We thank Dr. Morfeld for his comments on our updated mortality study of the U.S. coal miners study.[1] However, we disagree with his assertion that the excess of lung cancer we observed must be attributed to smoking alone. Firstly, despite the smoking prevalence being higher in our cohort than in the U.S. population in 1970, smokers in our population were significantly less likely to be heavy smokers (> 24 cigarettes daily) than men in the general US population (12.4% vs. 28.0%). Secondly, greater weight should be given to the findings from the internal analysis, which controlled for smoking, than to the SMR analysis. Here, as Dr. Morfeld acknowledges, there was a clear dose-response relationship between coal- mine dust exposure and lung cancer. Of note, there was an inverse relationship between radiographic CWP status and lung cancer mortality, implying that CWP and lung cancer were competing causes of death, and leading to weakening of the relationship between dust exposure and lung cancer for the older miners (i.e., those exposed to the high levels of dust existing prior to the 1969 Federal dust regulations).

Table 1 shows the mean cumulative coal dust exposures by year, as requested by Dr. Morfeld. These levels were lower in the last eight years of follow-up and were higher among lung cancer cases than among all subjects in that decade. This is probably explained by the fact that subjects with higher cumulative exposures were more likely to die during the earlier years of follow-up due to the effects of exposure and their age.

Dr. Morfeld correctly stated that we lacked work histories on our miners after enrolment (1969 to 1971), but incorrectly asserted that work histories were absent before enrolment. The lack of work histories in the post-enrolment period may have introduced some misclassification of exposures. However, as we and others have emphasized,[2,3] and as demonstrated by our sensitivity analysis, any such misclassification is likely minimal as most participants accumulated the bulk of their exposure before enrolment.

We agree with Dr. Morfeld, and stated in our article, that the British study of coal miners [4] had better exposure data than our study. While both studies show an excess of lung cancer in their most recent period of follow-up, [1,4] the British findings suggest that the excess is most strongly associated with silica, rather than with coal dust exposure as seen in our study. However, from both studies it is clear that there is an excess of lung cancer among coal miners which is unlikely explained by smoking alone.

Finally, we are confused by Dr. Morfeld's reference to the healthy worker survivor effect as a reason for dismissing our findings. As has been shown [3] the study was subject to a strong healthy worker survivor effect, causing a reduction in mortality in the early years of follow-up.

We remain firm in our conclusion that "Our findings and those from the British coal-miners cohort strongly suggest the need for continued investigation of lung cancer mortality and incidence among coal miners."[1]

Table I: Coal mine dust by year of death among all cohort members and among those for whom the underlying cause of death was lung cancer

Mean Cumulative Exposure

Coal Mine Dust

(mg/m3-yrs)

All Lung Cancer

(n= 8,829) (n= 568) Calendar year of death 1970-1989 89.7 86.9 1980-1999 82.0 75.5 2000-2007 42.8 51.5

Literature

1. Graber, J.M., Stayner, L.T., Cohen,R.A., Conroy, L.M., Attfield, M. D., Respiratory disease mortality among US coal miners; results after 37 years of follow-up. Occup Environ Med, 2014. 71: p. 30-39. 2. Kuempel, E.D., et al., Exposure-response analysis of mortality among coal miners in the United States. Am J Ind Med, 1995. 28(2): p. 167-84. 3. Attfield, M.D. and E.D. Kuempel, Mortality among U.S. underground coal miners: a 23-year follow-up. Am J Ind Med, 2008. 51(4): p. 231-45 4. Miller, B.G. and L. MacCalman, Cause-specific mortality in British coal workers and exposure to respirable dust and quartz. Occup Environ Med, 2010. 67(4): p. 270-6

Conflict of Interest:

None declared

We thank Dr Idrovo for his thoughts on our paper (1) regarding multilevel approaches to ecological studies (2). We agree with Dr Idrovo that incorporating different levels of aggregation to explore the impact of macro-determinants, or "cultural determinants", would be useful and could, in theory, illuminate important factors beyond causal hypothecation at the individual-level. In our study, however, we were unable to fully explore the effects of different levels of aggregation for the following reasons: (1) a priori it is difficult to define levels of aggregation based on (expected) homogeneity of ecological, social, cultural or economical macro -determinants that could be of importance in investigating associations between cancer incidence and population proxies of this exposure, other than solely based on geographical location or on arbitrarily defined cut- offs in indices like the Human Development Index. (2) Our study included only 165 nations, which limits statistical power on aggregation. Hopefully, national data will become available for more countries with time, but even then the total number of countries in the world is insufficient for exhaustive aggregation.

In fact, we carried out a multilevel analysis during the development of the methodology in (1) using geographical location (defined as "Continent") as the highest level of aggregation. The results were not described in (1), but a comparison between our final logistic model (using 1995 as the base year for the exposure proxy) and a multi-level logistic alternative showed similar results for the fixed-effects parameters (gross national income per capita, human development index, and 1995 mobile cellular subscriptions [per 100 people]). Aggregation did not provide any additional information, other than an indication that the between- continent variance is relatively small and only about 50% of the variance between countries within a continent. The effect size of the association between mobile cellular subscriptions (per 100 people) and brain cancer (national age-adjusted incidence rates) was similar for both methods but with reduced statistical power. Furthermore, the Bayesian Information Criterion (BIC) indicates that the single level logistic model had a better fit.

Although we broadly agree with Dr Idrovo's approach to analysing ecological studies we believe the concept of "diseases of civilization" as used by Milham (3) should not be used as an illustration of these macro- determinants. In fact, one of us has published a critique of Milham's paper (4), which we think is an example of the "ecological fallacy" (5) to which Dr Idrovo alludes (2).

References 1. de Vocht F, Hannam K, Buchan I. Environmental risk factors for cancers of the brain and nervous system: the use of ecological data to generate hypotheses. Occup Environ Med 2013; 70(5): 349-56. 2. Idrovo AJ. Three interpretations of an ecological study. Occup Environ Med 2013; in press. 3. Milham S. Historical evidence that electrification caused the 20th century epidemic of "diseases of civilization". Med Hypotheses 2010;74(2):337-45. 4. de Vocht F and Burstyn I. Historical "evidence" that electrification caused the 20th century epidemic of diseases of civilization and the ecological fallacy. Med Hyptheses 2010; 74(5): 957-8. 5. Morgenstern H. Ecological studies in epidemiology: concepts, principles, and methods. Annu Rev Pub Health 1995; 16: 61-81.

Conflict of Interest:

None declared

The very interesting article by de Vocht et al (1) is a good opportunity to discuss possible interpretations of results obtained in ecological studies. The study included 165 nations as observations and found an association between mobile/cellular telecommunications (per 100 people) and brain cancer (national age-adjusted incidence rates). Although in this case authors were interested in the generation of individual-level causal hypotheses, the same results can be interpreted in two more ways based on multilevel causal approach,(2) promulgated by social epidemiologists.

One first alternative interpretation can be called "full-population approach". According to the classic article by Rose, determinants of individual cases are not necessarily determinants of incidence rate.(3) In this approach there is not interest in individual or other level inferences, thus results of one ecological study could be valid in the same aggregation level of observations analyzed. Until I know only one study used national data from Nordic countries,(4) and its inconsistent results can be explained per difficulties to explore latency periods. For this, my conclusion is that de Vocht et al study is the most valid study at national aggregation level.

The second alternative interpretation is interested in an ecological approach but in different aggregation levels. For instance, it occurs when ecological studies based on national data are evidence for national sub- regions inferences. It can be possible but the presence of fallacy is a threat. In this case is needed to explore cross-level fallacies similar to ecological fallacy.(5) A discussion on this same topic is available in two commentaries,(6,7) and inferences on different aggregation-levels can be responsible of heterogeneity observed.

Explanations of these alternatives approaches should not to be based on biomedical concepts. Macrodeterminants and population-level outcomes act according to ecologic, social, cultural or economical processes. Thus an initial explanation of results can be based on a previous study by Milham, where "civilization" is the main determinant of some diseases with high occurrence in recent years. (8)

In conclusion, I agree with the authors that in occupational and environmental health ecological studies can be a useful source of evidence. However their results can offer more evidence if they are analyzed according to different aggregation-level approaches.

References

1. de Vocht F, Hannam K, Buchan I. Environmental risk factors for cancers of the brain and nervous system: the use of ecological data to generate hypotheses. Occup Environ Med 2013 (in press).

2. Diez-Roux AV. A glossary for multilevel analysis. J Epidemiol Community Health 2002;56(8):588-94.

3. Rose G. Sick individuals and sick populations. Int J Epidemiol 1985;14(1):32-8.

4. Deltour I, Auvinen A, Feychting M, Johansen C, Klaeboe L, Sankila R, Schuz J. Mobile phone use and incidence of glioma in the Nordic countries 1979-2008: consistency check. Epidemiology 2012;23(2):301-7.

5. Idrovo AJ. Three criteria for ecological fallacy. Environ Health Perspect 2011;119:A332.

6. Soderqvist F, Carlberg M, Hansson Mild K, Hardell L. Childhood brain tumour risk and its association with wireless phones: a commentary. Environ Health 2011;10:106.

7. Aydin D, Feychting M, Schuz J, Roosli M; CEFALO study team. Childhood brain tumours and use of mobile phones: comparison of a case- control study with incidence data. Environ Health 2012;11:35.

8. Milham S. Historical evidence that electrification caused the 20th century epidemic of "diseases of civilization". Med Hypotheses 2010;74(2):337-45.

Conflict of Interest:

None declared

The finding that nightshift work is linked to increased risk of ovarian cancer1 is one of a long series of studies finding that nightshift work is associated with increased risk of cancer [e.g., Ref. 2]. While reduced production of melatonin is a possible explanation, a better explanation is that since those on night shift sleep during daytime, they spend less time in the sun when they could be making vitamin D. Solar ultraviolet-B (UVB) irradiance is the primary source of vitamin D for most people.

Based on a study of night shift work and the risk of cancer in men,2 it was pointed out that a much better explanation than low melatonin was low solar UVB and vitamin D [Grant, Am J Epi, in press]. Additional support is found in the fact that night shift work is also associated with reduced risk of skin cancer.3 Also, both solar UVB and vitamin D have been found inversely correlated with risk of ovarian cancer.4,5

Thus, those working night shifts should consider taking vitamin D supplements in amounts sufficient to raise serum 25-hydroxyvitamin D concentrations to at least 75 nmol/l if not 100 nmol/l.4 To reach these concentrations could take 1000 to 4000 IU/d vitamin D3.

References

1.Bhatti P, Cushing-Haugen KL, Kristine G, et al. Nightshift work and risk of ovarian cancer. Occup Environ Med 2013 70:231-7.

2. Parent ME, El-Zein M, Rousseau MC, et al. Night work and the risk of cancer among men. Am J Epidemiol. 2012;176:751-9.

3. Schernhammer ES, Razavi P, Li TY, et al. Rotating night shifts and risk of skin cancer in the nurses' health study. J Natl Cancer Inst. 2011;103:602-6.

4. Grant WB. Update on evidence that support a role of solar ultraviolet-B irradiance in reducing cancer risk. Anticancer Agents Med Chem. 2013;13:140-6.

5. Toriola AT, Surcel HM, Calypse A, et al. Independent and joint effects of serum 25-hydroxyvitamin D and calcium on ovarian cancer risk: A prospective nested case-control study. Eur J Cancer. 2010;46:2799-805.

Conflict of Interest:

I receive funding from Bio-Tech Pharmacal (Fayetteville, AR), and the Sunlight Research Forum (Veldhoven) and have received funding from the UV Foundation (McLean, VA), the Vitamin D Council (San Luis Obispo, CA), and the Vitamin D Society (Canada).