12.0 EPIDEMIOLOGY

Epidemiology is the study of the occurrence of disease in relation to factors affecting the individual, his environment and his lifestyle. Epidemiology may be used to infer a causal link between these factors and the disease, although the biological mechanism responsible may not be identified. This, however is a difficult and often controversial process. To help distinguish causal from non-causal associations, Hill (1965) suggested nine criteria:

1. strength of the association,
2. consistency,
3. specificity,
4. temporality,
5. biological gradient,
6. plausibility,
7. coherence,
8. experimental evidence, and
9. analogy.

However, meeting these criteria need not be seen as necessary prerequisite to accept the association as causal (Rothman 1986). In particular, criterion 3 (specificity), is regarded by Rothman as "useless and misleading" and experimental evidence is seldom available for human population. Hill himself admitted that "none of my nine viewpoints can bring indisputable evidence for or against the cause-and-effect hypothesis and none can be required as a sine qua non". Lanes (1985) argued that causal inference is not part of science, but a question for public policy. While this may be an extreme position, it highlights how this question is not easily addressed using rigorous scientific thought. However, it is incumbent upon the scientist to investigate how the quality of the study may have reflect upon Hill's criteria.

Epidemiological studies of exposure to radiofrequency radiation

A number of epidemiological studies have conducted among occupational groups believed to be exposed to above-average levels of RF radiation. Some studies have reported health effects consistent with the well understood thermal hazards. For example, RF levels in excess of the recommended limits have been measured in proximity of diathermy units used in physiotherapy (Stuchly et al 1982, Delpizzo and Joyner 1987), an occupation that has been found to have an above average risk of adverse pregnancy outcomes (Kallen et al 1982).

More intriguing are the studies suggesting the possibility of hazards resulting from low-level environmental fields. From a public perception's view point, these may be roughly divided into two categories: general health surveys and cancer studies. The first group (for reviews see WHO 1981 and WHO 1993) have not established any clear evidence of adverse health effects, with the exception of some studies (Gordon 1966, Marha et al 1971, Dumanski et al 1975, Serdjuk 1977, Deroche 1971, Moscovici 1974) that reported an excess of some autonomic and central nervous system complaints and symptoms generally consistent with those of mild depression, that came to be collectively known as "microwave sickness". Because of the subjective nature of these symptoms, the reliability of these studies is difficult to assess. Moreover, most of these studies have not been reported in the mainstream scientific literature. The possibility of a true association between exposure to RF radiation and psycho-physiologic nervous system reaction cannot be ruled out. For epidemiology to fruitfully address this question, retrospective cohort studies need to be carefully designed to prevent recall bias.

Of much greater concern in the public eye is the claim that exposure to low levels of RF radiation may result in an increase in the incidence of at least some cancers (particularly leukemia and brain tumours). This rest on three elements:

• very few in vivo and in vitro studies
• some anecdotal evidence of brain tumors that because of their position and time of diagnosis were linked to the use of cellular telephones.
• a number of epidemiological studies that suggest an association between employment in occupation presumed to be exposed to above average electromagnetic radiation

With the exception of one study (Lester and Moore 1982), all epidemiological studies on RF exposure were conducted on occupationally exposed subjects. Lester and Moore (1982) reported an increase in cancer mortality in counties within which and Air Force base was located. Their finding was rejected by a subsequent analysis of the same data by Polson and Merritt (1985). Both of these studies have limited credibility. The ecological approach used by Lester and Moore is of intrinsically low quality and is fraught with dangers. Many other potentially relevant factors were not (and could not be) taken into proper consideration. On the other hand, the type of post-hoc re-evaluation carried out by Polson and Merritt needs to be regarded cautiously, since it is open to the possibility of observer bias.

The studies on occupational EMR exposure have been the subjects of several reviews (WHO 1981, 1993). One of the most recent and complete evaluations was done by the Advisory group on Non-ionizing Radiation for the National Radiation Protection Board (NRPB) of Great Britain. With respect to occupational exposure, NRPB speculated that "the large number of observations relating to leukemia... may well be due to bias in favor of publishing results that suggest a hazard, rather than the reverse". However it regards the evidence of a brain cancer risk as "a little more consistent" and comments that "there is some to suggest that the excess increases with duration of employment". It concluded that "the evidence suggests that occupational exposure to EM fields may cause a hazard of brain cancer, but it is far from certain that it does". This statement, together with NRPB's statement that "animal studies conducted at frequencies above about 100 kHz have provided some evidence for effects on tumor incidence" may be seen as supporting the anecdotal evidence of brain tumors arising from the use of cellular phones. The NRPB Advisory Group Chairman, Sir Richard Doll was quoted as saying, that with regard to the possibility of a cancer risk above 100 kHz, "there is room for more doubt" (MW News March/April 1992, p 9).

While doubt cannot be entirely eradicated at this stage, it is hard to see how epidemiological results could support this view. The first difficulty a reviewer faces is that of determining which studies should be considered. Although some of the occupations used as markers for exposure may entail exposure to RF radiation, this has not been verified. It most cases, EMF exposure is limited to ELF electric and magnetic fields. Therefore it is reasonable to argue that studies on electronics and telecommunications employees are studies on ELF exposure and not RF exposure. In the following tables, summarizing cohort and case-control studies of occupational exposure, the likelihood of exposure to RF/µw radiation and to ELF fields are tentatively assessed for each occupational group. Of the groups for which RF exposure is likely, only radio amateur operators and military personnel have been associated with an increase in the cancer incidence (these groups have been marked ® in the tables). In a few studies (marked ß ) the occupation description was broad enough that the possibility of some of the subjects being exposed cannot be ruled out. Nevertheless, it seems reasonable to conclude that, to the extent that these studies carry any information, when placed into the wider picture, they seem to indicate that a possible association between ELF and cancer may have confounded the RF studies, but not vice-versa. It is also interesting to note that Thomas et al (1987) found that subjects exposed to RF radiation in jobs other than those involving design, manufacture, installation or maintenance of electronic or electrical equipment, did not have an elevated brain tumor risk. This argues against exposure to µw/RF radiation being the causal factor for the observed association, although stratification of samples that are already small may easily lead to observations that appear to be inconsistent, purely because of sampling variations.

TABLES TO BE ADDED LATER

Table 12.1 Occupational groups investigated in cohort studies

Table 12.2 Occupational groups investigated in case-control studies

In summary, the epidemiological evidence of possible adverse health effects of electromagnetic fields specifically in the RF frequency band, as distinct from EMR in general and ELF fields in particular, is almost non-existent. The few associations between employment in the so called 'electrical/electronics occupations' and cancer may raise the hypothesis that exposure to µw/RF radiation is associated with some cancers, but the evidence is so weak that it barely makes a case for additional research. Nevertheless, continuing public concern, the absence of good quality epidemiological studies that may allay this concern, the results of a few animal studies and, more importantly, the in vitro studies showing that biological processes are affected by RF fields at levels below those required for significant tissue heating, will ensure that future studies will be carried out. It is therefore important to identify the major methodological issues that need to be addressed to ensure that these studies may make a meaningful contribution to understanding this issue.

Criteria for future studies

The results of an epidemiological study may be affected by bias, misclassification of exposure or of endpoint and limited sample size.

Sample size considerations

Epidemiological studies, particularly observational studies are essentially exercises in statistics. Their results, whether suggesting the presence or the absence of an association, must be qualified by the possibility that they may simply represent a chance finding. Reporting an association when one does not truly exists is called a type I error, while reporting no association when one truly exists is termed a type II error. For a given sample size, reducing the possibility of one type of error increases the probability of the other type. Which course of action is more appropriate is a matter for debate. Scientists traditionally presume the null hypothesis (no effect) and are required to prove that a health effect exists; however, public health professionals err on the side of safety, and tend to take protective action. If studies are intended to reassure the public must have a low probability of type II errors.

Unfortunately, the choice of significance tests is usually dictated by convention, rather than a case-by-case critical analysis. Usually, the level of significance is set at 0.05, two-sided and a confidence interval is calculated accordingly. This means that an association (usually expressed as a relative risk) is accepted as significant only if it is estimated that there is no more than a 5% chance that the confidence interval does not include the true relative risk. This practice leads to study results falling into two distinct groups: significant and non-significant. This is appropriate in situations where decisions (for example, to use an experimental treatment) are to be made on the basis of the study result. However, this approach is not always suitable in epidemiological research. Only if the confidence interval contains values that lies near unity, can a study result argue against an effect (Ahlbom 1993). If the interval is large, encompassing elevated relative risks as well as unity, the result is merely uninformative.

In practice, many (probably most) studies fall well short of having both a high power and high significance level. As an example, consider a study of 250 cases and as many controls in which a subject is classified as 'exposed' if he is in the top 5% of the exposure range. Assume that we are interested in detecting whether being exposed entails a relative risk of at least 2 (ie, we regard a smaller relative risk as clinically non-important). If we require the study to be significant with a confidence level of 95% (two-sided), such a study has only a 50% chance of satisfy this requirement if the risk truly exists. Conversely, if one wanted to limit the probability of a false-negative result to no more than 10%, there would be only a 50% probability that the confidence interval included the true relative risk. The wisest approach would be to acknowledge the limit of such a study and settle for a confidence level of about 77% which gives the study a similar statistical power. Confidence level decisions need to be taken a priori and convincingly explained, if the study is to be credible.

Exposure measurement errors

Direct measurement of thousand of subjects for the whole exposure period is usually impossible. Exposure is usually inferred from a point-in-time measurement or even less direct methods as job descriptions and activity diaries. Consequently, exposure is inevitably affected by non-negligible measurement errors. When exposure is measured on a categorical scale, measurement errors may lead to misclassification of exposure. Measurement errors and misclassification may be differential (if they affect cases and controls differently) or non-differential. Provided that measurement errors are non-differential, they will not result in an artefactual association and in general will produce conservative risk estimates. Exposure measurement errors also reduce the statistical power of the study (Quade et al 1980, Freedman et al 1990, Delpizzo & Borghesi 1993) and the dose response relationship (Dosemici et al 1990, Armstrong 1990, Delpizzo 1992).

It is difficult to determine a priori what are acceptable exposure measurement errors and it may be very costly and not very effective to attempt to reduce them. At this stage, when the most important aim is to establish reliably whether an association between exposure and cancer does exists, it would be advisable to use available resources to increase the sample size to counteract the loss of statistical power.

Confounding and Bias

The term "bias" may have different meanings in epidemiology. In this case, we use the term to indicate inaccurate inferences resulting from flaws in study design, data collection, analysis and interpretation, review and publication of results. A typical bias encountered in occupational studies is the so-called "healthy worker effect". Full time workers are less likely to suffer from a number of conditions that would prevent their participation in the workforce and of a number of conditions (eg, depression, alcoholism, poverty, homelessness) sometimes associated with unemployment; as a consequence, cancer mortality may be proportionally more common than in the rest of the population. "Confounding" can also have several meanings. In the present context, it indicates a situation in which a measure of the effect of an exposure on risk is distorted because of the association of another factor(s) that influence the outcome under study. It follows that a confounder is a factor that is associated with both the exposure and the outcome of interest. Both confounding and bias can result in the overestimate or underestimate of a true risk or even create an artefactual association.

Bias may be prevented by a careful study design. Confounding may not always be prevented, but may be controlled by stratified analysis, a technique requiring the subdivision of the sample. This underscores again the importance of a large sample size.

Study design

Normally, a case-control study is preferable for cancer studies, because, given the relative rarity of the endpoint, it is more efficient to select diseased subjects before exposure is assessed. The cohort design, in which subjects are initially selected on the basis of their exposure status, is the preferable option when exposure is uncommon.

No study design is ideally suited to this situation, since both the endpoint of interest and the exposure are rare. However, given the obvious difficulties of exposure assessment, a cohort design appears to be the lesser of the two evils.

Since exposure to above-average RF fields is often accompanied by above-average exposure to ELF fields, identifying cohorts for which this is not the case is of paramount importance. Thus, feasibility studies with this aim should receive the first priority. Suggested possibilities are military personnel, RF plastic welders and diathermy operators. Electronics and telecommunication workers, for whom elevated ELF exposure have been observed (Bowman et al 1988) are not suitable subjects.

In summary, if epidemiology is to provide any insight into the question of RF exposure and cancer, very large studies will need to be carried out. Given Australia's small population and limited resources, this may be difficult to accomplish. The possibility of international coordinated research should be given consideration.