Commentary: Epidemiology on the side of the angels

Jack Siemiatycki

INRS—Institut Armand Frappier, Université du Québec, 531 Boul. des Prainés, Laval, Quebec H7V 1B7, Canada. E-mail: jack.siemiatycki{at}inrs-iaf.uquebec.ca

Hurtig and San Sebastián1 have examined cancer incidence rates in a region of the Amazon basin of Ecuador and ostensibly found higher cancer rates in the area closer to oil extraction sites than in the area further away; they concluded that this should lead to the establishment of systems of environmental monitoring and control, and of cancer surveillance. The conduct and publishing of this work raise several issues, of which I will comment on three: the strength of evidence that this study affords, the replicability of this study; and the public health recommendations that can be made.

Strength of evidence from this study

Research on cancer in developing countries is difficult. Among the major problems is the fact that diagnosis of cancer is a fairly high-tech and expensive enterprise, not readily available to the majority of inhabitants of many developing countries. Available statistics on cancer incidence and mortality in most developing countries are probably incomplete, of questionable validity, and biased in their representation of variation by social class, geographical sub-region and other factors that may influence access to diagnostic services.

This study’s limitations are clear and are partly acknowledged by the authors. Stripped to its essence, it is a geographical correlation study with: an n of 2, a real possibility of bias in ascertainment of the outcome between the two study areas, a real possibility of confounding by a plethora of ethnic and social factors, and the crudest of measures of exposure. While the overall cancer incidence was ostensibly higher in the ‘exposed’ area, the cancer site distributions did not exhibit a pattern that would obviously throw suspicion on aetiological agents coming from the oil industry pollution. Namely, the greatest excess in the exposed area, representing virtually all of the observed excess among females, was for cancer of the cervix. The only other cancer site which exhibited more than a handful of excess cases in the exposed area was stomach cancer. Given the limitations of the study design and the lack of clear, strong results, this study provides no more than a hint that there may be a cancer problem in the area around oil fields.

Replication of studies—consistency of evidence

Because of the capricious nature of biases that can arise in epidemiological research, the need for multiple pieces of epidemiological evidence to support a hypothesis is well grounded and well recognized.2 But what about causal inferences when there is little opportunity for replication?

No observational epidemiology study is perfectly replicable, in the sense that a laboratory experiment should be perfectly replicable. Differing social/environmental contexts and differing study methods are inevitable even when two studies seem to be similar in objectives and study design. Still, some issues entail a fair degree of replicability. Smoking and lung cancer can be studied in a variety of populations and despite some real differences in the nature of the exposure variable (e.g. type of tobacco, smoking habits) and differences in the nature of the genetic make-up and social covariables, the pattern of results from different studies can be juxtaposed to draw some generalized inferences about smoking and lung cancer. There are many such circumstances that allow us to invoke the criterion of consistency of findings in evaluating causality. However, there are other issues, and this is particularly a problem in environmental epidemiological studies, where the nature of the exposure circumstances or of the covariable circumstances is so unique, that there is no realistic expectation that other epidemiological evidence could be assembled that would be directly informative for the hypothesis under study.

Rare environmental disasters and residential proximity to toxic waste dumps or industrial sources of pollution can be so unique in their chemical make-up and in their ability to pollute neighbouring human habitats, that there may be little or no hope of finding circumstances similar enough elsewhere to conduct studies to replicate what is found in one location.

Most epidemiological studies fall somewhere on the continuum from a completely unique non-replicable exposure and covariate situation to a completely generic and generalizable exposure and covariate situation. In general, the closer the study is to the ‘unique situation’ end of the continuum, the stronger the evidence needs to be on other causal criteria (strength of association, ‘dose-response‘, likely absence of known biases, plausibility, temporality). These various causal criteria should not be seen as a checklist of disconnected items that need to be satisfied in abstraction, one after the other. Each one provides an opportunity to support the causal hypothesis on a continuous rather than binary scale. That is, weaker evidence on one or more of these criteria can be compensated by stronger evidence on other criteria. It is possible to derive an inference of causality even when there is no opportunity to assess consistency; but the evidence should be compelling. Cancer risks following the atomic bombs in Japan provide one example.3 A more recent example comes from an outbreak of tumours of the ureter found among a group of Belgian women who had taken high doses of a Chinese herbal medicine as part of a weight loss regimen.4

Insofar as pollution due to oil extraction processes is concerned, the nature of any human exposure that might occur would depend critically on the engineering processes, control measures, the local geography, and the local human population habits and characteristics. It is questionable whether studies could be mounted in different places that would provide a reasonable estimate of the consistency of the epidemiological evidence for the hypothesis that residence near oil fields in Ecuador causes cancer. In any case, at this time there does not appear to be such evidence. What we are left with is this study in this locale on this population, and some indirect evidence concerning exposure to petroleum products in very different (mainly occupational) circumstances.

Justification for public health recommendations

If the evidence from this study does not provide convincing evidence of health effects due to an environmental exposure, should one use the study as a basis for public health recommendations?

Hurtig and San Sebastian call for further research, possibly at the individual level, to examine cancer risks in relation to oil industry pollution, and for the establishment of a cancer surveillance system in the area. These are reasonable recommendations that flow from the inconclusive study that they carried out.

They also call for ‘an environmental monitoring system to assess, control and assist in elimination of sources of pollution in the area’. Given the graphic descriptions of the extent of environ-mental pollution in the area, this recommendation is difficult to resist. However, it does not flow as a consequence of this study.

There is some tension between two views of the essential nature of epidemiology. One view holds that epidemiology is a scientific discipline whose raison d’être is to describe nature in its sphere of competence (distribution and determinants of health and illness) and another holds that epidemiology is that branch of public health which provides the empirical underpinnings of public health policy. These views are certainly not mutually exclusive, but neither are they perfectly congruent. The ‘scientific’ and ‘activist’ visions occasionally confront each other.

Environmental monitoring and control of pollution from oil fields seem like elementary public health needs irrespective of whether pollution can be demonstrably linked to cancer among local residents. At the very least, the case for monitoring should not depend on whether or not the link to disease has been demonstrated. Indeed, one of the reasons why it is so hard to demonstrate a link is precisely because the epidemiological database does not include adequate information on human exposures.

Epidemiological research is sometimes used as a cover of scientific legitimacy in calling for sensible public health precautions. While this definitely puts epidemiologists ‘on the side of the angels‘, it also risks compromising the scientific credibility of epidemiology. The paper by Hurtig and San Sebastian does not represent the most egregious example of such a tendency. But the apparent reach for suggestive results where such suggestions are at best hints, and the ease with which public health recommendations are made, suggest that the authors may have been leaning on the recommendations before the data were in and the evidence assessed. On the other hand, in the real world of lobbying and public policy, it seems that epidemiologists are sometimes ‘caught between a rock and a hard place’ when they try to simultaneously satisfy their rigorous scientific principles and their public health principles.

General remarks

Research on cancer in developing countries can be very helpful in elucidating potential environmental hazards, either in situations where the observed associations are so strong that they are unlikely to be explained by potential biases,5 or in situations where the quality of data and strong design permit reasonable inferences to be drawn even in the absence of strong associations.6 Epidemiology is an eclectic discipline, using an ever-expanding panoply of methods. In assessing methodological quality, we must make allowances for the resources and local conditions in which the investigators find themselves. To require the same standards of research design everywhere would lead to pockets of the world where there is no information at all on various issues.

The study by Hurtig and San Sebastian represents a bold attempt to use imperfect data to derive scientific knowledge; it is useful in highlighting the issue and drawing attention to the limitations of the data. But it does not provide strong evidence in favour of the hypothesis. Nevertheless, given the complexity of disease aetiology, and the need to discover both universal and local facets of disease aetiology, we should encourage the conduct of research such as this.

References

1 Hurtig A-K, San Sebastián M. Geographical differences in cancer incidence in the Amazon basin of Ecuador in relation to residence near oil fields. Int J Epidemiol 2002;31:1021–27.[Abstract/Free Full Text]

2 Hill AB. The environment and disease: association or causation? Proc R Soc Med 1965;58:295–300.[ISI][Medline]

3 Thompson DE, Mabuchi K, Ron E et al. Cancer incidence in atomic bomb survivors. Part II: Solid tumors, 1958–1987. Radiat Res 1994; 137:S17–67.[ISI][Medline]

4 Nortier JL, Martinez MC, Schmeiser HH et al. Urothelial carcinoma associated with the use of a Chinese herb (Aristolochia fangchi). N Engl J Med 2000;342:1686–92.[Abstract/Free Full Text]

5 Peers F, Gilman G, Linsell C. Dietary aflatoxins and human liver cancer. A study in Swaziland. Int J Cancer 1976;17:167–76.[ISI][Medline]

6 Wesseling C, Antich D, Hogstedt C, Rodriguez AC, Ahlbom A. Geographical differences of cancer incidence in Costa Rica in relation to environmental and occupational pesticide exposure. Int J Epidemiol 1999;28:365–74.[Abstract]