Inability to randomly assign exposures means that investigators must design their study such that the cases resemble the controls, except for exposure, in order to limit possible bias. Control selection bias is introduced if exposure is related to characteristics that would make persons more or less likely to be sampled, or, once sampled, to participate. In the Nordic countries, comprehensive national population registries can be used for selecting controls. All residents of each country are listed in these population registries, and participation rates are typically high, such that epidemiological studies of these populations are not compromised by control selection bias. In the USA, investigators often use random-digit dialing methods to identify controls, because no comprehensive registries are available. Random-digit dialing leads to difficulty in identifying, contacting, and recruiting controls of low socioeconomic status. The impact and possible bias introduced by enrolling fewer controls of low socioeconomic status is described by Poole and Trichopolous (Poole & Trichopoulos, 1991).
Case selection bias may occur in studies that are based on mortality records (death certificates). Studies of mortality rather than incidence are subject to bias if the survival rate of the exposed and unexposed subjects differs. This may occur if, for example, the exposure is related to socioeconomic status, and different socioeconomic groups have different survival rates for the studied disease. In addition, for diseases that are easily cured or allow patients to live with the disease for a long time and die of some other cause, use of death certificates leads to severe limitations for the identification of cases. Apart from limiting the power of the study (a severe limitation in studies of rare diseases), there is a large potential for selection bias in the identification of cases.
The inability to randomly assign exposures also introduces the possibility of confounding. Confounding is a mixing of effects between the exposure of interest and extraneous risk factors; it is not a product of the design or conduct of the study but results from an association among risk factors (Rothman, 1986). In order to be a confounder, a risk factor must be related to both the studied exposure and the disease. For example, consider a study to determine whether alcohol drinkers have a greater incidence of oral cancer than abstainers. Smoking is related to the incidence of oral cancer; it is also associated with alcohol consumption, and there is a greater proportion of smokers among alcohol drinkers than among nondrinkers. Since smoking increases the incidence of oral cancer, alcohol drinkers will have a greater incidence than nondrinkers, quite apart from any effect of alcohol drinking itself but simply as a consequence of the greater prevalence of smoking among alcohol drinkers. Thus, the apparent effect of alcohol drinking is distorted by the effect of smoking, and smoking is thus a confounding factor. Confounding can produce bias in either direction, artificially increasing or decreasing relative risk estimates, depending on the direction of the association between the exposure, the disease, and the confounder. When identified, confounding can be controlled through statistical methods.
Another limitation of observational epidemiological studies is that exposure occurs through the natural course of events rather than being assigned and controlled for by the investigator. Thus, determination of exposure is subject to inaccuracies, i.e. exposure misclassification. Exposure misclassification may distort measures of association seen in the study and may occur at several levels. For example, in occupational epidemiological studies, errors may occur in assigning job titles. Measurement errors that are dependent on either disease or exposure are termed 'differential misclassification', i.e. the exposure assignment differs for diseased and non-diseased subjects or disease classification differs for exposed and unexposed persons. Information on exposure can be obtained either prospectively (before the disease has occurred) or retrospectively (after the occurrence of the disease). In the former situation, there is no potential for differential misclassification of the exposure. In the latter situation, the recall of exposures that occurred before diagnosis may be influenced by the disease, if the patient's recollection of the exposure is required for its assessment.
In this report, the Working Group critically reviewed observational epidemiological studies of exposure to EMF that were of sufficient quality and reliability in respect of the aforementioned limitations.
After a number of these studies showed associations between cancer and electrical work, epidemiologists in several countries undertook studies with personal magnetic field monitors. All of the studies included measurements of exposure to ELF magnetic fields, and in some studies exposures to ELF electric fields and pulsed EMF (PEMF) at higher frequencies were studied. Accompanying the improvements in assessment of exposure to EMF, these newer studies had more reliable epidemiological designs: cohort and case-control. Since the literature is so voluminous, this review covers only the studies that meet the criteria for exposure assessment listed in Table 4.9.
|
|
|
| All cancers | |
| Leukemia and brain cancer | Full-shift monitoring* |
| Breast cancer | Job titles of electrical workers** |
| Other cancers | Full-shift monitoring |
| Central nervous system tumors from parental exposure | Job titles of electrical workers |
* Worker wears personal EMF monitor for a complete work
shift
** Includes one or more studies with full-shift measurements
Sahl et al. (Sahl et al., 1993) studied a cohort of utility workers in the USA comprising the 36,221 permanent employees who had worked for Southern California Edison (Edison) for at least one year between 1960 and 1988. Vital status was ascertained through the National Death Index, Social Security Administration, and the California Automated Mortality Linkage System. Workers were classified as 'electric' or 'non-electric' workers on the basis of their longest-held occupation at Edison. Electrical workers were defined as 'craft occupations who work near energized equipment'; non-electrical workers included all other non-managerial, non-electrical jobs. The age-adjusted relative risk (RR) for electrical occupations was calculated for all cancers combined (261) and was not significant (1.1; 95% CI, 0.92-1.3). [The limitations of the study are its relatively small sample size and potentially incomplete follow-up of the small percentage of former employees who left California before 1979.]
Savitz and Loomis (Savitz & Loomis, 1995) conducted a retrospective cohort mortality study of 138,905 men who had been full-time permanent employees of one of five participating electric utility companies for at least six months during the period 1950-86. A complete work history was assembled from company records for each worker. Men who had worked for their entire career in a utility's nuclear division were excluded from the cohort. Vital status to 1988 was determined for more than 99% of the men by searching Social Security Administration, Death Benefit Records, Health Care Financing Administration Records, and Drivers License records. Death certificates were obtained for 97% of the deceased. Subjects lost to follow-up were considered lost at the date of the last record showing they were alive. Exposures, measured with an AMEX-3D, were assessed for 2842 full shifts for a random sample of workers in 28 occupational categories. Exposure to potential confounding factors was evaluated by asking industrial hygienists and expert panels to assess exposures to a number of known or suspected carcinogens, including solvents, polychlorinated biphenyls, sunlight, and wood preservatives. In all analyses, the investigators adjusted for age, calendar year, race, social class, and work status (active versus inactive).
For total cancers (4833), a slight significant increase in risk was observed in the higher exposure categories (Table 4.10), adjusted for age, calendar year, race, social class, work status, and exposure to polychlorinated biphenyls and solvents. [The main limitation is the reliance on death certificates for diagnosis and the use of a magnetic field meter that could record only TWA exposures and not other metrics.]
Thériault et al. (Thériault et al., 1994) examined cancer occurrence in relation to exposure to magnetic fields among workers at three large electric utility companies: Hydro Québec and Ontario Hydro in Canada, and Electricité de France (EDF). A nested case-control approach within these three cohorts was used to study all cancers combined among the male population of the three companies (total: 223,000; EDF, 170,000; Ontario Hydro, 31,543; Hydro Québec, 21 749). The cohorts at the two Canadian companies included both active and retired workers employed between 1970 and 1988, although the Ontario Hydro cohort included only retired workers from 1970-73. The EDF cohort included only actively employed workers and covered the period 1978-89. There were 4151 new cases of cancer in the combined cohort between 1970 and 1989. At Ontario Hydro, cases were ascertained through the Ontario Cancer Registry; at Hydro Québec, ascertainment was made through the company's medical files, the Québec Tumor Registry and death certificates; at EDF, only cancers in active workers were identified through the company's specific epidemiological data base of sickness benefits.
To estimate cumulative exposure to magnetic fields, the authors combined the job history of each case and control with TWA exposure to magnetic fields estimated from a JEM for each job (see section 2.4). The JEM for each utility was based on an extensive program of measurements among its employees. For this study, 2066 workers in the three utilities wore a Positron exposure to EMF meter for a five-day work week. The Positron measures ELF EMF and was also thought to measure PEMF (see section 2.3.1). Only the data on ELF magnetic field were used in this study; the other fields were analyzed in subsequent publications. In the protocol used by Hydro Québec and EDF, workers were selected for measurements in job categories determined a priori on the basis of EMF sources, and the sample sizes were representative of the utility's worker distribution among those categories. In Ontario Hydro's measurement protocol, the sample of workers was representative of the job titles and work locations of the cases and controls.
Potential confounders examined included smoking and exposure to ionizing radiation, chemical agents, and sunlight. Data on smoking were available from medical histories for some subjects at Hydro Québec but not for the other two companies. Information on cumulative individual occupational dose of ionizing radiation was obtained from company radiation surveillance records. Occupational hygienists at each of the companies then developed JEMs for chemicals and sunlight. In an approach similar to that used to estimate exposure to EMF, each individual's job history was combined with the JEM to obtain a cumulative exposure estimate for each chemical agent. When available, data from exposure monitoring were used; otherwise, exposure to confounders was assessed by expert judgment. No association was found between the risk for all cancers combined and cumulative exposure to ELF magnetic fields (Table 4.10), overall (OR, 1.0; 95% CI, 0.91-1.1 for exposures > median compared with < median) or in any of the cohorts.
Guénel et al. (Guénel et al., 1996) reanalyzed the data from this cohort study with information on exposure to electric fields. There was a marginally significant decrease in risk for all cancers combined (Table 4.10). [As a person's exposure to electric fields depends on posture, grounding, and nearby objects, accurate measurement requires more sophisticated instruments and measurement protocols than those used in this and other epidemiological studies.]
Exposure to magnetic fields was determined from JEMs for the longest-held job during the 10 years prior to diagnosis of the case, and 1015 full-shift measurements were conducted at the subject's job or a similar one. Because measurements were not available for everyone, all subjects were assigned exposure based on the averages for all measurements for their job. JEMs were constructed in 169 occupational categories for the TWA, median, standard deviation (SD), and time exposed to > 0.2 µT, computed from these one-day measurements. Exposures to potential confounders were evaluated from workplace interviews,.
Age-adjusted odds ratios were calculated from an unmatched analysis of all controls for both leukemia and brain cancer cases. The relative risks for leukemia were significantly associated with the estimated TWA exposure to magnetic fields in the job held longest during the 10 years before diagnosis (Table 4.11). Although no formal testing of the trend is reported, the relative risks increased with quartiles of exposure, reaching significance for the highest exposure. For all leukemias combined, the highest risk was seen in the group with TWA exposure > 0.41 µT (OR = 1.7; 95% CI, 1.0-2.7). Analysis by leukemia subtype revealed no increased relative risk for acute myeloid leukemia (AML) but an increase in the risk for chronic lymphocytic leukemia (CLL) with greater exposure. The odds ratios for CLL in the highest TWA exposure categories were even higher and statistically significant (TWA 0.41 µT: OR = 3.7; 95% CI, 1.8-7.7). The relative risks for CLL in the highest category of exposure to magnetic fields increased somewhat when adjusted for exposure to benzene (OR = 4.1; 95% CI, 2.0-8.5), solvents (4.2; 1.9-9.4), and ionizing radiation (4.7; 2.2-9.7). Because the authors were concerned about a possible information bias due to the large number of proxy respondents among the cases, they also carried out analyses based on average daily exposure estimated from occupational information from the 1980 census. Although the odds ratios were slightly reduced, increases were observed with increasing exposure category: 1.0 (0.5-2.1), 1.9 (1.0-3.6), 2.3 (1.2-4.3), and 2.6 (1.3-5.4), respectively, for exposure to 0.16-0.19, 0.20-0.28, > 0.29, and > 0.41 µT. Concern about a possible bias linked to the high non-response rate was addressed by repeating the analyses with the occupation stated in the census for the entire population of cases and controls, including non-respondents. The odds ratios were further reduced, and, although increased, were no longer statistically significant: 0.9 (0.5-1.6), 1.6 (0.9-2.7), 1.6 (0.9-2.7), and 1.7 (0.9-3.3). These results are due to the larger proportion of unexposed subjects among non-respondents and indicate the presence of a differential bias.
[This study is unique in assessing EMF cancer risks in the general working population. In industry, magnetic fields are likely to be more diverse in their characteristics than the single-frequency fields usually found around electric power facilities. Another unique feature of this study is the effort to measure magnetic fields at the same job and workplace (or a close proxy) where the subjects had worked before their cancer diagnosis. For logistical reasons, the measurement protocol was changed during the course of the study, resulting in more measurements in case work places than in those of controls. Exposures were not based on an entire work history. The results of analyses during two different jobs, however, are similar. The analyses based on census occupational information, including non-responders, indicate the presence of a bias which could lead to an overestimate of risk among respondents.] (Bowman & Methner, 1998)
Matanoski et al. (Matanoski et al., 1993) conducted a case-control study of mortality from leukemia nested in a cohort of 1,300,000 active workers and 200,000 retirees of the American Telephone & Telegraph Co. (AT&T). Cases of primary leukemia (except CLL) were determined from AT&T's mortality records for 1975-80. To be included in the study, cases had to be in white males who had worked at AT&T for at least two years. For each case, three controls without leukemia were matched on retirement status at the age of the patient's death, gender, date of birth, and date of hire. Sporadic data on the subjects' employment histories were obtained from AT&T data tapes and from records of some participating regional phone companies which were separated from AT&T after the study period. Although 124 matched sets of cases and controls were identified, some analyses were based on smaller numbers since records specifying the last job could be obtained for only 75 sets of cases and controls (60%), and complete job histories were found for 35 sets (28%).
Full-shift measurements of magnetic fields were made with an EMDEX-C monitor from samples taken on 15-61 present-day workers in four categories of telephone line jobs (cable splicers, installers, central office technicians, and supervisors) and a sample of non-line jobs. For central office technicians, measurements were taken at switching facilities where either solid-state techniques or electromechanical relay crossbars were used; the latter began to be replaced during 1975-80. The older relay switching results in higher exposure than the new solid-state technique. The data on magnetic fields were used to construct JEMs with the TWA and peak exposures averaged for each job category. Odds ratios were calculated by conditional logistic regression for the last job held (relative to non-line workers) and for cumulative exposure (relative to values below the median). To assess exposure-response relationships, one-tailed tests for positive trend were performed on odds ratios for quartiles of cumulative exposure. For cumulative TWA exposures, the relative risk was elevated but not significant (OR = 2.5; 95% CI, 0.7-8.6), and there was no exposure-response relationship (p for trend = 0.27; based upon means of logs of exposures in each quartile). A marginally significant exposure-response relationship was found among workers using the older style relay switching (p for trend = 0.06). [A large fraction of subjects were excluded because of missing job records, which is a potential source of bias and resulted in small sample size.]
Sahl et al. (Sahl et al., 1993) conducted a nested case-control study in the cohort of employees at Southern California Edison (Edison) described in section 4.2.1.1, comprising 44 cases of leukemia identified from death certificates. Ten controls were selected for each case, frequency matched on age, gender, and race. For the case-control analysis, exposure to magnetic fields was assessed with a JEM based on 776 person days of exposure measurements made with an EMDEX II monitor among present-day Edison employees in 35 occupational categories. Several exposure metrics were computed from the measurements, including the arithmetic mean, geometric mean, 95th and 99th percentiles, and fraction of time exposed to > 1.0 or 5.0 µT. An exposure score was obtained for each subject by summing the product of the exposure metric from the JEM times years of employment. The odds ratios for leukemia in relation to the various exposure scores were close to 1 (Table 4.11) Further analyses at various time windows of exposure and latency did not reveal positive associations. [The limitations are discussed in section 4.2.1.1. They also include lack of information on potentially important confounding exposures to chemicals, ionizing radiation, and smoking and a reliance on death certificates for diagnosis. The failure to track the deaths of subjects outside California could also have introduced some slight negative bias.]
London et al. (London et al., 1994) performed a case-control study of leukemia which focused on nine electrical occupations originally suggested by Milham (Milham, 1985); see Table 2.4. Cases were all 2355 working men, aged 20-64, in whom leukemia had been diagnosed in Los Angeles County during 1972-90 and whose last occupation was entered in the County's cancer registry. Of these, 121 were electrical workers. The controls were 67,212 men from the same study base in whom other tumors, except of the central nervous system (CNS), had been diagnosed. They included 2665 electrical workers. Exposure to magnetic fields was estimated with a variety of meters from full-shift measurements taken on 278 Los Angeles workers in electric occupations and a random sample of 105 men in 18 other occupations. TWA magnetic fields were calculated for 24 different tasks with exposure to EMF. The proportion of time that electric workers usually spent at those tasks was estimated by interviews with expert panels at each workplace. From these data, task-weighted estimates of mean and standard deviation were calculated for each electrical occupation, for all electric occupations combined, and for all non-electrical occupations combined. Exposure to potential leukemogens in these occupations was also assessed by the expert panels.
Odds ratios were calculated by logistic regression adjusted for age. Men in electrical occupations as a whole had significantly greater exposure to magnetic fields than the sample of those in non-electrical jobs (mean TWA = 0.96 ± 0.13 µT vs. 0.17 ± 0.01 µT), except for electrical engineers. A weak trend in risk for leukemia of marginal significance was observed with exposure to the TWA magnetic field (OR increase per 1 µT = 1.2; 95% CI, 1.0-1.5). The trend with TWA was concentrated in the risk for chronic myelogenous leukemia (significant) and acute nonlymphocytic leukemia (nonsignificant); no increase in relative risk was seen for CLL. The limited information on exposure to confounders did not appear to explain the risk patterns. [The following limitations were identified: reliance on the single job specified in the registry to estimate exposure and a study design based on the proportion of leukemia among all cancers.]
Kheifets et al. (Kheifets et al., 1997b) used the exposure measurements of London et al. (London et al., 1994) to assess the risks for leukemia associate with exposure to ELF electric fields. Some of the EMF monitors also had electric field sensors, so that exposure to electric fields was known for 28% of the Los Angles workers monitored in the original study. Although no measurements were taken on power line workers in Los Angeles, they were assigned to have high exposures (> 20 V/m) on the basis of measurements done elsewhere. The authors noted additional difficulties in assessing exposure to electric fields, in particular that a person's exposure varies with posture and grounding. [The fraction of workers with the electric field sensor was 14-93% across occupations, which is a large potential source of bias.]
In their study of three large electric utility companies, Thériault et al. (Thériault et al., 1994) carried out a nested case-control study of leukemia. No significant association was found between cumulative exposure to magnetic fields and the risk for leukemia as a whole. In analyses of specific leukemia subtypes, only the risk for acute non-lymphocytic leukemia was found to be related to exposure to magnetic fields (OR = 2.4; 95% CI, 1.1-5.4) among workers with TWA exposures above the median compared to those below the median. When only AML was considered, the odds ratio among the half of workers with higher exposures increased to 3.2 (95% CI, 1.2-8.3). No significant exposure-response relationship was observed. The association with magnetic fields greater than or equal to the median was strongest for exposures received 20 or more years previously (OR = 4.6; 95% CI, 0.22-94), based on seven exposed cases. For other leukemia subtypes, no significant associations were reported. When analyses for all leukemias were restricted to individual cohorts, a significant association was observed between cumulative exposure above the median in Ontario Hydro (OR = 3.1; 95% CI, 1.1-9.7) but not in the other cohorts (Hydro Québec, OR = 0.29; 95% CI, 0.04-1.8; EDF, OR = 1.4; 95% CI, 0.61-3.1). The association was due mainly to AML. [The cohorts were defined and followed up differently: the EDF cohort included only active workers, while the Ontario Hydro cohort included only retirees for the first three years of the study. The levels of exposure were notably lower in the EDF cohort, probably due to the inclusion of gas workers. The JEMs were derived differently for each cohort, which may affect the interpretation of the results for the combined cohorts. The small numbers of cases by leukemia subtypes make interpretation of these results difficult.]
Guénel et al. (Guénel et al., 1996) analyzed exposure to ELF electric fields in the EDF cohort and found no association with leukemia risk (OR = 0.4; 95% CI, 0.13-1.2). [See section 4.2.1.1 for comments on the limitations of this study and that of Thériault et al.]
Miller et al. (Miller et al., 1996) re-analyzed the data on the Ontario Hydro cohort for exposure to both ELF electric and magnetic fields. An association with increasing exposure to electric fields was reported for all leukemias and leukemia subtypes; an increase was also seen in combination with magnetic fields (see Table 4.11). The JEMs at Ontario Hydro and the other two utilities were different, as noted above. Miller et al. found an association when exposures were derived from the JEM of the Ontario Hydro protocol, which disappeared with the JEM derived from the same measurements with the protocol used at the other two utilities. [The large 95% confidence intervals on the odds ratios prevent interpretation of these results.]
In their retrospective cohort study of exposure to magnetic fields and cancer in utility workers (described in section 4.2.1.1), Savitz and Loomis (Savitz & Loomis, 1995) identified 164 deaths from leukemia. No association was observed between leukemia risk and increasing cumulative exposure to magnetic fields (Table 4.11). Analysis by subtype showed no association, although a nonsignificantly increased risk for AML was seen (based on five cases) in the highest exposure category (> 4.3 µT-year). [Reliance on death certificates limits diagnoses, particularly for leukemia subtypes and cancers with long survival such as CLL; furthermore, the magnetic field meter used could record only TWA exposures and not other metrics. See section 4.2.1.1 for other comments on the limitations of this study.]
In a population-based case-control study, Feychting et al. (Feychting et al., 1997) examined leukemia risk in relation to occupational and residential magnetic fields. They estimated occupational exposure to magnetic fields from information from census data on the last job held before cancer diagnosis and from the JEM constructed by Floderus et al. (Floderus et al., 1996) in another study in Sweden. When analyses were restricted to subjects with little or no residential exposure, nonsignificant increases in the risk for leukemia as a whole and for AML and CLL were observed for occupational TWA exposure > 0.2 µT. The odds ratio for AML was greater for subjects with both residential and occupational exposure, based on three exposed cases (Table 4.11). Adjustment for possible confounding by motor fuel or exhaust, benzene, oil products, and welding fumes reported to interviewers did not change these associations. [The limitations are use of an exposure assessment based on a JEM for another population of working males and the large percentage of missing information on occupation, particularly for women, and the small numbers of subjects in the analysis by subtype and by joint occupational and residential exposures. See also section 4.2.2.1 for comments on the limitations on the residential study.]
Johansen & Olsen (Johansen & Olsen, 1998) conducted a retrospective cohort mortality study of electric utility workers in Denmark. The cohort consisted of 32,006 men and women who had been employed for at least three months at any of the 99 electric utilities in Denmark. Their employment history was obtained from utility records going back to 1909. Vital status was obtained from the Central Population Register and the National Death Certificate files for 1968-93, and cancer incidence was obtained from the Cancer Registry over the same period. A JEM was constructed by a panel of utility engineers who used 127 magnetic field measurements taken earlier (Skotte, 1994) to assign a TWA magnetic field exposure category (background, low, high, or unknown) to 475 combinations of job titles and work areas. Each subject was assigned the magnetic field exposure at their first job because < 1% of the employees had changed occupations within a utility, according to the company records. The SIRs for leukemia in men were calculated for different exposure with respect to the Danish male population, with adjustment for age, gender, and calendar year. No association between level of exposure to magnetic fields and leukemia risk was observed (Table 4.11). [The measurement database is far less comprehensive than those of most other studies, which had access to full-shift measurements.]
Sahl et al. (Sahl et al., 1993) also considered brain cancer in their case-control study nested in the cohort of employees at Southern California Edison (see section 4.2.1.2). The analyses included 32 deaths from brain cancer. The odds ratios by cumulative exposure category were close to 1.0. Further analyses at various time windows of exposure and latency did not reveal any association. [See section 4.2.1.2 for comments on the limitations of the study.]
Thériault et al. (Thériault et al., 1994) also examined the risk for brain cancer in their case-control study nested within a cohort of workers at three large electric utility companies (see section 4.2.1.1). In all, 250 cases of brain cancer were included in the study. A non-significant increase in risk of brain cancers as a whole was observed for workers with cumulative exposure to magnetic fields greater than the median (3.15 µT-years: OR, 2.0; 95% CI, 0.98-3.9) and the 90th percentile (15.7 µT-years: OR, 2.1; 95% CI, 0.80-5.7; see Table 4.12). A significantly increased risk was seen for astrocytoma among workers in the 90th percentile of cumulative exposure, based on five exposed cases. This association greatly diminished, however, when exact conditional logistic analyses were used [the appropriate method for analyzing data when such small numbers of cases are observed]. The analyses showed no significant differences across cohorts. [See section 4.2.1.2 for comments on the limitations of this study.]
In their re-analyses of the EDF cohort, Guénel et al. (Guénel et al., 1996), found an increased risk for brain tumors (based on 69 cases) among utility workers exposed to electric fields. The increase was significant in the group that had exposure to electric fields in the 90th percentile of the exposure distribution (OR, 3.1; 95% CI, 1.1-8.7). The association remained significant after adjustment for confounding exposures (including magnetic fields) and was strongest in workers with > 25 years' employment. [See section 4.2.1.2 for comments about the limitations of this study.]
In their reanalyzes of the Ontario Hydro cohort (see section 4.2.1.2), Miller et al. (Miller et al., 1996) also considered 69 cases of brain cancer. No association was seen with exposure to electric fields, based on two exposed cases in the 172-344 V/m-year category and three in the > 345 V/m-year category. [See section 4.2.1.1. for comments on the limitations of this study.]
In their retrospective cohort study of magnetic fields and cancer in utility workers (described in section 4.2.1.1), Savitz and Loomis (Savitz & Loomis, 1995) estimated exposure to magnetic fields for 144 workers who had died from malignant neoplasms of the brain and nervous system. An increased risk was seen in all exposure categories, which was statistically significant only in the highest category (> 4.3 µT: RR = 2.3; 95% CI, 1.2-4.6). The RR per µT-year was 1.1 (95% CI, 1.0-1.1). The association was slightly greater for cumulative exposures received 2-10 years previously (RR per µT = 1.9; 95% CI, 1.3-2.8). [See section 4.2.1.1 for comments on the limitations of this study.]
Feychting et al. (Feychting et al., 1997) also considered brain cancer in their case-control study of cancer and occupational and residential exposure to magnetic fields in Sweden (see section 4.2.1.2). They identified 223 cases of central nervous system tumors from the Swedish Cancer Registry. No association was seen between CNS tumors and occupational exposure to magnetic fields among people with little or no residential exposure. Among those with both residential and occupational exposures > 0.2 µT, no association was seen for CNS tumors as a whole, but a non-significant increase (OR, 2.2; 95% CI, 0.6-8.5) was reported for astrocytomas grades III and IV, based on three exposed cases. [See section 4.2.1.2 for comments on the limitations of this study].
Harrington et al. (Harrington et al., 1997) conducted a case-control study of death from brain cancer nested in a cohort of 84 018 male and female electric utility workers who had been employed for at least six months between 1972 and 1984 at the Central Electricity Generating Board in the United Kingdom. The cohort was defined from computerized employment records, which started at different times over the period 1972-79 in various regions of England and Wales. Deaths from brain cancer were identified from death certificates, and 112 diagnoses of primary brain cancer were verified through the National Cancer Registry. Six controls per case were chosen from the cohort, matched on gender, who were alive at the time of death of the case and closest in age to the case. Workers' computerized job histories were coded blindly in terms of 11 job groups by two engineers with long experience in the industry. Exposure in different jobs and locations was assessed from measured exposures from a survey of 258 staff in the British electricity supply industry (Merchant et al., 1994), in which both IREQ and Positron meters were used. In each job for which measurements were made, exposure was estimated as the TWA and the time-weighted geometric mean of the measurements made in this job. Various measures of the exposure of the cases and controls were calculated by combining job classifications with employment records (when available) and the estimated exposure in the relevant job. Exposure estimates could not be made for 86 subjects (18 cases and 68 controls) because of insufficient employment history. Information on 24 potential workplace carcinogens and neurotoxic agents was obtained from a JEM generated for this study. Statistical analyses were performed by conditional logistic regression. No association was observed between brain cancer and cumulative exposure to magnetic fields (measured either as TWA or time-weighted geometric mean) over the total career or over the five years before death. A significant increase in risk was, however, observed among people whose exposure was not classifiable owing to insufficient employment history. No association between exposure to any of the 24 potential confounders and brain cancer was observed. [The limitations in the exposure assessment resulted in no exposure estimates for 16% of cases and 10% of controls.]
Johansen and Olsen (Johansen & Olsen, 1998) also considered brain cancer in their retrospective cohort mortality study of electric utility workers in Denmark (see section 4.2.1.2). No association was seen between death from brain cancer (72 cases) and exposure to magnetic fields (see Table 4.12). [See section 4.2.1.2 for comments on the limitations of the study.]
Demers et al. (Demers et al., 1991) assessed exposure to EMF for the men in a population-based case-control study of epithelial breast cancer (Jauchem et al., 1992). The cases were determined from five years of records at cancer registries in six states and four metropolitan areas in the USA. Controls matched on age were selected by random-digit dialing for younger subjects and from Medicare records for those over 65. An interviewer asked about the subject's two longest-held occupation and possible risk factors for breast cancer. A subject was considered to have been exposed to EMF if his longest-held job was in one of five groups (Table 2.4). All exposed men had a twofold increase in relative risk, which increased in the electric utility trades (OR = 6.0; 95% CI, 1.7-21). Adjustment for known risk factors for breast cancer did not change these estimates appreciably. The relative risks were further elevated in subjects who were first exposed before the age of 30 and continued to be exposed for at least 30 years before cancer diagnosis. [This finding confirms the hypothesis that exposure to EMF is an early-stage carcinogen.] The study is the largest to look at male breast cancer and could therefore assess the timing of exposure to EMF and possible confounders, both of which supported the association. [The job titles used as a surrogate for exposure to EMF were not validated by field measurements in the study population. In addition, the refusal rate was substantial, especially among controls, and the occupational histories were limited to jobs held the longest.]
Tynes et al. (Tynes et al., 1992) studied a cohort of working-age men in Norway by linking census data on work histories to cancer registry data covering a 25-year period. Exposure to EMF was attributed for a list of 13 electrical job titles (Table 2.4). The SIRs for the electrical workers were calculated from the expected numbers of cancers for the entire cohort. The relative risks for male breast cancer were significantly elevated, being highest for men with jobs in electrical transport (only four cases), although the exposure to EMF is generally below the power frequency (e.g. 16.7 Hz for electric railways and DC magnetic fields for tram operators in the three largest Norwegian cities). [The job titles used as a surrogate for exposure to EMF were not validated by field measurements in the study population, and no information on potential confounders was available.]
Guénel et al. (Guénel et al., 1993) conducted a population-based study of working men and women in Denmark, whose cancer incidence was examined from 18 years of data from the national registry. To estimate exposure to EMF, 3932 combinations of occupational and industrial categories were assigned by expert judgment to four exposure categories (none, intermittent ELF, continuous ELF, and other EMF frequencies). The highest exposure category was ELF magnetic fields continuously above 0.3 T, which were attributed to 20 industry-job combinations (Table 2.4). Sex-specific cancer risks are calculated relative to the entire cohort of economically active subjects and are effectively SIRs. Guénel et al. reported relative risks for male and female breast cancer, neither of which were significantly elevated for either intermittent or continuous exposure to ELF EMF (see Table 4.13). The occupations considered to imply continuous exposure to ELF magnetic fields is unusual, as it includes jobs such as female shop assistants in the dairy products and bread industries and omits electric line workers, power station operators, and welders, who are combined with other metal workers in the Danish job classification system. The authors noted that these possible misclassifications would tend to lower the risk estimates. [The job titles used as a surrogate for exposure to EMF were not validated by field measurements in the study population, and no information on potential confounders was available.]
Rosenbaum et al. (Rosenbaum et al., 1994) conducted a case-control study of male breast cancer in New York State. The cases were all 71 cases of histologically confirmed primary breast cancer in men from eight counties in western New York state reported to the New York State Tumor Registry between 1979 and 1988. Controls were 256 men from the same eight counties who participated in the Prevention-Detection Clinic, a voluntary cancer screening program; they were cancer-free and frequency matched to the cases by race, year of diagnosis (screening for the controls), and age. The 'usual occupation' was taken from hospital records (76%) for cases, and information on 'type of work done' was obtained by questionnaire from 95% of the controls before screening. Complementary information on occupation was obtained from city directories, resulting in occupational data for 89% of the cases and 99% of controls. JEMs were constructed for occupational exposures to heat and EMF (Table 2.4). The risk for breast cancer was elevated for men exposed to heat but not for those exposed to EMF, based on six exposed cases, and could be adjusted only for heat exposure, age, and county of residence. [The job titles used as a surrogate for exposure to EMF were not validated by field measurements in the study population, and no information on potential confounders was available. In addition, the source and completeness of information on occupation differed for cases and controls, and the controls may not be representative of the population at risk.]
Floderus et al. (Floderus et al., 1994) analyzed the risk for breast cancer among all Swedish men who had been employed as railway workers and were aged 20-64 in 1960. The cohort was defined on the basis of data on occupation from the 1960 Swedish census. Cases of breast cancer were ascertained from the 1960 Cancer Environment Registry, which comprises information from the Cancer Registry for the period 1960-79. SIRs for specific occupations were calculated with respect to the entire population of working men from the 1960 census data, adjusted for age in 10-year intervals. No account was made for cancer mortality or incidence during the follow-up period. Data for 1960-69 and 1970-79 were analyzed separately because changes in railway work practices had reduced the size of train crews and how often individuals worked on trains. Exposures were assessed by the railroad job titles that were associated with high exposure to ELF magnetic fields according to earlier personal monitoring (Floderus et al., 1993). Significantly increased relative risks were reported for men in the most exposed occupations (Table 2.4): engine drivers, engine drivers and conductors, and railway workers in the first decade, based on two, three, and four cases, respectively (Table 4.13). In the second decade, only four cases of breast cancer were seen, none of which was in men with the occupations listed above. The authors discussed the differences between the two time periods and proposed several hypotheses, in particular an effect of reduction of exposures. [The very small number of cases and the improper follow-up of the cohort make interpretation of these results difficult.]
In an extension of their previous case-control study of leukemia and brain cancer (Floderus et al., 1993), described in section 4.2.1.2, Stenlund and Floderus (Stenlund & Floderus, 1997) also considered male breast cancer and testicular cancer. They identified 56 cases of breast adenocarcinoma from the Swedish Cancer Registry for the period 1985-91 from all of Sweden. The 1700 controls used in the study were from the previous case-control study of brain and leukemia. The participation rate was 82% for cases and 72% for controls. Exposure assessment was based on the use of the JEM, as described above, which was obtained in the previous case-control study on a different study population (Floderus et al., 1993) and complemented by 15 additional measurements in this study population. No association was observed between the risk for male breast cancer and exposure to magnetic fields (see Table 4.14). [The limitations are transporting JEMs across study populations, the low participation rates, limited information on potential controls, and the fact that the controls came from a different study and were therefore not matched on age or region with the cases in this study.]
Loomis et al. (Loomis et al., 1994b) used a large database of death certificates in the USA for the period 1985-89. Among 27,814 cases of breast cancer and 112,749 controls, 68 cases and 199 controls had been employed in traditional electrical occupations, resulting in a small but significantly increased risk (mortality odds ratio = 1.4; 95% CI, 1.0-1.8) in the electrical occupations originally defined by Milham (Milham, 1982). Since this restrictive definition of electrical work applied to only about 0.2% of women, Loomis et al. (Loomis et al., 1994b) examined other occupations in which more women might be exposed to EMF (Table 2.4), but saw no elevated associations. [The study had the following limitations: reliance on death certificates for occupation and diagnosis, use of other deaths as controls, assessment of exposure to EMF by a group of electrical occupations, and an inability to control for other risk factors for breast cancer.]
Cantor et al. (Cantor et al., 1995a) used the same death certificate data set, adding one more year of follow-up. Exposure to EMF was assessed by an industrial hygienist [no details were given]; however, this new classification resulted in twice as many women being defined as 'highly' exposed. They found no association for white women but a significant increase for black women with medium or high probability of exposure to ELF fields (see Table 4.15). [The absence of detail about the more liberal ELF exposure assessment makes evaluation of the differences in results from Loomis et al difficult.]
Coogan et al. (Coogan et al., 1996) studied occupational exposure to 60 Hz magnetic fields in a large case-control study of breast cancer in the USA (Newcomb et al., 1994). The cases were over 6800 cases of breast cancer diagnosed between 1988 and 1991 among women 74 years of age or younger from the cancer registries in Maine, Wisconsin, and Massachusetts. The controls were frequency-matched on age from drivers' license records (if < 65) and the Health Care Financing Administration's list of Medicare beneficiaries (if 65-74). The occupation most representative of the subject's career was obtained by telephone interview, with information on reproductive history and other breast cancer risk factors. The participation rates were 81% (6888) for cases and 84% (9529) for controls. Usual occupations were grouped into three categories (low, medium, and high) of potential exposure to 60 Hz magnetic fields above background by an industrial hygienist (Table 2.4); the remaining occupations were aggregated into a 'background' category. Conditional logistic regression stratified on age and state was used to calculate odds ratios, with adjustment for risk factors for breast cancer. The odds ratios by category were 1.0 (95% CI, 0.91-1.2) for low exposure, 1.1 (0.83-1.4) for intermediate exposure, and 1.4 (0.99-2.1) for high exposure. [There are problems in assessing exposure to EMF by expert judgment for workers in broad sectors of the economy (see section 2.4 for further discussion).]
Johansen and Olsen (Johansen & Olsen, 1998) also considered breast cancer in women in their retrospective cohort mortality study of electric utility workers in Denmark (see section 4.2.1.2). No association was seen between death from breast cancer (96 cases) and exposure to magnetic fields, based on very small numbers of exposed cases: two in the low-exposure category (0.1-0.29 µT), none in the intermediate category, and one in the highest category (> 1.0 µT). [No information on risk factors for breast cancer was available. See section 4.2.1.2 for comments on the limitations of the study.]
Armstrong et al. (Armstrong et al., 1994) examined the associations between exposure to PEMF and cancer among the Hydro Québec and EDF cohorts included by Thériault et al. (Thériault et al., 1994). This analysis was based on about 1000 person-weeks of measurements of exposure to these high-frequency electromagnetic transients, 508 lung cancer cases, and 508 controls. Smoking status was available from medical histories for some of the subjects at Hydro Québec but not at EDF. A significant association was observed between the risk for lung cancer and cumulative exposure to PEMF (OR, 3.1; 95% CI, 1.6-6.0 among workers with exposure > 90th percentile). This association was confined to the Hydro Québec workers (OR, 2.4; 95% CI, 1.1-5.3 for PEMF exposures > median and OR, 5.6; 95% CI, 2.0-16 for exposures > 90th percentile), despite a low standardized mortality ratio (SMR, 0.85) for lung cancer in that cohort. The association among Hydro Québec workers became stronger after adjustment for current smoking status and exposure to asbestos and other occupational agents (OR = 10; 95% CI, 3.2-31 for the highest exposure). The authors noted that the levels of exposure to both PEMF and ELF magnetic fields were considerably lower among EDF workers than among Hydro Québec workers. They also noted the incomplete and apparently incorrect characterization of the Positron's PEMF channel, which is more sensitive to radiocommunication signals (see section 2.3.1) (Heroux, 1991). [The information on smoking was incomplete.]
Miller et al. (Miller et al., 1996) also analyzed the risk for lung cancer in the Ontario Hydro cohort of Thériault et al. in relation to exposure to both ELF electric and magnetic fields (see description of study in section 4.2.1.2). No significant association was seen between the occurrence of lung cancer (263 cases) and exposure to electric or magnetic fields. [See section 4.2.1.2 for comments on the limitations of this study; in addition, there was no information on smoking.]
Savitz et al. (Savitz et al., 1997) subsequently re-examined the cohort from the study of five US utilities of Savitz and Loomis (Savitz & Loomis, 1995) in order to test the association between lung cancer and exposure to PEMF reported by Armstrong et al. (Armstrong et al., 1994). A total of 1692 deaths from lung cancer had occurred over the study period. The PEMF JEM from the Canada-France study was adapted with some approximation to the job categories of the five US utilities. No association was seen between the risk for lung cancer and exposure to 60 Hz magnetic fields. The risks associated with exposure to PEMF were slightly but significantly elevated in all exposure categories (Table 4.16); there was no indication of a trend with exposure. No data on smoking were available. The risk for lung cancer was still significantly elevated after adjustment for socioeconomic status. [Transposition of a JEM from the Canada-France study to this population is problematic.]
In the case-control study described above (section 4.2.1.3), Stenlund and Floderus (Stenlund & Floderus, 1997) also used the Swedish JEM to analyze the relative risks for testicular cancer. A total of 144 cases (94 seminomas and 50 non-seminomas) were identified from the Swedish Cancer Registry for the period 1985-87 in central Sweden. A small increase in risk was observed in all quartiles of TWA exposure to magnetic fields. The analyses were adjusted for age, education, and exposure to solvents. The odds ratios were 1.3 (95% CI, 0.7-2.4) for TWA exposure in the range 0.16-0.19 µT, 1.4 (0.8-2.7) for 0.20-0.28 µT, and 1.3 (0.7-2.5) for > 0.29 µT. The risk was significantly increased among workers with exposures > 90th percentile (> 0.41 µT: OR = 2.1; 1.0-4.3), particularly among men under the age of 40 (OR = 3.9; 95% CI, 1.4-11). A significantly increased risk in the 90th percentile was seen for non-seminomas in men below the age of 40 (OR = 16; 2.7-95, based on seven exposed cases and overall when workers were also probably exposed to solvents (OR = 3.8; 95% CI, 1.1-13.1). The authors regretted their inability to test for confounding by exposure to heat, a possible risk factor for testicular cancer. [See section 4.2.1.3 for comments on the limitations of this study.]
Thériault et al. (Thériault et al., 1994) also examined the occurrence of a number of other cancer types in relation to exposure to magnetic fields in a case-control study nested within the cohort study of workers at three large electric utility companies (see section 4.2.1.1). No association was seen with lymphoma, multiple myeloma, melanoma, or any of the other 13 cancer groupings considered. No association with exposure to electric fields was seen for cancers other than those of the brain and lung cancer by Miller et al. (Miller et al., 1996). [See section 4.2.1.1 for comments on the limitations of these studies.]
Schroeder and Savitz (Schroeder & Savitz, 1997) studied mortality from lymphoma and multiple myeloma in association with the JEM for magnetic fields and the cohort mortality data from Savitz and Loomis (Savitz & Loomis, 1995). In total, 154 deaths from non-Hodgkin's lymphoma, 87 from CLL (classified as low and intermediate or high grade), 29 from Hodgkin's disease, and 84 from multiple myeloma occurred during the study period. No association was found between levels of exposure to magnetic fields and the risk for CLL, Hodgkin's disease, or multiple myeloma. An association was found for non-Hodgkin's lymphoma, however, which was significant in the two intermediate exposure categories. The relative risks, adjusted for exposure to solvents, were 1.5 (0.9-2.4), 1.8 (1.1-2.9), 1.8 (1.1-3.1), and 1.3 (0.7-2.8) for cumulative exposure categories of 0.6 < 1.2, 1.2 < 2.0, 2.0 < 4.3, and > 4.3 µT-years, respectively. No association was seen for cumulative exposures in the previous 2-10 years. [Information on other risk factors for lymphomas was lacking. See also section 4.2.1.1 for comments on the limitations of the Savitz and Loomis study.]
In an exploratory study, Spitz and Johnson (Spitz & Johnson, 1985) investigated death from neuroblastoma. The jobs of controls and parents were obtained from the Texas birth certificate registry. Controls were selected so as to match the case's birth year distribution. Exposure to EMF was assigned to electrical jobs, with both a 'narrow' and a 'broad' definition (see Table 2.4). The relative risks were elevated with both definitions of exposure but reached significance only with the broad definition. [The broad definition included electrical equipment salesmen and repairmen, who are unlikely to have high exposure to magnetic fields. No adjustment was made for potential confounding exposures.]
The same Texas records were examined by Johnson and Spitz (Johnson & Spitz, 1989) for mortality from all CNS tumors (intracranial and spinal cord), which include the neuroblastomas studied by Johnson and Spitz (Spitz & Johnson, 1985). The methods were similar to those of the earlier study, except that controls were frequency-matched on gender and race as well as birth year. The relative risks in electrical occupations were not significantly elevated. [Their new definition of electrical occupations is perhaps the most questionable devised, as it includes jobs such as radio and Television performer and electrical goods and appliance salesmen.]
Nasca et al. (Nasca et al., 1988) conducted a study of the incidence of CNS tumors; controls were selected from birth certificates for the same geographical region and were matched on gender, race, and year of birth. Parental job histories were obtained by telephone interviews with the mothers (85% participation rate for cases and 70% for controls). Exposures to EMF and ionizing radiation were assigned to industry-occupation combinations (see Table 2.4). Some increase in risk was observed among children whose parents were employed in industries with exposure to ionizing radiation, and they were excluded as controls in the EMF analysis. The relative risks from exposure to EMF were not significantly elevated for either the broad or narrow definition. Power calculations indicated that the sample sizes were large enough to permit detection of relative risks as low as 2.5-2.8.
Bunin et al. (Bunin et al., 1990) studied the incidence of neuroblastomas in comparison with controls selected by random-digit dialing and with occupational histories ascertained by phone interviews. The controls were matched to cases on race, date of birth, and the first five digits of their telephone numbers (i.e. locality). Exposure to EMF was assigned to the electrical occupations defined by Spitz and Johnson (Spitz & Johnson, 1985). No increase in risk was observed.
Wilkins and Hundley (Wilkins & Hundley, 1990) studied the incidence of neuroblastomas in Columbus, Ohio. Controls and parental occupations were determined from birth certificates, and controls were matched on birth year, race, gender, and mother's county of residence. Electrical occupations of the fathers were defined in several ways, including those of Deapen and Henderson (Deapen & Henderson, 1986) and Lin et al. (Lin et al., 1985). No significant associations were seen with any definition.
Wilkins and Wellage (Wilkins & Wellage, 1996) studied the same population in Ohio for the association between exposure to EMF and risks for CNS tumors. Random-digit dialing was used to recruit controls, who were matched on gender and race (but not telephone exchange). Occupational histories and exposures to potential risk factors were ascertained by interviewing both biological parents. Exposure of the fathers to EMF was ascribed for a group of electrical occupations (Table 2.4). No association was seen with tumor risk. When welders were analyzed separately, the relative risk, based on only six cases, was marginally significant (OR = 3.8; 95% CI, 0.95-16). No evidence was seen for confounding by childhood or maternal exposure to X-rays, parental smoking, or household pesticides. [Confounding from welding fumes and solvents is a potential explanation for the association.]
Tornqvist (Tornqvist, 1998) examined cancers in the children of members of retrospective and prospective cohorts of electrical workers in the power industry. Details of the study design are given in section 4.5.1. The retrospective study found an equal number of cancers among children of fathers who had had electrical jobs in the censuses before and after the birth ('exposed') and the remainder of the cohort ('unexposed'). [On the basis of the number of cancers expected in the paper, the Working Group calculated the risk ratio as 0.75 (95% CI, 0.24-2.3), showing no association with the father's apparent electrical work at the time of conception.] Six of the 12 cancers were located in the CNS and were also found equally in exposed and unexposed children. No cancers were reported among the children in the prospective study. [As the paper did not report the number of cancers expected in this population, this finding is hard to interpret.]
In order to assess heterogeneity, the authors scored 15 individual study characteristics that were thought to predict the results, including study quality and design. The characteristics were scored by two independent epidemiologists who were blinded to the authors and the results of the studies. Because most of the completed studies contained limited information on exposure, the authors attempted to evaluate the relationship between exposure and disease by using additional information or measured exposure to EMF by job classification obtained from several large surveys of exposure in the workplace. These data were used to rank job categories and to compare exposure to the risk estimates for those job categories. The risk estimates of the 29 studies and various subsets are summarized by weighted averages in which the weight is the inverse of the variance of the estimate. A heterogeneity test, q statistic, and random-effects and fixed-effects summaries were computed. To investigate the sensitivity of the results to the weights used, several additional weighting schemes were used.
The pooled odds ratio from the 29 studies for a broad group of electrical occupations was 1.2 (95% CI, 1.1-1.3). [The pooled results for the broad group of electrical occupations includes data from studies in which no measurements of magnetic fields were made and which are therefore not reviewed in this document.] The authors observed lower risks in the Nordic cohort and incidence-based studies included in the meta-analyses. To assess exposure-response relationships, the authors pooled the risk estimates for the high, medium, and low categories of exposure from the available studies in which magnetic fields were measured. These included the results of Floderus et al. (Floderus et al., 1993), Thériault et al. (Thériault et al., 1994), and Savitz and Loomis (Savitz & Loomis, 1995). Although an increased risk was seen in all exposure categories, there was no evidence of an exposure-response relationship across the three exposure categories. [A limitation of this analysis is the combination of results across studies with different exposure categorizations. For example, the high exposure category in the study of Floderus et al. in the general industry may correspond to a medium exposure level in an electrical industry, possibly resulting in exposure misclassification.]
The pooled odds ratio from the 38 data sets (in 22 of which exposure was assessed by job titles alone) for leukemia was 1.2 (1.1-1.3) for a broad group of electrical occupations. Among the leukemia subtypes, a significant association was seen for both AML (OR = 1.4; 95% CI, 1.2-1.7) and CLL (OR = 1.6; 95% CI, 1.1-2.2). There was no evidence of heterogeneity of risk across the studies; in particular, the studies judged to have included better exposure assessment and design did not show a different risk.
In order to assess possible exposure-response relationships, the authors pooled the risk estimates for the high, medium, and low categories of exposure from studies in which magnetic fields had been measured and which were published at that time. The analyses included results from Floderus et al. (Floderus et al., 1993), Thériault et al. (Thériault et al., 1994), Savitz and Loomis (Savitz & Loomis, 1995), and Matanoski et al. (Matanoski et al., 1993) and two studies that included spot measurements (Tynes et al., 1992; Tynes et al., 1994). For both brain cancer and leukemia, the pooled relative risks showed no exposure-response relationship across the three exposure categories. An increased risk was seen in all exposure categories, although there was no evidence of an exposure-response relationship. [See comments on the limitations of these analyses in the discussion of the meta-analysis of brain cancer, above.]
Separate evaluations were made for the two major leukemia subtypes, chronic lymphocytic leukemia (CLL) and acute myelogenous leukemia (AML), and for all leukemias.
Chronic lymphocytic leukemia: The association between exposure to magnetic fields and CLL was considered in three studies of incidence, two in Sweden (Feychting et al., 1997; Floderus et al., 1993) and one in Canada and France involving three separate cohorts (Thériault et al., 1994), and in one of mortality in the USA (Savitz & Loomis, 1995).
No association was found in the US mortality study. The diagnoses were, however, based on death certificates, which is problematic for leukemia subtypes and particularly for CLL, because of the long survival time.
In the Canada-France incidence study of electric utility workers, a nonsignificantly increased risk was seen overall and in two of the three cohorts. A significant increase was seen in both of the Swedish studies. One of these (Feychting et al., 1997) provides unique information on the potential importance of combining occupational and residential exposures for adults, but it suffers from small numbers. In addition, their exposure assessment was based on a job-exposure matrix derived from magnetic field measurements for a different population of male workers, so their occupational exposures were not validated, especially for female workers. In the other Swedish study (Floderus et al.) of male workers in all occupations, the risk increased with increasing exposure; the risk was particularly strong for the highest exposure category and was increased somewhat when adjusted for exposure to potential confounders. The refusal rate in that study, however, could have introduced bias into the results.
Although each of these studies has its limitations, the limitations are different across studies, as are the designs and exposure assessment methods. Taken together, the studies of incidence suggest an association between exposure to magnetic fields and CLL.
Acute myelogenous leukemia: The association between exposure to magnetic fields and AML was considered in the same studies as for CLL. A nonsignificant increase in risk was found in the US mortality study, although the use of diagnoses from death certificates is problematic, as mentioned above.
In the Canada-France study, a significantly increased risk was seen overall for exposures above the median; this association is due mainly to a very high risk in one cohort, whereas a much smaller risk was seen in another cohort. The differences in definition and follow-up between the three studies, however, limit interpretation of the results. A nonsignificant increase in risk was seen in the study of Feychting et al. (Feychting et al., 1997), which became significant when restricted to the very small number of subjects who had both high occupational and high residential exposures. Although the study of Feychting et al. provides unique information on the potential importance of combining occupational and residential exposures in adults, it suffers from small numbers and weaknesses in exposure assessment, particularly for women. In the study of Floderus et al. (Floderus et al., 1993), no association was seen between exposure to magnetic fields and the risk for AML.
Leukemia: The association between exposure to magnetic fields and risk for leukemia in general was considered in the same studies. No association was found in either of the two US studies of mortality. The limitations of death certificate diagnoses mentioned above are less critical for leukemia in general than they are for specific subtypes. In the Canada-France study, no significant association was seen overall, although a significant association was seen in one cohort. The differences in definition and follow-up among three studies, however, limit interpretation of the results. A marginally significant association was seen in both Swedish studies; in the study of Feychting et al. (Feychting et al., 1997), when the analyses were restricted to subjects with high occupational and residential exposures, a significant elevation in risk was seen, based on nine cases. Although the study of Feychting et al. provides unique information on the potential importance of combining occupational and residential exposures in adults, it suffers from small numbers and weaknesses in exposure assessment, particularly for women.
One US study found a significant association in the highest exposure category and evidence for an exposure-response trend. The smaller US study showed no association. Both studies are based on diagnoses from death certificates, which is problematic for brain cancer owing to the difficulty in distinguishing primary cancers from metastases.
A nonsignificant elevation in risk was seen in the Canada-France study and in each of the cohorts in that study. In the study of Floderus et al. (Floderus et al., 1993), an association was reported between exposure to magnetic fields and brain cancer, which was significant only in one of the intermediate exposure categories; no evidence for a dose-response relationship was observed. No association was observed in the study of Feychting et al.
Although each of these studies has its limitations, the limitations are different across studies, as are the designs and exposure assessment methods. Taken together, the studies suggest an association between exposure to magnetic fields and brain cancer, although the results are somewhat inconsistent.
Male breast cancer: The relationship between the risk for male breast cancer and exposure to magnetic fields has been examined in only one study, in Sweden, in which exposures were assessed with a JEM derived from full-shift measurements of magnetic fields. No association was observed, although no adjustment for potential confounders was made.
This association was also considered in nine studies in which only job titles were used to classify workers by exposure. Only one study involved large numbers of cases and took into account risk factors for male breast cancers. In that study, a two-fold increase of borderline significance was seen among men in all exposed occupations combined; a significant increase was seen for the category of workers in electrical trades. The exposure assessment based on job title was not validated by measurements. The other studies, which were based on smaller numbers and had various limitations, gave inconsistent results. Most of these studies were not designed a priori to test this hypothesis.
Female breast cancer: The relationship between the risk for breast cancer in women and exposure to magnetic fields assessed with a JEM derived from full-shift measurements has been examined in only one study, in Denmark. No association was observed, but no adjustment was made for potential confounders.
Three other studies, in the USA, were based on job titles; in two, these were classified by experts into categories of probable exposure to EMF. These studies, which have methodological limitations mainly because they were not designed a priori to test an association with EMF, had mixed results.
[This conclusion was supported by 14 members of the Working Group; there were 11 votes for 'inadequate' evidence, 2 abstentions, and 2 absent.]
There is inadequate evidence for all other cancers.
[This conclusion was supported by 22 Working Group members; there were 2 votes for 'limited' evidence, 1 vote for 'lack' of evidence, 2 abstentions, and 2 absent.]
Verkasalo et al. (1996a) (Verkasalo, 1996) tested the hypothesis of a relationship between cancer risk and exposure to magnetic fields from high-voltage transmission lines in a cohort study in Finland. Information was obtained by linkage between the Finnish Cancer Registry, data on residential exposure to magnetic fields from the Finnish Transmission Line Cohort Study, and residential data from the Central Population Register and the 1970 Population Census. The study cohort consisted of all persons who had resided in a building with a calculated magnetic field of > 0.1 µT for any period between 1970 and 1989. Follow-up was from 1 January 1974 to 31 December 1989. Internal analyses were carried out by Poisson regression on data grouped on age (five-year categories), cumulative exposure, and social class [cumulative exposure was defined as cumulative exposure at the end of follow-up and was not treated as a time-dependent variable in the analysis]. In addition, SIRs were calculated with respect to the Finnish general population. The study cohort was very large, consisting of 383, 700 persons (189, 300 men) who contributed 2.5 million person-years of follow-up after age 20. Overall, 8415 cancer cases were observed (4082 in men). There was no association between cumulative level of exposure and the risk for all cancers (incidence rate ratio, IRR, 0.98 per µT-year; 95% CI, 0.96-1.0) or for any specific cancer type studied. The authors reported that skin melanoma was the only cancer for which the risk was somewhat increased throughout the three highest cumulative exposure categories: the IRR per µT-year was 0.91 (0.58-1.4), 1.5 (1.0-2.3), 1.5 (0.80-2.9), and 1.3 (0.59-2.7) for exposure to 0.2-0.39, 0.4-0.99, 1.0-1.99, and 2.0 compared with < 0.2 µT, respectively, in people of each sex. A marginally significant increased IRR was seen for multiple myeloma in men (IRR, 1.2 per µT-year; 95% CI, 1.0-1.5), and a nonsignificant decrease was observed for women (IRR, 0.87 per µT-year; 95% CI, 0.57-1.3). A significant increase in the risk for colon cancer was seen in women (IRR, 1.2 per µT-year; 95% CI, 1.0-1.3) but not in men. [No measurements were made to validate the calculated fields, and no distinction was made between apartments and single-family dwellings. The lack of information on other sources of residential exposure to EMF might have resulted in substantial exposure misclassification. In addition, no information on other risk factors, apart from age and sex, was available].
Preston-Martin et al. (Preston-Martin et al., 1988) investigated the risk for leukemia associated with use of electric blankets (see section 2.1) by including questions in questionnaires that were being used in two recently started case-control studies of AML and chronic myelogenous leukemia (CML) in Los Angeles County in the USA. The patients were residents of the area, aged 20-69 years, with histologically confirmed AML or CML diagnosed between July 1979 and June 1985, identified through the Los Angeles County population-based cancer registry. Both cases and controls had to be alive and to be able to be interviewed in English. One neighborhood control was matched individually to each case on gender, race, and birth year (within five years). The cancer registry contained 858 eligible cases, of whom 485 were alive. A physician's permission to contact a patient was obtained for 415. The participation rate of patients was 61% (68 could not be located, and 52 refused to be interviewed). [The response rate of controls is not given.] Controls could not be found in three neighborhoods. In all, 293 matched pairs were included in the analysis, including 156 cases of AML and 137 cases of CML. As information on use of electric blankets was included after the beginning of the study, this information was available for only 224 matched pairs. Neither AML nor CML appeared to be related to any regular use of electric blankets: the ORs were 0.9 (95% CI, 0.5-1.6) for AML and 0.8 (95% CI, 0.4-1.6) for CML. Cases and controls did not differ by average duration of use, year of first regular use, or year since last use. Adjustment for other risk factors found in the study did not change the results. [There was no indication of whether blankets had been used only for warming the bed or continuously throughout the night.]
Severson et al. (Severson et al., 1988) carried out a population-based case-control study of ANLL in adults in relation to residential exposure to power-frequency magnetic fields in King, Pierce, and Snohomish counties in western Washington State in the USA. The cases were identified through the cancer registry of the Fred Hutchinson Cancer Research Center, which is part of the SEER program. Cases were restricted to those of people aged 20-79 in whom ANLL had been diagnosed between 1 January 1981 and 31 December 1984. Both living and deceased patients were included, for a total of 164 cases. Random-digit dialing was used to select 204 controls in the same three geographical areas, who were frequency matched to the patients by gender and age (in five-year groups). A detailed questionnaire was administered to patients (or their next-of-kin if the patients had died) and controls; the questionnaire covered complete residential history and use of electric appliances. Exposure was assessed in three ways. First, the wire coding method of Wertheimer and Leeper (Wertheimer & Leeper, 1979) was used to classify all addresses in the study area in which the subject had resided in the previous 15 years. On the basis of maps of the residences and the wiring configuration, magnetic fields inside the residence were also estimated by a method developed by Kaune et al. (Kaune et al., 1987) . In addition, one-time measurements of the 60 Hz magnetic fields were made inside and outside the residence using a Power-frequency Field Meter Model 111 at the time of the interview if the subject had lived there for a continuous period of one year or longer immediately prior to the reference date (date of diagnosis for the cases). The measurements were made in the kitchen, the subject's bedroom, and the family gathering room, in both 'low power' and 'high power' configurations (i.e. with all major appliances that could conveniently be turned on and off). Finally, 24-h measurements were made in a limited sample of residences.
The response rates were 70% for cases and 65% for controls. For cases, most of the non-responses were due to physician refusal (17%), and only 4.9% of the subjects refused to be interviewed. The percentage of refusals among controls was much higher: (27%). The final analyses included 114 cases, with 91 cases of AML, and 133 controls. The patients tended to be of lower socioeconomic status than the controls and to smoke more; these factors were therefore adjusted for in subsequent analyses. There was no association between the risk for leukemia and electrical wiring configuration by the Wertheimer and Leeper scheme, either in the residence inhabited the longest in the 3-10 years before the reference date or in the residence closest to the reference date. There was also no association with magnetic fields calculated by the model of Kaune et al. One-time measurements were available for only 56% of homes, since many subjects had moved after the reference date. A non-significant increase in odds ratio was observed for mean exposures of 0.2 µT or more, in both low-power (OR, 1.5; 95% CI, 0.48-4.7) and high-power conditions (1.6; 0.49-5.0). When weighted mean exposure was considered, the increase was no longer apparent in low-power conditions and was reduced in high-power conditions (1.3; 0.35-4.5). No association was found with use of electric blankets, electric water-bed heater, or electric mattress pads. [The participation rates were low, and the type of refusal differed between cases and controls. Also, information on electric blanket use was limited.]
Lovely et al. (Lovely et al., 1994) further analyzed the data obtained in this case-control study, focusing on information obtained by questionnaire on use of motor-driven personal appliances (electric razors, hair dryers, and massage units), for which it was judged that use would represent higher exposures to peak magnetic fields than most other home appliances. Information was available on whether subjects had ever used such appliances, on the number of years (in the 10 years before diagnosis) they had had such appliances, and on the daily length of reported use. The authors also characterized magnetic fields produced by several models of hair dryers, electric razors, and massage units and found that partial body exposure could exceed 0.5 µT at rates of change exceeding 10 T/s. Massage units resulted in the highest average peak magnetic flux densities (0.35 mT ± 0.15); electric razors were in the middle range (0.19 mT ± 0.17), and hair dyers in the lowest (0.03 mT ± 0.004).
There was no association between ever using one of these appliances and the risk for leukemia (unadjusted OR, 0.71; 95% CI, 0.41-1.2 for ever vs. never used). When the appliances were considered individually, massage units had been used more frequently by cases (3.0; 1.4-6.3, based on 24 exposed cases), while hair dryers had been used more frequently by controls (0.38; 0.22-0.66). There was a nonsignificantly increased risk for leukemia among ever users of electric razors (1.3; 0.80-2.2). A positive trend (p < 0.05) was seen for reported daily of use of electric razors; the unadjusted odds ratios were 0.70 (0.32-1.5) for use up to 2.5 min/d, 1.6 (0.76-3.25) for use 2.5-7.5 min/d, and 2.4 (1.1-5.5) for use > 7.5 min/d in comparison with never use. No pattern of risk with length of use was seen for massage units [the numbers of exposed cases and controls were quite small for this analysis] or hair dryers. The authors noted that information on use may have been limited, since the majority of leukemia patients in the study had died before the interview and information on reported duration of use was obtained from proxy respondents.
Data from this study were further analyzed by Sussman et al. (Sussman et al., 1996) to assess the likelihood that the observed association with duration of use of electric razors was due to a bias related to the fact that proxy respondents had been used only for cases and not for controls. Of 110 patients for whom information on duration of use of electric razors was available, only 24 had provided this information themselves. On the basis of this small number of cases, there was no association between ever use of electric razors and the risk for leukemia (OR, 0.7; 95% CI, 0.3-1.7). There was also no association with reported daily length of use: the odds ratios were 0.6 (0.2-2.3; three exposed cases) for use up to 2.5 min/d, 0.6 (0.1-2.6; two exposed cases) for 2.5-7.5 min/d, and 0.9 (0.2-4.7; two exposed cases) for > 7.5 min/d in comparison with never use. When analyses were restricted to proxy respondents, however, the odds ratio for ever use was 1.6 (0.9-2.8), and a significant trend with daily duration of use was observed. The authors suggested that the association reported by Lovely et al. was therefore an artifact due to proxy response. [The number of patients for whom information was obtained directly is too small to rule out similar odds ratios].
A population-based case-control study was carried out in Sweden by Feychting and Ahlbom (Feychting & Ahlbom, 1992; Feychting & Ahlbom, 1994) to assess the impact on the risk for leukemia of exposure to magnetic fields from high-voltage transmission lines. The methods used were similar to those used in their case-control study of childhood leukemia (Feychting & Ahlbom, 1993). The study base consisted of all persons aged 16 years or more who had lived at least one year on a property located within 300 m of any of the 220 and 400 kV transmission lines in Sweden during the period 1960-85. These individuals were identified from maps of the Central Board for Real Estate Data to identify properties at least partially located on the corridor and the Population Registry to identify the approximately 400,000 residents of these properties. The case-control study was nested in this study base. Cases of leukemia were identified by record linkage with the Swedish Cancer Registry for the period 1960-85. For each case, two controls were selected at random from the cohort among those who had lived in the power-line corridor at least one year before the reference date and who lived near the same power line as the case. The controls were matched to the case on age (within five years), gender, parish of residence, and year of the diagnosis.
A detailed description of the exposure assessment method is given in section 2. It included spot measurements and calculations of fields generated by the power line. For each case and control, the historical field was then calculated from the average load for the relevant years preceding the diagnosis. Calculations were made for the year of diagnosis, or closest to diagnosis if the subject had moved, and for 1, 5, and 10 years before diagnosis. These estimated levels were used to assign an average magnetic field exposure in microtesla for each year the subject had lived in the power corridor over the previous 15 years. Information on the following confounding factors was available: age, gender, year of diagnosis, residence in the county of Stockholm or not, type of housing (single-family vs. apartment), and socioeconomic status.
A total of 325 cases of leukemia (72 AML, 57 CML, 14 ALL, and 132 CLL) and 1091 controls were included in the analysis. There was no association between the risk for all leukemia and calculated exposure to magnetic fields closest to the time of diagnosis. For AML and CML, however, non-significantly increased odds ratios were seen for fields of > 0.2 µT in comparison with < 0.09 µT (AML: OR, 1.7; 95% CI, 0.8-3.5; based on nine exposed cases; CML: 1.7; 0.7-3.8; based on seven exposed cases). For analyses based on calculated cumulative exposure during the 15 years preceding the diagnosis, the odds ratios for all leukemia were 1.0 (0.6-1.8) for cumulative exposures of 1.0-1.9 µT-years, 1.5 (1.0-2.4) for > 2.0 µT-years, and 1.5 (0.9-2.6) for > 3.0 µT-years, in comparison with < 0.99 µT-years. Elevated odds ratios were seen at exposure > 2.0 µT-years for both AML (2.3; 1.0-4.6) and CML (2.1; 0.9-4.7). Although these results were not adjusted for age or socioeconomic status, the authors reported that they were not greatly changed when adjustment was carried out. The authors also presented the results of matched analyses, which were similar to those of the unmatched analyses. The results of analyses based on spot measurements yielded odds ratios close to unity for all categories of exposure and leukemia subtypes, except for CML in the > 0.2 µT category (1.5; 0.7-3.2). The authors commented on the small number of cases, which prevented further analyses of duration of exposure, and on the lack of information on exposure to EMF from other sources, particularly occupational. [It is difficult to estimate exposures over a long period of follow-up.]
In a follow-up to this study, Feychting et al. (Feychting et al., 1997) obtained information on occupation from the censuses performed by Statistics Sweden every fifth year. Exposure was assessed from the occupation held in the year prior to the reference date (date of diagnosis for the cases) and a JEM developed from workday measurements in a large number of jobs held by a sample of the general male population within the framework of an occupational case-control study (see section 4.2.1.2). No information was available on the occupations of 43% of the women.
No association was seen between the risk for all leukemia and calculated residential exposure to magnetic fields after exclusion of subjects who were not exposed residentially but who were exposed to > 0.13 µT occupationally. Again, however, elevated odds ratios were seen for AML (OR, 2.4; 95% CI, 0.9-5.7) and CML (2.1; 0.8-5.5) among patients in the highest exposure category (> 0.2 µT) compared with the lowest (< 0.1 µT). When the analyses were restricted to people who had only residential exposure in the highest category (compared with 'unexposed' subjects with < 0.1 µT residential and < 0.13 µT occupational exposure), the odds ratios for AML and CML were no longer elevated (AML: 1.3; 0.4-5.0; based on three exposed cases; CML: 0.5; 0.1-3.9), although the very small number of cases prevents any interpretation of these results. The odds ratios for patients who had both high occupational and high residential exposure were much higher than those who had either (6.3; 1.5-27 for AML and CML based on three exposed cases of each subtype). [The limitations of the previous study apply here as well. This study is the first to combine both residential and occupational exposures (see section 4.2.1 for more discussion). The validity of extrapolating occupational information, particularly for women, is questionable.]
In a very large cohort study in Finland, described in detail in section 4.2.2.1), Verkasalo (Verkasalo, 1996) considered the risk for leukemia associated with calculated exposure to magnetic fields from transmission lines. These data were analyzed in more detail in case-control analyses [mistakenly referred to as case-cohort analyses]. A total of 196 cases of leukemia were included, with 60 of AML, 12 of ALL, 30 of CML, 73 of CLL, and 21 of other or unknown subtype. For each case, 10 controls were selected from the cohort and matched on gender, age at diagnosis (within one year), and whether they were alive in the year of diagnosis of the case. The statistical analyses were based on conditional logistic regression, adjusted for type of municipality. Several exposure measures were used: cumulative exposure and exposure 0-4, 5-9, and > 10 years before diagnosis; annual average magnetic fields 1-20 years before diagnosis; highest annual average magnetic field and in windows of 0-4, 5-9, and > 10 years before diagnosis; and age at first exposure to annual average magnetic field greater than a specified level and duration and time since exposure to annual averages above that level.
No association was seen between the risk for all leukemia or for specific subtypes and cumulative exposure or highest annual average exposure. Adjustment for type of housing or for occupational exposure (none vs. possible or probable, based on expert judgment) did not affect the results. For CLL, a significant increase was seen with dichotomized cumulative exposures of > 0.2 µT-years and > 0.4 µT-years for > 10 years before diagnosis and for durations of exposure to fields of > 0.10 µT for > 12 years, based on three exposed cases. No association was seen for other types of leukemia. [The limitations noted for the study of Verkasalo et al. apply here as well; however, type of residence was examined in this analysis.] (Verkasalo, 1996)
In a case-control study described in detail in section 4.2.1.2, Li et al. (Li et al., 1997) studied 1135 histologically confirmed incident cases of leukemia in northern Taiwan which were identified from the National Cancer Registry between 1987 and 1992. Controls were chosen among people with cancers other than those potentially related to exposure to magnetic fields. Eleven case-control pairs were excluded after verification of hospital records and a further 416 because one of the members of the pair resided in one of the 14 districts for which maps were not availability and therefore exposure could not be assessed. The addresses of the remaining cases and controls were plotted on detailed maps with information on utility routes; distance to the nearest transmission line was then estimated to within 10 m. Average and maximum residential magnetic fields were assessed on the basis of distance from the lines and information on distance between wires, height of wires above the ground, annual and maximum loads on the lines in 1987-92, current phase, and geographical resistivity of earth. The calculated magnetic fields were validated by direct 30-40-min indoor measurements made at low-power conditions in 407 residences. With results grouped in three categories (< 0.1, 0.1-0.2, > 0.2 µT), the agreement (k) between measured and calculated fields was 0.64 (95% CI, 0.5-0.78), which increased to 0.82 (0.79-0.86) when analyses used cut off points 0.5-1 µT and were restricted to houses in which both the measured and the calculated fields were > 0.2 µT. Limited information was available on potential confounders because of restrictions on interviews for the study.
The odds ratios were increased for patients in the middle and highest categories of calculated exposure to magnetic fields in the year of diagnosis: 1.3 (0.8-1.9) for 0.1-0.2 µT and 1.4 (1.0-1.9) for > 0.2 µT, compared with < 0.1 µT. In analyses by leukemia subtype, the odds ratios varied between 0.8 and 2.8. [The numbers of exposed cases by subtype were too small for these results to be interpreted; calculations of magnetic field were limited to the year of diagnosis because previous addresses were not available (see section 2 for comments on the exposure assessment in this study). Only limited information on confounders was available. Although the authors excluded cancer types thought to be potentially related to exposure to EMF, use of cancer cases as controls is still a concern.]
The odds ratio for breast cancer with use of electric blankets was 0.89 when adjusted for age and education and 1.0 after further adjustment for risk factors for postmenopausal breast cancer: body mass index, age at first pregnancy, number of pregnancies, age at menarche, family history of breast cancer, and history of benign breast disease. There was no trend in odds ratio with increasing years or frequency of use. In the analysis of the general mode of use of the electric blankets, a slightly increased risk was found for women who reported using electric blankets continuously throughout the night compared with never users (OR adjusted for all risk factors, 1.5; 95% CI, 0.96-2.2). Further analyses showed a slightly increased risk in the heaviest users (only 8% of cases and 6% of controls), i.e. those who had used electric blankets continuously throughout the night, daily during the season over the previous 10 years (1.4; 0.77-2.4). The authors concluded that there was no association between use of electric blankets and risk for postmenopausal breast cancer, although the increased risk among heaviest users should be investigated further. [Interpretation of this study is complicated by the very low response rates, particularly among controls, the lack of information on the type and age of the electric blankets, and the absence of information on other source of exposure to ELF EMF.]
In a complementary study, Vena et al. (Vena et al., 1994) also studied the role of electric blankets in women aged 40 or more with premenopausal breast cancer diagnosed between 1986 and 1991. Women were considered premenopausal if they were still menstruating or not menstruating due to medical intervention yet retained one ovary and were under the age of 50. Controls were drawn from the drivers' license rosters and frequency matched on age to the cases. The response rates were again low: 66% among cases (72% of thenon-interviewed cases were due to denial of permission to interview by physician) and 62% among controls. Interviews were carried out at home with a lengthy questionnaire including questions about electric blanket use. The odds ratio for use of an electric blanket at any time in the previous 10 years was 1.2 (95% CI, 0.83-1.7) after adjustment for age, education, and other risk factors, as in the study of postmenopausal breast cancer. There was no trend in risk with number of years of use. A slight increase in risk was observed among women reporting daily use during a season in comparison with never users (1.3; 0.86-1.9) and among those who reported using electric blankets continuously throughout the night (1.4; 0.94-2.2). The odds ratio among the heaviest users (continuously throughout the night, daily during the season over the previous 10 years) was 1.1 (0.59-2.1). [The interpretation of this study is complicated by the low response rates, the lack of information on the type and age of the electric blankets, and the absence of information on other sources of exposure to ELF EMF.]
In a letter to the Editor, Stevens (Stevens, 1995) commented on the results of these two studies. He suggested that they should be combined, thus increasing their statistical power and the likelihood that the small increases in risk noted for continuous use throughout the night in both studies would reach statistical significance. He further commented on the possible mechanisms by which EMF could affect the risk for breast cancer, suggesting that if they reversed the oncostatic effect of melatonin at the site of action in breast tissue they would be more important for estrogen receptor-positive tumors. Vena et al. (Vena et al., 1995) responded that pre- and postmenopausal tumors had been separated because of their different etiologies. The a priori hypothesis of these studies was that EMF differentially affect premenopausal and postmenopausal tumors on the basis of Stevens' (Stevens, 1987) hypothesis that they affect risk through the pineal-melatonin pathway, which is more relevant for premenopausal women. Vena et al. (Vena et al., 1995) nevertheless presented the results of combined analyses. The odds ratio was 1.1 (95% CI, 0.85-1.4) for ever use, 1.2 (0.90-1.5) for daily use, and 1.5 (1.1-1.9) for continuous use throughout the night. Analysis by duration of continuous use throughout the night showed no association. The authors cautioned about over interpretation of these results because of the limited capacity of the study to address such issues and the retrospective nature of the exposure assessment.
A case-control study of several adult cancers and residential exposure to 60 Hz magnetic fields from transmission lines was carried out in northern Taiwan by Li et al. (Li et al., 1997) (see section 4.2.2.2). A total of 2407 histologically confirmed incident cases of breast cancer reported between 1990 and 1992 were identified from the National Cancer Registry. The same number of controls were drawn from the registry among women with cancer, excluding leukemia, brain cancer, breast cancer, cancers of the hematopoietic and reticuloendothelial system, skin, ovary, fallopian tube, and broad ligament. They were matched one-to-one to cases on date of birth (within five years) and date of diagnosis (within six months). Twenty-two case-control pairs were excluded after verification of hospital records and a further 823 because one of the members of the pair resided in one of the 14 districts for which maps were not available, leaving 1562 case-control pairs for analysis. An index of urbanization was derived on the basis of local population density, age, mobility, economic activity, family income, educational level, and sanitation facilities. There was no association with calculated exposure to magnetic fields in the year of diagnosis (OR, 1.1 for exposure to > 0.2 or 0.1-0.2 µT in comparison with < 0.1 µT). [See comments on the limitations of this study in section 4.2.2.2.]
In the Finnish cohort study described in section 4.2.2.1, Verkasalo et al. (Verkasalo et al., 1996) also examined the risk for breast cancer in relation to calculated residential exposure to magnetic fields. Overall, 1229 cases of breast cancer were found among women in the cohort. No association was seen with risk, with SIRs of 1.1 (0.98-1.1) for exposure to < 0.2 µT, 1.1 (0.88-1.3) for exposure to 0.20-0.39 µT, 0.89 (0.71-1.1) for exposure to 0.40-0.99 µT, 1.2 (0.89-1.6) for exposure to 1.00-1.99 µT, and 0.75 (0.48-1.1) for exposure to > 2.00 µT. [See comments on the limitations of this study in section 4.2.2.2.]
In their population-based nested case-control study described in detail above (section 4.2.2.2), Feychting et al. (Feychting et al., I998) also considered the effect of magnetic fields from high-voltage transmission lines on the risk for breast cancer, using 699 cases in women and nine in men, identified from the Swedish Cancer Registry. Controls were selected at random from the study base among people who had lived in the power-line corridor for at least one year before the reference date and who lived near the same power line as the case. The controls were matched to the cases on age (within five years), gender, and parish of residence in the year of the diagnosis. Matching was one-to-one for female breast cancer cases and one-to-eight for male breast cancer cases. A detailed description of the method of exposure assessment is given in section 4.3.1). Information on the estrogen-receptor status of the tumor was obtained from medical records and was available for only 82 of the 699 cases.
There was no association between the risk for breast cancer in women and calculated exposure to magnetic fields closest in time to the diagnosis (OR, 1.0; 95% CI, 0.7-1.5 for 0.2 µT compared with < 0.1 µT). A nonsignificantly elevated risk was observed among women with cumulative exposures 3.0 µT-years in the six years immediately preceding diagnosis (1.6; 0.8-3.2). A nonsignificant increase was also observed among women aged < 50 (1.8; 0.7-4.3) (0.2 µT vs. <0.1 µT) and among women who were estrogen receptor-positive (1.6; 0.6-4.1) ( 0.1 µT vs. <0.1 µT), with an exposure cut-point of 0.1 µT. Among estrogen receptor-positive women under 50, the odds ratio was 7.4 (1.0-180), based on six exposed cases and one exposed control. In males, a nonsignificant increase in risk was observed (2.1; 0.3-14) for calculated exposure to magnetic fields > 0.2 µT closest to the time of diagnosis. [In addition to the limitations described in section 4.2.2.1, no information was available on important risk factors for breast cancer, since all of the data were obtained from registry and hospital files. Information on estrogen-receptor status, moreover, was available for only a limited number of cases.]
In a follow-up to this study, Feychting et al. (Feychting et al., 1997) obtained information on occupation from the censuses performed by Statistics Sweden as described in detail in section 4.2.2.2. No association was seen between the risk for tumors of the CNS and calculated residential exposure to magnetic fields after excluding subjects who were not exposed residentially but were exposed to > 0.13 µT occupationally; and no association was seen when the analyses were restricted to people who had only residential exposure (OR, 0.7; 95% CI, 0.3-1.7; seven exposed cases).
In a case-control study described in detail insection 4.2.2.2,Li et al. (Li et al., 1997) studied 705 histologically confirmed incident cases of brain tumor (ICD-O 191) in northern Taiwan identified from the National Cancer Registry between 1987 and 1992. Controls were chosen among people with cancer, excluding cancer types potentially related to exposure to magnetic fields. Nine case-control pairs were excluded after verification of hospital records and a further 241 because one of the members of the pair resided in one of the 14 districts for which maps were not available. No association was seen with calculated exposure to magnetic fields in the year of diagnosis (OR, 0.9; 95% CI, 0.5-1.7 for exposure to 0.1-0.2 µT; 1.1; 0.8-1.6 for exposure to > 0.2 µT, compared with < 0.1 µT). In analyses by tumor subtype, the odds ratios varied between 0.6 and 2.8. [See comments in section 4.4.2 on the limitations of this study.]
In the study of Verkasalo et al. (Verkasalo et al., 1996) described in section 4.2.2.1, the risk for gliomas and meningiomas was also analyzed. Over the study period, 301 cases of tumors of the CNS were observed. The incidence was not different from that of the Finnish population, and no association with calculated cumulative exposure to magnetic fields was observed. The SIRs with respect to the general population were 0.94 (95% CI, 0.8-1.1, 238 cases) for < 0.2 µT, 1.1 (0.77-1.5; 35 cases) for 0.2-0.39 µT, 0.64 (0.37-1.0, 16 cases) for 0.4-0.99 µT, 0.55 (0.18-1.3, five cases) for 1.00-1.99 µT, and 0.92 (0.37-1.9, seven cases) for > 2.00 µT. Although separate analyses were carried out for gliomas and meningiomas, and the authors reported only that the results were consistent with those for cancer of the nervous system as a whole. [See comments on the limitations of this study in section 4.4.2.]
Magnetic fields were calculated on the basis of distance from transmission lines, historical loads, and line configuration in three case-control studies, from Finland, Sweden, and Taiwan. No information on other sources of residential exposure was available in any of these studies. In the Swedish study, no association was observed between the risk for any leukemia and exposure to magnetic fields, either for calculated fields closest to the time of diagnosis or for cumulative exposure over 15 years. A slight increase in risk for acute myelogenous leukemia (AML) and chronic myelogenous leukemia (CML), but not for chronic lymphocytic leukemia (CLL), was seen with calculated fields. The increase for AML disappeared when the analyses were restricted to subjects with no or very low occupational exposure, based, however, on a very small number of cases. The risk of people with both high residential and high occupational exposures was increased. In the Finnish study, no association was seen between either all leukemias or specific subtypes and highest annual or cumulative exposure to magnetic fields. A significant increase in the risk for CLL (but not for other leukemia subtypes) was seen in association with cumulative exposure to > 20 or > 0.40 µT-years for 10 years or more before diagnosis. In the study in Taiwan, a small increase in risk for leukemia was seen in association with exposure to 0.1-0.2 or > 0.2 µT of magnetic fields during the year of diagnosis.
In one study in the USA, in which wire coding was used to estimate exposure, no association was seen with the risk for leukemia.
Direct measures of magnetic fields in the homes of the study subjects were used in two studies. In the US study, a small, nonsignificant increase was seen in association with mean exposure but not with weighted mean exposure; measurements were made, moreover, in only a fraction of homes. In the Swedish study, no association was seen between spot measurements of exposure to EMF and the risk for leukemia.
Calculated fields were used in three studies, in Finland, Sweden, and Taiwan. In the Finnish and Taiwanese studies, no association was seen between exposure to magnetic fields and the risk for breast cancer. In the Swedish study, although no association was seen overall, nonsignificant increases were observed in the highest cumulative exposure category among young women, particularly those who were estrogen receptor-positive (based on very small numbers).
The Swedish study also considered the risk for male breast cancer. Although the number of cases was extremely small, a two-fold, nonsignificant increase was observed.
[This conclusion was supported by 24 Working Group members; there was one vote for 'lack' of evidence, 1 abstention, and 3 absent.]