The Working Group decided to exclude studies in which established epidemiological methods were not used, for example those in which advertisements were used to identify subjects or in which dwellings were analyzed rather than subjects. Studies in which the methods were too crude to assess exposure to magnetic fields were not considered; thus, results based on distance alone were not included, nor were those of one study in which 0.1 µT was the highest exposure cut-point (Myers et al., 1990). One study in which exposure was assessed for only 12% of the subjects was also not evaluated (Dockerty et al., 1998). The Working Group also decided not to consider studies that were re-analyses of primary data. As weight is given in the evaluation to individual studies with primary data, formal meta-analyses are included but are only briefly discussed.
The characteristics of all of the studies are summarized in Table 4.20, while the results are summarized in Tables 4.21-4.24, and the meta-analyses are summarized in Tables 4.25 and 4.26.
In this exploratory study, when wire codes were assigned, the case or control status of the homes was known to the coder. Thus, exposure assessment was not conducted in a blinded fashion. This may have introduced bias, although the investigators conducted two smaller studies to assess the possible effect of lack of blinding on the outcome (Wertheimer & Leeper, 1979). In part of that study, a separate investigator blindly coded the homes of 70 cases and 70 controls and found a 91% agreement, with about half of the disagreements favoring the association with higher wire codes and half countering the association. [This rate of agreement is similar to that observed in studies in which the process of wire coding was blinded] (Tarone et al., 1998). In a second, smaller study including birth addresses in Colorado Springs and Pueblo, 32% of the cases and 18% of the controls had lived in high-wire code homes, whereas in the larger study the numbers were 37 and 20%, respectively. Furthermore, accounting for potential confounding factors and effect modifiers (e.g. socioeconomic class, urban-suburban differences, traffic density, and gender) did not change the results. None of the relative risks reported in this study was adjusted for these factors. [Death certificate-based studies are subject to differential survival bias. Children in families of higher socioeconomic class have greater access to health care and may thus have higher rates of survival from cancer. Consequently, there may be a bias toward lower socioeconomic cases in this study. If children in families of lower socioeconomic class tended to be categorized in the high-current configuration homes, such bias would inflate the risk estimates.]
Savitz et al. (Savitz et al., 1988) conducted a case-control study of residential exposure in the same area as that of Wertheimer and Leeper (Wertheimer & Leeper, 1979). They studied Denver residents under 15 years of age and assessed the exposure in the homes occupied at the time of cancer diagnosis and two years before diagnosis. Cases were obtained from population-based cancer registries and hospital records for the years 1976-83. Controls were selected by random-digit telephone dialing after the close of the study period and were further restricted to have been living in the study area at the time their matched cases (by age ± three years, gender, and telephone exchange at the time of cancer diagnosis of the matched case) were diagnosed. A total of 356 childhood cancer cases were eligible, and 278 population controls were identified. [It is unclear why there are fewer controls than cases despite the matched selection of controls.] Exposure in the home occupied at the time of diagnosis was assessed by the dichotomous wire coding scheme of Wertheimer and Leeper (Wertheimer & Leeper, 1979) and the five-level wire code of Wertheimer and Leeper (Wertheimer & Leeper, 1982) for 90% of cases and 93% of controls and by spot measurements of EMF taken 1-9 years after diagnosis, by the front door, in the child's and parents' bedrooms, and in all rooms occupied by the child for at least 1 h/d; 36% of cases and 75% of controls participated. Both the electric and magnetic fields were measured with a model 111 or 113 Electric Field Meter. A weighted average of measurements in all measured rooms was computed and used as a summary exposure measure.
The relative risk for all cancers among children living in homes in the high-current classification was 1.5 (95% CI, 1.0-2.3). A detailed categorization of wiring configurations (very low, which included buried wires, ordinary low, ordinary high, and very high) was analyzed to elucidate a dose-response relationship. The estimated relative risks tend to increase in a linear fashion up to an almost tripling of risk in the highest exposure category (2.8; 0.9-8.4), representing a statistically significant linear trend. This association was not corroborated by spot field-strength measurements, regardless of low or high power use conditions. For leukemias, a relative risk of 1.5 (0.9- 2.6) was reported when comparisons were made between high- and low-current classifications; the relative risk with spot measurements of >0.2 µT was 1.9 (0.7-5.6) under low-power use conditions and 1.4 (0.6-3.5) under high-power use conditions. For brain cancer, a relative risk of 2.0 (1.1-3.8) was found for the high-current configuration exposure category. For homes occupied at the time of diagnosis, the relative risk for brain cancer associated with average spot magnetic field-strengths > 0.2 µT was 1.0 (0.2- 4.8). No increase in the risk for childhood lymphoma was found in a comparison of high-current and low-current configurations. The relative risk for lymphoma with spot measurements >0.2 µT was 2.2 (0.5-10) under low and 1.8 (0.5-6.9) under high power use conditions. Approximately 20 other potential risk factors for childhood leukemia were considered, and those found to be related to the risk for leukemia, e.g., socioeconomic status, traffic density, maternal age, and smoking during pregnancy, were controlled for in the analyses. Control for confounding did not change the risk estimates.
[An important methodological issue is the selection of controls that represent a residentially stable subset of the population in which the cases occurred after the study period. Control selection bias may have been introduced if exposure was related to characteristics of residential stability. Jones et al. examined the effect of the differential stability in a different population and concluded that the control selection bias would lead to an exaggerated estimate of risk. Wertheimer and Leeper, however, presented additional analyses of the Savitz et al. study to show that the effect of the bias was to attenuate risk. The very low participation rate for spot measurements among the cases limits the validity of the results based on these measurements. Control selection through random-digit dialing is also a limitation. A strength of the study is the evaluation of a large number of potential confounding factors, including socioeconomic status and traffic density.] (Jones et al., 1993; Savitz et al., 1988; Wertheimer et al., 1994)
London et al. (London et al., 1991) conducted a case-control study of 232 cases and 232 population controls, focusing strictly on the incidence of childhood leukemia, and assessed exposures in selected homes during an 'etiologic period' defined as the period beginning at the estimated time of conception and ending on the date of diagnosis for children aged one year or less at diagnosis, six months before diagnosis for children aged one to two years at diagnosis, and one year before diagnosis for children aged above two years at diagnosis. The population base in this study consisted of all children under 10 years of age living in Los Angeles County during the years 1980-87. The 331 eligible cases were obtained from a population-based tumor registry. Controls were friends (65 controls accrued during the period 1980-84) or obtained by random-digit telephone dialing (167 controls accrued during the period 1985-87) and were matched to the cases by age, gender, and ethnicity. [Whether friends of cases are representative of the entire population from which the cases were identified is not clear. Issues previously discussed with regard to random-digit telephone dialing are also applicable here.]
Exposure was assessed by three methods: five-level wire code configuration classifications developed by Wertheimer and Leeper (Wertheimer & Leeper, 1982); spot measurements in the center of the child's bedroom of EMF, static magnetic fields, and the harmonic content of the magnetic field; and 24-h measurements of magnetic fields under the bed in the child's bedroom. Spot measurements were made with a Deno Power Frequency Meter 120; 24-h measurements were made with an IREQ meter in the early parts of the study and with an EMDEX-100 meter for most of the study. The comparability of the data collected with the IREQ and EMDEX meters was ascertained. Harmonic content was determined during the spot measurement period with the broad-band mode of the Deno meter. The investigators also recorded the Earth's magnetic field during the spot measurements with a fluxgate magnetometer (Bartington MAG-01). The participation rates were 42% for both cases and controls for the spot measurements, 50% for cases and 56% for controls for 24-h measurements, and 66% for cases and 81% for controls for wire codes. The measurements were made 1-10 years after diagnosis. Comparisons across these various exposure metrics and between previous studies showed that the 24-h average measurements were considerably higher than the spot measurements in the same homes, and in contrast to the study of Savitz et al. (Savitz et al., 1988), more than twice the number of control homes were classified in the high-current configuration categories (45%), but spot-measured magnetic fields were lower within five-level wiring configuration code categories (the difference was especially pronounced in the high-current categories). London et al. (London et al., 1991) noted that this discrepancy may have been due to the differences in the electrical distribution system in the Los Angeles and Denver areas.
The relative risk for leukemia in relation to wire code configuration (high- versus low-current classification) was 1.7 (95% CI, 1.1-2.5). With a more detailed categorization of wiring configuration (very low, ordinary low, ordinary high, and very high), the estimated relative risks tended to increase in a linear fashion (statistically significant linear trend) up to more than a doubling of risk in the highest exposure category (2.2; 1.1-4.3). No associations were found with spot electric or magnetic field measurements or static magnetic fields. The results with the 24-h magnetic field measurements indicated an increased risk only for the highest cut-point used (> 0.27 µT) with a relative risk of 1.5 (0.7-3.3). The authors noted that similar results were obtained for the ALL and ANLL subtypes of leukemia, but they did not report them separately. Potential factors associated with cancer risk that were controlled for and included were age, gender, ethnicity, paternal use of pesticides, use of cigarettes, drugs, and incense, traffic density, and socioeconomic status. [A limitation of this study is the uneven ascertainment of wire code information between cases and controls: spot and 24-h measurements of magnetic fields resulted in much lower percentages of ascertainment than wire codes.]
Feychting and Ahlbom (Feychting & Ahlbom, 1993) conducted a population-based case-control study in Sweden to examine the association between exposure to magnetic fields generated by high-voltage power lines and cancer incidence in children. The population base for this study consisted of all children who resided on a property located in a high-voltage power-line corridor, defined as a property located at least partially within 3 m of any 220 or 400 kV power lines. The subjects were children under 16 years of age who resided within a high-voltage power-line corridor during 1960-85. They were followed from the time they moved into the corridor through the end of the study period. The Swedish Cancer Registry was used to identify the 142 cancer cases occurring within high-voltage corridors during the study period. Approximately four controls per case (total, 558) were selected randomly from the study base and were matched according to age, gender, parish residence during the year of diagnosis or the last year before the case moved, and proximity to the same power line.
Exposure to magnetic fields was assessed by spot measurements, contemporary calculated fields, and historical calculated fields (see section 2.4). The spot measurements were made closely following the protocol implemented by Savitz et al. (Savitz et al., 1988) with a meter constructed for the purpose of the study. Spot measurements were performed in the home within the power-line corridor in which the patients and corresponding controls had lived closest to the time of diagnosis and were obtained for 62% of cases and controls. The measurements were made 5-31 years after diagnosis, with a median of 16 years (Feychting et al., 1995). Most of the dwellings in which measurements were not made were located in Stockholm. Contemporary calculated field strengths were estimated from information about the height of the towers, distance between the towers, distance between phases, ordering of phases, and line load. The contemporary line load was obtained during the visit to each dwelling. Both transmission and distribution lines were considered; however, less than 20% of the homes were located near distribution lines. Historical annual average line load during the study period was used in calculating historical field strengths for all but one case and four controls. In their analysis, Feychting and Ahlbom (Feychting & Ahlbom, 1993) emphasized the use of historical fields calculated from exposure metrics, due to the presumed accuracy of using historically calculated field strengths to estimate exposures several decades previously for some subjects and the fact that averaged spot measurements are not a representative measure of long-term exposure. Cut-points for the analyses of historical calculations consisted of a three-level, ordinal scale with the following categories: < 0.1 µT , 0.1 µT, < 0.2 µT , and >0.2 µT; analyses were also made for exposures of >0.3 µT. Feychting et al. (Feychting et al., 1995) examined exposures >0.5 µT. With spot measurements, the highest exposure examined was >0.2 µT.
The results for historically estimated field strengths showed elevated risks for childhood leukemia with increase in exposure (statistically significant linear trend), up to more than a tripling of risk in the exposure category >0.3 µT (RR, 3.8; 95% CI, 1.4-9.3). When the data on leukemia were further stratified by age and gender, a dose-response trend was observed for single-family dwellings but not multiple-family dwellings. Controlling for potential confounders (age, gender, county, dwelling type, year of diagnosis, socioeconomic status, and levels of nitrogen dioxide as an estimate of exposure to motor vehicle exhaust) did not alter this relationship. For exposure to 0.5 µT, the relative risk was 4.6 (1.5-14). There was no evidence in this study of an association between historically calculated exposure to magnetic fields and all cancers combined, lymphomas, or central nervous system tumors. In analyses based on spot measurements, no increase in risk was observed for all cancers, leukemia, or central nervous system tumors.
In analyses to validate the various calculations and measurements of magnetic fields used by Feychting and Ahlbom (Feychting & Ahlbom, 1993), spot measurements showed poor agreement with calculated historical fields but good agreement with contemporary calculations. This result was interpreted by the authors as an indication that spot measurements are poor predictors of exposure many years earlier. [A strength of this study is the minimal potential for selection bias in the analyses of historical calculations. Another strength is that the historical calculations are based on established laws of physics and conditions prevailing prior to diagnosis. A limitation is the small number of exposed subjects.]
Olsen et al. (Olsen et al., 1993) conducted a population-based case-control study to investigate whether residence before and after birth near high voltage facilities was associated with an increased risk for childhood cancer. They studied 1707 cases of leukemia, malignant lymphoma, and central nervous system tumor in Danish children aged < 15 reported to the Danish Cancer Registry during the years 1968-86. Two to five controls, selected randomly from the Danish Central Population Register among cancer-free children, were matched to each case by gender and age (± one year), for a total of 4788 controls. They assessed exposure from calculated average magnetic field strengths. Residences located outside the area of potential exposure to high-voltage facilities (generally > 300 m away from overhead lines or transformer substations) were assumed to have an average calculated magnetic field of zero. Estimates of relative risk were obtained for cut-points associated with low (0.1 µT), intermediate (0.25 µT), and high exposure (0.4 µT) and were adjusted for gender and age at diagnosis.
Comparisons made between exposure to >0.4 µT and to < 0.1 µT resulted in crude relative risks of 6.0 (95% CI, 0.8-44) for leukemia, 6.0 (0.7-44) for central nervous system tumors, and 5.0 (0.3- 82) for malignant lymphoma. In analyses further adjusted for the potential confounding effects of population density, socioeconomic class, and family's mobility, no effect on the risk estimates was observed. [The current on the transmission lines was not recorded, but was estimated on an annual average basis by an expert group of utility planners. This process led to uncertainty in the calculated fields. In addition, the methods of calculation were not confirmed by measurements. Due to the low prevalence of exposure, the risk estimates were unstable.]
Verkasalo et al. (Verkasalo et al., 1993) conducted a population-based cohort study of 134 800 Finnish children (68 300 boys and 66 500 girls) aged < 20 who lived within 500 m of 110-400 kV overhead power lines and to magnetic fields calculated to be >0.01 µT during the period 1970-89. The 140 cases of childhood cancer were obtained from the Finnish Cancer Registry and included all primary tumors of the nervous system, leukemia, lymphoma, and all other cancers grouped. Exposure to magnetic fields was estimated from calculations of the annual average fields for all years between birth and diagnosis, based on information about typical line configurations, historical load on the lines, and distance (obtained from computerized sources). The historical load for the last third of the observation period was obtained from simulations, those for the middle third from existing records, and those for the first third from the last year with existing records. Exposure was assessed by two calculated estimates: average magnetic field and cumulative exposure. Cumulative exposure was defined as the average exposure per year multiplied by the number of years exposed (µT-years). The cut-points chosen for high exposure were 0.2 µT for average exposure and 0.4 µT-years for cumulative exposure to magnetic fields. These cut-points were selected a priori from the distribution of the number of exposed children and taking into account the typical residential magnetic field of 0.01 µT (referent exposure). [The calculated fields were not validated by actual measurements.] A cohort approach with person-years calculations was used to investigate the risk of cancer in children living close to power lines. The expected number of cases was calculated from Finnish national incidence rates.
Regardless of exposure metric, a four-fold increase in risk for nervous system tumors was seen among boys (SIR, 4.2; 95% CI, 1.4-9.9) for cumulative exposure, but not for girls for whom no cases were observed at high exposure. The increase in risk was largely attributable to one boy who had three primary tumors of the nervous system. If the analysis had been restricted to first primary cancers, the estimate would have been slightly elevated, with a larger confidence interval overlapping the null. The SIRs for leukemias, lymphomas, and cancers at other sites were close to unity for both average and cumulative exposure. The results for cumulative exposure of >1.0 µT-years has also been reported (Verkasalo et al., 1994). The relative risk estimates were 2.3 (95% CI, 1.0-1.3) for all cancers combined, 3.5 (0.7-1.0) for leukemia, and 2.8 (0.6-8.1) for nervous system tumors. These risk estimates were not adjusted for potential confounding factors. In particular, Verkasalo et al. (Verkasalo et al., 1993) reported that childhood cancer, other than leukemia, was commoner in urban than in rural settings. Thus, the crude risk estimates reported (excepting leukemia) are slightly higher than they would have been had the analysis been adjusted for this covariant. [Selection bias is not a concern in this study since it is population-based. The study is limited by the small number of exposed subjects. The calculated field strengths were not validated by measurements. The validity of including three brain tumors in one subject as three separate cases is questionable.]
The United States West Coast Childhood Brain Tumor Study was a multicenter, population-based case-control interview study to evaluate potential environmental and nutritional risk factors for diagnosis of brain tumor (benign or malignant tumor of the brain, cranial nerves, or cranial meninges) in children aged < 19 in 1984-90. To be eligible for participation in the study, each child's biological mother had to speak English, be available for interview, and have a telephone. Mothers of children with brain tumors were asked questions about exposure and conditions thought likely to be related to risks for pediatric tumor (e.g. ionizing radiation, predisposing genetic syndromes, exposure to magnetic fields during pregnancy). Subanalyses of this multicenter trial addressed the risk for childhood brain tumor in relation to residential exposure to magnetic fields in Los Angeles, California (Preston-Martin et al., 1996b), to residential power-line configurations, electric heating sources, and electric appliances in Seattle, Washington (Gurney et al., 1996), and the use of electric blankets and water-bed heaters for the entire study population (Preston-Martin et al., 1996a).
Preston-Martin et al. (Preston-Martin et al., 1996b) initiated a study of magnetic fields two years after the beginning of the multicenter interview study. The parents of cases were re-contacted by telephone in order to obtain measurements of magnetic fields at their residences. Of the 304 cases in the multicenter study, 298 cases were included in the substudy. A control group of 298 children within the same range of birth years and with the same distribution by gender as the cases were identified by random-digit dialing, were matched to the cases by gender, birth date (± one year), and had to be the same age at the time of interview as the case had been at the time of diagnosis. Cases and controls were accrued concurrently during 1989 to the end of the study period; before 1989, the controls were accrued by random-digit dialing [but further details are not given]. Exposure to magnetic fields was assessed from spot measurements taken outside the residence and from wire code configurations. Magnetic fields could be measured for 59% of the eligible cases and 54% of the eligible controls. The exterior residential measures included the fields over water meters and water pipes, static magnetic fields, front-door fields, and STAR magnetic field profiles (see section 2.3.1), including the front wall and perimeter of the dwelling. [The STAR meter does not measure magnetic fields at harmonic frequencies.] For cases whose current residence was also the residence in which they lived at the time of diagnosis (and for their matched controls), interior home measurements were taken, consisting of 24-h EMDEX measurements in the children's bedrooms and in a second room in which the children spent most of their time.
Wire codes were obtained for three types of residences: the residence occupied at the beginning of the study (nine months before birth), the residence occupied for the longest time, and the residence occupied at the date of diagnosis. Wire codes were obtained for 80% of cases for homes two years before diagnosis. The investigators found that too few subjects were within category of the usual reference wire code, 'underground', to serve as a stable reference, and so they pooled the 114 cases and 102 controls in the 'very low' and 'ordinary low' categories and used this as the reference category. [This categorization of wire codes is unusual.]
Preston-Martin et al. (Preston-Martin et al., 1996b) reported elevated risks for brain cancer in several exposure categories but no statistically significant trend, e.g. for all exposure metrics. When the analyses were restricted to very high exposure (> 0.3 µT ), the 24-h EMDEX measure revealed a relative risk of 1.7 (95% CI, 0.6- 5.0), and spot measurements and STAR profiles both led to a relative risk of 0.9 (95% CI, 0.3- 3.2 and 0.2-4.1, respectively). These estimates are highly unstable owing to the small numbers of subjects living in residences with magnetic field strengths > 0.3 µT, i.e. 12 or fewer cases and 7 or fewer controls. [Limitations associated with random-digit dialing telephone methods to select controls are applicable to this study. Notably, the participation rates for all EMF measurements except wire codes were low. Stratified analyses based on whether controls were concurrently accrued with cases yielded inconsistent risk estimates with wire codes.]
Gurney et al. (Gurney et al., 1996) initiated an epidemiological study to assess the relationship between childhood brain cancer and proximity to high-current power lines. The study population was derived from the Seattle, Washington, component of the multicenter United States West Coast Childhood Brain Tumor Study and comprised children < 20 years of age at diagnosis of a primary brain tumor in 1984-90, who were identified from a population-based cancer registry. Of the 179 eligible cases, 133 participated in the study (a participation rate of 74%). A control group of children stratified within the same range of birth years and county of residence as the cases was identified by random-digit dialing, which resulted in the participation of 270 controls out of 343 eligible for inclusion (a participation rate of 79%). Exposure was assessed by the five-level wire coding scheme developed by Wertheimer and Leeper (Wertheimer & Leeper, 1982).
Gurney et al. (Gurney et al., 1996) reported no association between the occurrence of pediatric brain tumors and residential exposure to magnetic field sources, which included analyses of five-level and two-level wire code configurations. The risk for brain tumor did not increase with increasing exposure (relative to underground wiring) when the five-level Wertheimer and Leeper wire code configuration was analyzed, the relative risks being 1.3 (95% CI, 0.7-2.1) for very low current configuration, 0.7 (0.3-1.6) for low current configuration, 1.1 (0.6-2.1) for high current configuration, and 0.5 (0.2-1.6) for very high current configuration. When the wire codes were dichotomized (high vs. low), the relative risk was approximately unity. After evaluating 13 potential risk factors, the authors found no associations, and thus crude risk estimates were reported. [The limitations associated with random-digit dialing to select controls are applicable to this study. A strength is the large number of potential confounding factors evaluated.]
Linet et al. (Linet et al., 1997) assessed the association between childhood ALL and residential exposure to magnetic fields for children aged < 15 who resided in Illinois, Indiana, Iowa, Michigan, Minnesota, New Jersey, Ohio, Pennsylvania, or Wisconsin and were registered with the Children's Cancer Group, in whom ALL was diagnosed during the period 1989-94. A total of 767 cases were eligible for inclusion. Random-digit telephone dialing was used to recruit the controls, who were individually matched to the cases by the first eight digits of their telephone numbers (including area code), age, and race, resulting in 725 controls who were eligible for inclusion. The authors reported a participation rate of 78% for cases and 63% for controls.
Magnetic fields were measured at all subjects' residences with EMDEX-C meters. The standardized measurement protocol followed included 24-h measurements in the child's bedroom (with the meter placed under or adjacent to the bed), 30-s measurements in the center of the child's bedroom, the family room, the kitchen, and the room in which the mother slept during the subject's pregnancy, and a 30-s outdoor measurement made within 1 m of the front door of the residence. A single summary exposure metric for each home was calculated from a weighted average of the room measurements (see section 2.3.1). The weights were based on the typical amount of time spent in each room according to the child's age (Kleinerman et al., 1997). For each child under the age of five, the investigators attempted to measure magnetic fields in all homes the subject had lived in for at least six months and required that at least 70% of the child's life had been spent in the measured homes. For children over the age of five, the investigators measured a maximum of two homes lived in during a reference period of five years immediately preceding diagnosis, and required that the child had lived for at least 70% of the reference period in the measured houses. A weighted average of all homes was used as the summary exposure metric. On the basis of the distribution of the measurements in the control homes, the following exposure categories were chosen a priori for the summary measurements of residential magnetic fields: 0.065 µT, 0.065-0.09 µT, 0.1-0.19 µT, and >0.2 µT . Higher exposures (up to 0.5 µT) were also measured. Wire code classifications (the five-level Wertheimer-Leeper classification and the modified three-level Kaune-Savitz scheme) were assigned to the subjects' main residences for a subgroup of 408 case-control pairs in which both the case and the control had been residentially stable, i.e. lived for at least 70% of the reference period in one home, and to those in which the family had lived during the mother's pregnancy (230 matched case-control pairs). [In most cases, the fields were measured within two years of diagnosis; thus, the measurements taken are more representative of exposure during the relevant etiologic period than fields many years and sometimes decades after diagnosis. There are apparent discrepancies in the reported participation rates of all eligible subjects: we calculate participation rates of 68% for cases and 48% for controls. Another concern is that wire codes were assessed for only 43% of the eligible cases and 32% of the eligible controls. Limitations associated with the use of random-digit dialing to select controls apply to this study.]
For TWA exposure to >0.2 µT, the matched and unmatched analyses give relative risks of 1.5 (95% CI, 0.91-2.6) and 1.2 (0.86-1.8), respectively. For exposure to >0.3 µT, matched analyses were not reported, but the unmatched analyses gave a relative risk of 1.7 (1.0-2.9). For the a priori measurement categories, when exposure was evaluated as a continuous variable, the p value for trend was 0.09 for the matched analysis and 0.15 for the unmatched analysis. When exposure was evaluated as a categorical variable, the p value for trend was 0.12 for the matched analysis and 0.22 for the unmatched analysis. In the matched analyses based on wire code configurations, the risk for childhood ALL was not associated with very high wire codes in the subject's main residence: relative risk = 0.88 (95% CI, 0.48-1.6) for the Wertheimer-Leeper wire code and 1.0 (0.65-1.7) for the Kaune-Savitz wire code. The results reported for the 225 matched case-control pairs for which wire codes were assessed for the mothers' residences during pregnancy showed a trend (p = 0.07) for the Wertheimer-Leeper wire code configuration.
All of the unmatched risk estimates reported by Linet et al. (Linet et al., 1997) were adjusted for the age of the subject at the reference date, the subject's gender, the mother's educational level, and family income. These adjustments for potentially confounding variables had little effect on the risk estimates. The authors note that a limitation of this study is the use of random-digit dialing, which was believed to have resulted in higher family incomes of controls as compared with cases. The authors interpret their study as providing little evidence for an association between exposure to magnetic fields and childhood leukemia. [It is unclear why the criteria for matching controls to cases did not include gender, since males are at greater risk for leukemia than females.] (Robison et al., 1995) [The heterogeneity of wire codes as estimates of exposure to magnetic fields across geographic areas (nine states) may also be of concern. The use of TWA measured fields in several different homes may have diluted the effect (e.g. high exposures for short times averaged with low exposures for longer times). The results based on 24-h measured fields strengthen the evidence for an association, especially in view of the pattern of trend in the estimates; however, the results for wire codes detract from this evidence.]
Tynes and Haldorsen (Tynes & Haldorsen, 1997) conducted a nested case-control study of Norwegian children < 15 years of age who had lived in a census ward crossed by power lines (voltages of 45 kV or more in urban areas and more than 100 kV in rural areas) during at least one of the years 1960, 1970, 1980, 1985, 1987, or 1989. Cases were identified from the Cancer Registry of Norway and consisted of children in whom cancer had been diagnosed during the years 1965-89. For each case, five controls were selected from the cohort who had been alive at the time of diagnosis of the case and were matched by gender, year of birth, and municipality. Calculated historical magnetic fields were the primary basis for classifying study subjects into different categories of exposure. All power lines 11 kV or greater were considered in the calculations of exposure. Underground cables were not taken into account because it was assumed that they were not a significant source of magnetic fields. Categories for the analyses of exposure were obtained for a three-level ordinal scale from average background levels in a typical Norwegian residence (< 0.05 µT) and the median TWA exposure of the controls (0.14 µT). [The calculated fields were not validated by actual measurements.]
The risk for cancers at all sites combined in relation to calculated TWA exposure to magnetic fields from birth to diagnosis was estimated to be 1.9 (1.2-3.3) for the category 0.5 < 0.14 µT and 0.9 (0.5-1.8) for exposure to >0.14 T; the corresponding risks for leukemia were 1.8 (0.7-4.2) and 0.3 (0.0-2.1), respectively. Adjustment for socioeconomic status and number of residences did not affect these results. [The results are based on a low exposure category (0.14 µT, and few subjects were exposed to magnetic field strengths of >0.14 µT: approximately 2% (TWA exposure) and 4% (calculated exposure closest to time of diagnosis) of the study population.]
Michaelis et al. (Michaelis et al., 1998; Michaelis et al., 1997) performed a population-based case-control study to explore the association between childhood leukemia and exposure to EMF in Lower Saxony, Germany, and later used the same study design in Berlin; they then pooled the two sets of results. Cases were recruited from the German Childhood Cancer Registry. The eligibility criteria for the study in Lower Saxony were newly diagnosed leukemia during the years 1988-93, date of birth after 1 July 1975, < 15 years of age at diagnosis, and a resident of Lower Saxony at the date of diagnosis. For the cases in Berlin, the diagnosis had to have been made between January 1991 and September 1994, and the children had to be residents of Berlin at the time of diagnosis. A total of 283 cases were eligible. Controls (919) were selected from the respective local government offices for registration of residents and were matched to the cases by gender, date of birth, and district within the city. Questionnaires were distributed to patients and controls to ascertain their residential history and potential confounding factors, i.e. socioeconomic status and degree of urbanization. The participation rates were 62% for cases and 45% for controls. [The age ranges and observation periods differed for the two cohorts, but these were controlled for in the analysis.]
Exposure was assessed by two methods: measurements of the magnetic field over a 24-h period and indoor spot measurements with an EMDEX II meter. The 24-h measurements were collected in the subject's bedroom and in the living room of the residence where the child had lived longest before the date of diagnosis. Measurements were made 1-7 years after diagnosis. The median magnetic field strength during the 24-h period in the child's bedroom was the primary measure used in the analyses. Median values were preferred to mean values since they are less likely to be influenced by outlying values. The relative risk estimates were adjusted for gender, age, socioeconomic status, and degree of urbanization.
The exposure was dichotomized at 0.2 µT, and the risk estimate was 2.3 (95% CI, 1.2-12). The risk was higher in Lower Saxony, 3.2 (0.7-15) than in Berlin, 1.3 (0.1-12). When the analysis was restricted to the median magnetic field measured during the night, the risk estimate for the 0.2 µT cut-point was 3.8 (1.2-12). No associations were reported between the incidence of childhood leukemia and indoor spot measurements. [The main limitations of this study are the low percentage of subjects exposed to magnetic fields at strengths > 0.2 µT, the reduction in the study size due to inability to obtain permission to measure residential magnetic fields, and the lower participation rate among controls than among cases.]
Savitz et al. (Savitz et al., 1990) examined the association between the incidence of childhood cancer and prolonged exposure to electrical appliances on the basis of information from interviews conducted during their previous case-control study (Savitz et al., 1988), described in the previous section. The appliances of primary interest in this study were electric blankets, heated water-beds, electric heating pads, and bedside electric clocks, and they studied both the mother's and the child's use, i.e. prenatal and postnatal exposure, respectively. Exposure was first examined as 'ever' or 'never' use, separately for prenatal and postnatal exposures; more detailed information obtained included electric blanket and heated water-bed settings, duration of use, and timing of use. Complete information on use of appliances was obtained for 233 mothers of patients (65% of those eligible) and 206 mothers of controls (74% of those eligible) and for 244 patients (69% of those eligible) and 216 controls (78% of those eligible).
Use of none of the appliances considered in the study was associated with a notably elevated relative risk (ever vs. never use) for cancers at all sites. The estimated, unadjusted risks with prenatal exposure to electric blankets were 1.8 (95% CI, 0.9-4.0) for childhood brain cancer, 1.3 (0.7-2.6) for leukemia, and 1.1 (0.4-3.6) for lymphoma. In analyses of prenatal exposure stratified by electric blanket use during the first trimester, the estimated risks were 1.6 (0.8-3.2) for cancers at all sites, 4.0 (1.6-9.9) for brain cancer, and 2.3 (1.0-5.8) for leukemia. For postnatal use of electric blankets, the crude risk estimates were 1.5 (0.6-3.4) for cancers at all sites, 1.5 (0.5-5.1) for leukemias, 1.2 (0.3-5.7) for brain cancer, and 1.0 (0.2-8.6) for lymphoma.
[The methodological limitations of this study include those with the control selection process, described by Savitz et al. (Savitz, 1988) and the fact that the study size was severely limited for assessing the exposures of interest, especially in the postnatal analyses, as very few children were exposed to electric blankets or electric water-bed heaters. Parents' recollection of appliance use was the only method used to assess exposure, and their recall may be incomplete; furthermore, the parents of patients might report differently from those of controls, resulting in recall bias. The questionnaire was not validated. Most of the statistical analyses in this study were based on 'ever/never' use of appliances. An underlying assumption is that the pattern of appliance use among users was similar for cases and controls; if the patterns of use were dissimilar, the sensitivity to detect an association would be diminished.]
London et al. (London et al., 1991) examined the association between the incidence of childhood cancer and prolonged exposure to electrical appliances on the basis of information collected at interviews during their study of residential exposure. The appliances of primary interest were those that could produce considerable magnetic fields. Children's exposure was estimated on the basis of regular use (at least once a week) of the appliance. The statistical analyses were based on regular vs. infrequent use (less than once per week). Mothers' exposure was based on use of the appliance at any time during pregnancy. Complete information on use of 15 appliances by mothers and children was obtained for 232 cases and 232 controls; the participation rates were 70% for cases and 90% for controls.
The relative risks associated with use of 11 of the appliances were greater than the null. The risks for leukemia among children who used black-and-white televisions and electric hair-dryers at least once a week were 1.5 (95% CI, 1.0-2.2) and 2.8 (1.4-6.3), respectively, as compared with nonusers and children who used these appliances less than once a week. The largest relative risks were associated with use of electric blankets and curling irons, but these estimates were unstable because of the small number of subjects. The risks associated with use by the mothers of three of the five appliances that were evaluated were greater than the null. [The limitations of this study are similar to those described for the study of Savitz et al.] (Savitz et al., 1990).
As part of the study of Preston-Martin et al. (Preston-Martin et al., 1996b), a questionnaire was administered to mothers and fathers about exposures and conditions thought likely to be related to the risk for pediatric brain tumors. Mothers were asked about exposure to appliances during their pregnancy and about their child's daily exposure to specific appliance-related sources of magnetic fields. The exposures were examined as 'ever' or 'never' used, separately for the mother's exposure during pregnancy and the child's exposure. Complete information on appliance use by mothers and children was obtained for 304 cases and 304 controls.
The risks associated with use by the mothers of six of the seven appliances that were evaluated were greater than the null. An approximately two-fold increase in risk for brain tumors was reported among children whose mothers had slept in electrically heated water-beds during pregnancy (2.1; 1.0-4.2). For the 12 appliances that were evaluated for children, the relative risks were greater than the null for six. As for the mothers, there was a two-fold increase in risk among children who slept in electrically heated water-beds (2.0; 0.6-6.8). [With such low percent use of these appliances, it is difficult to assess their impact with respect to cancer risk.]
Preston-Martin et al. (Preston-Martin et al., 1996a) evaluated the effects of use of electric blankets and water-bed heaters on the risk for pediatric brain tumors. Since there were not enough cases in Los Angeles to adequately address this hypothesis, cases were included from two other regions that participated in the multicenter US West Coast Childhood Brain Tumor Study: Seattle, Washington and San Francisco, California. The cancer registry at each location was used to identify 813 cases, of which 540 children were considered to be eligible on the basis of information in the registry and the study criteria and were enlisted in the study as cases. Controls were identified and recruited from the same geographic regions in which the cases arose by a two-step random-digit telephone dialing procedure. This resulted in a comparison group that was similar to the cases with regard to gender and age and at ratios of approximately two controls per case in Seattle and San Francisco and one control per case in Los Angeles, for a total of 801 eligible controls. Of those eligible to participate, 73% of cases and 74% of controls were interviewed for the study.
Maternal use of electric blankets during pregnancy with the patient, regardless of trimester or use, was not associated with subsequent risk for brain tumor; thus, the relative risk as compared with nonusers of these appliances was 0.9 (95% CI, 0.6-1.2). Similar results were reported for prenatal use of water-beds with electric heaters, but the risk varied substantially with geographic area. The relative risk for pediatric brain tumor associated with maternal use of heated water-beds during pregnancy was 2.1 (1.0-4.4) for Los Angeles participants and 0.7 (0.4-1.0) for participants from San Francisco and Seattle. [This study includes information about use of electric blankets and heated water-beds that was reported by Preston-Martin et al. and Gurney et al.] (Preston-Martin et al., 1996b) and (Gurney et al., 1996). [Although data from three study sites was pooled, exposure to electric blankets and heated water-beds was relatively uncommon, especially among children. Furthermore, reported use of these appliances from the questionnaires may not have been a sensitive enough measure to detect differences in brain cancer risk due to prolonged exposure to magnetic fields.]
Gurney et al. (Gurney et al., 1996) assessed exposure to magnetic fields from electric blankets and heated water-beds on the basis of responses to a questionnaire administered in person to 133 patients and 270 controls, with participation rates of 74% among cases and 79% among controls. A mailed questionnaire was used subsequently to collect information on use of electric heating and electric appliances other than electric blankets and water-beds. Questionnaires were returned for 98 cases and 208 controls. The analyses of use of electric heaters (within three years before diagnosis of the cancer) and appliances were based on whether the subject had ever used the appliance in question or never used it.
Mothers' use of seven appliances and heat sources was evaluated. The only increase in relative risk was that associated with prenatal use of an electric water-bed. Of the relative risks associated with use of 19 appliances and heat sources that were evaluated for children, eight were greater than the null. [The study size was severely limited for assessing the exposures of interest. Recall may have been incomplete, and it is unclear whether the parents of the patients report differently from those of controls, resulting in recall bias. Most of the statistical analyses in this study were based on 'ever/never' use of appliances. An underlying assumption is that the pattern of appliance use among users was similar for cases and controls; if the patterns of use were dissimilar, the sensitivity to detect an association would be diminished.]
As part of a study of residential exposure to magnetic fields, Linet, et al. (Linet et al., 1997), Hatch, et al. (Hatch et al., 1998), evaluated the association between childhood ALL and use of electrical appliances during pregnancy and childhood. The study population is described in section 4.2.2. Personal interviews were conducted with 788 patients and 699 controls, corresponding to participation rates of 84% among cases and 54% among controls. [The authors report participation rates of 88% and 64%, but the number of eligible cases was 942 and that of eligible controls, 1232.] Of these participants, 640 matched pairs were used in the analysis. Questions were asked about the mother's use during pregnancy and the child's use of electric blankets, mattress pads, heating pads, water-beds, stereos or other sound systems, television, video machines in arcades, computers, microwave ovens, sewing machines, hair-dryers, curling irons, humidifiers, and electric clocks. Use of stereo systems without headsets, night lights, and ceiling fans was considered unlikely to have resulted in substantial exposure to magnetic fields, but they were included as 'red herring' variables to evaluate the potential for recall bias. The interviewers asked for the child's age when starting and stopping use of a specific appliance and the frequency of use during the year before diagnosis. For three appliances associated with potentially high exposure to magnetic fields (electric blankets, water-beds, and hair-dryers), questions were asked about the frequency of use during the last year of use if the child had stopped using the appliance. The authors analyzed the use of each of the appliances separately and did not create an overall estimate of exposure to magnetic fields from appliances. Mattress pads and electric blankets were analyzed together. Matched analyses were conducted with adjustment for income and maternal education. Potential confounding from parental age, occupation, smoking, type of dwelling, urbanization, number of siblings, and breast feeding were evaluated. Age- and gender-specific analyses were done, and the mother's use was analyzed according to trimester.
The estimated risks were 1.6 (95% CI, 1.1-2.3) for mother's use of electric blankets or mattress pads and 1.5 (1.0-2.1) for their use of heating pads and humidifiers, but with no consistent dose-response pattern. When frequency of use was evaluated, a reduced risk was found for use of sewing machines (0.76; 0.59-0.98). The risk estimates for use of other appliances were close to unity.
Elevated risks were found for children's use of electric blankets (2.8; 1.5-5.0), hair-dryers (1.6; 1.2-2.1), curling irons (1.7; 0.91-3.3), video arcade machines (1.7; 1.2-2.3), sound systems with headset (1.3; 0.97-1.8), and video games connected to a television (1.9; 1.4-2.7). Consistent dose-response patterns were found for use of sound systems with a headset, video arcade machines, and video games connected to a television, but not for use of electric blankets, hair-dryers, or curling irons. The risk estimates for other appliances were close to unity, except for night-lights (0.81; 0.63-1.0). None of the 'red herring' variables was associated with an increased risk. Time spent watching television was associated with disease, and the risk increased with increasing time. Watching television at closer than 6 feet (2 m) was also associated with disease, but the highest risk was found for distances 4 to 6 feet (>1.3 and < 6 m; relative risk, 1.7; 1.2-2.4). For distances < 4 feet (< 1.3 m), the relative risk was 1.6 (1.1-2.4).
When the time spent watching television and the distance from the television were combined, inconsistent dose-response patterns were seen. The authors emphasized the potential for differential recall bias, especially when patterns of use changed after diagnosis, as may be the case for television viewing patterns. They also mention the possibility that mothers of children with ALL may be more prone to remember use of appliances discussed in the media as potential risk factors for leukemia, such as electric blankets. Another source of error noted by the authors is potential selection bias due to the use of random-digit dialing for control selection and the lower participation rate among controls, both being associated with socioeconomic status. The authors also mention the possibility of confounding from some factor related to the type of life-style associated with, for example, watching television for many hours per day. They found it unlikely that the associations found reflect a causal association between exposure to magnetic fields and childhood ALL. [The inconsistent dose-response patterns may also be the result of nondifferential exposure misclassification.] Selection bias is unlikely to explain the findings for use of electric blankets. Differential recall bias may have affected the results, but a comparison between answers from an earlier telephone interview and those obtained at the later personal interview about electric blanket use did not indicate differences in the recall between cases and controls. [Questions on other appliances were not validated. Another limitation is the considerable lower participation rate among controls as compared to cases. Finally, the magnetic fields of the appliances were not measured.]
In reviewing the individual epidemiological studies, it is difficult to summarize and draw overall conclusions about a possible association between exposure to ELF EMF and the incidence of childhood cancer. These difficulties arise due to differences in study design, case selection, identification of controls, exposure assessment methods, and accounting for such factors as confounders and effect modifiers. Furthermore, a common feature of most of the studies was the lack of a large population of highly exposed children.
Meta-analysis provides a means for summarizing the results of individual studies into a single measure of effect (Fleiss, 1993). The objectives of a meta-analyses often include identification and review of all studies conducted on a specific topic; assessment of the consistency and comparability of the results of each identified study; estimation of average measures of effect, if studies are sufficiently similar; and assessment of reasons for heterogeneity (Blair et al., 1995).
Four meta-analyses have been conducted of studies of the possible association between exposure to magnetic fields and the incidence of childhood cancers, in which summary effect measures were stratified by exposure metric (wire codes, distance from electrical facility, calculated magnetic fields, and measured magnetic fields) (NRPB, 1992); (Ahlbom et al., 1993); (Miller et al., 1995); (Meinert & Michaelis, 1996). These are summarized in Tables 4.25 and 4.26. In general, the summary risks for each of the metrics considered were elevated but less than 2.0 at all cut-points. [In general, as the number of studies included in the meta-analysis increased, the confidence intervals narrowed, indicating increasing statistical power; however, the effect estimates did not change substantially. Use of different exposure metrics, such as 24-h measures vs. spot measures, or cut-points, such as 0.2 µT vs. 0.3 µT, can result in quantitative but not qualitative differences.]
Wartenberg and colleagues (Wartenberg et al., 1998) and National Academy of Sciences (NRC et al., 1997) conducted a more extensive meta-analysis of the studies of the possible association between exposure to magnetic fields and the incidence of childhood leukemia and brain tumors. In addition to calculating a summary effect measure, they conducted a limited sensitivity analysis, assessed the heterogeneity among the studies, evaluated possible dose-response relationships, evaluated the likelihood of publication bias, and assessed the robustness of results to inclusion of additional studies. They also provided a more carefully considered rationale for grouping studies.
The reported summary effect measures are similar to those reported in previous meta-analyses: generally > 1.0 but < 2.0. The results were moderately robust to exclusion of individual studies. Substantial heterogeneity was reported for wire codes and proximity to electrical facilities for both leukemia and brain tumors. Analyses to characterize the source of the heterogeneity gave equivocal results. The relative risk estimates for dose-response relationships for spot measures were weakly elevated, those for calculated fields were slightly higher, and those for wires codes (converted to magnetic fields by using the mid-range of the wire code category) were still slightly higher. There is little evidence for publication bias. Calculations showed that inclusion of an extremely large study would be required to substantially change the reported summary estimates of effect.
[This meta-analysis included several studies that this Working Group excluded from consideration in their deliberations.]
Selection bias introduced by the control selection technique is unlikely in the Wertheimer and Leeper study, because a birth registry was used as the population source. In the other three studies, random-digit dialing was used to select controls, which may have introduced some bias toward higher socioeconomic status among the controls. If higher socioeconomic status is related to lower wire codes, use of random-digit dialing may have led to an overestimation of the risk. As Savitz et al., London et al., and Linet et al. all used random-digit dialing to select controls, it is unlikely that the associated bias would have affected only the risks found by Savitz et al. and London et al. The study of Savitz et al. has a further limitation in the way in which controls were selected, requiring them to be residentially stable; this may have introduced bias leading to an overestimation of the risk, but it is unlikely that this could entirely explain the larger risk estimate. The observed elevated risks and dose-response patterns cannot be explained by selection bias.
Confounding from other risk factors for childhood leukemia must also be considered. The etiology of this disease is largely unknown. The most frequently discussed factors that may be related to both wire codes and childhood leukemia are traffic density and socioeconomic status. Confounding from traffic density was evaluated in all three studies that showed increased relative risks and was found not to explain the observed association. It was not evaluated in the study of Linet et al. The potential impact of socioeconomic status was evaluated in all four studies: it did not explain the observed results. Furthermore, Savitz et al. evaluated a substantial number of additional potential risk factors, which were also shown to have little effect on the risk estimates.
The lack of an association with wire codes in the study by Linet et al. is difficult to explain, given the associations observed in the other studies. The validity of using wire codes in regions other than in Denver is not clear, but the potential shortcomings should have applied to the studies of both Linet et al. and London et al. Linet et al. included several regions in their study, however, and the validity of using the wire codes as estimates of exposure to magnetic fields may vary among the regions. The fact that the proportion of subjects exposed to magnetic fields > 0.2 µT was considerably higher in the study of Savitz et al. than in those of Linet et al. or London et al. may explain some of the disparity in the results. Linet et al. included only cases of acute lymphoblastic leukemia, while the other studies included all types of leukemia. The results of the studies, when taken together, support the association between classification of exposure from wire codes and the incidence of childhood leukemia. This is further supported by the results of the formal meta-analysis, which found, on average, a 40% excess risk for leukemia (OR = 1.5; 95% CI, 1.0-2.2) (Wartenberg et al., 1998).
Among the studies in which calculated fields were used to assess exposure to magnetic fields, four Nordic studies were considered to be of sufficient quality to be used in the evaluation. Three of the studies found increasing leukemia risk with increasing calculated fields (Feychting & Ahlbom, 1993; Olsen et al., 1993; Verkasalo et al., 1993), and a smaller study found no effect (Tynes et al., 1992). All four studies were population-based, with minimal potential for selection bias both in terms of control selection and participation rates. The exposure assessment method used in these studies is based on the laws of physics and engineering design and provides estimates of the exposure for a relevant etiologic period from historical information about line loads and configurations. Thus, the exposure estimates may be less subject to misclassification than those in the studies based on wire codes. The main limitations of all four of the studies are the small number of cases and the low prevalence of exposure. Potential confounding from traffic exhaust was controlled in the Swedish study and did not change the effect estimate. Adjustment for socioeconomic status was made in all of the studies except that in Finland, again with no effect on the observed risk estimates. The results of these studies, when taken together, support an association between exposure to calculated magnetic fields and the incidence of childhood leukemia. This conclusion is further supported by the results of the formal meta-analysis which found, on average, a 63% excess risk for leukemia (OR = 1.6; 95% CI, 0.8-3.0) (Wartenberg et al., 1998).
Of the studies in which spot measurements were used to assess exposure to magnetic fields, three were considered to be of sufficient quality to be used in the evaluation (London et al., 1991; Michaelis et al., 1998; Savitz et al., 1988). The results of these three studies are inconsistent, two being close to unity and the third (Savitz et al.) showing increased risks. The very low participation rate among cases in the study of Savitz et al. limits its validity. The study of Feychting and Ahlbom (Feychting & Ahlbom, 1993) is not included in this assessment because the spot measurements were made too long after the relevant etiologic period.
Neither selection bias nor confounding had a major impact on the reported results. The study of Michaelis et al. is limited by the small number of exposed subjects. Exposure misclassification is a potential limitation in all three studies. The usefulness of spot measurements for retrospective assessment of exposure to magnetic fields during the etiologically relevant period has been questioned. These studies do not provide sufficient information to evaluate the association between exposure to magnetic fields evaluated by spot measurements and the incidence of childhood leukemia. Furthermore, the formal meta-analysis did not find an appreciable excess risk for leukemia (OR = 1.2; 95% CI, 0.7-2.1) (Wartenberg et al., 1998).
Three studies in which 24-h measured magnetic fields were used to assess exposure to magnetic fields were considered to be of sufficient quality to be used in an evaluation of the association between the incidence of childhood leukemia and exposure to magnetic fields (Linet et al., 1997; London et al., 1991; Michaelis et al., 1998). The results of all three studies showed an increased risk for children in higher exposure classes. The data reported by Linet et al. showed an exposure-response relationship, which was not found by London et al. and was not assessed by Michaelis et al.
The studies of both Linet et al. and London et al. had potential limitations due to use of random-digit dialing to select controls. Control selection in the study of Michaelis et al. is unlikely to be subject to selection bias. The low participation rates in all three studies might have resulted in selection bias. Confounding due to socioeconomic status was controlled in all three studies. Confounding by traffic exhaust was addressed in previous studies in which other exposure assessment methods were used, and was found unlikely to affect the results. The study of Michaelis et al. is limited by the small number of exposed subjects, and there were few highly exposed subjects in all three studies. It is not clear how well a 24-h magnetic field measurement reflects exposure during the relevant etiologic period, nor how representative it is of long-term exposure, given the weekly, seasonal, and secular patterns. This method is, however, an improvement over spot measurements of magnetic fields. The study of Linet et al. is an improvement over the previous studies because of the markedly shorter time between diagnosis and exposure assessment.
The results of these studies provide some support for a possible association between exposure based on 24-h measured magnetic fields and the incidence of childhood leukemia. This conclusion is further supported by the results of the formal meta-analysis, which showed, on average, a 50% excess risk (OR = 1.5; 95% CI, 1.0-2.3).
Three studies of appliance use were considered to be of sufficient quality to be used in an evaluation of the association between the incidence of childhood leukemia and exposure to magnetic fields (Hatch et al., 1998; London et al., 1991; Savitz et al., 1990). The results do not fit a coherent pattern, but elevated risks were reported for a variety of appliances in different studies. Many increased risk estimates were found by Hatch et al.; however, it is interesting to note that no associations were found for the three appliances that do not to involve significant exposure to magnetic fields (i.e. those that were included in the study to assess the possible role of recall bias). The possibility of recall and reporting bias in the study of Hatch et al. and in any of the other studies cannot, however, be ruled out. Furthermore, chance cannot be ruled out as an explanation for the observations. In addition, low participation rates and use of random-digit dialing in all three of the studies may have influenced the results in any direction. These studies provide inadequate evidence to assess an association of use of appliances and the incidence of childhood leukemia.
Equally inconclusive results were observed in studies of the possible association between childhood brain tumors and spot-measured magnetic fields, 24-h measured magnetic fields, and use of appliances. The data from studies on childhood brain tumors and calculated magnetic fields provide some evidence that there is no association; however, these results are based on a very small number of cases, and chance cannot be ruled out as an explanation for the lack of association. The results of the meta-analysis results are inconclusive with regard to all exposure metrics.
Confounding seems to be an unlikely explanation for the observed results. A confounder must be associated with a sufficiently large relative risk to overcome that associated with magnetic fields. Given the extensive search for possible confounders and the fact that no strong candidates have been identified, the impact of confounding appears to be minimal.
Selection bias cannot be ruled out in several of the studies; however, increased risk estimates and consistent dose-response patterns were found also in the Nordic studies, in which selection bias is unlikely.
As research on EMF evolved, both exposure assessment and study designs have improved. The results of studies would thus have been expected to become more consistent. In fact, this has not occurred, which raises questions about whether the 'improvements' in exposure assessment have more accurately captured the relevant EMF exposure.
In sum, although the exposure metrics used as surrogates for exposure to magnetic fields are of varying precision, it is difficult to find an explanation other than exposure to magnetic fields for the consistency of the reported excess risks for childhood leukemia in studies conducted in different countries under different conditions, with different study designs. Overall, the Working Group gave preference to the Nordic studies in which bias in the selection of study subjects can be ruled out and in which the most sophisticated exposure assessment methods were used. The Working Group considers, with some minor reservations, that the strengths and consistency of these study results are suggestive in spite of their limitations.
[This conclusion was supported by 20 Working Group members; there were 6 votes for 'inadequate' evidence, 2 abstentions, and 1 absent.]
There is inadequate evidence with respect to childhood nervous system tumors.
[This conclusion was supported by 25 Working Group members; there were 2 abstentions and 2 absent.]
There is inadequate evidence with respect to childhood lymphoma.
[This conclusion was supported by 25 Working Group members; there were 2 abstentions and 2 absent.]
| Wertheimer and Leeper (1979) | Cancer mortality records (1950-1973) of persons less than 19 years of age, that were born in Colorado and
resided in Denver Largest number of cases used in the analysis: 328 | Denver-area birth certificates Total number of controls enrolled: 344 | 2-level wire code (HCC vs. LCC) Non-blinded wire code assessment | Cancer onset age Urban/suburban Socioeconomic status Birth order Maternal age Traffic congestion Gender | Crude relative risk estimates cited Potential confounders were individually analyzed |
| Savitz et al. (1988) | All cancer incidence cases reported in Denver, Colorado during the years 1976-1983 of persons less
than 15 years of age 356 cases identified 320 cases had 5-level wire codes assessed 252 cases were interviewed 128 cases had magnetic field measurement data | Controls selected via random digit telephone dialing methods Matched to cases by age, gender, and telephone exchange area 278 controls identified 259 cases had 5-level wire codes assessed 222 cases were interviewed 207 cases had magnetic field measurement data | 5-level wire codeIn-home electric and magnetic field spot measurements under low and high power use conditions | Gender, age Type of housing Socioeconomic status Smoking during pregnancy Traffic density Parental age Race and education Income Family cancer history, In utero exposure to Alcohol use X-rays, influenza, and medications Birth defects Birth order Birth weight Illness Residential stability Medication x-rays | Control selection procedures resulted in the controls being more residentially stable than the cases Matched analysis not performed Adjusted relative risk estimates were described but not presented in a table and did not change the results. |
| London et al. (1991) | All leukemia incidence cases reported to the Los Angeles County Cancer Surveillance Program (1980-1987) of persons
less than 10 years of age 331 cases identified and 232 interviewed 169 cases had 24-hour magnetic field measurements recorded 140 cases had spot measurements recorded 219 cases had 5-level wire codes assessed | Controls were obtained through friends of cases (1980-1984) and via random digit telephone dialing methods
(1980-1987) Matched to cases by age, gender, and ethnicity 257 controls identified and 232 interviewed 149 controls had 24-hour magnetic field measurements recorded 109 controls had spot measurements recorded 207 controls had 5-level wire codes assessed | 5-level wire code Outdoor and in-home electric and magnetic field spot measurements under low and normal power use conditions 24-hour magnetic field measurements underneath the bed in the child's bedroom Self-report of appliance use | Various factors associated with cancer that were reported in previous studies Appliance use Socioeconomic status | Exposures assessed in homes during an "etiologic period" defined as the period beginning at the estimated time of
conception and ending on the date of diagnosis inset "for children aged 1 year or less at time of diagnosis date minus
6 months" for children aged 1-2 years at diagnosis; and one year prior to diagnosis for children diagnosed at ages
greater than 2 years Matched and unmatched analyses were carried out since the mean 24-hour magnetic field exposure was lower among matched controls as compared to unmatched |
| Olsen et al. (1993) | All leukemia, tumor of the central nervous system, or malignant lymphoma incidence cases reported to the Danish Cancer Registry (during the period from April 1, 1968 to December 31, 1986) of persons less than 15 years of age 1707 cases identified | Two to five controls were selected at random from among people who had survived without cancer until the date of
diagnosis of the case Controls matched to the case by gender and date of birth 4788 eligible controls identified | Distance from transformer substations, overhead lines, and underground cablesAverage calculated magnetic field
exposureaveraged over residence period Cumulative calculated magnetic field exposure = (number of months exposed) multiplied by (average calculated level of magnetic field at the residence) | Gender Age at diagnosis Socioeconomic status Population density at place of residence Number of changes of address | Analyses controlled for gender and age at diagnosis for each cancer grouping At cut-point 0.4 µT and greater the numbers of exposed cases and controls is too small to indicate significant increases in cancer incidence for individual cancer types. |
| Feychting and Ahlbom (1993) | All cancer incidence cases reported to the Swedish Cancer Registry of persons under 16 years of age living on a
property located within 300 meters of any 220 or 400 kV power lines in Sweden (1960-1985) 142 cases identified 141 cases had calculated fields assessed 89 cases had spot measurements recorded | Controls per case were randomly selected from all persons under 16 years of age living on a property located within
300 meters of any 220 or 400 kV power lines in Sweden (1960-1985) Controls matched to cases by birth year, gender, residence in the same parish during the year of diagnosis or the last year before the case moved, lived near the same power line as the case 558 controls identified 554 controls had calculated fields assessed 344 controls had spot measurements recorded | Distance to power lines from residence In-home magnetic field spot measurements under low and high power use conditions Calculations of the magnetic fields generated by the power lines at the time spot measurements were assessed (calculated contemporary fields) and for the year closest in time to diagnosis (historical calculated fields) | Age Gender Year of diagnosis Whether or not the subject lived in the county of Stockholm Type of residence (single- family home or apartment) Nitrogen dioxide content as an index of air pollution from road traffic Socio-economic status | Matched analysis was conducted Magnetic field spot measurements taken 5-31 years after diagnosis. Median 16 years |
| Verkasalo et al. (1993) | All primary cancer cases reported to the Finnish Cancer Registry ( 1970-1989) of persons less than 20 years of age living within 500 m of overhead power lines of 110-400 kV in magnetic fields calculated to be 0.01 µT and greater140 cases identified | Entire cohort consisted of 68,300 boys and 66,500 girls less than 20 years of age living during 1970-1989 within 500 m of overhead power lines of 110-400 kV in magnetic fields calculated to be 0.01 µT and greater | Calculated magnetic field exposure average exposure cumulative exposure | Gender Age | Analysis based upon cohort approach (standardized incidence ratios) with person years at risk stratified by gender,
age (grouped in 5-year age categories), and exposure category Analysis included all primary cancers rather than first primary cancers, which resulted in multiple cancers per person being counted |
| Preston-Martin et al. (1996a) | Patients under 20 years of age who were residents of Los Angeles County and had a benign or malignant primary
tumor of the brain, cranial nerves, or cranial meninges of any histologic type diagnosed during the period
January 1,1984-June 30, 1991 437 eligible cases were obtained 304 cases had complete interview data 292 cases had wiring maps assessed for at least one residence 255 cases had spot measurement data 236 cases had at least one STAR magnetic field profile 110 cases had 24-hour measurements assessed in at least one room | Controls were obtained via random digit telephone dialing methods Matched to cases by birth date, age at time of diagnosis 433 eligible controls were obtained 304 controls had complete interview data 269 controls had wiring maps assessed for at least one residence 206 controls had spot measurement data 181 controls had at least one STAR magnetic field profile 101 controls had 24-hour measurements assessed in at least one room | Wire coding according to the 5-level Wertheimer-Leeper classification Spot measurements inside and outside the residence STAR magnetic field profiles 24-hour magnetic field measurements taken in the child's bedroom and other rooms Self-report of appliance use | Demographic variables Parental occupation Exposures during gestation Building type Appliance use Maternal occupational exposure to high magnetic fields during pregnancy | Matched and unmatched analyses performed Compared to parents of cases, a higher proportion of control mothers and fathers were Latino Compared to cases, a higher proportion of controls were in the highest social class |
| Gurney et al. (1996) | Patients under 20 years of age who were diagnosed with a benign or malignant primary tumor of the brain, cranial
nerves, or cranial meninges of any histologic type diagnosed during the period 1984-1990 195 cases were eligible Largest number of cases used in the analysis: 120 | Controls were obtained via random digit telephone dialing methods Approximately two cases per control were stratified by age, gender, and area of residence 270 eligible controls were identified Largest number of controls used in the analysis: 240 | Wire coding according to the 5-level and 2-level Wertheimer-Leeper classifications Self-report of heating sources and appliance use | Age, gender, race County at reference date Reference year Mother's education Family history of brain tumors Passive tobacco smoke exposure in the home Whether the child lived on a farm Whether or not the child had a history of head injury x-ray to the head or neck, Epilepsy, or Fits from severe fever | A small percentage of cases were classified in the very high current classification (3%) |
| Linet et al. (1997) | All acute lymphoblastic leukemia incidence cases registered with the Children's Cancer Group (1989-1994) of persons
less than 15 years of age 942 cases were eligible for the study 629 unmatched cases had magnetic field measurements recorded (463 case-control pairs included in the analysis) 408 matched cases had wire codes assessed in their main residence 225 matched cases had wire codes assessed in the residence of pregnancy | Controls were obtained via random digit telephone dialing methods Matched to cases by the first eight digits of
the telephone number, age, and race 1292 controls were eligible for the study 619 unmatched controls had magnetic field measurements recorded (463 case-control pairs included in the analysis) 408 matched controls had wire codes assessed in their main residence 225 matched controls had wire codes assessed in the residence of pregnancy | 24-h magnetic fields measurements in the child's bedroom 30-s measurements in the center of the child's bedroom, family room, kitchen, the room in which the mother slept during the index pregnancy, and near the front door of the residence Wire coding according to the 5-level Wertheimer-Leeper classification and the modified 3-category Kaune-Savitz scheme 24-h measurement 1-bedroom | Age Gender Race Socioeconomic status Temporal factors Urbanization Type of residence Gender Race Mathew's educational level | Magnetic field measurements generally measured within two years of leukemia diagnosis Single summary exposure for magnetic field measurements was calculated based upon a weighted average of the room measurements with weights based upon estimated time spent in each room according to the child's age. Matched and unmatched analyses cited. Measurements summarized as time-weighted-average over five years preceding diagnosis. |
| Tynes and Haldorsen (1997) | All cancer incidence cases reported to the Cancer Registry of Norway (1965-1989) of persons less than 15 years of
age who had lived in a census ward crossed by high-voltage power lines (45 kV or more in urban areas and more than 100 kV
in rural areas) during at least one of the years 1960, 1970, 1980, 1985, 1987, or 1989 532 cases identified Largest number of cases used in the analysis: 500 | 5 controls per case were selected at random from among Norwegian children living in a census ward crossed by
high-voltage power lines (45 kV or more in urban areas and more than 100 kV in rural areas) during at least one of the
years 1960, 1970, 1980, 1985, 1987, or 1989 that were alive at the time of diagnosis of the case Matched to cases by gender, birth year, and municipality 2122 controls identified Largest number of controls used in the analysis: 2004 | Distance to power line from residence Calculated magnetic field exposure estimate Time-weighted average calculated magnetic field exposure Time-weighted average cumulative exposure Average maximal exposure Mother's exposure at the time of conception Child's exposure in the year closest in time to diagnosis Average exposure during the first year of the child's life Average exposure during the first 4 years of the child's life | Socioeconomic status Type of building Number of dwellings | Few subjects were exposed to calculated, magnetic fields greater than 0.14 µT The narrow distribution of exposures resulted in limited discrimination of the effects of different exposure indices |
| Michaelis et al. (1997) | All leukemia incidence cases reported to the German Childhood Cancer Registry (1988-1993) of persons less than
15 years of age and a resident of Lower Saxony at the date of diagnosis 219 cases were identified 1 29 cases had 24-h magnetic field measurements recorded | Controls were obtained from the files of local government offices for registration of residents Two controls per case were selectedone control from the same registration office as the caseone control selected from a randomly chosen registration office in Lower Saxony 328 controls had 24-h magnetic field measurements recorded | 24-h magnetic field measurements in the child's bedroom 24-h magnetic field measurements in the living room Spot measurement at the residence where the child lived the longest Non-blinded magnetic field assessment | Gender Age Age at diagnosis Socioeconomic status Urbanization | Matched analysis conductedA prior cut-point of µT G was chosen Low percentage of subjects exposed to measured magnetic field strengths greater than 2 µT (1.5% of the entire study population) |
classification | lymphoblastic no. of cases | ||||
| Wertheimer & Leeper (1979) | Birth address: LCC HCC Death address: LCC HCC |
84 52 92 63 |
reference 2.28 (1.34-3.91) reference 2.98 (1.78-4.98) | ||
| Savitz et al. (1988) | HCC/LCC VHCC/Buried | 27 / 70 7 / 28 | 1.54 (0.90-2.63) 2.75 (0.94-8.04) |
<19 / 59 6 / 24 | 1.28 (0.70-2.34) 2.75 (0.90-8.44) |
| London et al. (1991) | UG+VL OLCC OHCC VHCC |
31 58 80 42 | references 0.95 (0.53-1.69) 1.44 (0.81-2.56) 2.15 (1.08-4.26) | ||
| Linet et al (1997) | UG+VLCC OLCC OHCC VHCC | 175 116 87 24 |
references 1.07 (0.74-1.54) 0.99 (0.67-1.48) 0.88 (0.48-1.63) | ||
| Calculated fields | |||||
| Feychting & Ahlbom (1993) | Unmatched analyses (µT) <0.1 0.1-0.19 >0.2 >0.3 Matched analyses:(µT) 0.1-0.19 >0.2 | 27 4 7 7 | reference 2.1 (0.6-6.1) 2.7 (1.0-6.3) 3.8 (1.4-9.3) 4.3 (1.0-8.9) 3.5 (0.9-13.6) | ||
| Olsen et al. (1993) | (µT) < 0.1 0.1-0.24 >0.25 >0.40 |
829 1 3 3 |
reference 0.5 (0.1-4.3) 1.5 (0.3-6.7) 6.0 (0.8-44) | ||
| Verkasalo et al. (1993, 1994) | Cumulative exposure (µT-years) 0.01-0.39 >0.40 >1.0 Average exposure (µT) 0.01-0.19 >0.2 | 32 3 3 32 3 |
0.90 (0.62-1.3) 1.2 (0.26-3.6) 3.5 (0.7-10) 0.89 (0.61-1.3) 1.6 (0.32-4.5) | ||
| Tynes & Huldersen (1997) | Average exposure (µT) < 0.05 0.05-0.13 >0.14 Closest to diagnosis (µT) <0.05 0.05-0.13 >0.14 >0.2 | 139 8 1 134 10 4 2 |
reference 1.8 (0.7-4.2) 0.3 (0.0-2.1) reference 1.5 (0.7-3.3) 0.8 (0.3-2.4) 0.5 (0.1-2.2) | ||
| Spot measurements | |||||
| Savitz et al. (1988) | Low Power conditions (µT) < 0.2 >0.2 High power conditions (µT) < 0.2 >0.2 Electric fields (µT) < 12 V/m >12 V/m | 31 5 30 7 31 6 | reference 1.93 (0.67-5.56) reference 1.41 (0.57-3.50) reference 0.75 (0.29-1.91) | 23 3 23 4 23 4 | reference 1.56 (0.42-5.75) reference 1.05 (0.34-3.26) reference 0.67 (0.22-2.04) |
| London et al (1991) | Low power conditions (µT) < 0.032 0.032-0.067 0.068-0.124 >0.125 | 67 34 23 16 | reference 1.01 (0.61-1.69) 1.37 (0.65-2.91) 1.22 (0.52-2.82) | ||
| Michaelis (1997b) |
Short-term measurement (µT) < 0.2 >0.2 | 170 6 | reference 0.7 (0.3-1.8) | ||
| London, et al. (1991) | 24 hour measurements (µT) 0-0.067 0.068-0.118 0.119-0.267 >0.268 | 85 35 24 20 | reference 0.68 (0.39-1.17) 0.89 (0.46-1.71) 1.48 (0.66-3.29) | ||
| Michaelis, et al. (1997a) | Median of measurements (µT) < 0.2 >0.2 Mean of measurements (µT) < 0.2 >0.2 Median during the night (µT) < 0.2 >0.2 | 125 4 125 4 124 5 | reference 3.2 (0.7-14.9) reference 1.5 (0.4-5.5) reference 3.9 (0.9-16.9) | ||
| Michaelis et al. (1997b) | Median of measurements (µT) < 0.2 >0.2 Median during the night (µT) < 0.2 >0.2 | 167 9 167 9 | reference 2.3 (0.8-6.7) reference 3.8 (1.2-11.9) | ||
| Linet et al. (1997) | Unmatch Analysis (µT) < 0.065 0.065-0.099 0.1-0.199 0.2-0.299 0.3-0.399 0.4-0.499 >0.5 >0.2 >0.3 Matched Analysis (µT) <0.065 0.065-0.099 0.1-0.199 0.2-0.299 0.3-0.399 0.4-0.499 >0.5 >0.2 |
267 123 151 38 22 14 9 83 45 206 92 107 29 14 10 5 58 | reference 1.1 (0.81-1.50) 1.1 (0.83-1.48) 0.92 (0.57-1.48) 1.39 (0.72-2.72) 3.28 (1.15-9.39) 1.41 (0.49-4.09) 1.24 (0.86-1.79) 1.7 (1.0-2.9) reference 0.96 (0.65-1.40) 1.15 (0.79-1.65) 1.31 (0.68-2.51) 1.46 (0.61-3.50) 6.41 (1.30-31.73) 1.01 (0.26-3.99) 1.53 (0.91-2.56) | ||
classification |
|||
| Wire codes | |||
| Wertheimer and Leeper (1979) |
Birth address HCC/LCC |
22 / 35 |
2.36 (1.03-5.41) |
| Death address HCC/LCC |
30 / 36 |
2.40 (1.15-5.01) | |
| Savitz et al. (1988) |
HCC/LCC | 20 / 39 | 2.04 (1.11-3.76) |
| VHCC/Buried | 3 / 17 | 1.94 (0.47-7.95) | |
| Preston-Martin et al. (1996a) |
UG VLCC/OLCC OHCC VHCC |
39 114 97 31 |
2.3 (1.2-4.3) reference 0.8 (0.6-1.2) 1.2 (0.6-2.2) |
| Gurney et al. (1996) |
High/low UG |
23 / 97 47 |
0.9 (0.5,1.5) reference |
| Calculated fields | |||
| Feychting & Ahlbom (1993) |
Unmatched analyses (µT) < 0.1 0.1-0.19 >0.2 >0.3 |
29 2 2 2 |
reference 1.0 (0.2-3.8) 0.7 (0.1-2.7) 1.0 (0.2-3.9) |
| Matched analyses 0.1-0.19 >0.2 |
0.8 (0.1-4.9) 0.7 (0.1-3.2) | ||
| Olsen et al. (1993) |
< 0.1 0.1-0.24 >0.25 >0.4 |
621 1 2 2 |
reference 1.0 (0.1-9.6) 1.0 (0.2-5.0) 6.0 (0.7-44) |
| Verkasalo et al. (1993) |
Cumulative exposure (µT-years) 0.01-0.39 >0.40 >1.0 |
32 7 3 |
(SIR) 0.82 (0.56-1.2) 2.3 (0.94,4.8) 2.8 (0.6-8.1) |
| Average exposure (µT) 0.01-0.19 >0.2 |
34 5 |
0.85 (0.59-1.2) 2.3 (0.75-5.4) | |
| Tynes & Haldorsen (1997) |
Average exposure (µT) < 0.05 0.05-0.13 >0.14 |
144 8 4 |
reference 1.9 (0.8-4.6) 0.7 (0.2-2.1) |
| Closest to diagnosis (µT) < 0.05 0.05-0.13 >0.14 |
142 5 9 |
reference 0.9 (0.3-2.5) 1.1 (0.5-2.5) | |
| Measurements | |||
| Savitz et al. (1988) |
Spot measurements Low power conditions (µT) < 0.2 >0.2 |
23 2 |
reference 1.04 (0.22-4.81) |
| High power conditions < 0.2 >0.2 |
22 3 |
reference 0.82 (0.23-2.93) | |
| Electric fields (V/m) < 12 >12 |
22 3 |
reference 0.53 (0.15-1.81) | |
| Preston-Martin et al. (1996) |
Spot measurements (µT) > 0.2 > 0.25 > 0.3 p for trend |
13 11 7 |
0.7 (0.3-1.5) 0.9 (0.3-2.3) 0.9 (0.3-3.2) 0.29 |
| 24-h measurements (µT) > 0.2 > 0.25 > 0.3 p for trend |
16 13 12 |
1.2 (0.5-2.8) 1.4 (0.5-3.8) 1.7 (0.6-5.0) 0.79 | |
| STAR profiles (µT) > 0.2 > 0.25 > 0.3 p for trend |
13 10 5 |
1.2 (0.5-3.3) 1.5 (0.5-5.1) 0.9 (0.2-4.1) 0.82 | |
cases |
|||
| Wire codes | |||
| Wertheimer and Leeper (1979) |
Birth address HCC/LCC |
10 / 21 |
2.48 (0.73-8.37) |
| Death address HCC/LCC |
18 / 26 |
2.08 (0.84-5.16) | |
| Savitz et al. (1988) |
HCC/LCC VHCC/Buried |
5 3 |
0.80 (0.29-2.18) 3.30 (0.80-13.65) |
| Calculated fields | |||
| Feychting & Ahlbom (1993) |
Unmatched analyses (µT) < 0.1 0.1-0.19 >0.2 >0.3 |
16 1 2 1 |
reference 0.9 (0.0-5.2) 1.3 (0.2-5.1) 0.9 (0.0-5.4) |
| Matched analyses 0.1-0.19 >0.2 |
0.8 (0.1-7.8) 0.9 (0.2-5.0) | ||
| Olsen et al. (1993) |
< 0.1 0.1-0.24 >0.25 >0.4 |
247 2 1 1 |
reference 5.0 (0.7-36) 5.0 (0.3-82) 5.0 (0.3-82) |
| Verkasalo et al. (1993) |
Cumulative exposure (µT-years) 0.01-0.39 >0.40 |
14 1 |
(SIR) 0.88 (0.48-1.5) 0.64 (0.02-3.6) |
| Average exposure (µT) 0.01-0.19 >0.2 |
15 0 |
0.91 (0.51-1.5) 0.0 (0.0-4.2) | |
| Tynes & Haldorsen (1997) | Average exposure (µT) < 0.05 0.05-0.13 >0.14 |
27 1 2 |
reference 1.0 (0.1-8.7) 2.5 (0.4-15.5) |
| Closest to diagnosis (µT) < 0.05 0.05-0.13 >0.14 |
27 1 2 |
reference 0.8 (0.1-6.7) 1.2 (0.2-6.4) | |
| Measurements | |||
| Savitz et al. (1988) |
Spot measurements Low power conditions (µT) < 0.2 >0.2 |
11 2 |
reference 2.17 (0.46-10.31) |
| High power conditions < 0.2 >0.2 |
10 3 |
reference 1.81 (0.48-6.88) | |
| Electric fields (V/m) < 12 >12 |
11 2 |
reference 0.70 (0.15-3.27) | |
system tumors | |||||||
| Savitz et al. (1990)1 |
Electric blanket | 1.3 (0.7-2.6) | 1.5 (0.5-5.1) | 1.1 (0.4-3.6) | 1.0 (0.2-8.6) | 1.8 (0.9-4.0) | 1.2 (0.3-5.7) |
| Electric water bed | 0.3 (0.1-1.2) | 0.7 (0.2-2.5) | - | - | 0.5 (0.2-2.0) | 0.3 (0.1-2.7) | |
| Bedside electric clock | 0.9 (0.5-1.6) | 1.4 (0.7-2.9) | 0.5 (0.2-1.2) | 1.5 (0.6-4.5) | 0.8 (0.4-1.7) | 1.1 (0.5-2.8) | |
| Heating pad | 0.9 (0.4-2.2) | - | 2.0 (0.7-5.9) | - | 0.9 (0.4-2.7) | - | |
| Hair dryer | - | 0.5 (0.2-1.3) | - | 0.7 (0.2-2.5) | - | 0.6 (0.3-1.7) | |
| London et al. (1991)2 |
Bedroom. air conditioner | 0.91 (0.51-1.66) | 0.54 (0.21-1.25) | ||||
| Electric blanket | 1.21 (0.66-2.29) | 7.0 (0.86-121.8) | |||||
| Electric fan | 1.16 (0.77-1.75) | 1.20 (0.81-1.80) | |||||
| Electric space heater | 1.18 (0.62-2.32) | 1.45 (0.82-2.66) | |||||
| Electric water bed | 0.67 (0.34-1.28) | ||||||
| B&W television | - | 1.49 (1.01-2.23) | |||||
| Electric clock (all) | - | 1.33 (0.90-1.97) | |||||
| Electric clock-dial | - | 1.86 (0.97-3.83) | |||||
| Electric clock-digital | - | 1.10 (0.71-1.72) | |||||
| Color television | - | 1.06 (0.66-1.74) | |||||
| Curling iron | - | 6.0 (0.72-104.8) | |||||
| Electric clippers | - | 1.0 (0.06-19.60) | |||||
| Electric hair dryer | - | 2.82 (1.42-6.32) | |||||
| Microwave oven | - | 0.81 (0.48-1.36) | |||||
| Video game | - | 1.57 (0.80-3.27) | |||||
| Preston-Martin et al. (1996a)1 |
Electric blanket | 1.2 (0.6-2.2) | 1.2 (0.5-3.0) | ||||
| Electric water bed | 2.1 (1.0-4.2) | 2.0 (0.6-6.8) | |||||
| Electric clock (all) | 1.0 (0.8-1.3) | 0.7 (0.4-1.0) | |||||
| Electric clock-dial | 1.1 (0.7-1.8) | 0.6 (0.3-1.4) | |||||
| Electric heat | 1.6 (0.8-3.0) | 1.3 (0.7-2.4) | |||||
| Electric heat-radiant | 1.3 (0.2-8.3) | 1.4 (0.4-5.0) | |||||
| Microwave | 1.4 (0.9-2.3) | 1.0 (0.6-1.5) | |||||
| Ham radio | - | 2.1 (0.2-23.7) | |||||
| Hair dryer | - | 1.2 (0.7-2.1) | |||||
| Curling iron | - | 1.0 (0.4-2.5) | |||||
| B&W television | - | 0.7 (0.4-1.4) | |||||
| Baby monitor | - | 0.6 (0.2-0.7) | |||||
| Preston-Martin et al. (1996b)1 |
Electric blanket | 0.9 (0.6-1.2) | 1.0 (0.6-1.7) | ||||
| Electric water bed | 0.9 (0.6-1.3) | 1.2 (0.7-2.0) | |||||
| Hatch et al. (1998) |
Electric blanket Ever used |
1.59 (1.11-2.29) | 2.75 (1.52-4.98) | ||||
| Electric water bed Ever used |
0.9 (0.67-1.21) | 1.19 (0.87-1.62) | |||||
| Hair dryer Ever used |
1.14 (0.8-1.61) | 1.55 (1.18-2.05) | |||||
| Curling iron Ever used |
1.06 (0.83-1.36) | 1.74 (0.91-3.31) | |||||
| Reference: Never used or not within 3 feet |
Reference: Not used during reference year |
||||||
| Electric clock Digital display |
0.98 (0.73-1.31) | 1.20 (0.83-1.76) | |||||
| Dial display | 0.81 (0.52-1.28) | 1.69 (0.61-4.65) | |||||
| TV video game Ever used |
1.91 (1.36-2.68) | ||||||
cancers |
system tumors | ||||
| NRPB (1992)1 |
Wertheimer and Leeper (1979) Fulton et al. (1980) Tomenius (1986) Savitz et al. (1988) Coleman et al. (1989) Lin and Lu (1989) Myers et al. (1990) London et al. (1991) |
Measured EMFs:1.82 (1.09-3.04) Distance from EMF source:1.11 (0.71-1.73) Wire codes (HCC vs. LCC):1.53 (1.04-2.25) |
Measured EMFs: 1.16 (0.65-2.08) Distance from EMF source: 1.31 (0.72-2.21) Wire codes (HCC vs. LCC): 1.39 (1.08- 1.78) |
Measured EMFs: 1.85 (0.91-3.77) Distance from EMF source: 1.09 (0.50-2.37) Wire codes (HCC vs LCC): 2.04 (1.11-3.76) | |
| Ahlbom et al. (1993) |
Feychting and Ahlbom (1993) Olsen et al. (1993) Verkasalo et al. (1993) |
Calculated EMFs: 1.3 (0.9-2.1) | Calculated EMFs: 2.1 (1.1-4.1) |
Calculated EMFs: 1.0 (0.3- 3.7) | Calculated EMFs: 1.5 (0.7-3.2) |
| Washburn et al. (1994) 2,3 |
Wertheimer and Leeper (1979) Fulton et al. (1980) Tomenius (1986 ) Savitz et al. (1988) Coleman et al. (1989) Myers et al. (1990) London et al. (1991) Lowenthal et al. (1991) Fajardo-Gutierrez (1993) Feychting and Ahlbom (1993) Olsen et al. (1993) Petridou et al. (1993) Verkasalo et al. (1993) |
Distance from EMF source: 1.49 (1.11-2.00) |
Distance from EMF source: 1.58 (0.91- 2.76) |
Distance from EMF source: 1.89 (1.34-2.67) |
|
| NAS Report (1994)2,4 |
Wertheimer and Leeper (1979) Fulton et al. (1980) Tomenius (1986 ) Savitz et al. (1988) Coleman et al. (1989) London et al. (1991) Fajardo-Gutierrez (1993) Feychting and Ahlbom (1993) Olsen et al. (1993) Petridou et al. (1993) Verkasalo et al. (1993) |
Wire codes (HCC vs. LCC): 1.48 (1.18-1.85) - fixed 1.52 (1.08-2.14) - random Wire codes and distances less than 100 m: 1.36 (1.13-1.63) - fixed 1.38 (1.08-1.76) - random Spot measurements (2 mg exposures and greater): 0.92 (0.57-1.49) - fixed 0.89 (0.51-1.57) - random |
|||
| Feychting et al. (1995) 5 | Feychting and Ahlbom (1993) Olsen et al. (1993) | 0.1 - 0.19 µT: 1.4 (0.6-2.9) >0.2: 1.5 (0.9-2.7) >0.5: 3.5 (1.7-7.3) | 10.1 - 0.19 µT: 2.0 (0.7-5.3) >0.2: 2.0 (1.0-4.1) >0.5: 5.1 (2.1-12.6) | 0.1 - 0.19 µT: 0.7 (0.1-5.6) >0.2: 2.1 (0.8-5.5) >0.5: 3.3 (0.7-15.0) | 1.0 - 1.9 mG: 1.1 (0.3-3.6) >2.0: 0.8 (0.3-2.4) >5.0: 2.3 (0.6-8.0) |
| Meinert and Michaelis (1996)2,6 | Wertheimer and Leeper (1979) Fulton et al. (1980) Tomenius (1986 ) Savitz et al. (1988) Coleman et al. (1989) Myers et al. (1990) London et al. (1991) Fajardo-Gutierrez (1993) Feychting and Ahlbom (1993) Olsen et al. (1993) Petridou et al. (1993) Verkasalo et al. (1993) Preston-Martin et al. (1994) |
Wire code (HCC vs. LCC): 1.37 (0.94-2.00) Distance: < 100 m: 1.09 (0.89-1.35) < 50 m: 1.10 (0.86-1.40) < 25 m: 1.42 (0.88-2.29) EMF measures: > 0.1 µT: 0.97 (0.82-1.15) > 0.2 µT: 1.23 (0.96-1.57) > 0.3 µT: 1.62 (1.10-2.39) |
Wire code (HCC vs. LCC): 1.66 (1.11-2.49) Distance: < 100 m: 1.13 (0.79-1.62) < 50 m: 1.31 (0.92-1.87) < 25 m: 1.85 (0.98-3.49) EMF measures: > 0.1 µT: 1.55 (0.88-2.73) > 0.2 µT: 1.89 (1.10-3.26) > 0.3 µT: 1.27 (0.28-5.76) |
Wire code (HCC vs. LCC):1.32 (0.52-3.37) EMF measures: > 0.1 µT: 2.18 (0.51-9.34) > 0.2 µT: 2.21 (0.72-6.80) > 0.3 µT: 1.69 (0.43-6.59) |
Wire code (HCC vs. LCC): 1.50 (0.69-3.26) Distance: < 50 m: 1.53 (0.19-12.0) EMF measures: > 0.1 µT: 0.89 (0.39-2.05) > 0.2 µT: 1.30 (0.78-2.19)> 0.3 µT: 1.89 (0.80-4.43) |
| Calculated Fields | |||||||
| Measured Fields | |||||||
| Wire Codes | |||||||
| Proximity to electrical facilities | |||||||
| Leukemia: dose-response analysis | No. of studies | Random effects relative risk (95% CI) per 0.1 µT | Regression slope | Standard error | P-Value for heterogeneity test | ||
| Spot measurements | |||||||
| Calculated fields 0.2 or 0.25 µT 0.3 or 0.4 µT |
4 | 1.2 (1.0-1.5) | 0.20 |
0.11 | 0.2 | ||
| Wire codes | |||||||
| Scored by spot measurements | |||||||
| Scored by 24-hour Bedroom measurements | |||||||
| Brain cancer: meta analysis | No. of studies | No. of exposed cases | Random effects odds ratio (95% CI) | Fail-safe N | Sample size needed | P-value for heterogeneity | Range of odds ratios from sensitivity analysis |
| Calculated fields | |||||||
| Measured fields | |||||||
| Wire codes | |||||||
| Proximity to electrical facilities | |||||||
| Spot measurements | |||||||
| Calculated fields 0.2 or 0.25 µT 0.3 or 0.4 µT |
4 | 1.1 (1.0-1.3) | 0.10 | 0.07 | |||
| Wire codes | |||||||
| Scored by spot measurement | |||||||
| Flat model | |||||||
| Linear model | |||||||