH) Number 2003-111) Cincinnati, Ohio, National Institute for Occupational Safety and Health, 2003

Dear Editor,

Kojo et al. [1] report their results on breast cancer risk among airline cabin attendants in a nested case-control study. Increased incidence of breast cancer has been repeatedly found among Finnish and other airline cabin attendants and that is the motive of the study. The results do not support the hypothesis that cosmic radiation exposure as measured in the study is strongly linked to the induction of breast cancer among the cabin attendants. I would like to comment on some aspects of this study as well as how the results are interpreted. A cohort of Finnish cabin attendants serves as a study base, cases of breast cancer were found in the Finnish Cancer Registry and controls were chosen with matching on year of birth from non-cases of the cohort. This setting seems to be ideal and it should be possible to identify all cases of breast cancer that occur in the study base. However, exposure information in the study was collected after the cases of breast cancer have been diagnosed. This procedure will influence validity because it opens up the possibility that breast cancer may influence the access and the quality of the information on exposure. This is exactly what happens in the study and I will come to it later.

In the paper Kojo et al. state the following policy implications: “There is no need to take occupational factors into account in breast cancer prevention among cabin attendants.”[1] This is a surprisingly determined generalisation in the light of the small material of the study. The study has not convincingly demonstrated the absence of effect of an occupational exposure on breast cancer risk among cabin attendants. There is a well known definition of a negative study but the study of Kojo et al. [1] does not fit into that definition. A true negative study must be large and sensitive, and it must have accurate exposure data.[2] It seems to me that the study of Kojo et al. is lacking in all three aspects.[1] I would like to point out the following: The authors themselves discuss the smallness of the material consisting of 27 cases. It also seems a rather crude method to estimate cumulated radiation dose on basis of, among other parameters, five questions on number of round-trip flights per month divided in two or three decades. Further there was only 52% participation rate in the questionnaire survey on exposure. The limitation of collecting exposure information retrospectively has been addressed in a recent study on airline cabin attendants.[3] That study did not suffer from low participation.

Kojo et al. [1] have a problem with possible selection bias because of poor participation when estimating the exposure to cosmic radiation. In an attempt to evaluate this problem they calculate the odds ratio for breast cancer for all subjects in the cohort (44 cases and 921 non-cases, all subjects with known start and end of cabin work according to information obtained from Finnair and the Finnish Cabin Crew Union). Based on this information they calculate active work year and combine this with the estimated mean annual cosmic radiation dose by calendar period [4] to obtain a crude estimate of cosmic radiation dose for every person explained in the Methods.[1] Not surprisingly they get a similar odds ratio in the matched case-control study as in the analysis when they calculate the odds ratio using the crude cosmic radiation exposure data in the Results.[1] Both exposure estimates involve information from the questionnaire survey with the poor participation rate. However, this fact does not keep the authors from concluding on the similarity of the two odds ratio in the Discussion.[1] Here we seem to be facing a phenomenon called arguing in a circle. Arguing in a circle occurs when two or more unproved propositions are used to establish each other. In this aspect I would like to point out the interesting difference in the mode of expression concerning these odds ratio in the Result section.[1] In the main study with the smaller number of subjects in the univariate analysis it is: “..cumulative radiation dose …..showed no effect on breast cancer”, and in the multivariate analysis it is: "..cumulative radiation dose...showed negligible effects on breast cancer", whereas the odds ratio based on calculation on crude work years gets: “..using crude cosmic radiation exposure data, the occupational radiation dose was not associated with breast cancer…”.

The authors mention the limitation of their study due to retrospective collection of the information on exposure. The case ascertainment was retrospective and therefore eight cases (18%) were deceased and lost from the study and authors assume the loss to be greater among breast cancer cases due to excess mortality from breast cancer compared to non-cases. This can indeed be calculated from the data given in the paper to be 24 deaths or 2% of the non-cases. Thus proportionally more cases than non-cases were lost because they were deceased and thus it was not possible to collect information from them in regard to exposure. In the study of Kojo et al. [1], breast cancer influences the access to information on exposure and this has possibly introduced bias.

Kojo et al. [1] do not mention the possibility that the disease, breast cancer, can influence the subjects answers, which may introduce errors.[2, 5] This may particularly occur if cabin attendants who get breast cancer or the non-cases suspect that there is an association between occupation and cancer and such suspicion can easily arise based on information via mass media if not from other sources. No information is given in the paper on measures taken to avoid possible influence by the authors knowledge of the aim of the study. There is no comment on whether investigation on the exposure conditions were conducted blind as to the case-control status of the cabin attendants. For example it is not very clear who selected and how the representative routes were chosen. These routes were later used to calculate radiation dose.[4] One can only speculate whether these have introduced bias. However, the authors excluded from the study those who had worked for less than two years as airline cabin attendants and it appears not to have involved cases. This exclusion based on exposure variable introduces bias towards the null hypothesis.

In the conclusion of the Abstract and in the last sentence of the Main messages it says that three occupational factors are studied and it is stated that there is no clear evidence that they affected the breast cancer risk. It is not a simple task to find where in the paper the authors give a clear account of all three occupational factors, however, it is easy to identify the cumulative radiation dose in mSv as an occupational factor. The other two occupational factors are the disruption of the sleep rhythm and the disruption of the menstrual cycle. Inexperienced reader may be confused whether these factors belong to outcome or exposure and the same may be valid for cases and non-cases. Are these disturbances, identified with the other exposure data retrospectively, so clearly related to the occupation as to serve as a surrogate of exposures? Is it possible to escape sleep disturbances as a consequence of long haul flight? Is it possible that breast cancer cases get sleep disturbances of causes other than occupational? Is it possible that these disturbances are not suitable exposure indicators?

Kojo et al. [1] divided the material into two parts in an attempt to evaluate the possible longer recall period concerning flight activity, disturbances of menstrual cycle and sleeping among those over 50 years of age as compared to younger women. The next step was to calculate odds ratio for breast cancer in the two groups (50 years of age and younger, and over 50 years of age) associated with each of the three factors separately i.e. cumulative radiation dose, sleep disturbances, and menstrual disturbances. These calculations, with less than 27 cases in each group, (the number in each group is not available in the paper) yield six different odds ratio and wide 95% confidence intervals, which all include unity. However, the authors conclude that the estimates were comparable suggesting that the lower participation among those older than 50 years did not bias the results. The longer recall for the older women is not mentioned in this respect, which was the goal in the outset. Here the authors conclude firmly based on small material arguing for the validity of their study. In this discussion, on what Kojo et al. [1] call modifying effect of age, we are shown the range of exposure in the groups, 0-103.5 mSv for women 50 years of age and younger, and 0-136.8 mSv for women over 50 years of age. It is rather confusing to see the range go down to zero, given that airline cabin attendants with less than two years career were excluded from the study.

In the Discussion Kojo et al. [1] inform us on the fact that the excess risk in the incidence of breast cancer among Finnish cabin attendants has persisted based on updated follow up. They suggest that this risk is related to well known risk factors of breast cancer such as family history of breast cancer and possibly to moderate or heavy alcohol consumption. And the authors do not compare their findings with information from other studies on breast cancer among cabin attendants ignoring the benevolent recommendation given long ago.[5]

References

1. Kojo K, Pukkala E, Auvinen A. Breast cancer risk among Finnish cabin attendants: a nested case-control study. Occup Environ Med, 2005:62;488-493.

2. Hernberg S. Introduction to Occupational Epidemiology. Chelesea: Lewis Publisher, 1992.

3. Grajewski B, Atkins DJ, Whelan EA. Self-reported flight hours vs. company records for epidemiologic studies of flight attendants. Aviat Space Environ Med, 2004;75:806-810.

4. Kojo K, Aspholm R, Auvinen A. Occupational radiation dose estimation for Finnish aircraft cabin attendants. Scand J Work Environ Health, 2004;30:157-163.

5. Breslow NE, Day NE. Statistical methods in cancer research, Vol. I. The analysis of case-control studies. Lyon: International Agency for Research on Cancer, 1980.

Dear Editor,

We thank Dr. Rafnsson[1] for valuable comments on our paper.[2] Rafnsson finds our policy implications surprising. In the light of present evidence, we do not find further measures justified for reducing radiation exposure among cabin crew. The justification for this view is the fact that exposure limits common for all radiation workers, also apply for the cabin crew. Dose monitoring indicates that the cosmic radiation doses are within the exposure limits. We see no reason to depart from the general radiation protection principles.

Cohort studies have shown an excess risk of breast cancer in cabin crew, in particular among those with long employment, with 1.5 – 3.4 fold incidence compared with the general population.[e.g. 3] Nevertheless, the radiation doses received are low and the expected effect based on previous literature is very small, with RR well below 1.1.[4] Neither previous studies nor our study have been able to identify the cause for the excess incidence of breast cancer. Lack of association in our study does not exclude the contribution of cosmic radiation in the development of breast cancer, but it implies that other risk factors are likely to have a greater role.

Rafnsson[1] finds our approach to occupational radiation dose estimation crude. For cabin attendants, the only available source of information on the number of flights during their career is the cabin attendants themselves and thus, the questionnaire approach in exposure assessment was appropriate. We used self-reported numbers of flights by route and calendar period. We feel this is an improvement compared to previous studies[e.g. 5-7], none of which have had any estimates of the individual cosmic radiation dose. They have been based on surrogate indicators such as length of employment or flight route type assignment.

Rafnsson[1] claims that we did not consider the possibility that breast cancer can influence the subjects’ answers. Recall bias is intrinsic in all case-control studies with subjects as source of information and the issue was discussed in our paper.[2]

We excluded cabin attendants who worked for less than two years because they had negligible exposure (due to very short period of cabin work). In addition, several studies have shown that short-term employees differ in terms of mortality and cancer risk from those with more stable employment. Therefore, they are commonly excluded from occupational cohort studies to avoid bias.

Our study has shortcomings inherent to retrospective case-control study and to sparse data. Therefore, it cannot provide conclusive evidence but does nevertheless supply new information. A large prospective follow- up study with a large data set would be valuable. Currently, a retrospective study, combining all the Nordic cabin crew cohorts with comprehensive cancer incidence registration systems and improved dose estimation algorithm is ongoing, and may be able to provide further insight to the issue.

References

1. Rafnsson V. Retrospective assessment of exposure. Occup Environ Med, electronic letter 25 Jul 2005.

2. Kojo K, Pukkala E, Auvinen A. Breast cancer risk among Finnish cabin attendants: a nested case control study. Occup Environ Med 2005;62:488-493.

3. Pukkala E, Auvinen A, Wahlberg G. Incidence of cancer among Finnish airline cabin attendants. BMJ 1995;311:649-652.

4. Boice JD Jr, Blettner M, Auvinen A. Epidemiologic studies of pilots and aircrew. Health Phys 2000;79:576-684.

5. Haldorsen T, Reitan JB, Tveten U. Cancer incidence among Norwegian airline cabin attendants. Int J Epidemiol 2001;30:825-830.

6. Rafnsson V, Sulem P, Tulinius H, et al. Breast cancer risk in airline cabin attendants: a nested case-control study in Iceland. Occup Environ Med 2003;60:807-809.

7. Reynolds P, Cone J, Layefsky M, et al. Cancer incidence in California flight attendants (United States). Cancer Causes Control 2002;13:317-324.