I enjoyed reading the study of Huizink et al(1)about the Amsterdam air disaster. As pointed out in their table 2, airway disease was one of the symptoms found to be more prevalent in the exposed group. Reactive airway disease can develop after a single large exposure to an irritant. This type of injury has also been noted in the nasal airway. One of the reasons the reactivity persists for many years or never goes away is the fact that following the onset of the hyperreactive state even exposure to very low levels of an irritant will trigger the reactivity. Each time that occurs the inflammatory process that developed with the initial exposure is perpetuated. If this can occur in the airways, why can’t one experience reactive “stress” disease? Stress activates the hypothalamic- pituitary-adrenal axis (HPA). More Recent studies demonstrate that emotional stress is associated with inflammation (2,3). Huizink et al(1) measured C-reactive protein, however, they didn’t control for age, body mass index, total cholesterol etc. The serum ferritin is confounded since it is an acute phase reactant.

If one postulates the development of a hyperreactive HPA axis then lesser stressors, be they physiological, but especially emotional, will trigger HPA reactivity. How frequently this may occur in any individual would depend on many factors, including genetic, pre-exposure personality, subsequent life events, etc. Depending on the frequency and duration of this overstimulation it could lead to the development of all degrees of HPA dysfunction, which in turn could effect the various systems (endocrine, immune, emotional, etc.) leading to multiple health problems and complaints. This would include chronic fatigue, fibromyalgia and other non-explainable, as well as explainable diseases(4,5).

1. Huizink AC, Slottje P, Witteveen AB et al. Long term health complaints following the Amsterdam Air Disaster in Police Officers and Fire-Fighters. Occup and Environ Med 2006; 657-662.

2. Clays E, DeBacquer D, Delanghe J et al. Associations between dimensions of job stress and biomarkers of inflammation and infection. J. Occup Environ Med 2005; 47:878-883.

3. Ford DE, Erlinger TP. Depression and C-Reactive Protein in US adults. Arch Intern Med 2004; 161:1010-1014.

4. McEwen B, Norton Lasley E, eds.: The End of Stress as we Know It. Washington DC: Joseph Henry Press, 2002: 61-65.

5. Chrousos GP, Elenkov IJ. Interactions of the Endocrine and Immune Systems In: De Groot LJ, Jameson JL, eds. Endocrinology, 4th Edition. Philadelphia, USA, W.D. Saunders Company, 2001: 571-583.

(author response)

We appreciate the interest in our recent meta-analysis of occupational trichloroethylene (TCE) exposure and non-Hodgkin’s Lymphoma (NHL)(1). Three criticisms were mentioned as, “serious limitations”: 1) that the alternative descriptions of the Group I occupational cohort studies (multiple industry vs. aerospace, incidence vs. mortality and Europe vs. U.S. studies) should have been characterized and discussed a priori 2) that mortality and incidence data should not be combined in a meta-analysis, and 3) our interpretation that the epidemiologic data are not supportive of a causal relationship is wrong and it is suggested that our analysis provides more evidence of a causal effect between TCE exposure and NHL. These criticisms, however, are not serious limitations and have little relevance in interpreting the meta-analysis findings. We also disagree that our meta-analysis provides further support for a causal association.

The author would have preferred that we consider interpretation of potential sources of heterogeneity a priori. The evaluation of heterogeneity across individual studies is a critical component of any meta-analysis and has not been addressed in previous studies (2,3). We did consider different approaches to stratification a priori, but preferred to discuss the interpretation of this after we had analyzed these (and other) potential sources of heterogeneity. Prior to conducting a quantitative analysis of heterogeneity, interpretation of potential findings would be mere speculation.

With respect to combining incidence and mortality data, we agree that incidence data are generally preferable if available. In our meta- analysis of subgroups of studies, we stratified studies on this characteristic and, in fact, the analysis of mortality studies was more homogeneous (Table 2). More importantly, if TCE exposure were causally associated with NHL, we would expect both incidence and mortality rates to reflect this.

Finally, with respect to causal interpretation of the available data, it is our opinion that causal inference needs to be made based on a comprehensive evaluation of the data. Such an evaluation is not limited to the calculation of relative risk estimates and confidence intervals (or p-values), but also includes evaluation of exposure-response, consistency, and other factors, in addition to potential sources of bias (4,5). Results were inconsistent across the different groups of studies (e.g. considering Group I, Group II and the case-control studies), available data in the Group I studies did not indicate exposure response trends, there were significant limitations with respect to exposure classification across all studies, and there was variation in findings across subgroups within the Group I cohort studies. These observations considered together led us to conclude that epidemiologic data of occupational TCE exposure and NHL were not consistent with a causal association.

Jeffrey H. Mandel, MD, MPH Michael Kelsh, Ph.D. Pamela J. Mink, Ph.D. Dominik D. Alexander, Ph.D.

Exponent Health Sciences

Conflict of interest: The authors have consulted for a number of private and governmental clients on health issues related to occupational and environmental TCE exposure.

References: 1. Wartenberg D. TCE exposure and NHL-supportive evidence. OEM.bmjjournals.com/cgi/eletters/oem.2005.022418v1#349 2. Wartenberg D, Reyner D, Scott CS. Trichloroethylene and cancer: The epidemiologic evidence. Environ Health Perspect 2000;108 (suppl 2):161-76. 3. Wartenberg D, Scott CS. Carcinogenicity of trichloroethylene (Letter). Environ Health Perspect 2002;110:A13-A14. 4. Susser M. 1973. Causal Thinking in the Health Sciences: Concepts and Strategies in Epidemiology. New York: Oxford University Press. 5. Bradford Hill A. 1965. The environment and disease: association or causation? Proc R Soc Med 58:295-300.

Dear Editor,

This carefully conducted investigation confirms the idea that paternally mediated reproductive effects are possible in the painting trades (1,2).

Perhaps, a better indicator of exposure than toluene and its metabolites would be the use of urinary alkoxyacetic acids which stem from aliphatic ethylene glycol ethers typically used in modern water miscible paints (3) and which are confirmed testicular toxins.

The critical toxic mechanism may be the inhibition of succinate dehydrogenase activity (4) which links the mitochondrial respiratory chain with the tricarboxylic acid cycle. The accumulating succinate has important metabolic effects as it impedes the degradation of the hypoxia inducible factor 1 (HIF-1) in proteasomes thereby allowing upregulation of several growth factors relevant in apoptosis and cell growth (5).

References

1 Welch LS, Schrader SM, Turner TW, Cullen MR. Effects of exposure to ethylene glycol ethers on shipyard painters: II. Male reproduction. Am J Ind Med, 1988; 14: 509-528.

2 Lähdetie J. Occupational- and exposure-related studies on human sperm. J Occup Environ Med, 1995; 37: 922-930.

3 Laitinen J, Liesivuori J, Savolainen H. Urinary alkoxyacetic acids and renal effects of exposure to ethylene glycol ethers. Occup Environ Med, 1996; 53: 595-600.

4 Laitinen J, Liesivuori J, Turunen T, Savolainen H. Urinary biochemistry in occupational exposure to glycol ethers. Chemosphere, 1994; 29: 781-787.

5. Brière JJ, Favier J, Benit P. Mitochondrial succinate is instrumental for HIF 1 alpha nucler translocation in SDHA-mutant fibroblasts under normoxic conditions. Hum Mol Genet, 2005; 14: 3263-3269.

Dear Editor,

The authors conclude that 'there was evidence of a persisting risk among process workers first employed since 1953' in the Clydach, South Wales, refinery. Unfortunately, they were unable to incorporate the confounding influence of such well-known predictors of health as attained education, income level and smoking status. An analysis of their data, stratified finely by smoking status, would yield useful information on the independent effect of exposure. The confounding role of smoking behaviour on inferences concerning nickel exposure can be demonstrated by examining a recently published case-control study by one of the authors of lung cancer risk in Norwegian nickel refinery workers.[1]

Table 1, abridged from a corresponding table in Grimsrud et al. (2002)¹, shows clearly that lung cancer risk increases with amount smoked, reaching a 30-fold higher level for the Norwegian study's heaviest smokers.

Table 2 shows that total nickel exposure risk decreases with increasing smoking level, clearly illustrating the complex interaction of disease, exposure and smoking behaviour. The highest exposure risk appears for the Never/Former smokers. However, for this group, the counts of cases in Table 2 (3,23) are almost the same as the counts (4,22) of the cases in Table 1. The significant odds ratios (ORs) for lung cancer were 3.8 for smoking and 5.0 for exposure. Without adjusting for smoking in this group, it is not possible to assess the separate contribution of either exposure or smoking. However, as will be seen, most of the risk associated with exposure comes from this small subset.

This confounding of smoking and exposure effects is not surprising since there was a higher proportion of non-smokers among the controls in the Never/Former smoker group (93 of 234 controls or 40%) than among the cases (4 of 26 cases or 15%) (Table 1). The Light/Medium smoker group also shows relatively more light smokers among its controls (106 of 253 controls or 42%) than among its cases (49 of 146 cases or 34%). The OR for total nickel exposure in this group rises from 12.3 to 22.1 or 1.8 (Table 2). This estimate can be compared with the smoking OR for this group, which rises from 11.7 in Light smokers to 17.7 in Medium smokers or 1.5 (Table 1). These OR values are also not significantly different. The lung cancer risk in the Light/Medium smokers of the study could be attributed to smoking behaviour. Finally, the odds ratio of 0.9 in the heavy smoking group (Table 2) also provides no evidence of risk associated with nickel exposure.

The complete Norwegian case/control dataset would have been required to duplicate exactly the odds ratios estimated using conditional logistic regression models to analyze lung cancer risk as a function of smoking status and total nickel exposure in Table 7 of Grimsrud et al. (2002). However, a simple logistic analysis (Table 3) closely approximates those ORs and their confidence intervals. This good agreement allows us to use our logistic models to examine the relationships between smoking and exposure in the Norwegian study.

These odds ratios by smoking status lead to a conclusion very different from Grimsrud et al.'s. Significantly elevated lung cancer risks are seen for the Never/Former and Light/Medium smoker groups and, for the reasons discussed above, the attribution of this risk to smoking and/or exposure is unknown. These findings could be due to differences in the smoking histories of cases and controls.

A comparison of two logistic models, each containing additive terms for total nickel exposure (2 levels) and smoking (3 levels) as explanatory factors, but one of which also includes a linear interaction term for exposure and smoking, shows that the interaction term is significant (p=0.016).[3] The significance of the interaction term indicates that an analysis that omits it will yield measures of exposure risk that include interaction effects.

Smoking and exposure are strongly related in the Norwegian study. Its authors used a simple model in which the effects of smoking and exposure on lung cancer risk were presumed additive. Our analysis has shown the exposure/smoking interaction effect is significant. When one factor (smoking) has a large effect relative to that of the factor of interest (nickel exposure), the analysis is more informative if done separately for each level of the influential factor.[4] The failure to include this step in the analysis makes it impossible to assess the risk of nickel exposure. This calculation, which could be simply done with available Norwegian data, would inform our understanding of exposure risk in this and other studies, including the current Clydach study.

D.K. Andrews and J.G. Heller

Authors' affiliations

D.K. Andrews, Dept. of Statistics, University of Toronto

J.G. Heller, James G. Heller Consulting Inc., Toronto, Canada, and Dept. of Public Health Sciences, University of Toronto

Competing interests

Inco Ltd financially supported preparation of this letter by the authors. Falconbridge Ltd financially supported research² by the authors underlying this letter.

References

1 Grimsrud TK, Berge SR, Haldorsen T, et al. Exposure to different forms of nickel and risk of lung cancer. Am J Epidemiol 2002; 156: 1123- 1132.

2 James G. Heller Consulting Inc. Report for Falconbridge Limited on Occupational Exposure Limits for Nickel. Falconbridge Ltd., Toronto, Canada. April 8, 2005. [unpublished]

3 The SAS code yields output for two logistic models with and without an interaction term. The first model without the term has a chi squared likelihood ratio of 99.1857 with 4 degrees of freedom (df). The second with the term has a likelihood ratio of 93.4049 with 3 df. Consequently, the addition of the interaction term accounts for a decrease of (99.1857 - 93.4049) = 5.5808 with 1 df. The probability that a chi- square with 1 df exceeds this value is 0.016202, rounded to 0.016 in the text. Alternatively, SAS code yields a p-value of 0.0211 for the test of significance of the interaction term, based on an asymptotic approximation of normality for the distribution.

4 Cox DR, Planning of Experiments. John Wiley and Sons, New Jersey. April 1992. ISBN: 0-471-57429-5.