Letter to the Editor: Re Fordyce et al. (2019) Vermont Talc Miners and Millers Cohort Study Update

Readers are invited to submit letters for publication in this department. Submit letters online at http://joem.edmgr.com. Choose “Submit New Manuscript.” A signed copyright assignment and financial disclosure form must be submitted with the letter. Form available at www.joem.org under Author and Reviewer information. To the Editor: I read with interest the exchange of letters in your journal between Egilman and colleagues (Journal of Occupational and Environmental Medicine, Publish Ahead of Print DOI: 10.1097/JOM.0000000000001783) and Fordyce and colleagues (Journal of Occupational and Environmental Medicine, Publish Ahead of Print DOI: 10.1097/JOM.0000000000001784) concerning the latter's publication on mortality among Vermont talc miners.1 An important part of the exchange had to do the use of Reference Rates in the calculation of Standardized Mortality Ratios. Fordyce at al wrote: ‘Many of the criticisms leveled at us by Egilman et al. make it evident that they do not understand the basic principles of occupational cohort standardized mortality ratio (SMR) studies. Their letter provides little evidence that any of the authors has either conducted or analyzed an occupational cohort study.’ I performed a PubMed search and discovered that Fordyce and Moolgavkar had done only a few occupational cohort studies between them, and that the Vermont cohort was the first asbestos-exposed cohort that they had looked at. They thus demonstrate an incomplete understanding of the issues when dealing with the SMR for mesothelioma in a cohort study. I have been publishing the results of studies of cohorts occupationally exposed to asbestos since 1981 when I described the mortality experience of a cohort of workers who received compensation for asbestosis in Ontario..2 Subsequent studies included those of mortality in a cohort of asbestos cement factory workers3; a cohort of employees at an amosite insulation manufacturing plant4; mortality among employees manufacturing electrical conduit from coal tar pitch and asbestos,5 a cohort of Ontario insulators6; a cohort of employees at an Ontario brake manufacturer7; and, cohort studies of construction workers including pipe fitters8 and brick layers.9In none of these studies did I compute an SMR for mesothelioma. Instead I tabulated the number of cases of mesothelioma, or calculated mortality rates. The reasons for this go to the heart of the disagreement between Egilman et al and Fordyce et al. and are discussed below. Fordyce at al. write: ‘As an occupational cohort is followed forward in time, the observed number of deaths from specific causes (as classified by the International Classification of Diseases [ICD] code in use at the time of death) is compared with an expected number of deaths derived from an appropriate comparison group. Egilman et al. appear to have great difficulty with understanding the requirements for an appropriate comparison group for SMR occupational cohort studies. In SMR studies, it is customary to choose the general underlying population from which the cohort members are drawn as the comparison group.’ I agree that it is customary to choose the general underlying population as the comparison group, but in the case of mesothelioma it is far from obvious that this is an appropriate comparison group. When studying a population exposed to a potentially toxic substance, one could ask: 1) How does the disease risk in this population compare to the disease risk in another population also exposed to this toxic substance? Or, 2) How does the disease risk in this population compare to the disease risk in an unexposed population? Fordyce and colleagues wrote ‘It is absurd to suggest, as Egilman et al. do, that rates of asbestos-related diseases in the Vermont talc miners and millers should be compared with rates in populations that were never exposed to asbestos.’ It is not absurd at all to make that comparison. Animal experimenters testing for the carcinogenicity of a new substance do not use a group of control animals that have been exposed to carcinogens. In comparing the Vermont cohort to the general population, Fordyce and colleagues chose to answer question 1: How did the rate of mesothelioma in the Vermont cohort compare to the rate of mesothelioma in the general population of Vermont and the United States? The general population is an asbestos-exposed population which includes individuals with occupational exposures such as insulators, construction workers, manufacturing workers, refinery workers, shipyard workers, men with naval exposures, and others. They address the question as to whether the risk among the Vermont miners is the same as, greater than, or lesser than the risk is a population including these asbestos-exposed individuals. It is my opinion that when attempting to discover whether exposure in an environment containing an alleged toxic substance increases the risk of cancer it is more informative to ask Question 2: How does the risk of mesothelioma in the Vermont cohort compare to the risk in a population unexposed to asbestos? Fordyce and colleagues state that ‘the reason for choosing local and national general populations as the comparison groups is that the members of the occupational cohort are drawn from the general population and would be expected to have the same employment profiles as the general population before and after membership of the occupational cohort under study.’ I have never seen this rationale stated during my 40 years as an epidemiologist and I think that this statement doesn’t bear up to scrutiny. I have done a cohort study of cancer risk in workers with silicosis.10 I used Provincial rates for comparison, but I had no expectation that the workers would have the same employment profiles as the general population (eg miners, doctors, lawyers, school teachers, airline pilots etc.) before and after membership of the occupational cohort under study. If I were conducting a study of school teachers to explore any cancer risk associated with the use of asbestos modeling materials in the classroom, I would not want to compare them to a population that included insulators, asbestos factory workers, and shipyard workers. The real reason for choosing the general population for comparison is because there are few other choices for a comparison group that is large enough to give relatively stable outcome rates, particularly for less common outcomes. Recognizing that there may be rate differences caused by systematic differences between the populations, particularly with respect to important confounders such as smoking, it is sometimes possible to make post hoc adjustments if additional information about the confounder, such as data about differential smoking rates, are available10 Mesothelioma is a rare disease with three known causes. These include asbestos, other mineral fibers such as erionite, and ionizing radiation. It is not possible to find, for epidemiologic use, an American population without asbestos exposure since ambient exposure to asbestos has been ubiquitous. It is my opinion, that the best we can do to approximate an unexposed population is to use mesothelioma rates among women as the comparison. Some women have had occupational exposure to asbestos, others have had domestic exposures, and many have used cosmetic talc, but it is almost certainly true that the average population asbestos dose among women will be lower than among men. It is thus useful to use American women as a proxy for the comparison of the miners to an asbestos-unexposed population. I recognize that the mesothelioma rates among American women are likely to be higher than the rates among women who truly have no asbestos exposure. Boffetta and colleagues have published an International Analysis of Age-Specific Mortality Rates From Mesothelioma.11 Age-standardized mesothelioma rates among women in the United States have been 1.7 per million from 2002 – 2013. Fordyce at al reported one case of mesothelioma among 17,170 person-years of observation in the Vermont cohort.1 The rate ratio, using American women as the reference, is thus approximately 1/17000 over 2/1,000,000 = 29. I used Stata Statistical Software to compute 95% (0.50 – 561) and 90% (1.01 – 1.73) Confidence Intervals on this Rate Ratio. An alternative approach, using the data of Fordyce is to note that the mesothelioma rates of women tabulated by Boffetta and colleagues are about 20% of those of men. Fordyce et al calculated the expected number of mesotheliomas for men in the Vermont cohort to be 0.17. If we use female rates as the comparison, then the expected number of mesotheliomas would be about 0.04. This leads to an SMR of 20 and a Fisher Exact 95% Confidence Interval of (0.6 – 130). The 90% Confidence Interval, corresponding to the 1-tailed hypothesis that work in the Vermont talc industry increases the risk of mesothelioma, is 1.3 - 120. Fordyce and colleagues wrote ‘Even if an SMR analysis demonstrated an increased risk of mesothelioma in the Vermont talc workers cohort – counterfactually, as we do not acknowledge that there is any evidence of increased risk – further investigation would need to be performed before it could be concluded that such increased risk could be attributed to any exposure sustained while working as a member of the cohort.’ There is, of course, evidence of increased risk in the cohort with an elevated SMR. The cohort is so small (as evidenced by the width of the Confidence Interval) that the statistical Power to find statistical significance is very low. The observation of only a single case in Vermont makes the assessment of risk very imprecise, but the point estimate of risk is high and is compatible with an elevated occupational risk of mesothelioma among these miners and millers. I agree with the authors that further investigation would need to be performed before it could be concluded that such increased risk could be attributed to any exposure sustained while working as a member of the cohort.

Récupéré en direct depuis OpenAlex et désinversé. Les résumés ne sont pas conservés dans cette base de données : les index inversés représentent 8,6 Go des 9,3 Go de texte de la base, et le serveur dispose de 13 Go libres.