From the Department of Epidemiology, School of Public Health, University of Alabama, University Station, Birmingham, AL.
Address correspondence to: Elizabeth Delzell, University of Alabama at Birmingham, 1665 University Boulevard, Birmingham, AL 35294.
The chemical industry’s vast economic and technologic achievements pervade modern society, and it is not surprising that its operations have caused serious health and environmental problems. Although the extent of these problems is unknown, recognition that 20% or more of established human carcinogens derive from chemical industry activities, 1,2 episodes of unintentional but disastrous toxic emissions from manufacturing facilities 3–5 and other events have focused concern on understanding and minimizing the health and environmental impact of the production, storage, use and disposal of chemicals.
In this issue, Greenberg et al.6 present the first comprehensive meta-analysis of data from follow-up studies of mortality and cancer incidence among chemical industry workers in the United States, Canada and Western Europe. Their research encompasses 32 years of epidemiologic research. It covers 185 papers published from 1966 through 1997 and includes over one million workers with over 15 million person-years of follow-up. Their results confirm that exposure to carcinogens in the industry has increased workers’ risk of certain forms of cancer and indicate that exposures in the industry have not had a discernible adverse effect on other fatal diseases.
The approach used by Greenberg et al. to select appropriate publications from the pertinent research is reasonable, their evaluation of study quality is laudable, and their analytic procedures are for the most part suitably probing. The results most likely to be controversial are those pertaining to cancer. Greenberg et al. report a meta-standardized mortality ratio (meta-SMR) above 100 for 15 of 22 cancer categories. The increases in observed over expected numbers of deaths are modest, ranging from 2 to 62% for cancers of the larynx, lung, bladder, brain, thyroid, lymphohematopoietic tissue and several other sites. The meaning of these increases is uncertain because of their small size and because of other limitations noted by Greenberg et al., the most problematic of which are exposure diversity, publication and reporting bias and uncontrolled confounding. The small meta-SMR elevations could be due to bias, to workers’ exposure to established carcinogens, to chemical industry agents that have not been identified as human carcinogens, to nonoccupational confounders or even to chance.
Persuasive interpretation is not possible for most of the increases. Nevertheless, Greenberg et al. attribute their results for lung and bladder cancers to exposure to known human carcinogens. Their interpretation, based on examining the exposure setting of the industrial groups having a lung or bladder cancer SMR of 2.0 or more, is not completely convincing. A possibly useful supplemental approach would have been to compute a residual meta-SMR, derived by removing from the meta-analysis studies of workers exposed to established lung and bladder carcinogens. A residual meta-SMR near 1.0 would have supported their interpretation more directly.
Other cancer increases, particularly those based on relatively small numbers of observed deaths, could plausibly be due to bias. Cancer of the thyroid, which has the highest meta-SMR, is a prominent example. Few of the mortality follow-up studies reported results for this form of cancer. This void in reporting could have resulted because both the observed and the expected number were nonzero but small (eg, <3) or simply because there was no observed death. Particularly if the latter theory explains most of the omissions, the meta-SMR for thyroid cancer, based on data from just 11 of 173 mortality studies and on only 28 observed and 17 expected deaths, would be an overestimate of the true SMR for chemical industry workers. Results for multiple myeloma, reported in only 18 studies, also are likely to be due at least in part to reporting bias.
Greenberg et al. note publication bias as a limitation of the meta-analysis and state that they were unable to investigate its impact. One way to have reduced this problem would have been to obtain unpublished reports submitted to federal agencies for regulatory compliance purposes, to ask chemical companies and epidemiologists to identify other unpublished studies and to include all of these in the meta-analysis after evaluating quality. Under the topic of publication bias, Greenberg et al. also mention that their results could be distorted because some of the published studies were motivated by preliminary data suggesting an excess of a disease: “Thus, in general the published literature, and hence our meta-analyses, are weighted with cohorts at higher suspicion for elevated morbidity or mortality rates.”6 No procedure has been devised to assess bias due to the omission from a meta-analysis of studies that were never done. On the other hand, the possibility that a positive association is due, not to a causal relation, but to chance or to bias derived from observing a group of workers in a selected time period during which a problem was suspected, can eventually be addressed by performing analyses of time periods covered by updates of studies having positive associations.
Other interesting and possibly controversial aspects of the meta-analysis are the definition of the chemical industry, the specification of study designs for inclusion in the analysis and limitations that prevented a comprehensive evaluation of subgroups of chemical industry workers with long duration of employment and many years since entering the industry. Greenberg et al. included a follow-up study in the meta-analysis if its subjects worked for a company that was eligible to be a member of the American Chemistry Council. This procedure produced an analysis that covers diverse industrial settings, ranging from the production of chemicals to operations that use large amounts of chemicals to make products such as pesticides, reinforced plastics, munitions, paints, synthetic textile fibers and rubber tires. The definition, however, excludes some industries that are heavy users of chemicals.
Greenberg et al. excluded nested case-control studies from their analysis because such studies are done to determine the effects of exposure to particular chemicals or processes, are slanted by their focus on occupational settings where a medical problem is known to exist and do not provide information directly suitable to the objective of assessing workers’ disease rates in relation to general population rates. This rationale is acceptable, but it highlights the fact that many of the follow-up studies included in the meta-analysis, like the nested case-control studies, could have been done to investigate an association that was suspected before the study began. Nested case-control studies may eventually prove useful in evaluating the impact of employment in the chemical industry on workers’ health, as these studies can estimate the proportion of a disease excess at the industry level that is attributable to chemical exposure. Greenberg et al.’ s exclusion of all population-based case-control studies is more difficult to understand. The stated reason was “because our goal was to evaluate the general health of chemical-industry workers and not risks related to specific chemical exposures.”6 Many population-based case-control studies, however, classify industry and occupation broadly and present results for chemical industry workers as a broad category.
The meta-analysis of Greenberg et al. enhances our perspective on the chemical industry’s impact on human health to some extent but does not, and was not intended to, provide a comprehensive assessment of disease patterns due to chemical exposure in the industrial exposure setting. The analysis does not include all occupational groups heavily exposed to chemicals, and it does not address nonfatal medical conditions other than cancer. Greenberg et al. do, however, accomplish their limited objectives of developing an overview of mortality and cancer incidence patterns among chemical workers and of identifying several relations whose interpretation needs clarification.
2. NTP (National Toxicology Program). Report on Carcinogens, 9th ed. Carcinogen Profiles, 2000. U.S. Department of Health and Human Services, Public Health Service, 2000.
3. Bertazzi PA, Bernucci I, Brambilla G, Consonni D, Pesatori AC. The Seveso studies on early and long-term effects of dioxin exposure: a review. Environ Health Perspect 1998; 106: 625–633.
4. Editorial. Has the world forgotten Bhopal? Lancet 2000; 356: 1863.
5. Dhara R, Dhara VR. Bhopal - a case study of international disaster. Int J Occup Environ Health.; 1995; 1: 58–69.
6. Greenberg RS, Mandel JS, Pastides H, Britton NL, Rudenko L, Starr TB. A meta-analysis of cohort studies describing mortality and cancer incidence among chemical workers in the United States and Western Europe. Epidemiology 2001; 12: 727–740.
© 2001 Lippincott Williams & Wilkins, Inc.