We have read the paper by Erren and Morfeld (2011) [1] with interest and offer the following comments on the concerns expressed by them about our work estimating the burden of occupational cancer in Britain [2]. We agree with Erren and Morfeld that burden of disease studies inform public health decision making by providing useful indicators of the contribution of different risk factors and indeed the results from our study are already being used for this purpose.
(i) Causality
Burden of disease studies always face the key issue of which diseases and hazards to include. In our study, a pragmatic decision, after discussion at one of the three international workshops held during the project, was to include all occupationally-related carcinogens and occupational circumstances defined by the International Agency for Research on Cancer (IARC) as group 1 (definite human carcinogen i.e. sufficient evidence of carcinogenicity in humans, causality established) and 2A (probable human carcinogen i.e. limited evidence of carcinogenicity in humans, causality credible but bias and confounding cannot be ruled out, and sufficient evidence of carcinogenicity in animals); these classifications are made following review and assessment of all available evidence of carcinogenicity by an expert working group and are respected worldwide. A supplementary table in our BJC paper [2] gives results for IARC group 1 carcinogens only and we discuss the fact that omission of shift (night) work and breast cancer (classified as group 2A) greatly reduces the overall burden of occupational cancer for women. However, our results, following on from the IARC classification, have increased the awareness of and interest in breast cancer and shift work resulting in more research both in the UK and elsewhere to investigate the nature of the risk e.g. from different shift work patterns; this will contribute to development of optimum shift work patterns to reduce risk.
(ii) Methodological issues
We agree that the methods may be subject to biases and uncertainty and that interpretation of results needs to take these into account. We included a table of biases in one of our papers and the likely impact these would have [3]. We are currently carrying out sensitivity analyses of different sources of bias on our results and expect to publish these soon. In response to Erren and Morfeld¿s particular points:
Levin¿s equation
We are aware of the well documented biases that may result from using Levin¿s estimator for the attributable fraction with adjusted relative risks. For Great Britain, population based estimates of exposed proportions amongst cases that would allow the use of Miettinen¿s AF estimator with adjusted relative risk estimates were rarely, if ever, available, so that using Levin¿s estimator which requires estimates of exposed proportions in the population as a whole was the only currently available option. We have investigated the size and direction of the bias that this introduces into our estimates of AF in practice, based on the range of proportions exposed and relative risks used in our study. If adjusted relative risks are lower than unadjusted, the bias in the AF estimate is downwards, i.e. AFs are underestimated, and vice versa. The relative (%) bias is consistently slightly less than the % difference between the (observed) adjusted RR (RRa) and unknown unadjusted RR (RRu). So for example if RRa/RRu = 0.9 [(RRa-RRu)/RRu=10%], the mean bias across all our estimates would be 9.8%, if RRa/RRu = 0.5 mean bias = -48.4%, etc. Our results indicate that this source of bias is in general small compared with other sources of bias, for example from assumptions about cancer latency and staff turnover and the effect on the estimates of numbers ever exposed, the absence of estimates of the proportion in the workplace exposed where CAREX data was unavailable, and the general scarcity of exposure level measurement data and dose response relative risk estimates [4].
We emphasise in all publications that the results are useful for assessing relative contributions to the cancer burden of the different occupational carcinogens and work in different industry sectors, given that all estimates are subject to some, generally similar, magnitude and direction of bias.
AF estimation based on broad definition of exposure
Introducing bias by collapsing categories as described in the discussions of broad definitions of exposure was not an issue in this study. However categorising industry sectors into broad categories regarding overall `higher¿ and `lower¿ exposures is indeed a limitation of our study. Again this was a pragmatic decision to address data limitations and is discussed in our paper [2]. Although for many carcinogens some dose-response risk estimates may be available there is a paucity of reliable British data on which to estimate the proportion exposed at different levels between and within industry sectors. We have been careful to use estimates of relative risk from studies that were conducted in populations where exposure levels are representative of the level of risk to which British workers are likely to have been exposed. In our extension of the methodology for predicting the future burden under different intervention scenarios we have generally expanded the categories to four [5].
Non-additivity of AFs
As we say in our earlier paper, multiple exposures and other non-occupational risk factors need to be considered when combining the AFs for different risk factors [3]. Cancer is a multifactorial and multistage disease that may not be due to any single sufficient cause but rather a sequence of `hits¿ over a life course. For example, smoking alone may not be sufficient to cause lung cancer and those who get it are likely to have been exposed to several lung carcinogens and possess other characteristics such as some form of inherited susceptibility. The mathematical implication of this is that the sum of attributable fractions for several exposures may be greater than 100%, with the amount exceeding 100% being partly due to synergistic interactions among the risk factors [6].
Two separate issues are included here, namely the bias introduced when combining the AFs for disjoint (non-overlapping) exposures, and the problem of allocating attributable fractions between competing risks when the overall AF estimate may exceed 100%.
Upward bias is introduced from directly summing marginal AFs for disjoint exposures. This bias is reduced, although not eliminated, if the AFs are combined using a product sum [7,4]. We have estimated the size of this bias and it is again small compared to the other potential sources of bias that we have identified above.
As we were only considering occupational exposures and not attempting to identify all contributing carcinogenic risks, for no cancers did our combined estimates of attributable fraction approach 100%. Therefore the necessity to adopt a procedure to allocate synergistic interaction effects between competing exposures did not arise.
Excess deaths versus premature deaths
We agree of course that death is inevitable and that burden of disease estimates per se should not be interpreted strictly as indicating that removal of the exposure will reduce permanently the annual number of deaths attributed to a risk factor [8]. The UK Committee on the Medical Effects of Air Pollutants in their report on the mortality effects of long-term exposure to particulate air pollution suggest that attributable burden in terms of deaths should be considered as the number of excess deaths together with their associated loss of life [9]. They use life table approaches in their estimation. However, we have also used our approach to estimate disability-adjusted life-years (DALY) which consist of years of life lost through death and years of life lived with a disability. We will be publishing a paper reporting this aspect of our study in the future.
In Summary:
We feel that the estimation of attributable disease burden is an important tool in public health for informing risk reduction. We would encourage decision makers to consider all the summary measures studies such as ours provide i.e. disease/hazard risk estimates, proportions of the population exposed, attributable fractions, attributable numbers of deaths and incidence and DALYs, when developing their risk reduction strategies and to take into account biases and uncertainties where possible.
Full details of all the data sources, justification for choice of data and the results of our study will be available shortly in a dedicated supplement of the British Journal of Cancer (Under review).
References
1. Erren C and Morfeld P. Attributing the burden of cancer at work: three areas of concern when examining the example of shift-work. Epidemiologic Perspectives and Innovations 2011, 8:4.
2. Rushton L, Bagga S, Bevan R, Brown TP, Cherrie JW, Holmes P, Fortunato L, Slack R. Occupation and cancer in Britain. B J Cancer 2010, 102: 1428¿1437
3. Rushton L, Hutchings S, Brown TB. The burden of cancer at work; first steps to prevention. Occup Environ Med 2008, 65: 789-800
4. Hutchings S and Rushton L. The burden of occupational cancer in Britain: statistical methodology. B J Cancer (suppl), in press.
5. Hutchings S and Rushton L. Towards risk reduction: predicting the future burden of occupational cancer. American J Epidemiology 2011, 173: 1069-1077.
6. Vineis P and Kriebel D. Causal models in epidemiology: past inheritance and genetic future. Environ Health Global Access Science Source 2006, 5:21
7. Steenland K and Armstrong B. An overview of methods for calculating the burden of disease due to specific risk factors. Epidemiology 2006, 17: 512-519
8. Brunekreef B, Miller BG and Hurley JF. The brave new world of lives sacrificed and saved, deaths attributed and avoided. Epidemiology 2007, 18: 785-788.
9. Committee on the Medical Effects of Air Pollutants. The mortality effects of long-term exposure to particulate air pollution in the United Kingdom. COMEAP, 2010 [http://comeap.org.uk/images/stories/Documents/Reports/]
Competing interests
None
Dr. Hutchings` response after estimating the burden of occupational cancer in Britain: all three areas of concern remain when examining the example of shift-work
Thomas C. Erren, Institute and Policlinic for Occupational Medicine, Environmental Medicine and Prevention Research, University of Cologne
21 February 2012
Are the (i) causal, (ii) practical and (iii) methodological areas of concern [1] with regard to attributing breast cancer risks to shift-work, in particular, and the burden of disease (BOD) calculations, in general, taken care of by Dr. Hutchings` recent response [2]?
`Yes` and `no`.
Here is why:
(a) We appreciate that Dr. Hutchings kindly provided comments on our commentary [1].
(b) We appreciate that Dr. Hutchings provided preliminary information after additional analyses.
(c) We appreciate that Dr. Hutchings envisaged further systematic bias analyses.
(d) We note that Dr. Hutchings concluded that current ambiguities to which we pointed such as Levin`s estimator `is in general small compared with other sources of bias`.
(e) We appear to agree on the fact that body counts are not a reliable measure to evaluate the effect of exposure and exposure reduction efforts.
(f) In view of open questions and loose ends from statements (b), (c), (d) and (e), it appears surprising that results from Rushton et al.`s BOD work [3] are already being used as a basis of public health decision making [2].
(g) In our commentary we referred to several papers by several authors, including Dr. Hutchings. We note that the comments to which we refer here are provided under Dr. Hutchings` name [2].
Overall, with regard to the `yes` above, we await the envisaged further analyses and discussions [inter alia `Full details of all the data sources, justification for choice of data and the results of our study will be available shortly in a dedicated supplement of the British Journal of Cancer`; 2].
With regard to the `no` above, we have further methodological and the same causal and practical questions [1] following Dr. Hutchings` response [2].
To exemplify further methodological challenges, the suggested calculation of disability-adjusted life-years [DALYs] for a specific cause of death [e.g. 4] is potentially biased because the years-of-life-lost component cannot be calculated for a specific cause of death without relying on unprovable assumptions [5, 6].
Moreover, the implication that `synergistic interaction` should only be considered when the combined fraction is higher than 100% would appear to be a non-sequitur (please see concepts of interaction in [7]).
Dr. Hutchings suggested that ambiguities to which we pointed may be small compared with other problems. Whatever the influence of `other potential sources of biases that we have identified` [2], unless proven otherwise by diligent analyses, the potential biases listed in our commentary [1] can not be considered as negligible. In fact, the estimated distortions due to the current use of Levin`s formula, a broad definition of exposure or the application of non-additive versions of the attributable fractions are likely to be non-negligible. With this background, if Dr. Hutchings were right that the potential biases we listed in our commentary [1] are small in comparison to `other potential sources of bias` [2], the overall accuracy of burden of BOD estimates published by Hutchings and Rushton [8] and in the SHEcan report [9] may have to be looked at and interpreted with considerable care.
We do, of course, agree that there are further important uncertainties involved which we did not mention in our commentary. As one example we like to draw the readers` attention to the `background level` of exposure used in the estimation of the number of future excess lung cancer cases due to crystalline silica dust exposure in Great Britain, calculated up to the year 2060 in [10]. This `background level` is taken from Table 4 of a study by Steenland and co-workers [11], a pooled analysis of 10 occupational cohorts exposed to crystalline silica. In our view, it is a strong assumption that the second of the quintiles of the calculated cumulative exposure is a reliable estimate of `background exposure` in relation to the low and high exposure levels defined by Hutchings and Rushton [10] on the basis of other investigations [12, 13]. Indeed, the latter varied significantly from the 2001 IARC multicentre study [11] both with regard to the study populations and the applied methodologies.
With regard to the same causal and practical questions, what we and others [14] consider as necessary conditions before computing attributable caseload estimates and what we addressed in some detail with regard to shift-work and cancer in 2011 [1], namely that `causality` and `measures to intervene or prevent` must be established, continue to be unfulfilled. Indeed, the IARC classification of shift-work that involves circadian disruption as a `probable` human carcinogen implies a critical degree of uncertainty [15] and `optimum shift-work patterns to reduce risk` [2] are not just around the corner.
Finally, with regard to `public health decision making` we see two imperatives: First, sustainable prevention or intervention measures must be based on data and analyses that have stood and can stand the tests of time. Second, with specific regard to cancer, we should focus limited resources on those `culprits` which have been established as carcinogens [16].
With curiosity we look forward to additional information and research in due course.
Thomas C Erren (1) and Peter Morfeld (2)
(1) Institute and Policlinic for Occupational Medicine, Environmental Medicine and Prevention Research, University of Cologne, Germany.
(2) Institute for Occupational Epidemiology and Risk Assessment (IERA), Evonik Industries AG,
Germany.
References:
1. Erren C and Morfeld P. 2011. Attributing the burden of cancer at work: three areas of concern when examining the example of shift-work. Epidemiologic Perspectives and Innovations 8:4.
2. Hutchings S. Response to Erren and Morfeld, 2011. Available under: http://www.epi-perspectives.com/content/8/1/4/comments#680698
3. Rushton L, Bagga S, Bevan R, Brown TP, Cherrie JW, Holmes P, Fortunato L, Slack R. 2010. Occupation and cancer in Britain. B J Cancer 102: 1428-1437.
4. Nelson DI, Concha-Barrientos M, Driscoll T, Steenland K, Fingerhut M, Punnett L, Prüss-Ustün A, Leigh J, Corvalan C. 2005. The global burden of selected occupational diseases and injury risks: Methodology and summary. Am J Ind Med 48(6):400-418.
5. Robins JM, Greenland S. 1991. Estimability and estimation of expected years of life lost due to a hazardous exposure. Statistics in Medicine 10:79-93.
6. Morfeld P. 2004. Years of Life Lost due to exposure: Causal concepts and empirical shortcomings. Epidemiol Perspect Innov 1:5.
7. Rothman KJ, Greenland S, Lash TL. 2008. Modern epidemiology. Philadelphia: Lippincott Williams and Wilkins 3 ed.
8. Hutchings S, Rushton L.2011. Toward risk reduction: predicting the future burden of occupational cancer. Am J Epidemiol 173(9):1069-1077.
9. Cherrie JW, Gorman Ng M, Shafrir A, van Tongeren M, Mistry R, Sobey M, Corden C, Rushton L, Hutchings S. 2011. Health, socio-economic and environmental aspects of possible amendments to the EU Directive on the protection of workers from the risks related to exposure to carcinogens and mutagens at work. SHEcan summary report. Research Project: P937/IOM. Institute of Occupational Medicine, Edinburgh, UK.
10. Hutchings S, Rushton L: 2011. Toward risk reduction: predicting the future burden of occupational cancer. Am J Epidemiol 173(9):1069-1077.
11. Steenland K, Mannetje A, Boffetta P, et al. 2001. Pooled exposure response analyses and risk assessment for lung cancer in 10 cohorts of silica-exposed workers: an IARC multicentre study. Cancer Causes Control 12(9):773-784.
12. Pelucchi C, Pira E, Piolatto G, et al. 2006. Occupational silica exposure and lung cancer risk: a review of epidemiological studies 1996-2005. Ann Oncol 17(7):1039-1050.
13. Kurihara N, Wada O. 2004. Silicosis and smoking strongly increase lung cancer risk in silica-exposed workers. Ind Health 42(3):303-314.
14. Poole C. 2002. The darkness at the end of the tunnel: summary and evaluation of an international symposium on light, endocrine systems and cancer. Neuro Endocrinol Lett 23 (Suppl 2):71-78.
15. Straif K, Baan R, Grosse Y, Secretan B, El Ghissassi F, Bouvard V, et al. 2007. Carcinogenicity of shift-work, painting, and fire-fighting. Lancet Oncol 8:1065-1066.
16. Erren TC, Gross JV, Morfeld P. 2011. In favor of controlling proven, but not probable, causes of cancer. Environ Health Perspect 119(11):A469; author reply A469-70.
Response to Erren and Morfeld, 2011
13 January 2012
We have read the paper by Erren and Morfeld (2011) [1] with interest and offer the following comments on the concerns expressed by them about our work estimating the burden of occupational cancer in Britain [2]. We agree with Erren and Morfeld that burden of disease studies inform public health decision making by providing useful indicators of the contribution of different risk factors and indeed the results from our study are already being used for this purpose.
(i) Causality
Burden of disease studies always face the key issue of which diseases and hazards to include. In our study, a pragmatic decision, after discussion at one of the three international workshops held during the project, was to include all occupationally-related carcinogens and occupational circumstances defined by the International Agency for Research on Cancer (IARC) as group 1 (definite human carcinogen i.e. sufficient evidence of carcinogenicity in humans, causality established) and 2A (probable human carcinogen i.e. limited evidence of carcinogenicity in humans, causality credible but bias and confounding cannot be ruled out, and sufficient evidence of carcinogenicity in animals); these classifications are made following review and assessment of all available evidence of carcinogenicity by an expert working group and are respected worldwide. A supplementary table in our BJC paper [2] gives results for IARC group 1 carcinogens only and we discuss the fact that omission of shift (night) work and breast cancer (classified as group 2A) greatly reduces the overall burden of occupational cancer for women. However, our results, following on from the IARC classification, have increased the awareness of and interest in breast cancer and shift work resulting in more research both in the UK and elsewhere to investigate the nature of the risk e.g. from different shift work patterns; this will contribute to development of optimum shift work patterns to reduce risk.
(ii) Methodological issues
We agree that the methods may be subject to biases and uncertainty and that interpretation of results needs to take these into account. We included a table of biases in one of our papers and the likely impact these would have [3]. We are currently carrying out sensitivity analyses of different sources of bias on our results and expect to publish these soon. In response to Erren and Morfeld¿s particular points:
Levin¿s equation
We are aware of the well documented biases that may result from using Levin¿s estimator for the attributable fraction with adjusted relative risks. For Great Britain, population based estimates of exposed proportions amongst cases that would allow the use of Miettinen¿s AF estimator with adjusted relative risk estimates were rarely, if ever, available, so that using Levin¿s estimator which requires estimates of exposed proportions in the population as a whole was the only currently available option. We have investigated the size and direction of the bias that this introduces into our estimates of AF in practice, based on the range of proportions exposed and relative risks used in our study. If adjusted relative risks are lower than unadjusted, the bias in the AF estimate is downwards, i.e. AFs are underestimated, and vice versa. The relative (%) bias is consistently slightly less than the % difference between the (observed) adjusted RR (RRa) and unknown unadjusted RR (RRu). So for example if RRa/RRu = 0.9 [(RRa-RRu)/RRu=10%], the mean bias across all our estimates would be 9.8%, if RRa/RRu = 0.5 mean bias = -48.4%, etc. Our results indicate that this source of bias is in general small compared with other sources of bias, for example from assumptions about cancer latency and staff turnover and the effect on the estimates of numbers ever exposed, the absence of estimates of the proportion in the workplace exposed where CAREX data was unavailable, and the general scarcity of exposure level measurement data and dose response relative risk estimates [4].
We emphasise in all publications that the results are useful for assessing relative contributions to the cancer burden of the different occupational carcinogens and work in different industry sectors, given that all estimates are subject to some, generally similar, magnitude and direction of bias.
AF estimation based on broad definition of exposure
Introducing bias by collapsing categories as described in the discussions of broad definitions of exposure was not an issue in this study. However categorising industry sectors into broad categories regarding overall `higher¿ and `lower¿ exposures is indeed a limitation of our study. Again this was a pragmatic decision to address data limitations and is discussed in our paper [2]. Although for many carcinogens some dose-response risk estimates may be available there is a paucity of reliable British data on which to estimate the proportion exposed at different levels between and within industry sectors. We have been careful to use estimates of relative risk from studies that were conducted in populations where exposure levels are representative of the level of risk to which British workers are likely to have been exposed. In our extension of the methodology for predicting the future burden under different intervention scenarios we have generally expanded the categories to four [5].
Non-additivity of AFs
As we say in our earlier paper, multiple exposures and other non-occupational risk factors need to be considered when combining the AFs for different risk factors [3]. Cancer is a multifactorial and multistage disease that may not be due to any single sufficient cause but rather a sequence of `hits¿ over a life course. For example, smoking alone may not be sufficient to cause lung cancer and those who get it are likely to have been exposed to several lung carcinogens and possess other characteristics such as some form of inherited susceptibility. The mathematical implication of this is that the sum of attributable fractions for several exposures may be greater than 100%, with the amount exceeding 100% being partly due to synergistic interactions among the risk factors [6].
Two separate issues are included here, namely the bias introduced when combining the AFs for disjoint (non-overlapping) exposures, and the problem of allocating attributable fractions between competing risks when the overall AF estimate may exceed 100%.
Upward bias is introduced from directly summing marginal AFs for disjoint exposures. This bias is reduced, although not eliminated, if the AFs are combined using a product sum [7,4]. We have estimated the size of this bias and it is again small compared to the other potential sources of bias that we have identified above.
As we were only considering occupational exposures and not attempting to identify all contributing carcinogenic risks, for no cancers did our combined estimates of attributable fraction approach 100%. Therefore the necessity to adopt a procedure to allocate synergistic interaction effects between competing exposures did not arise.
Excess deaths versus premature deaths
We agree of course that death is inevitable and that burden of disease estimates per se should not be interpreted strictly as indicating that removal of the exposure will reduce permanently the annual number of deaths attributed to a risk factor [8]. The UK Committee on the Medical Effects of Air Pollutants in their report on the mortality effects of long-term exposure to particulate air pollution suggest that attributable burden in terms of deaths should be considered as the number of excess deaths together with their associated loss of life [9]. They use life table approaches in their estimation. However, we have also used our approach to estimate disability-adjusted life-years (DALY) which consist of years of life lost through death and years of life lived with a disability. We will be publishing a paper reporting this aspect of our study in the future.
In Summary:
We feel that the estimation of attributable disease burden is an important tool in public health for informing risk reduction. We would encourage decision makers to consider all the summary measures studies such as ours provide i.e. disease/hazard risk estimates, proportions of the population exposed, attributable fractions, attributable numbers of deaths and incidence and DALYs, when developing their risk reduction strategies and to take into account biases and uncertainties where possible.
Full details of all the data sources, justification for choice of data and the results of our study will be available shortly in a dedicated supplement of the British Journal of Cancer (Under review).
References
1. Erren C and Morfeld P. Attributing the burden of cancer at work: three areas of concern when examining the example of shift-work. Epidemiologic Perspectives and Innovations 2011, 8:4.
2. Rushton L, Bagga S, Bevan R, Brown TP, Cherrie JW, Holmes P, Fortunato L, Slack R. Occupation and cancer in Britain. B J Cancer 2010, 102: 1428¿1437
3. Rushton L, Hutchings S, Brown TB. The burden of cancer at work; first steps to prevention. Occup Environ Med 2008, 65: 789-800
4. Hutchings S and Rushton L. The burden of occupational cancer in Britain: statistical methodology. B J Cancer (suppl), in press.
5. Hutchings S and Rushton L. Towards risk reduction: predicting the future burden of occupational cancer. American J Epidemiology 2011, 173: 1069-1077.
6. Vineis P and Kriebel D. Causal models in epidemiology: past inheritance and genetic future. Environ Health Global Access Science Source 2006, 5:21
7. Steenland K and Armstrong B. An overview of methods for calculating the burden of disease due to specific risk factors. Epidemiology 2006, 17: 512-519
8. Brunekreef B, Miller BG and Hurley JF. The brave new world of lives sacrificed and saved, deaths attributed and avoided. Epidemiology 2007, 18: 785-788.
9. Committee on the Medical Effects of Air Pollutants. The mortality effects of long-term exposure to particulate air pollution in the United Kingdom. COMEAP, 2010 [http://comeap.org.uk/images/stories/Documents/Reports/]
Competing interests
None
Dr. Hutchings` response after estimating the burden of occupational cancer in Britain: all three areas of concern remain when examining the example of shift-work
21 February 2012
Are the (i) causal, (ii) practical and (iii) methodological areas of concern [1] with regard to attributing breast cancer risks to shift-work, in particular, and the burden of disease (BOD) calculations, in general, taken care of by Dr. Hutchings` recent response [2]?
`Yes` and `no`.
Here is why:
(a) We appreciate that Dr. Hutchings kindly provided comments on our commentary [1].
(b) We appreciate that Dr. Hutchings provided preliminary information after additional analyses.
(c) We appreciate that Dr. Hutchings envisaged further systematic bias analyses.
(d) We note that Dr. Hutchings concluded that current ambiguities to which we pointed such as Levin`s estimator `is in general small compared with other sources of bias`.
(e) We appear to agree on the fact that body counts are not a reliable measure to evaluate the effect of exposure and exposure reduction efforts.
(f) In view of open questions and loose ends from statements (b), (c), (d) and (e), it appears surprising that results from Rushton et al.`s BOD work [3] are already being used as a basis of public health decision making [2].
(g) In our commentary we referred to several papers by several authors, including Dr. Hutchings. We note that the comments to which we refer here are provided under Dr. Hutchings` name [2].
Overall, with regard to the `yes` above, we await the envisaged further analyses and discussions [inter alia `Full details of all the data sources, justification for choice of data and the results of our study will be available shortly in a dedicated supplement of the British Journal of Cancer`; 2].
With regard to the `no` above, we have further methodological and the same causal and practical questions [1] following Dr. Hutchings` response [2].
To exemplify further methodological challenges, the suggested calculation of disability-adjusted life-years [DALYs] for a specific cause of death [e.g. 4] is potentially biased because the years-of-life-lost component cannot be calculated for a specific cause of death without relying on unprovable assumptions [5, 6].
Moreover, the implication that `synergistic interaction` should only be considered when the combined fraction is higher than 100% would appear to be a non-sequitur (please see concepts of interaction in [7]).
Dr. Hutchings suggested that ambiguities to which we pointed may be small compared with other problems. Whatever the influence of `other potential sources of biases that we have identified` [2], unless proven otherwise by diligent analyses, the potential biases listed in our commentary [1] can not be considered as negligible. In fact, the estimated distortions due to the current use of Levin`s formula, a broad definition of exposure or the application of non-additive versions of the attributable fractions are likely to be non-negligible. With this background, if Dr. Hutchings were right that the potential biases we listed in our commentary [1] are small in comparison to `other potential sources of bias` [2], the overall accuracy of burden of BOD estimates published by Hutchings and Rushton [8] and in the SHEcan report [9] may have to be looked at and interpreted with considerable care.
We do, of course, agree that there are further important uncertainties involved which we did not mention in our commentary. As one example we like to draw the readers` attention to the `background level` of exposure used in the estimation of the number of future excess lung cancer cases due to crystalline silica dust exposure in Great Britain, calculated up to the year 2060 in [10]. This `background level` is taken from Table 4 of a study by Steenland and co-workers [11], a pooled analysis of 10 occupational cohorts exposed to crystalline silica. In our view, it is a strong assumption that the second of the quintiles of the calculated cumulative exposure is a reliable estimate of `background exposure` in relation to the low and high exposure levels defined by Hutchings and Rushton [10] on the basis of other investigations [12, 13]. Indeed, the latter varied significantly from the 2001 IARC multicentre study [11] both with regard to the study populations and the applied methodologies.
With regard to the same causal and practical questions, what we and others [14] consider as necessary conditions before computing attributable caseload estimates and what we addressed in some detail with regard to shift-work and cancer in 2011 [1], namely that `causality` and `measures to intervene or prevent` must be established, continue to be unfulfilled. Indeed, the IARC classification of shift-work that involves circadian disruption as a `probable` human carcinogen implies a critical degree of uncertainty [15] and `optimum shift-work patterns to reduce risk` [2] are not just around the corner.
Finally, with regard to `public health decision making` we see two imperatives: First, sustainable prevention or intervention measures must be based on data and analyses that have stood and can stand the tests of time. Second, with specific regard to cancer, we should focus limited resources on those `culprits` which have been established as carcinogens [16].
With curiosity we look forward to additional information and research in due course.
Thomas C Erren (1) and Peter Morfeld (2)
(1) Institute and Policlinic for Occupational Medicine, Environmental Medicine and Prevention Research, University of Cologne, Germany.
(2) Institute for Occupational Epidemiology and Risk Assessment (IERA), Evonik Industries AG,
Germany.
References:
1. Erren C and Morfeld P. 2011. Attributing the burden of cancer at work: three areas of concern when examining the example of shift-work. Epidemiologic Perspectives and Innovations 8:4.
2. Hutchings S. Response to Erren and Morfeld, 2011. Available under: http://www.epi-perspectives.com/content/8/1/4/comments#680698
3. Rushton L, Bagga S, Bevan R, Brown TP, Cherrie JW, Holmes P, Fortunato L, Slack R. 2010. Occupation and cancer in Britain. B J Cancer 102: 1428-1437.
4. Nelson DI, Concha-Barrientos M, Driscoll T, Steenland K, Fingerhut M, Punnett L, Prüss-Ustün A, Leigh J, Corvalan C. 2005. The global burden of selected occupational diseases and injury risks: Methodology and summary. Am J Ind Med 48(6):400-418.
5. Robins JM, Greenland S. 1991. Estimability and estimation of expected years of life lost due to a hazardous exposure. Statistics in Medicine 10:79-93.
6. Morfeld P. 2004. Years of Life Lost due to exposure: Causal concepts and empirical shortcomings. Epidemiol Perspect Innov 1:5.
7. Rothman KJ, Greenland S, Lash TL. 2008. Modern epidemiology. Philadelphia: Lippincott Williams and Wilkins 3 ed.
8. Hutchings S, Rushton L.2011. Toward risk reduction: predicting the future burden of occupational cancer. Am J Epidemiol 173(9):1069-1077.
9. Cherrie JW, Gorman Ng M, Shafrir A, van Tongeren M, Mistry R, Sobey M, Corden C, Rushton L, Hutchings S. 2011. Health, socio-economic and environmental aspects of possible amendments to the EU Directive on the protection of workers from the risks related to exposure to carcinogens and mutagens at work. SHEcan summary report. Research Project: P937/IOM. Institute of Occupational Medicine, Edinburgh, UK.
10. Hutchings S, Rushton L: 2011. Toward risk reduction: predicting the future burden of occupational cancer. Am J Epidemiol 173(9):1069-1077.
11. Steenland K, Mannetje A, Boffetta P, et al. 2001. Pooled exposure response analyses and risk assessment for lung cancer in 10 cohorts of silica-exposed workers: an IARC multicentre study. Cancer Causes Control 12(9):773-784.
12. Pelucchi C, Pira E, Piolatto G, et al. 2006. Occupational silica exposure and lung cancer risk: a review of epidemiological studies 1996-2005. Ann Oncol 17(7):1039-1050.
13. Kurihara N, Wada O. 2004. Silicosis and smoking strongly increase lung cancer risk in silica-exposed workers. Ind Health 42(3):303-314.
14. Poole C. 2002. The darkness at the end of the tunnel: summary and evaluation of an international symposium on light, endocrine systems and cancer. Neuro Endocrinol Lett 23 (Suppl 2):71-78.
15. Straif K, Baan R, Grosse Y, Secretan B, El Ghissassi F, Bouvard V, et al. 2007. Carcinogenicity of shift-work, painting, and fire-fighting. Lancet Oncol 8:1065-1066.
16. Erren TC, Gross JV, Morfeld P. 2011. In favor of controlling proven, but not probable, causes of cancer. Environ Health Perspect 119(11):A469; author reply A469-70.
Competing interests
None.