Secondary Logo

Journal Logo

Original Articles

Radiofrequency Exposure and Mortality from Cancer of the Brain and Lymphatic/Hematopoietic Systems

Morgan, Robert W.; Kelsh, Michael A.; Zhao, Ke; Exuzides, K. Alex; Heringer, Shirley; Negrete, Wendy

Author Information
  • Free

Abstract

Concern over the safety of radiofrequency (RF) energy from cellular telephones and transmitting facilities for cellular systems has grown alongside the potential increase in public RF exposure resulting from the proliferation of wireless communication systems. RF communication bands, which include amplitude- and frequency-modulated (AM/FM) radio, very high frequency (VHF) and ultra-high frequency (UHF) television, cellular telephones, pagers, two-way radios, radar, and satellite communications, make up part of the nonionizing portion of the electromagnetic spectrum. Extremely low-frequency energies (0–300 Hz, with 50 and 60 Hz used for electric power), infrared light, and visible light make up the remainder of the nonionizing energies. RF exposure is defined as the electromagnetic fields resulting from frequencies ranging between 300 Hz and 300 GHz. 1 Microwaves fall within the RF range with frequencies between 300 MHz and 300 GHz. Analog cellular telephones operate in the range of 800–960 MHz, and newer digital personal communication devices operate in the range of 1600–2000 MHz.

The thermal effects of high exposures to RF electric and magnetic fields are the basis for existing safety and health exposure guidelines. 1–3 Recent public health debate has focused on the potential health effects of low-level athermal RF field exposure. 1,2,4–6 The current scientific consensus is that there is insufficient evidence from animal and epidemiologic research to demonstrate adverse human health effects. 1,2

Studies have reported excess leukemia rates associated with proximity to AM/FM radio and television facilities, 7–9 but these results have not been confirmed by other reports. 10–12 Such ecologic studies have substantial limitations from poor exposure assessment, reliance on group rather than individual data, and lack of data regarding potential confounders.

Occupational studies have examined cancer morbidity and mortality in association with RF exposures among military personnel, including U.S. Navy veterans of the Korean War, 13 U.S. Air Force personnel, 14 Polish servicemen, 15 embassy workers, 16 electric utility workers, 17 telecommunication workers, 18–21 and plasticware workers. 22 Exposures in these cohorts often consisted of a mixture of RF, microwave, and power-frequency (50 and 60 Hz) electric and magnetic fields. These studies and other registry-based studies found excess cancer mortality or morbidity for cancers of the brain, 23–26 lymphatic and hematopoietic systems, 15,27–29 respiratory system, 13,17 stomach, 15,17 breast, 20 and skin (melanoma). 15,30 All studies relied on qualitative classifications of RF exposure, with four 13,15–17 supported by limited RF measurements. In a study of amateur radio operators living in Washington State and California, Milham 27 observed increased mortality for brain cancer, lymphatic and hematopoietic tissue cancers, and other lymphatic malignancies. In a cohort of cellular telephone users, Rothman et al31 reported no excess all-cause mortality; mortality risk for all cancers or specific cancers was not reported. Associations with brain cancer and leukemia were most frequently reported, although other studies did not find elevated relative risks for these cancers. 13,16,17 Recent reviews suggest no consistent elevation in cancer risk associated with RF exposures. 1,32,33

We studied Motorola employees, because their probability of RF exposure is higher than the general population. Motorola has been active in the design, manufacture, testing, and use of wireless devices for more than 50 years. Motorola’s products include two-way radios, defense and space communication devices, pagers, cellular telephones, and related antenna infrastructure. Workers’ exposure occurred at RFs of 30, 150, and 450 MHz for two-way radios (beginning in the 1960s); 800 MHz for cellular telephones; and higher frequencies for microwave communications. Most Motorola employees have no significant occupational exposures to extremely low (50–60 Hz) electromagnetic fields or ionizing radiation. Although solvent or chemical exposures may have occurred among a small percentage of employees in some sectors of this workforce, such exposures were not assessed for this study.

We specified two a priori hypotheses: (1) RF exposure is a risk factor for brain cancer, and (2) RF exposure is a risk factor for all lymphomas and leukemias.

Subjects and Methods

Cohort Definition and Ascertainment of Vital Status

The cohort comprised all U.S. Motorola employees with at least 6 months of cumulative employment who were employed for at least 1 day between January 1, 1976 and December 31, 1996. Our primary data source was the company’s electronic personnel files, which provided information on each employee’s job code and description, work location, employment dates, termination date (if applicable), employment status (active, terminated, retired, or deceased), and standard demographic information (date of birth, sex, and race/ethnicity). We followed cohort members’ work histories and vital status through December 31, 1996. We limited our study to the 1976–1996 period because of the availability of electronic personnel data, the availability of mortality data from the National Death Index, and the availability of mortality data from the Social Security Administration (SSA) for the 1975–1980 period. During this 5-year period, identification of deaths using the SSA data is very high. 34 Because 4 years of employee data were missing from the electronic files, we reconstructed those years from original personnel records. To validate cohort completeness, we randomly sampled 2,153 hard-copy records from archived personnel files (stratified by termination year and location). We compared the selected records against the cohort assembled from the electronic personnel data.

Mortality Ascertainment

To establish vital status as of December 31, 1996, we compared the cohort with the SSA Master Mortality File and the National Death Index. For deceased cohort members, we obtained death certificates from states’ vital statistics offices and company benefits records. Trained nosologists coded underlying and associated causes of death according to the International Classification of Diseases, 9th revision (ICD-9). To evaluate the accuracy of our ICD-9 coding, we recoded a random sample of death certificates after blinding our nosologists to the original coding.

Exposure Assessment

Exposure assessment was limited by the lack of personal monitoring data. On the basis of expert opinion, however, it was possible to classify relative exposures and identify large occupational groups with little or no RF exposure. We created a job exposure matrix to categorize each of 9,724 job titles into one of four RF exposure groups labeled as background (coded as 0), low (coded 1), moderate (coded 6), and high (coded 100). We developed the classification system from numerous discussions and review meetings among the study team, Motorola experts (research engineers, industrial hygienists, and technical managers from each of the eight business sectors), and exposure assessment experts. We based exposure classifications on business sectors, work site, job codes and descriptions, calendar period, and the expert assessments. Our expert panel considered the sources of RF exposure that these jobs would have as part of their usual work activities. The experts were unaware of the disease status of individuals when developing the RF exposure matrix.

The numeric values (0, 1, 6, and 100) for RF exposure groups were derived from an exposure validation study. In the validation study, RF exposure sources were classified into the following groups: group 0, background exposure; group 1, 0 to <0.6 W; group 2, 0.6 to <2.0 W; group 3, 2.0 to <5.0 W; group 4, 5.0 to <50 W; and group 5, 50+ W. Using the midpoint of these validation study categories, we calculated the average score values across job titles classified in the job exposure matrix. The approximate relative values we estimated from the validation study for the job exposure matrix categories were 0, 1, 6, and 100. We used these values as well as the simple scale (0, 1, 2, and 3) to represent the relative exposure rankings.

Business sectors were defined by the products or services they produced, for example, pagers, cell phones, semiconductors, automotive products, and government and defense communications. RF exposures occurred primarily among employees in paging, government, and cellular telephone business sectors. The current communication products (pagers and cellular telephones) involve less energy output than previous products (hand-held radios and government communication systems), and worker exposures have decreased over time. Because personal cellular telephone use was variable, spanned across job titles, and was not well documented, we did not incorporate cellular telephone use per se into the occupational RF exposure matrix.

Similar job titles (for example, engineer) could vary in potential exposure depending on business sector. The four RF exposure groups included the following types of jobs and activities. Unexposed workers (similar to levels received by the general public) included administrative and support personnel who rarely encountered workplace RF exposures above background levels. Sample job titles in this group included managers, clerks, analysts, and marketing and administrative staff. Low RF exposure workers had infrequent low-level RF exposures resulting from their work or proximity to assembly areas. Job titles in this group included assemblers, inspectors, operators, supervisors, assembly technicians, and engineers not directly involved with RF technologies. Moderate RF exposure workers included individuals who routinely used hand-held radios or worked with RF product development. This group included security, maintenance, and some supervisory staff and, for some business sectors, engineers and technicians (Table 1). High RF exposure workers included technicians, testers, and engineers involved with RF product testing, antenna systems, and government communications projects. Employees in these jobs could have higher and more frequent workplace RF exposures than employees in the low or moderate RF exposure groups (Table 1).

Table 1
Table 1:
Types of Job Titles Classified into Moderate or High Radiofrequency (RF) Exposure Groups by Business Sector at Motorola

Limited RF area measurement data have been sporadically collected for some sectors of the company. We reviewed these data and concluded that they could not be used in any quantitative way to estimate historical workplace RF exposures. The RF area measurement data were collected primarily for leakage detection purposes and were not comprehensive or complete, and most of the data were collected very recently and were not reflective of historical exposures.

Statistical Analysis

We compared mortality rates of the Motorola cohort and RF-exposed subcohort against populations internal and external to Motorola. For external comparisons, we combined the mortality rates of the states of Arizona, Florida, Illinois, and Texas, where most Motorola facilities are located. We used the OCMAP 35 analysis program to calculate standardized mortality ratios (SMRs) and 95% CIs, adjusting for age, gender, and race. For internal comparisons, we compared mortality rates of workers with moderate or high RF exposure against the mortality rates of workers with either background or low RF exposure. Using Poisson regression models, we cal- culated adjusted rate ratios and 95% CIs. We used SAS software (proc GENMOD) to perform all statistical modeling procedures. 36 This procedure calculated likelihood ratio-based CIs. Final Poisson models controlled for age, gender, and period of hire. As part of model development, we examined the influence of race/ethnicity and length of employment. Race/ethnicity did not influence relative risk estimates, nor could we identify sufficient socioeconomic or biologic rationale to include this factor as a potential confounder. 37–39 In our final multivariate models, we excluded length of employment, because it was highly correlated with age and did not substantially affect relative risk estimates.

We summarized exposure assignments in three ways: (1) usual categories, (2) peak categories, and (3) cumulative exposure scores. The usual assignment reflected RF exposure level for the job held longest while at Motorola. The peak assignment reflected the job with the highest RF level. The cumulative RF exposure score was calculated as the sum of the products of the RF levels assigned to each job title multiplied by duration of employment in all jobs throughout an employee’s Motorola work history. If an employee had background exposure, the RF exposure score for that person was 0. We computed SMRs for the entire cohort and then for the moderate and high RF-exposed subcohorts. We also computed SMRs by duration of employment and latency assumptions (categorized as 0–9.9, 10–19.9, and 20+ years). In Poisson regression analyses, we used the cumulative exposure score to divide the cohort into two exposure groups:high (equal to or above the median) and low (less than the median). We then calculated rate ratios comparing the two groups. For usual and peak categories, we compared high-, moderate-, and low-exposure groups vs low and background. We also individually compared the high and moderate RF groups with the nonexposed group.

We examined the rate ratios for brain and lymphatic/hematopoietic cancers by duration of exposure and latency assumptions. We divided exposure duration into categories of 0–5, 5.1–9.9, and 10 years and greater RF exposure. Because of small numbers with each cancer outcome by duration of exposure, however, we compared those with more than 5 and those with 0–5 years of RF exposure. To assess the potential impact of different latency assumptions, we calculated rate ratios for exposure groups reclassified using 5-, 10-, and 20-year lag periods. We defined the lag period by counting back from the date of death or end of study and then calculated cumulative exposure up to the lag date. We also examined the potential impact of period of hire by subdividing the cohort into three groups: (1) those hired before 1975, (2) those hired between 1975 and 1985, and (3) those hired after 1985. We then calculated rate ratios from the Poisson models within each of these hire groups.

In addition, we reanalyzed data using different assumptions concerning exposure assignment, calculation of cumulative exposure scores, and statistical models. Where debate persisted among experts regarding the most appropriate RF assignment for a job title, we reanalyzed data using alternative exposure assignments in which we reassigned job titles to a lower RF exposure group. This component represented approximately 2.6% of our exposure matrix records and affected 5–9% of individual exposure assignments, depending on the exposure classification. Different RF levels were used to calculate cumulative RF exposure scores (ie, 0, 1, 6, and 100 or 0, 1, 2, and 3 to represent background, low, moderate, and high RF levels, respectively). Results presented in this study are derived from the 0, 1, 6, and 100 exposure scores. In our third type of sensitivity analysis, we used Cox regression 40,41 and Mantel-Haenszel stratified analysis instead of Poisson regression methods to assess whether the results changed with alternative statistical procedures. 41,42

Results

Descriptive Statistics

This occupational cohort of 195,775 workers contributed 2.7 million person-years of follow-up. At the end of the study, the cohort consisted of 6,296 deceased workers, 72,775 active workers, and 116,704 retired or terminated workers still alive at end of study. In our validation study of cohort completeness, we found that 99.4% of 2,153 randomly sampled hardcopy personnel records were included in the electronic data files. Of the 13 records not found in the electronic files, 10 did not meet our criteria for inclusion in the cohort.

The Motorola workforce had a high proportion of women employees (44%;Table 2) and was relatively young compared with most previously reported occupational cohorts. Women tended to work more in operator/assembly jobs (low or no RF exposure), whereas there were proportionally more men who were engineers (moderate or high RF exposure) among this workforce (Table 2). The racial/ethnic composition closely mirrored that of the U.S. population (75.1% white, 9.8% Hispanic, 9.3% African-American, 5.3% Asian, and 0.5% Native American;Table 2). Most employees (90.3%) worked in four states: Arizona, Florida, Illinois, and Texas. Most workers (66.2%; N = 129,591) in this cohort were born after 1950. There is a substantial number of workers (14.7%, N = 28,847), however, who were born before 1940, with another 37,329 (19%) born in the 1940s (Table 2). Although a majority (80%) of the Motorola cohort were hired after 1970, there were approximately 48,239 (24%) who were hired on or before 1976, which would allow for the potential for at least 20 years of follow-up.

Table 2
Table 2:
Demographic and Work History Characteristics of Motorola Employees, 1976–1996

On the basis of the usual classification, most of the workforce (N = 141,235, 72.1%) had no RF exposure above background level, and 8.6% had moderate or high usual RF exposure (Table 3). For peak exposure classification, the numbers of workers in each RF exposure group (low, moderate, and high) were somewhat higher than the usual exposure classification (Table 3).

Table 3
Table 3:
Distribution by Radiofrequency (RF) Exposure Classification among Motorola Employees, 1976–1996

Standardized Mortality Ratio Results

In the total cohort, we observed a pronounced healthy worker effect, with an all-cause SMR of 0.66 (95% CI = 0.64–0.67) and SMR for all cancers of 0.78 (95% CI = 0.75–0.82) (Figure 2). Of 60 specific causes of death or groups of related causes (25 are presented in Figures 2 and 3), no SMR exceeded 1.28, and only 5 were ≥1.0. All other SMRs were near or less than 1.0. Other digestive cancers had an SMR of 1.12 (N = 15), thyroid and other endocrine cancers had an SMR of 1.10 (N = 10), and Hodgkin’s disease had an SMR of 1.14 (N = 19). All other SMRs were near or below 1.0 (Figures 2 and 3). Men and women had similar patterns of mortality (data not shown). SMRs for all cancers of the lymphatic/hematopoietic systems and central nervous system (CNS)/brain were 0.77 (95% CI = 0.67–0.89) and 0.60 (95% CI = 0.45–0.78), respectively (Figure 2). Of the 53 CNS malignancies, 51 were brain cancers; therefore, the SMRs for CNS malignancies essentially reflect brain cancer mortality. FIGURE

FIGURE 1
FIGURE 1:
Standardized mortality ratios, 95% CIs, and number of cases (N) for selected mortality outcomes: entire Motorola cohort (N = 195,775), 1976–1996. Comparison group comprises the combined populations of Arizona, Florida, Illinois, and Texas.
FIGURE 2
FIGURE 2:
Standardized mortality ratios, 95% CIs, and number of high and moderate radiofrequency exposed cases (N): selected cancer outcomes, Motorola employees, 1976–1996. Comparison group comprises the combined populations of Arizona, Florida, Illinois, and Texas.

SMR analyses for the 24,621-member RF-exposed subcohort (moderate and high by peak exposure classification) produced a pattern similar to that of the overall cohort (Figure 3). In this subcohort, SMRs for all cancers of the lymphatic and hematopoietic system were 0.54 (95% CI = 0.33–0.83); for leukemias only, 0.77 (95% CI = 0.38–1.38); and for CNS/brain cancer, 0.53 (95% CI = 0.21–1.09). With only three deaths among the exposed group, we observed an SMR of 1.11 (95% CI = 0.23–3.24) for Hodgkin’s disease. For melanoma, the SMR was 1.13, and for cancers of the breast, stomach, and respiratory system, SMRs were below 1.0 (Figure 3).

We further analyzed the RF-exposed subcohort by duration of employment for all causes, all cancers, and the primary outcomes (brain cancer, all lymphatic/hematopoietic system cancers combined, and leukemia). The duration-of-employment calculations incorporated employment time from work histories before January 1, 1976, the date used to begin the start of person-year computations and cohort follow-up. For CNS/brain cancers, all lymphatic/hematopoietic cancers, leukemia, and Hodgkin’s disease, the SMRs did not show an increase with increased duration of employment. In the highest-duration group (20+ years), the SMRs were all below 1.0, ranging from 0.58 to 0.90 (data not shown).

Internal Cohort Analyses

To reduce bias due to the healthy worker effect, we performed internal cohort comparisons using Poisson regression models to estimate relative risks for usual, peak, and cumulative RF exposure scores. The rate ratios were near or below 1.0 for brain cancers, all lymphatic/hematopoietic cancers, leukemia, and non-Hodgkin’s lymphomas.

There were 51 total brain cancer cases, using underlying cause of death only, and 71 cases for both underlying and associated causes. We present the findings for underlying cause of death only because many of the associated causes were the result of metastatic cancer. There were ten exposed cases (≥ median) for the cumulative model, six (high or moderate) exposed cases for the usual model, and seven for the peak model (Table 4). All rate ratios for brain cancer were at or near 1.0, with upper CIs ranging from 1.59 to 2.84 depending on the exposure model. There was no increase in rate ratios by exposure duration or latency assumption.

Table 4
Table 4:
Rate Ratios (RR) and 95% Confidence Intervals (CI): Brain Cancer by Radiofrequency Exposure Classification, Duration of Exposure and Latency Assumptions

We observed 193 cases of lymphatic or hematopoietic cancers (leukemia and lymphomas, including both Hodgkin’s disease and non-Hodgkin’s lymphoma) using underlying cause of death and 203 cases for underlying and associated causes. There were 34 exposed cases under the cumulative exposure model and 18 and 21 exposed cases under the usual and peak exposure models, respectively. All rate ratios were well below 1.0. Again, there was no increase in the rate ratios when we considered exposure duration or latency assumptions (Table 5).

Table 5
Table 5:
Rate Ratios (RR) and 95% Confidence Intervals (CI): All Lymphatic/Hematopoietic Cancers by Radiofrequency Exposure Classification, Duration of Exposure, and Latency Assumptions

For leukemia, a subset of the all-lymphatic/hematopoietic cancer group, there were 79 cases coded for underlying cause and 87 cases coded for underlying and associated causes. There were 13, 10, and 11 cases in the RF-exposed groups for the cumulative, usual, and peak models, respectively. The rate ratios were lowest for the cumulative and peak exposure models (0.57 and 0.66) and 0.99 for the usual exposure model. Rate ratios for duration of exposure and latency analyses were all below 1.0 (Table 6).

Table 6
Table 6:
Rate Ratios (RR) and 95% Confidence Intervals (CI): Leukemia by Radiofrequency Exposure Classification, Duration of Exposure, and Latency Assumptions

Non-Hodgkin’s lymphoma, which accounted for 91 (45%) of the all-lymphatic/hematopoietic cancers, showed no positive association with RF exposure (Table 7). All rate ratio estimates were below 1.0, and there was no trend evident by duration of exposure or latency analyses (Table 7). There were 14 cases of non-Hodgkin’s lymphoma with RF exposure scores higher than the median value for cumulative exposure, and there were 4 and 6 cases classified as high or moderate RF exposed for the usual and peak exposure models, respectively.

Table 7
Table 7:
Rate Ratios (RR) and 95% Confidence Intervals (CI): Non-Hodgkin’s Lymphoma by Radiofrequency Exposure Classification, Duration of Exposure, and Latency Assumptions

There were 19 cases of Hodgkin’s disease in the entire cohort, with 7 exposed cases under the cumulative model and 3 exposure cases for the usual or peak exposure models. For peak and usual exposure models, the rate ratios were 1.7 and 3.2, with wide CIs (Table 8). For the cumulative exposure model, the rate ratios were near 1.0 (Table 8). The rate ratios for analysis by duration of exposure or latency assumptions based on the cumulative exposure model were all near or below 1.0. There was no case of Hodgkin’s disease that occurred at least 20 years after employment and no case among the highest exposure group for the peak and usual models (ie, all exposed cases were in the moderate RF exposure group). Hodgkin’s disease was the only cancer we examined that showed a modestly elevated rate ratio (2.25; 95% CI = 0.4–10.4) for one of the hire periods (1975–1985). None of the other cancer outcomes had elevated risk estimates for different periods of hire (data not shown).

Table 8
Table 8:
Rate Ratios (RR) and 95% Confidence Intervals (CI): Hodgkin’s Disease by Radiofrequency Exposure Classification, Duration of Exposure, and Latency Assumptions

The large number and percentage of women in this cohort allowed us to examine gender-specific trends for brain cancer, all lymphatic cancers, leukemia, and non-Hodgkin’s lymphoma (Table 9). There was no or only one exposed case for peak and usual exposure models for these outcomes, and therefore we present results only for the cumulative exposure models. Rate ratios for male and female workers were similar for leukemia, higher for females for all lymphatic/hematopoietic cancers and non-Hodgkin’s lymphoma, and higher for males for brain cancer. All gender-specific rate ratios were near or below 1.0. These findings were based on only two exposed female brain cancers and three RF exposure leukemia cases, which led to wide CIs (Table 9).

Table 9
Table 9:
Rate Ratios (RR) and 95% Confidence Intervals (CI): Brain Cancer and Cancers of the Lymphatic and Hematopoietic Systems by Gender for Cumulative Radiofrequency Exposure

The study results were consistent across sensitivity analyses in which we changed our exposure assignments or statistical models. We observed similar findings using Mantel-Haenszel stratified analyses and Cox regression models (data not shown). Our results also indicated that reassigning certain job groups, for which there was a lack of consensus about exposure, to lower RF exposure levels did not appreciably change relative risk estimates. Similarly, using different RF levels to calculate cumulative exposure scores did not substantially change our findings for brain cancers, lymphomas, and leukemias (data not shown).

Discussion

This is the first report of cancer mortality trends among Motorola employees, a large cohort of workers, with employees who have received higher RF exposures than the general public. The lack of elevated mortality risk for brain cancers and all lymphatic/hematopoietic cancers combined suggests that occupational RF exposure, at the frequencies and field levels experienced within this cohort, are not associated with an increased risk for these diseases. Recent reviews of occupational studies have concluded that epidemiologic research has not consistently identified an excess of cancer risk among occupations exposed to low-level RF energy. 1,32,33 Nevertheless, there have been reports of elevated cancer risk in several cohorts.

Our results highlight a pronounced healthy worker effect among this cohort. This observation is common in occupational groups with stable employment and relatively high wages. 43–46 In particular, the low SMRs for cirrhosis of the liver (SMR = 0.46), external causes of death (SMR = 0.52), heart disease (SMR = 0.68), and nonmalignant respiratory disease (SMR = 0.66) suggest a cohort with healthful lifestyles.

For malignant melanoma of the skin, ovarian cancer, and Hodgkin’s disease, SMRs were slightly above 1.0 for the high and moderate RF-exposed groups (Figure 3). The findings for ovarian cancer and Hodgkin’s disease are each based on three exposed cases, thus generating wide CIs and limiting any firm conclusions regarding these outcomes and RF exposure. There is no established occupational or environmental risk factor for ovarian cancer. 47 Most known or suspected risk factors involve genetics, reproductive history, and hormonal factors. 48–50 Risk factors for melanoma include the presence of normal or dysplastic nevi, skin type (white), hair and eye color (blond, blue eyes), and exposure to sunlight. 51–57 All who died of melanoma in this cohort were whites. Previous studies have reported elevations in melanoma and Hodgkin’s disease among potential RF exposed workers or groups (for example, amateur radio operators). 15,21 Given the small number of cases in our study, the modestly elevated SMRs, and the lack of any compelling biologic rationale, we consider the melanoma findings inconclusive.

We had to rely on estimates of relative exposure rather than personal exposure measurements. Exposure assignments were based on the opinion of Motorola and non-Motorola experts. In addition, we have combined all frequencies and types of RF exposure (for example, near field and far field). In our qualitative RF job exposure matrix, it was not possible to separate job titles into historical RF exposure categories based on specific RF frequency ranges. Given the extensive review process and input from the experts, however, we believe that the RF job exposure matrix is reasonably accurate in ranking relative, overall RF exposures. Additionally, we conducted a validation study to evaluate the RF exposure matrix by comparing relative rankings with detailed exposure information collected from a sample of our cohort. This comparison confirmed that our RF job exposure matrix reliably classified job titles into relative RF exposure levels. Although it is difficult to compare the RF exposure levels with other recently studied military and occupational cohorts, 13–15,17,22,32,33 the occupational RF levels among Motorola workers are lower than military and plastics manufacturing workers, in which occupational guidelines have been documented as having been exceeded. 15,22

Unmeasured confounding factors always are a potential concern in epidemiologic studies. Nevertheless, we are not aware of any other risk factors for brain cancers and lymphatic/hematopoietic system cancers strong enough and associated, positively or negatively, with RF exposure that would change our findings and produce elevated risk estimates. Exposure misclassification, if present in this study, is most likely to be nondifferential and would tend to bias the results toward the null value. The impact of exposure misclassification depends on the extent of misclassification and the level of the true relative risk. If we assume that the true relative risk is low (for example, a rate ratio in the range 1.3–1.5), the impact is substantially less than if the true relative risk is higher (for example, rate ratio >3.0). If any risk effect exists, it is likely to be small, and nondifferential exposure misclassification would likely produce only minimal or moderate bias.

On the basis of the range of upper confidence limits that we observed for cancer mortality, our findings generally do not support threefold or higher relative risks for brain cancers, lymphomas, and leukemias due to RF exposure. We did not observe indications of excess relative risk, but we cannot rule out the possibility of potential effects in the range of 1.5–2.0 relative risk. Although the cohort was large, statistical power was limited by the relatively young age, the small portion of the cohort (3.2%) who have died, the rarity of the cancers of interest, and the small percentage of the cohort who were RF exposed. This study is the first examination of brain cancer and leukemia/lymphoma risk in a large occupational cohort of wireless communication manufacturing workers. The majority of the cohort is relatively young, however, and because of the assumed long latency between exposure and cancer, it may be too early to detect a potential RF health effect in this cohort.

Acknowledgment

We thank Elizabeth A. Holly, Jack S. Mandel, Robert J. McCunney, and Patricia A. Buffler for their thoughtful review and commentary. We thank Motorola Engineering, Industrial Hygiene, and Human Resources staffs for their assistance with data collection and exposure assessment, especially Quirino Balzano, Al Davidson, David Gunn, George Kalafayan, Richard Kiesell, Ed Kovacik, and William Wagner for their technical review and guidance. Finally, we thank Steve Bloom and William Kaune for assistance in exposure assessment and Exponent staff members Beverly Smith, Pamela Chapman, Daniel Cher, Heidi Anderson, Regina Dyson, Kathy Dyson, Elizabeth Heilman, Ramzi Mrad, and Brian Davis for their research assistance.

References

1. World Health Organization. Electromagnetic Fields (300 Hz–300 GHz). Geneva: World Health Organization, 1993.
2. Repacholi MH. Low-level exposure to radiofrequency electromagnetic fields: health effects and research needs. Bioelectromagnetics 1998; 19:1–19.
3. Kuster N, Balzano Q. Experimental and numerical dosimetry. In: Kuster N, Balzano Q, Lin J, eds. Mobile Communications Safety. New York: Chapman & Hall, 1997.
4. Michaelson SM, Lin JC. Biological Effects and Health Implications of Radiofrequency Radiation. New York: Plenum Press, 1987.
5. Michaelson S. Biological effects of radiofrequency radiation: concepts and criteria. Health Physics 1991; 61:3–14.
6. Hitchcock R, Patterson R. Radio-Frequency and ELF Electromagnetic Energies. New York: Van Nostrand Reinhold, 1995.
7. Dolk H, Shaddick G, Walls P, Grundy C, Thakrar B, Kleinschmidt I, Elliott P. Cancer incidence near radio and television transmitters in Great Britain. I. Sutton Coldfield transmitter. Am J Epidemiol 1997; 145:1–9.
8. Hocking B, Gordon IR, Grain HL, Hatfield GE. Cancer incidence and mortality and proximity to TV towers. Med J Aust 1996; 165:601–605.
9. Maskarinec G, Cooper J. Investigation of a childhood leukemia cluster near low-frequency towers in Hawaii. Proceedings of the 26th annual meeting of the Society for Epidemiologic Research, Keystone, Colorado, June 16–18, 1993. Am J Epidemiol 1993;666.
10. Dolk H, Elliott P, Shaddick G, Walls P, Thakrar B. Cancer incidence near radio and television transmitters in Great Britain. II. All high power transmitters. Am J Epidemiol 1997; 145:10–17.
11. Selvin S, Schulman J, Merrill DW. Distance and risk measures for the analysis of spatial data: a study of childhood cancers. Soc Sci Med 1992; 34:769–777.
12. Knox EG, Gilman EA. Hazard proximities of childhood cancers in Great Britain from 1953–1980. J Epidemiol Community Health 1997; 51:151–159.
13. Robinette CD, Silverman C, Jablon S. Effects upon health of occupational exposure to microwave radiation (radar). Am J Epidemiol 1980; 112:39–53.
14. Grayson JK. Radiation exposure, socioeconomic status, and brain tumor risk in the US Air Force: a nested case-control study. Am J Epidemiol 1996; 143:480–486.
15. Szmigielski S. Cancer morbidity in subjects occupationally exposed to high frequency (radiofrequency and microwave) electromagnetic radiation. Sci Total Environ 1996; 180:9–17.
16. Lilienfeld A, Tonascia J, Tonascia S, Libauer C, Cauthern GM. Evaluation of Health Status of Foreign Service and Other Employees from Selected Eastern European Posts: Final Report. Pub. No. NTIS PB-288 163. Washington DC: Department of State, 1978.
17. Armstrong B, Theriault G, Guenel P, Deadman J, Goldberg M, Heroux P. Association between exposure to pulsed electromagnetic fields and cancer in electric utility workers in Quebec, Canada, and France. Am J Epidemiol 1994; 140:805–820.
18. McDowall M. Leukemia mortality in electrical workers in England and Wales. Lancet 1983; 1:246.
19. Pearce N, Sheppard R, Howard J, Fraser J, Lilley B. Leukaemia in electrical workers in New Zealand. Lancet 1985; 1:811–812.
20. Tynes T, Hannevik M, Andersen A, Vistnes AI, Haldorsen T. Incidence of breast cancer in Norwegian female radio and telegraph operators. Cancer Causes Control 1996; 7:197–204.
21. Vagero D, Ahlbom A, Olin R, Sahlsten S. Cancer morbidity among workers in the telecommunications industry. Br J Ind Med 1985; 42:191–195.
22. Lagorio S, Rossi S, Vecchia P, De Santis M, Bastianini L, Fusilli M, Ferrucci A, Desideri E, Comba P. Mortality of plastic-ware workers exposed to radiofrequencies. Bioelectromagnetics 1997; 18:418–421.
23. Lin R, Dischinger P, Conde J, Farrell K. Occupational exposure to electromagnetic fields and the occurrence of brain tumors. J Occup Med 1985; 27:413–419.
24. Preston-Martin S, Mack W, Henderson B. Risk factors for gliomas and meningiomas in males in Los Angeles County. Cancer Res 1989; 49:6137–6143.
25. Speers M, Dobbins J, Mill V. Occupational exposures and brain cancer mortality: a preliminary study of East Texas residents. Am J Ind Med 1988; 13:629–638.
26. Gallagher R, McBride M, Band P, Spinelli J, Threlfall W, Tamaro S. Brain cancer and exposure to electromagnetic fields. J Occup Med 1991; 33:944–945.
27. Milham S. Increased mortality in amateur radio operators due to lymphatic and hematopoietic malignancies. Am J Epidemiol 1988; 127:50–54.
28. Robinson C, Lalich N, Burnett C, Sestito J, Frazier T, Fine L. Electromagnetic field exposure and leukemia mortality in the United States. J Occup Med 1991; 33:160–162.
29. Calle E, Savitz D. Leukemia in occupational groups with presumed exposure to electrical and magnetic fields. N Engl J Med 1985; 313:1476–1477.
30. Vagero D, Swerdlow A, Beral V. Occupation and malignant melanoma: a study based on cancer registration data in England and Wales and in Sweden. Br J Ind Med 1990; 47:317–324.
31. Rothman K, Loughlin J, Funch D, Dreyer N. Overall mortality of cellular telephone customers. Epidemiology 1996; 7:303–305.
32. Bergqvist U. Review of epidemiological studies. In: Kuster N, Balzano Q, Lin J, eds. Mobile Communications Safety. New York: Chapman & Hall, 1997.
33. Elwood J. A critical review of epidemiologic studies of radiofrequency exposure and human cancers. Environ Health Perspect 1999; 107suppl:155–168.
34. Schnorr T, Steenland K. Identifying deaths before 1979 using the Social Security Administration death master file. Epidemiology 1997; 8:321–323.
35. Marsh G, Preininger M. OCMAP: a user-oriented occupational cohort mortality analysis program. Am Stat Soc 1980; 34:245–246.
36. SAS Institute. SAS/STAT software. Cary, NC: SAS Institute, 1996.
37. Williams DR. Race and health: basic questions, emerging directions. Ann Epidemiol 1997; 7:322–333.
38. Kaufman J, Cooper R, McGee D. Socioeconomic status and health in blacks and whites: the problem of residual confounding and the resiliency of race. Epidemiology 1997; 8:621–628.
39. van Loon AJ, Brug J, Goldbohm RA, van den Brandt PA, Burg J. Differences in cancer incidence and mortality among socio-economic groups. Scand J Soc Med 1995; 23:110–120.
40. Cox D. Regression models and life tables. J R Stat Soc Ser B 1972; 34:187–202.
41. Callas P, Pastides H, Hosmer D. Empirical comparisons of proportional hazards, Poisson, and logistic regression modeling of occupational cohort data. Am J Ind Med 1998; 33:33–47.
42. Greenland S. Modeling and variable selection in epidemiologic analysis. Am J Public Health 1989; 79:340–349.
43. Wen C, Tsai S. Anatomy of the healthy worker effect: a critique of summary statistics employed in occupational epidemiology. Scand J Work Environ Health 1982; 8:48–52.
44. Arrighi H, Hertz-Picciotto I. The evolving concept of the healthy worker survivor effect. Epidemiology 1994; 5:189–196.
45. Checkoway H, Pearce N, Crawford-Brown D. Research Methods in Occupational Epidemiology. New York: Oxford University Press, 1989.
46. Howe G, Chiarelli A, Lindsey J. Components and modifiers of the health worker effect: evidence for three occupational cohorts and implications for industrial compensation. Am J Epidemiol 1998; 128:1364–1375.
47. Hartge P, Stewart P. Occupation and ovarian cancer: a case-control study in the Washington, DC Metropolitan Area, 1978–1981. J Occup Med 1994; 36:924–927.
48. Whittemore AS, Harris R, Itnyre J. Characteristics relating to ovarian cancer risk: collaborative analysis of 12 U.S. case-control studies. IV. The pathogenesis of epithelial ovarian cancer. Collaborative Ovarian Cancer Group. Am J Epidemiol 1992; 136:1212–1220.
49. Whittemore AS, Harris R, Itnyre J. Characteristics relating to ovarian cancer risk: collaborative analysis of 12 U.S. case-control studies. II. Invasive epithelial ovarian cancers in white women. Collaborative Ovarian Cancer Group. Am J Epidemiol 1992; 136:1184–1203.
50. Whittemore AS, Harris R, Itnyre J, Halpern J. Characteristics relating to ovarian cancer risk: collaborative analysis of 12 US case-control studies. I. Methods. Collaborative Ovarian Cancer Group. Am J Epidemiol 1992; 136:1175–1183.
51. Elwood J, Jopson J. Melanoma and sun exposure: an overview of published studies. Int J Cancer 1997; 73:198–203.
52. Garbe C, Buttner P, Weiss J, Soyer HP, Stocker U, Kruger S, Roser M, Weckbecker J, Panizzon R, Bahmer F, Tilgen W, Guggenmoos-Holzman I, Orfanos CE. Risk factors for developing cutaneous melanoma and criteria for identifying persons at risk: multicenter case-control study of the Central Malignant Melanoma Registry of the German Dermatological Society. J Invest Dermatol 1994; 102:695–699.
53. Lee J. Factors associated with melanoma incidence and prognosis. Semin Surg Oncol 1996; 12:379–385.
54. Langley R, Sober A. A clinical review of the evidence for the role of ultraviolet radiation in the etiology of cutaneous melanoma. Cancer Invest 1997; 15:561–567.
55. Katsambas A, Nicolaidou E. Cutaneous malignant melanoma and sun exposure: recent developments in epidemiology. Arch Dermatol 1996; 132:444–450.
56. Herzfeld P, Fitzgerald E, Hwang S, Stark A. A case-control study of malignant melanoma of the trunk among white males in upstate New York. Cancer Detect Prev 1993; 17:601–608.
57. Weiss J, Bertz J, Jung E. Malignant melanoma in southern Germany: different predictive values of risk factors for melanoma subtypes. Dermatologica 1991; 183:109–113.
Keywords:

brain neoplasms; healthy worker effect; leukemia; lymphoma; occupational health; radiofrequency exposure; neoplasms

© 2000 Lippincott Williams & Wilkins, Inc.