The health effect of magnetic fields (MFs) of extremely low frequency has remained controversial despite efforts to reach consensus.^1,2 The main challenges in studying MF are (1) accurately measuring MF exposure level during the relevant time period and (2) identifying susceptible populations.

Ever since the first report of a potential effect of electromagnetic fields (EMFs) on the risk of childhood leukemia,³ studying the health effect of EMF has mainly been focused on cancer risk.^4–9 Although the correct measurement of MF exposure should be personal exposure during the etiologically relevant time period, MF exposure in most studies was measured by surrogate, including wire code classification of the residence and residential spot measurement, frequently measured retrospectively.^3,5,8 Residential spot measurement does not capture all personal MF exposure at home and ignores exposure outside the residence. Wire code classification correlates poorly with actual residential MF level.¹⁰ Imprecise measurement of MF exposure coupled with mis-specification of the relevant exposure period could lead to significant misclassification of MF exposure level, which, if nondifferential, would dilute any true effect. Consequently, it was not surprising that many studies failed to detect an effect of MF exposure, if one exists. More recent studies with more accurate measurement of MF exposure in the relevant time period have tended to report an association with the exposure.^7,11–16

With rare exceptions,¹⁷ no attempt has been made to identify a population susceptible to MF. It is conceivable that the biological effects of MF will most likely be felt among the population most vulnerable to environmental insults such as MF. If a true MF effect is difficult to detect owing to exposure misclassification, then a failure to identify susceptible populations further reduces the ability to detect an MF effect, especially if the susceptible population consists of only a small part of the study population.

The association between MF exposure and the risk of miscarriage has been studied only to a limited extent, and the examination has mostly been for exposure to video display terminals (VDTs). Because of the limited amount of MF emitted from VDTs,¹⁸ however, VDTs are unlikely to be a major source of MF in a woman’s daily life. Therefore, it would be difficult to detect an association of miscarriage with VDT use, even if one does exist.^19,20 One study with actual measurement of VDT MFs, however, indicated that when a woman was exposed to a VDT with a high MF level [a peak level >9 milligauss (mG)] during pregnancy, she had a more than 3-fold increased risk of miscarriage.²¹ Another case-control study reported an association between an increased residential spot MF level obtained retrospectively and risk of miscarriage including subclinical abortion determined by measuring serum human chorionic gonadotropin level.²² Use of electric blankets has also been associated with risk of miscarriage.²³

We carried out a prospective cohort study to examine the association between 24-hour personal MF exposure and miscarriage. A previous study had suggested that a time-weighted average (TWA) MF exposure above 2 mG conveyed an excess risk.²⁴ The current study was funded by the California EMF Program to test this hypothesis. The funding authorities agreed that we were free to evaluate the association of other exposure metrics with miscarriage. Accordingly, in addition to TWA, we also examined one metric of interest to us, the maximum MF (MMF) encountered during the day.

In-Person Interview

All participating women were interviewed in person by a well-trained interviewer to obtain detailed information on known risk factors for miscarriage and other adverse pregnancy outcomes, as well as potential confounders. The women were also asked about their residential and occupational exposures to MF including the use of appliances, as well as their daily activities during pregnancy.

Magnetic Field Measurements

Measurement of Personal Magnetic Field Exposure

To measure her MF exposure during pregnancy, each participating woman was asked to wear an EMDEX-II meter for 24 hours starting immediately after the in-person interview. The EMDEX-II was initiated in advance with a custom program to collect MF measurements every 10 seconds and store both broadband (40–800 Hz) and harmonic (100–800 Hz) resultant MF levels. The meter was specifically programmed only to show the time of day on the display without revealing any MF exposure level so that participants would remain blinded to the MF exposure level. Subjects were also asked to keep a diary recording their activities during this period.

At the end of the measurement period, a technician from Enertech Consultants Inc (Campbell, CA), the contracting firm for conducting MF measurements, examined the data both alone and in combination with the subject’s diary. The technician resolved any concerns about the data or diary with the subject at this time. The diary and a copy of the data then were forwarded to T. Dan Bracken Inc (Portland, OR), the contracting firm for performing data management on MF exposure, for further review and incorporation into the final MF database to create summary exposure measurements for analyses. After these examinations, women whose EMDEX II data did not match the activity diary or whose EMDEX II data revealed that they had failed to wear the meter (no MF recording) were excluded from the analysis (a total of 73 subjects).

To determine whether the daily activity pattern at the 24-hour measurement represented her typical day during pregnancy, we asked each participant at the end of the 24-hour measurement whether the patterns of the following activities were “fairly similar” or “quite different”: home in bed, home not in bed, at work, during travel, and other activities. If a participant answered that the daily activity pattern was “quite different” for any of these five activity categories, her measurement day was considered nontypical; thus, her MF measurements on that day may not reflect her true exposure level during her pregnancy.

Residential Spot Magnetic Field Measurements

Spot measurements were taken in the subject’s bedroom, the kitchen, and the most frequently occupied room that was neither a bedroom nor a kitchen. Measurements were made at the abdominal level in the center of each room as well as the location that the subject typically occupied. In addition, measurements were taken at the front entrance of the residence and at approximately 15-foot intervals proceeding clockwise around the residence. A measurement was also made at the outside location nearest the subject’s bedroom.

Wire Code and External Wiring Information

The Enertech Consultants technicians performed wire coding and collected information on external wiring by producing an aerial sketch of the residence and all overhead electric system lines within 150 feet of the residence. This information was used to determine the Wertheimer-Leeper wire-code categories, which were classified as underground, very low-current configuration, ordinary low-current configuration, ordinary high-current configuration, and very high-current configuration.^3,10

Pregnancy Outcomes

The pregnancy outcomes for all participants were ascertained through one of the following methods: linking various automated KPMCP databases, reviewing medical charts, and telephoning those whose outcomes could not be identified through the previous two methods. Among 1,063 women who participated in the study, pregnancy outcomes for 1,058 (99.5%) participants were identified. Although the final outcomes were unknown for the remaining five subjects because of their moving out of the area, they were included in the final analysis and their pregnancy was censored at the gestational age at which they were known to have remained pregnant (all beyond 20 weeks of gestation). After excluding 21 additional women with missing data on personal exposure information or with incomplete interviews, 969 subjects were left in the final analysis. Figure 1 summarizes the recruitment and participation of the study.

---

FIGURE 1 Recruitment process.

---

Because the MF exposure was measured after the interview, neither interviewers nor participants knew their MF exposure level at the time of interview. In most cases, they were both also blinded to participants’ pregnancy outcomes at the interview. Nevertheless, because our study recruited participants at an early gestational age (median gestational age at entry was 40 days) when miscarriage occurs at a higher frequency, 97 participants had already had a miscarriage at the time of interview. Of them, 78 had a miscarriage before the initial contact and the remaining 19 had a miscarriage after having given their consent to participate in the study but before their interview. They were included in the study because they resided in the same residence when the miscarriage occurred as well as meeting other eligibility criteria. Nonetheless, these women remained unaware of their MF exposure level.

Statistical Analysis

We used the Cox proportional hazards model to examine the miscarriage risk associated with MF exposure during pregnancy while controlling for potential confounders and taking into account different gestational ages at entry. A woman was considered at risk of miscarriage as soon as she had a positive pregnancy test (entry time). Gestational age in days was used as the time variable. The woman continued to be considered at risk until either she had a miscarriage or was censored. Women who had other pregnancy outcomes including ectopic pregnancy or induced abortion (3.6%) were censored at the time when those outcomes occurred. Women who remained pregnant beyond 20 weeks of gestational age (80%) were censored at 20 weeks of gestation because by definition, no miscarriage occurs after 20 weeks of gestation.

To take into account the entry at various gestational ages, the time variable (gestational age) with left-truncation was used in the proportional hazards model.^25,26 The association between MF exposure during pregnancy and miscarriage risk was evaluated at any specific gestational age only among those women who were pregnant and had entered into the study at that time. Using the left-truncation of the time variable to reflect participants’ actual contribution of their person-time to the risk assessment in the Cox proportional hazards model allowed control of any potential biases caused by the association of gestational age at entry with MF exposure and miscarriage risk. The potential confounders included in the Cox proportional hazards model were based on the known or potential risk factors for miscarriage as well as on common sociodemographic variables.

Because the mechanism of the potential effect of MF during pregnancy and the risk of miscarriage was not clear, we decided to examine the effect of the MMF level exposed for a potential threshold effect, in addition to the effect of average dose (TWA). It seemed more plausible to us that MF exposure has a threshold below which any exposure is biologically irrelevant. Thus, we postulated that MMF is a better measure for detecting the MF biological effect than TWA which, combining MF doses at all levels, is a diluted and insensitive measure.

To evaluate a potential threshold effect of MF exposure, we first examined the relation between MMF level in deciles and the risk of miscarriage. Figure 2 shows that a woman’s MMF level during the 24-hour measurement period appeared to be associated with an increased rate of miscarriage, starting around 12–18 mG. The rate remained elevated with increasing MMF exposure level. Therefore, we chose 16 mG as the cutoff for all subsequent analyses because it was also the cutoff for the first quartile. The cutoff was also chosen for practical reasons because, before the data collection, we had selected several exposure levels for which other parameters of exposure dose (for example, total sum of MF, duration, and number of times above the specifically selected level) were constructed. Between 10 and 20 mG, 16 mG was the only such cutoff point that was preselected. Therefore, by choosing 16 mG, we would be able to examine other parameters of exposure.

---

FIGURE 2 Miscarriage rate by maximum magnetic field (MF) exposure.

---

Table 1 and Table 1A presents the characteristics of the exposed (MMF 16 mG) and unexposed (MMF <16 mG) subjects. Overall, there was little difference between the two cohorts in demographic characteristics, potential risk factors for miscarriage, reproductive history, and gestational age at entry to the study. The exposed women (MMF 16 mG) were more likely to have been employed before conception, to have had fever during pregnancy, and to have drunk tapwater, but they were less likely to have had a history of subfertility defined as failure in conceiving after having had regular intercourse without contraception for more than 12 months.

---

Table 1. Characteristics of Study Population by Daily Maximum Magnetic Field (MMF) Exposure Level (<16 mG or 16 mG)

---

---

Table 1A. Continued

---

A few known risk factors for miscarriage, including a lack of nausea and vomiting, vaginal bleeding, maternal age 35 years, and prenatal smoking, were also associated with risk of miscarriage in our study population.

Prenatal exposure to MMF 16 mG was associated with an 80% increased risk of miscarriage. This observed association was robust against potential confounders, for the estimate barely changed after adjustment for about 30 known risk factors for miscarriage or potential confounders listed in Table 1; crude RR = 1.81 vs adjusted RR (aRR) = 1.80. Using total sum of MF amount 16 mG as a measure of dose above the threshold (taking into account both MF level and duration above the threshold), the risk of miscarriage remained elevated for higher doses of MF exposure (Table 2). Using other dose parameters including MMF in quartiles, and duration or number of times above the threshold (16 mG), showed a similar relation.

---

Table 2. Daily Maximum Magnetic Field Exposure during Pregnancy and the Relative Risk (RR) of Miscarriage

---

To determine whether the exposure to MMF 16 mG was simply a marker for certain activities, we examined the location of the exposure. About half of the exposed women were exposed to MMF 16 mG from multiple locations/activities. Among the single location of the exposure, sleeping in bed, which likely encompassed a relatively large percentage of the 24-hour measurement period, only contributed less than 1% of MMF exposure 16 mG. On the other hand, travel, which likely covered a relatively short time period, conveyed about 14% of the MMF exposure. The risk of miscarriage associated with MMF 16 mG did not vary much by the location/activity of the exposure; the risk of miscarriage was 17.7% for those who were exposed from multiple locations, 18.1% for those who were exposed only from the period at home but not in bed, 18.8% for those who were exposed only from workplace, 19.4% for those who were exposed only during travel, and 20.6% for those who were exposed from other locations/activity periods.

To evaluate whether fetuses at an early gestational age are more susceptible to MMF exposure, we examined the association separately for fetal loss before and after 10 weeks of gestation. Table 3 shows that the risk of miscarriage associated with MMF was higher for fetal loss before 10 weeks of gestation (aRR = 2.2, 95% CI = 1.2–4.0). If a fetus had survived to 10 weeks or more, the association was noticeably reduced (aRR = 1.4, 95% CI = 0.8–2.5).

---

Table 3. Daily Maximum Magnetic Field Exposure during Pregnancy and the Relative Risk (RR) of Miscarriage by Gestational Age

---

To examine whether the effect of prenatal MMF exposure was greater for women who might be more susceptible to environmental insults, we restricted analyses to women who had a history of either multiple miscarriages (2 or more) or subfertility. Table 4 shows that the association of MMF with miscarriage was stronger in this group of women than in the overall population; aRR = 3.1 (95% CI = 1.3–7.7) for the exposure MMF 16 mG and aRR = 4.7 (95% CI = 1.4–15.9) for the exposure before 10 weeks of gestation.

---

Table 4. Daily Maximum Magnetic Field Exposure during Pregnancy and the Relative Risk of Miscarriage among Susceptible Populations: Women with a History of Subfertility and/or Multiple Miscarriages

---

To examine further the effect of the misclassified MF exposure measurement on the association, we stratified our participants by whether their activity patterns at the measurement day represented their typical daily activity patterns during pregnancy. Presumably an MF measurement obtained on a nontypical day was less likely to represent the overall MF exposure during pregnancy, resulting in more misclassification of the true MF exposure level, than an MF measurement obtained on a typical day. Table 5 shows that the association was strengthened among women whose MMF measurement was obtained during a typical day (aRR = 2.9; 95% CI = 1.6–5.3), whereas the association disappeared among women whose MMF measurements were obtained on a nontypical day (aRR = 0.9; 95% CI = 0.5–1.8). Compared with Table 3 and Table 4, Table 6 also shows that after excluding the subjects with any aspect of their day characterized as nontypical, a stronger association with risk of miscarriage was consistently observed under various examinations.

---

Table 5. Daily Maximum Magnetic Field Exposure during Pregnancy and the Relative Risk (RR) of Miscarriage by Women Whose Daily Activities at Measurement Were and Were Not Their Typical Daily Activities during Pregnancy

---

---

Table 6. Various Measures of the Amount of Daily Magnetic Field Exposure during Pregnancy and the Relative Risk of Miscarriage among Women Whose Daily Activities at Measurement Were Their Typical Daily Activities during Pregnancy

---

Spot measurements did not show a consistent pattern of an association between increased exposure level (in quartiles) and the rate of miscarriage. In our study, the residential wire-code category was not associated with either MMF or risk of miscarriage (the results can be obtained upon request).

The potential misclassification of MF exposure was likely to be influenced by two factors: temporal variation in MF level and daily activity pattern. Few studies have evaluated the temporal variation of MF exposure level. One such study used repeated measurements over 12–26 months and concluded that MF level is relatively stable over the study period and that MF measurement on a single visit is a good indicator of average personal exposure levels over time, although the temporal stability of the MMF metric was not specifically examined.¹⁰

To examine the potential influence of a change of activity patterns on our results, we stratified the analysis of the effect on women depending on whether the measurement day was a typical day during this pregnancy. If MMF exposure is truly associated with the risk of miscarriage, one would expect the association to be stronger among women whose measurement day reflected their typical day during pregnancy. Table 5 shows that the MMF association was indeed greater among women whose MMF measurement likely reflected their true exposure during pregnancy (aRR = 2.9; 95% CI = 1.6–5.3), whereas there was no MMF association observed among women whose MMF measurements were not likely to have reflected their true exposure during pregnancy (aRR = 0.9; 95% CI = 0.5–1.8). After excluding women whose MF measurement was obtained on a nontypical day, various other measures also indicated a stronger association (Table 6). This observation provides further evidence that prenatal MMF exposure may be genuinely related to the risk of miscarriage.

Although the overall participation rate (39%) was low, this was a prospective cohort study and MMF exposure level was largely unknown to the general public. Thus, the low participation rate was unlikely to be associated with MMF exposure. In addition, although we do not know the MMF level for nonparticipants, our data records revealed that the rate of miscarriage among nonparticipants was 17.2%, compared with 16.4% among participants (Table 2), indicating comparability between participants and nonparticipants with regard to their risk of miscarriage. Because we recruited women at an early gestational age (median gestational age of 40 days), 78 subjects had already had a miscarriage (49% of all miscarriage cases) at the time of initial contact for their participation. They were included in the study because measurements taken soon after miscarriage (median interval of 22 days) were considered still representative of their MMF exposure level before miscarriage. Separate analyses stratifying miscarriage cases depending on whether their measurements were taken before or after their miscarriage showed essentially the same results for both types of cases; for miscarriage <10 weeks of gestation, aRR = 5.6 (95% CI = 0.7–42.4) and 6.1 (95% CI = 1.9–20) for cases measured before and after miscarriage, respectively; for miscarriage 10 weeks, aRR = 1.7 (95% CI = 0.7–3.9) and 1.6 (95% CI = 0.3–7.6), respectively.

Owing to the limited studies of the MF effect on the risk of miscarriage,^18–23,27 a comparison of our results with the literature may be difficult. Nevertheless, examining the literature of the epidemiologic studies of the MF effect on other health outcomes, especially childhood leukemia, reveals that the inconsistency of results from previous studies might be attributed to a lack of adequate exposure measurement and a failure to identify a susceptible population. Most previous studies were case-control in design and the MF exposure was often measured retrospectively, using the exposure level many years after the relevant time period to represent the actual MF level of interest in the past. Many studies only used indirect measurements of MF level such as wire code configuration. Although more recent studies have attempted direct measurements, frequently only residential spot measurements were obtained to represent a participant’s overall personal MF exposure level. Residential spot measurements do not necessarily capture residential exposure, let alone overall personal exposure from all sources. All of these may compromise MF measurements and could lead to misclassification of MF exposure level (for both cases and controls), which would tend to mask an underlying MF effect. More recent studies that captured personal MF exposure and measured MF exposure closer to the relevant time period seem more likely to demonstrate an association between MF exposure and health outcomes such as childhood leukemia.^11,12,14,16

Our study was prospective in design and measured MF exposure level at, or close to, the relevant time of interest. We used personal measurement that captured MF exposure from all sources encountered by a woman. Therefore, the MF exposure level obtained in our study better reflected the true MF exposure level in the time period of interest than most previous studies of the MF effect, thus providing a better chance to detect the adverse MF effect. Our study also demonstrated that if we stratified our analyses by whether the daily activity pattern at measurement reflected a participant’s typical pattern during pregnancy, the associations with various measurements of MMF exposure were strengthened among women whose daily activity pattern at measurement was typical (Table 5 and Table 6). At the same time, no association could be detected among those whose daily activity pattern at measurement was not their typical pattern during pregnancy and, thus, less likely to reflect their true MF exposure during pregnancy. This observation suggests that the lack of appropriate measurement of MF exposure during the appropriate time period may reduce the ability to detect an MF effect and may have contributed to the absence of an association in other studies.

A second factor that may be important in detecting an MF effect is the identification of a susceptible population that includes sensitive endpoints, susceptible time periods, and vulnerable populations. So far, few studies have focused on this issue.¹⁷ Our study examined the MF effect on early and late miscarriage (<10 vs 10 weeks of gestation), which may be different in their sensitivity to MF exposure. Second, we evaluated the MF effect among those with a history of multiple miscarriages or subfertility, a population that suggested an underlying reproductive difficulty, and thus perhaps a high susceptibility to environmental insults. Our results suggest that MF exposure was more strongly related to early miscarriage (Table 3 and Table 6) and demonstrated a stronger association with the risk of miscarriage among the susceptible population (Table 4 and Table 6). It is conceivable that an embryo or fetus at early gestational age is much more sensitive to environmental insults. One of the reasons why a previously reported Finnish study was able to detect an MF association despite their crude MF exposure assessment (retrospectively obtained spot measurement) may have been that their endpoint was very early miscarriage including subclinical miscarriage.²² Using this endpoint may have allowed the detection of a greater EMF effect owing to the increased susceptibility of embryos/fetuses at an early gestational age. Therefore, an association was detected despite the misclassified MF exposure due to the crude MF measurement. A recent study of MF and childhood leukemia also reported that the association was greater among young children (<6 years of age).¹² A higher risk among young children seems plausible if one considers the vulnerability of early childhood development and its relation to possible fetal exposure during pregnancy. Therefore, a greater ability to identify a susceptible population could enhance ability to detect an MF effect.

This population-based cohort study with prospectively measured MF exposure level revealed an increased risk of miscarriage associated with an MMF exposure level 16 mG. This association appeared to have a threshold around 16 mG and persisted regardless of the locations/activities of MMF exposure. Prenatal MMF exposure was more strongly associated with early miscarriage (<10 weeks of gestation) when embryos or fetuses are likely much more sensitive to environmental insults, and among women who may be more susceptible to environmental exposures. The association was much stronger when women whose 24-hour MF measurements may not reflect their true prenatal MF exposure were excluded. These biologically coherent observations, all based on a priori hypotheses, provide evidence that prenatal MF exposure above a certain level (possibly around 16 mG) may increase risk of miscarriage.

Our study did not have information on the exact sources of measured MMF 16 mG. Fields of such magnitude can be found near electric appliances (for example, microwave ovens and fluorescent desk lamps); very close to devices with electrical motors (for example, hair dryers, can openers, and fans), electric equipment in the work place, and electrically powered transit systems; and under or above certain types of power lines.

The robustness of the association between MMF and miscarriage risk against potential confounders was supported by evidence that despite adjustment for more than 30 variables of known or suspected risk factors for miscarriage, the estimates were barely altered. Moreover, prompted by the findings in this study, Lee et al²⁴ reanalyzed the data from the study in which the findings related to TWA exposure led to funding the current study and confirmed our observed association between MMF and risk of miscarriage.

The MMF exposure level in our study population was comparable with that found in a nationwide survey²⁸ and our study population was racially/ethnically and socioeconomically diverse.

Although the potential mechanisms of a possible MMF effect on the risk of miscarriage are not currently well understood, early fetuses are known to be sensitive to environmental insults. A disruption of early fetal development at the cellular or molecular level by external MFs could conceivably result in fetal death. Despite the lack of clear understanding of the underlying mechanisms, these findings raise the question of a possible effect of MMF on early fetal loss.

Acknowledgments

We thank Raymond Neutra, Vincent DelPizzo, and Geraldine Lee for their contribution to the study design, data collection, and comments on the manuscript. Prepublication peer reviews from Abdelmonen Afifi, Michael Criqui, Lowell Sever, and Nancy Wertheimer were greatly appreciated. We also thank Cathy Schaefer for helping with study design, and Luana Acton, Diane Galligan, Melissa Parker, Nancy Rieser, Heather Washington, and Stephanie Webb for conducting interviews, as well as Richard L. Collett and William M. Zoerner for obtaining EMF measurements. The staff at the laboratories and the departments of obstetrics and gynecology at the Kaiser Permanente San Francisco and South San Francisco facilities provided generous support during the study period. The following utility companies and people generously provided us with the EMDEX II meters for exposure measurement: Southern California Edison Company, Sacramento Municipal Utility District, Seattle City Light, Greg Chang and Pacific Gas and Electric, and Margaret Wrensch from the University of California at San Francisco.