A Critical Review of Epidemiologic Studies of Radiofrequency Exposureand Human CancersJ. Mark ElwoodDepartment of Preventive and Social Medicine, University of Otago, Dunedin, New Zealand

This paper reviews studies that have assessed associations between likely exposure toradiofrequency (RF) transmissions and various types of human cancer. These studies includethree cluster investigations and five studies relating to general populations; all of these studiesconsider place of residence at the time of cancer diagnosis in regard to proximity to radio ortelevision transmitters. There are also five relevant occupational cohort studies and severalcase-control studies of particular types of cancer. These studies assessed a large number ofpossible associations. Several positive associations suggesting an increased risk of some types ofcancer in those who may have had greater exposure to RF emissions have been reported.However, the results are inconsistent: there is no type of cancer that has been consistentlyassociated with RF exposures. The epidemiologic evidence falls short of the strength andconsistency of evidence that is required to come to a reasonable conclusion that RF emissionsare a likely cause of one or more types of human cancer. The evidence is weak in regard to itsinconsistency, the design of the studies, the lack of detail on actual exposures, and the limitationsof the studies in their ability to deal with other likely relevant factors. In some studies there maybe biases in the data used. - Environ Health Perspect 107(Suppl 1):155-168 (1999).http.//ehpnetl.niehs.nih.gov/docs/1999/Suppl-1/155-168elwood/abstract.html

Key words: electromagnetic radiation, radiofrequency radiation, cancer, epidemiology,leukemia

Radiofrequency radiation (RFR) occupiesthe range from 100 kHz to 300 GHz in theelectromagnetic spectrum. RFR is a higherfrequency (shorter wavelength) than theextremely low frequency (ELF) radiationused in electric power sources, and a lowerfrequency than infrared radiation. Thisrange is used for radiofrequency (RF) com-munications and microwave sources, includ-ing in approximate order by increasingfrequency, amplitude modulation (AM)radio, frequency modulation (FM) radio,very high frequency (VHF) radio and televi-sion (TV), ultrahigh frequency (UHF) TVand cellular telephone transmissions, andmicrowave ovens, radar, and satellite com-munications. Natural RFR is negligible, so

Manuscript received at EHP 17 July 1998; accepted 2October 1998.

Address correspondence to J.M. Elwood,Department of Preventive and Social Medicine,University of Otago, Hugh Adam Cancer Epide-miology Unit, PO Box 913, Dunedin, New Zealand.Telephone: 64 3 479 7220. Fax: 64 3 479 7164.E-mail: melwood@gandalf.otago.ac.nz

Abbreviations used: ALL, acute lymphoblastic lym-phoma; ELF, extremely low frequency; EMF, electro-magnetic field; FM, frequency modulation; ICNIRP,International Commission on Non-ionizing RadiationProtection; LGAs, local government areas; OR, oddsratio; PEMFs, pulsed electromagnetic fields; RF,radiofrequency; RFR, radiofrequency radiation; RR,relative risk; SAR, specific energy absorption rate; TV,television; UHF, ultrahigh frequency; VHF, very highfrequency.

population exposure is a phenomenon ofthe current century, and has increasedgreatly in recent years. Public concerntends to focus on new types of emitters,and there have been public concerns aboutradio and TV transmitters, TV receivers,and microwave ovens coincident with theirdevelopment and utilization.

Recently in several countries there hasbeen considerable public concern and legalproceedings about cellular telephone sys-tems (cell phones) in regard to potentialrisks both to users of cell phones and frompopulation exposure to cell phone transmit-ters. Although cell phone users have a muchhigher potential dose exposure because thedevice is held close to the head (1,2), publicconcern has often been greater regardingcell phone transmitters, where although thepotential dose exposure is much less, theexposure is seen as involuntary; health con-cerns may also be raised as a support foraesthetic and other objections.

International guidelines for RF expo-sures have recently been revised by theInternational Commission on Non-Ionizing Radiation Protection (ICNIRP)(3). These guidelines for body exposurewere set on the basis of avoiding thermaleffects. The basic restriction for whole-bodyexposure is a specific energy absorption rate(SAR) of 0.4 W/kg for occupational expo-sure and 0.08 W/kg for general population

exposure. The ICNIRP review of humanand animal data for RF ranges shows thatthe threshold for irreversible effects in eventhe most sensitive tissues is > 4 W/kg undernormal environmental conditions (3). As aresult, the occupational exposure restrictionis based on a safety factor of 10, and thegeneral population basic restriction on afurther reduction factor of 5, resulting in0.08 W/kg. These reference levels in equiv-alent power densities are 200 pW/cm2 at 10to 400 MHz; fl2 gW/cm2 at 400 to 2000MHz, where f= frequency in MHz; and1000 PW/cm2 at 2 to 300 GHz. Somenational standards hold to 200 gW/cm2throughout this range. Power outputs fromcell phone transmitters are low and expo-sure levels decline with the square of dis-tance from the source, so that even atdistances as short as 30 m, exposure levelsare likely to be < 5% of the public exposurelimit. Thus, interest in potential humaneffects from cell phone transmitter expo-sures depends on the existence of relevantbiologic effects at levels much below thoseproducing thermal effects-so-called ather-mal effects. Higher limits were set for expo-sure of smaller body parts: the occupationallocalized SAR limit for exposure of the headwas set at 10 W/kg, averaged over any 10-gmass of tissue and over any 6-min period;the general public exposure limit for thehead is 2 W/kg averaged similarly. Theselevels are relevant to cell phone users.

The ICNIRP report (3), which alsoreviewed several other expert reports(1,4-9), concluded that exposure to thesefields is therefore unlikely to initiate car-cinogenesis. These expert reports notemany negative results from in vitro studieson DNA damage, mutation frequency, andchromosome aberration frequency. Inaddition there are data suggesting biologiceffects, some of which are potentially rele-vant to cancer causation, at low exposurelevels. Strand breaks in DNA in rodent tis-sues have been described at SAR levelsaround 1 W/kg (10-12), although themethodology of these studies has beenquestioned (13). Excesses of malignancieshave been noted in rats exposed tomicrowaves (14). Some studies suggestedpromotion effects on preinitiated cells (15)

Environmental Health Perspectives * Vol 107, Supplement l * February 1999 155

J.M. ELWOOD

but did not clearly exclude thermal effects,whereas other studies at athermal levelsreported no effects (16,17). An importantrecent study showed an increase (RR = 2.4,95% confidence limits 1.3-4.5) in lym-phoma incidence in genetically pre-disposed mice compared to controls atdosage levels of pulsed 900 MHz fields,which for most animals would have beenbelow thermal levels (18).

There is evidence of effects of athermallevels of RFs on calcium release or surfacebinding in cells (19,20)-with some nega-tive reports (21)-and on brain electricalactivity (19), T-lymphocyte cytotoxicactivity (22), changes in non-cyclic-AMP-dependent kinases (23), and on orthoninedecarboxylase activity (24,25), althoughthere are also many negative reports (3).Increased neoplastic transformation in cellsalso treated with a chemical promoter hasbeen noted (26).

Thus there are various experimentalresults that are consistent with biologiceffects, some of which might be related tocarcinogenic mechanisms, of RFs atstrengths below those that produce thermaleffects. However, the results are inconsistentand no clear mechanism has been shownconsistently in a variety of cell systems andanimals. As a result the consideration of theepidemiologic literature relating RFs tocancer occurrence in humans must be bal-anced against this uncertain background:There is no clear evidence for a carcino-genic action relevant to intact humans, butsimilarly, it is difficult to argue that acarcinogenic mechanism can be ruled out.

Neither the ICNIRP (3) nor any earliermajor reports discussed epidemiologicstudies in any detail. The most interestingepidemiologic studies of general popula-tion RF exposure have been publishedrecently and were not included in the epi-demiologic review by Bergqvist (27).Therefore, although the ICNIRP and ear-lier reports concluded that there was nogood evidence for an epidemiologic associ-ation, continuing public concern and therecent epidemiologic studies stimulated thecurrent review.

MethodsPublished reports on studies of RFs andcancer in humans were identified by a liter-ature search using Medline (28) to searchfrom 1988 to June 1998, by searching ref-erences given by the major reviews(1,3-9), and by other studies and reviews.The studies used were limited to studies ofindividuals or communities classified aslikely to have been exposed to RF emis-sions; studies of ELF radiation or electro-magnetic fields (EMFs) in general, wherethe main exposure was likely to be to ELF,were not included. Nonhuman studies,laboratory studies, individual case reports,and studies without any comparison groupwere not considered.

Further calculations were needed toobtain results from Robinette et al. (29) ina format similar to that of the others: 95%confidence limits of the given observed toexpected ratios were calculated, and a trendtest was applied using the method ofBreslow and Day (30).

ResultsThe epidemiologic studies dealing with RFemissions and human cancer fall into fourgroups: studies of clusters of cases; studiesof general populations exposed to TV,radio, and similar emissions; studies ofoccupational groups with exposures tosuch emissions; and case-control studies.

Cauer StudiesA cluster study is based on the unplannedobservation of an apparent excess numberof cases of a disease in a particular placeand time period. The occurrence of clus-ters of any uncommon disease is an

expected phenomenon due to chance vari-ation and small numbers. There is no sta-tistical or other method to determine,once a cluster has occurred, whether ornot it is due to chance; investigations ofclusters are often unhelpful, and are oftendone only as a public relations effort to

allay community concern (31). The appro-priate approach to a cluster is to treat it as

raising a new hypothesis to be tested inindependent data.

Three cluster investigations of cancerand RF radiation have been reported inpeer-reviewed literature (Table 1).A cluster of 12 cases of acute leukemia

in children in Hawaii led to a case-controlstudy of a slightly larger number of cases

(32), which showed an excess among thoseliving within 2.6 miles of radio towers. Thiswas not significant (odds ratio [OR] = 2.0,95% confidence limits 0.06-8.3). Theauthors conduded that "the clustering may

Table 1. Radiofrequency emissions and cancers: cluster studies.

Cluster studiesHawaii (32) U.S. policemen (33) Sutton Coldfield (34)

Study characteristicExposure Residence within 2.6 miles Use of hand-held radar gun Residence close to TV or

of radio towers radio transmittersAscertainment Residence Self-report ResidenceExposed group General population U.S. policemen General populationStudy type Case-control: 14 cases Cohort: 6 cases Small area incidence analysisType of cancer Acute leukemia, children Testicular VariousCancer type/OR/test for trend

All leukemia 1.83 (1.22-2.74)ab,* 0.001 a,*Acute leukemia 2.0 (0.06-8.30)c 1.86 (0.89-3.42)ab 0.004a*Acute myeloid leukemia 1.02 (0.26-2.60)ab 0.045a8*Acute lymphatic leukemia 3.57 (0.74-10.4)ab NSaChronic myeloid leukemia 1.23(0.15-4.43)ab NSaChronic lymphatic leukemia - 2.56(1.11- 5.05)ab* 0.007a*Non-Hodgkin lymphoma 0.66 (0.28-1.30)ab NSaMultiple myeloma 1.54 (0.74-2.83)ab NSaTestis Only shared risk factoraAll cancer 1.09 (1.01-1.17)ab.* NSa

- iviluuiit,,c,l, nedi, rersp,ecves * vo IO/, zuppiement *aFebruary 1999

imo, nonsigniiicant, p>u.uU; -, no intormation was given in the study; OR, odds ratio. 'Adult only. bOR within 2 km. cChild only. *Statistically significant, p< 0.05.

156 Fnvirnnm_ntAl "i-AIth P -renmr-ixine - X1l I n-7 1. .

RFAND HUMAN CANCERS: EPIDEMIOLOGY REVIEW

have been a chance event, but because ofits particular characteristics, we feel itshould be noted" (32).

Another cluster investigation followedthe observation of what appeared to be ahigh number of cancers of the testis (sixcases) among 340 U.S. policemen whoused radar guns and often kept them ontheir lap in their patrol car (33). Thisstudy showed a positive association, butbecause it was based on a cluster situationit cannot be interpreted further.

An appropriate type of cluster investi-gation is the study of a suspected cluster ofleukemias and lymphomas in adults livingclose to the Sutton Coldfield TV and radiotransmitter near Birmingham, UK (34).The authors used data over a 12-yearperiod to compare the residential postcodeof patients with cancer and the censusnumber of residents in that postcode area(allowing adjustment for age, gender,regional variations within the country, andan index of socioeconomic level).

For the types of cancer suspected onthe initial cluster, there was an excess oftotal adult leukemia within 2 km, with therisk declining from there out to the edgeof a 10-km circle. Similar results were seenfor some subtypes of leukemia. Part of thetrend of decreasing risk from distancefrom the transmitter was because theobserved number of cases in people livingclose to the 10-km boundary was less thanexpected. For lymphomas-also part of theinitial cluster-there was an excess riskwithin the 10-km circle, but the risk wasless in those within the inner 2-km circle.The authors appropriately concluded that"no causal implications regarding radio andTV transmitters can be drawn from thisfinding, based as it is on a single clusterinvestigation" (34).

Studies ofGeneral PopultionsExXosed to Radio andTransmitterT evision TransmissionsFive studies look at general populationgroups living close to RF transmitters(Table 2).

Sutton Coldfield Study. In the furtheranalysis of the Sutton Coldfield study, sev-eral other types of cancer in adults wereassessed-types that were not part of theinitial cluster (33). Of 11 types of cancerassessed, two showed a significant trend inincidence with distance from the trans-mitter (melanoma and bladder cancer).Although not affected by the dustering bias,the difficulty here is one of multiple testing;some positive findings will arise by chance.

These associations are new observations notbased on a prior hypothesis; therefore theyrequire further assessment.

The results for childhood cancer canalso be assessed because the original clusterwas related to adult cancer. There was nosignificant trend with distance for allcancer or for leukemia in children.

Study of Twenty-One U.K. Trans-mitters. Dolk et al. (35) then identified all21 radio and TV transmitters in Britainwith transmission power of over 500 kWfor TV or 250 kW for FM radio, includingSutton Coldfield. They again assessed aninner circle of 2 km, an outer circle of 10km of residence, and the trend in risk withdistance from the transmitter up to 10 km.They used data for up to 12 years, whichincluded 3300 adult leukemia cases andwere based on a total population livingwithin 10 km of any transmitter of 3.39million people. The associations suggestedby the Sutton Coldfield study (with adultleukemia, skin melanoma, and bladdercancer) were tested by a new analysis basedonly on all the other transmitters, excludingSutton Coldfield. For childhood cancer,where there was no significant excess inSutton Coldfield, the data from all 21transmitters were combined.

The most important aspect of the studywas to assess if the increase in adultleukemia seen near the Sutton Coldfieldtransmitter was also seen near the othertransmitters. That was not shown. Thetotal number of cases living within theinner 2-km circle from the other 20 trans-mitters was 3% less than expected, whereasthe number living within the overall10-km circle was 3% more than expected.The detailed results for all adult leukemiasare complex, however. Close to the trans-mitter the observed-to-expected ratio was< 1, although it was based on a smallnumber of cases. The ratio rose to 1.15 atdistances of 2 to 3 km, then graduallydecreased to become close to 1 at dis-tances beyond 7.5 km. Because there werefew cases close to the transmitter, the sta-tistical test used shows a weak decline inrisk due to the risk falling from distancesof 2 to 3 km out to the edge of the circle,with borderline significance (p= 0.05).

These results do not confirm thefindings in Sutton Coldfield. The result iswhat we would expect if the SuttonColdfield results were influenced by thefact that that study was based on an initialcluster of cases close to the transmitter.Similarly, for most subtypes of leukemiaand lymphoma, the risk in those living

closest to the transmitter (within 2 km)was less than expected. Only one type,chronic lymphatic leukemia, showed a pat-tern of some increased risk closer to thetransmitter, but this was nonsignificant.

Results for individual transmittersshowed significant declines in risk of adultleukemia with increasing distance for threetransmitters; results from two of thesetransmitters showed that the risks at theedge of the 10-km circle were less thanexpected. The overall declining trend withdistance was mainly due to the data for theCrystal Palace transmitter near London; ithad most of the observed cases because ofthe high population. The Crystal Palacetransmitter is a high-powered TV transmit-ter (> 870 kW) without FM transmission;the other two transmitters showing trendsin adult leukemia were combined TVtransmitters (500 and 287 kW) plus FMtransmissions of 250 kW. The SuttonColdfield transmitter transmits both TVand FM frequencies. The authors groupedtransmitters by whether they carried TV orFM emissions, and by power, but no cleardifferences emerged (35).

The authors suggest three possibleexplanations of the trend of reducing riskfrom 2 to 3 km out to 10 km from thetransmitters. It may be due to chance or to"other unmeasured sociodemographic orenvironmental factors" (35). Alternatively,if reduced risk is causally related to thetransmissions, the usual exposure model onwhich exposure declines with the square ofdistance would not explain the results; theauthors state: "if there were a true associa-tion with radio transmission, the lack ofreplication of the pattern and magnitude ofexcesses near Sutton Coldfield may indi-cate that a simple radial decline exposuremodel is not sufficient" (35). The authorsconclude that in general their study is neg-ative, and "at most, gives no more thanvery weak support to the Sutton Coldfieldfindings" (35).

Of the two other cancer types related todistance in the Sutton Coldfield study, theresults for the other 20 transmittersshowed no trend with distance for bladdercancer, and only a small irregular variationin melanoma. Leukemias and brain can-cers in childhood were also examinedusing data from all 21 sites. There was nosigniflcant increased risk in either diseasewithin the inner 2-km circle or in the over-all 10-km circle, and no regular variationwith distance.

Study in Sydney, Australia. Thisstudy (36) assessed cancer incidence and

Environmental Health Perspectives * Vol 107, Supplement t * February 1999 157

J.M. ELWOOD

Table 2. Radiofrequency emissions and cancers: general population studies.

General population studiesSutton Coldfheld (34) U.K. 21 transmitters (35) Sydney 1 (36,37) Sydney 2 (40) Sydney 2 (40) San Francisco (41)

Study characteristicExposure TV, radio TV, radio TV TV TV TV, radioAscertainment Residence Residence Residence Residence Residence ResidenceFrequency, MHz 30-1000 30-1000 60-215 60-215 60-215

All areas ExcludingLane Cove

Cancer type/risk/test for trendAdult

All leukemia - 0.97 (0.78-1.21)a 0.052 1.18 (0.98-1.42)b - -Acute leukemia - 0.94 (0.67-1.31 )a NSAcute myeloid leukemia - 0.77 (0.50-1.19)a NSAcute lymphatic leukemia - 0.90 (0.39-2.11()a NS 1.32 (1.09-1 .59)b*Chronic myeloid leukemia - 0.63 (0.30-1.29)( NS 1.09 (0.91-1.32)bChronic lymphatic leukemia - 1.20 (0.83-1.46)a NSOther leukemia - - 1.67 (1.12-2.49)b.*

AdultMultiple myeloma 1.54 (0.74-2.83)a NSBrain cancers 1.31 (0.75-2.29)a NS 0.89(0.71-.11)b -Melanoma skin 1.43 (0.83-2.44)a 0.018* 1.11 (0.84-1.46)a NSBladder 1.52 (1.13-2.04)a.* 0.04* 1.08 (0.94-1 .24)a NSMelanoma eye 0 cases NS -

Male breast 1.64 (0.04-9.13)a NS -

Female breast 1.08 (0.90-1.31)a NS -

Lung 1.01 (0.84-1.21)a NS --Colorectal 1.13 (0.94-1.35)a NS - -Stomach 0.75 (0.54-1.06)a NS -

Prostate 1.13 (0.82-1.55)a NS -

ChildAll leukemia NS 1.12 (0.61-2.06)a NS 1.58 (1.07-2.34)b.* 1.38 (0.99-1.91)c 0.90 (0.56-1.44)C 0.73 NSdLymphatic leukemia - - 1.55 (1.00-2.41)b* -

Myeloid leukemia - 1.73 (0.62-4.81 )bOther leukemia - 1.65 (0.33-8.1 9)bAll cancer 1.8e NS -

Brain - 0.50 (0.10-1.46)8 NS 1.10 (0.59-2.06)b 1.16 NSdHodgkin - - 1.23 NSdNon-Hodgkin lymphoma - 1.03 NSdLymphoma and leukemia - - 0.89 NSdAcute lymphatic leukemia -- - 1.45 (0.96-2.19)b 0.83 (0.45-1.55)c -No information was given in the study; NS, not significant; RR, relative risk. 'OR within 2 km. bRR: high to low exposure areas. ORR: regression, for 1 pW/cm2 exposure

change. dRR for residence within 3.6 km of tower. *Confidence limits are not given. *Statistically significant, p< 0.05.

mortality around three TV transmitters innorthern Sydney. There was no suggestionof a cluster of disease. The three towersemit TV transmissions in the range of 60to 215 MHz with one additional channelat approximately 500 MHz. The maxi-mum power density was estimated as8 jsW/cm2 at approximately 1 km from thetowers, reducing to 0.2 gW/cm2 at 4 km;estimates at 8 km were approximately0.05 pW/cm2 and were substantially less at12 km. The analysis compared an innerarea within 4 km with an outer area thatextended from approximately 4 km to 15km away. Incidence and mortality datafrom leukemia in total, by subtype, andfrom brain tumors were collected from1972 to 1990. These data were subdividedby gender and by broad age groups.

For adult cancers (15+ years of age)(37) there was a small nonsignificant

increase in the incidence of total adultleukemia (RR= 1.18, 95% confidence lim-its 0.98-1.42) and no significant increasein leukemia mortality.

There was an increased incidence ofchildhood leukemia (RR= 1.58, 95% con-fidence limits 1.1-2.3, based on 134 cases)(36). This excess was seen separately forlymphatic, myeloid, and other leukemias inchildhood, although the latter two are basedon small numbers and are not significant.The mortality rates for leukemia in childrenwere increased, more so than for incidence(RR= 2.3, limits 1.4-4.0). There was noexcess of brain tumors in adults or children.

Hocking et al. (36) comment that con-trol for other factors was limited. Theyclaim that the socioeconomic distributionof the inner and outer areas is generallysimilar. The broad age range used raises thepossibility of age confounding; a younger

age distribution within the group 0 to 15years of age would confer a higher leu-kemia risk. They point out that the innerarea has higher traffic density, which couldbe related to increased benzene exposure;the inner area also has higher populationdensity, but they note there is a hypothesisthat leukemia in children might beincreased in sparsely populated areas intowhich many people have recently moved(38). Hocking et al. (36) suggest that theincrease could be related to the modula-tions at 50 Hz to 5 MHz as well as to thecarrier wave.

The Hocking et al. (36) study is con-siderably weaker than the British studies(34,35): It deals only with one area. Itmakes only one comparison-betweenareas defined at one arbitrary distance fromthe transmitter-rather than exploring theoverall relationship with place of residence.

Environmental Health Perspectives * Vol 107, Supplement 1 * February 1999158

RF AND HUMAN CANCERS: EPIDEMIOLOGY REVIEW

There is little control for other potentiallyrelevant factors. The results, showinga stronger association for childhoodleukemia than for adult leukemia, contrastwith the British data, which show no dearexcess in children. The authors conclude"more detailed studies (e.g., relating casesto power density contours) are required toreplicate any association and to look fordose-response relationships before anyconclusions can be drawn" (36).

Comparisons between U.K. andSydney Studies. The authors of theSydney study (37) have commented thattheir study found an excess of childhoodleukemia, which was not found in the UKstudy. Hocking et al. (37) point out thatthe UK transmitters were 3 to 10 timesmore powerful than those used in Australiabut generally used UHF (430-890 MHz),whereas the Sydney transmitters used VHF(63-217 MHz). Although it would beexpected that any effect should be greater inthe UK, Hocking et al. (37) hypothesize awindow effect, where biologic effects mightshow a complex relationship with strengthof exposure.

In reply, the principal author of theUK study notes that "our results aroundmultiple transmitters were more equivocalthan is reflected in the letter by Hockinget al." (39). She emphasizes that the UKstudy found no excess of leukemia inthose living closest to the transmitters,and although they found some decline inincidence with increasing distance, theycould not distinguish whether this wasassociated with either TV or FM trans-mission. They found a weak trend nearthe Crystal Palace transmitter, which doesnot transmit FM, as well as near FMtransmitters. In regard to the windoweffect, Dolk (39) queries the suggestion ofHocking et al. (37) and comments that"when assessing the evidence from justtwo studies, it is easy to make the theoriesfit the data post hoc and very prematureto conclude that the two studies 'suggest'such effects" (39).

Further Analysis ofLeukemia Ratesin North Sydney in Children and LikelyExposure to Radiofrequency Emissionsfrom Transmitters. A further analysis ofthe data from North Sydney has beenpublished (40). McKenzie et al. (40)point out that Hocking et al. (36) com-pared an inner group of three aggregatedlocal government areas (LGAs) close to thetransmission tower with an outer ring ofaggregated LGAs. Hocking et al. (36)excluded on socioeconomic grounds some

other LGAs also close to the towers,although they stated that there was littleevidence of a relationship between socioe-conomic levels and acute lymphoblasticleukemia (ALL) in New South Wales,Australia. Therefore, the new study exam-ined 16 LGAs: 3 closest to the towers(Willoughby, Lane Cove, and NorthSydney), as used by Hocking et al. (36), 7in North Sydney, and 6 in other areas ofSydney; the distances of these LGAs arecomparable to the outer ring used in theHocking et al. study. This study and thestudy of Hocking and his co-workers useddata collected from 1972 to 1990.

McKenzie et al. (40) calculated the sig-nal strengths by a formula dependent onthe distance from the tower and the angleof depression from the horizontal, and alsocarried out site measurements using aHoladay Instruments broad band isotropicfield strength meter. The calculated signalstrengths showed reasonable correlation forfree space conditions at roof height, withother sites subject to shadowing havinglower observed strengths, and two pointsnear the base of the antennae havinggreater emissions than calculated. AverageRFR exposure levels for the LGAs are pre-sented based on calculated signal strengthsfor a random sample of 20 residences inthe high signal strength areas, and at otherdistances by an average of fewer measure-ments. Signal strengths in the three closestLGAs were 1.46 iW/cm2 in Lane Cove,1.40 pW/cm2 in Willoughby, and 0.66PW/cm2 in North Sydney. Most otherareas showed average strengths of < 0.25pW/cm2, apart from Hunters Hill, whichshowed 0.46 jW/cm2. Maximum levels instreets immediately below the antennaewere up to 100 1W/cm2.

The incidence rate ofALL was highestin Lane Cove, a high exposure area, but therates in North Sydney and Willoughby, theother two high exposure areas, were similarto those for the less-exposed areas. Theregression trend for ALL incidence againstestimated RFR exposure based on all LGAsgave a relative risk of 1.45 (95% confi-dence limits 0.96-2.19) for a change inRFR exposure of 1 jiW/cm2; that is, analmost significant positive association. Fortotal childhood leukemia the result wassimilar, with a relative risk of 1.38 (95%confidence limits 0.99-1.91). However,the exclusion of Lane Cove removed thistrend, giving relative risks of 0.83 (95%confidence limits 0.45-1.55) for ALL and0.90 (95% confidence limits 0.56-1.44)for total leukemia.

McKenzie et al. (40) concluded thatleukemia rates vary in different localities inSydney, and in particular, the leukemia inci-dence in Lane Cove was abnormally highduring the years studied. This is unlikely tobe due to RFR exposure because the othertwo areas with high RFR exposures,Willoughby and North Sydney, did notshow any increase in leukemia incidence.The relationship reported by Hocking et al.(36) depends on this excess in Lane Cove.The excess in Lane Cove must be due eitherto chance or to some factor apart from RFradiation. For the excess to be related toRFR exposure, actual exposures of childrenwho live in Lane Cove would have to bemuch higher than those estimated; corre-spondingly, actual exposures in Willoughbywould have to be much lower than thosecalculated. The authors suggest that furtherstudies are needed to examine the excessrates in Lane Cove (40).

The second North Sydney study (40) isan improvement on the first North Sydneystudy by Hocking and colleagues (36)because the extra detail gives greater abilityto assess a regular association betweenleukemia occurrence and estimated RFexposures. The estimation of exposure isstill based primarily on distance, althoughthe measurements performed confirm thatthe three areas closest to the transmittershave higher exposure levels when measuredat a few open air sites than do the otherareas. The second study (40) still shows anassociation between leukemia incidenceand estimated RFR exposure if all areas areincluded. However, of three areas withhigh RFR exposure levels, only one ofthese three has an elevated risk. This makesit more likely that some factor other thanRFR is responsible for the excess.

San Francisco Study. Another study ofgeneral population exposures assessedcancers in those younger than 21 years ofage in San Francisco between 1973 and1988 (41). This study assessed leukemia,lymphatic cancer, and brain cancer byexamining the distribution of cases interms of residence in relationship to theSutro Tower, a large tower that carries TVand radio transmitters. No information onemission levels is included. Adjustmentsfor population density in small areas weremade. The results showed no evidence thatcancers were more common in those wholived close to the TV and radio emissionstower. The risk ratio for residence within3.5 km was 0.73 for all leukemia (p= 0.86).

Honolulu Study. A further study notedby Dolk et al. (35) was an unpublished

Environmental Health Perspectives a Vol 107, Supplement 1 * February 1999 159

J.M. ELWOOD

report. It showed an increase in adultleukemias in areas of Honolulu withbroadcasting antennae, but Dolk et al. (35)comment that "interpretation of that studywas complicated both by the ecologicnature of the design and the problem ofpotential confounding."

Studies ofOcc tonal GroupsThe third set of studies applies to occupa-tional exposures (Table 3). The results arecomplex because in cohort studies alltypes of cancer can be assessed and thecategorization of cancer type varies greatly.

Polish Military Study. This cohortstudy (42) assessed military personnel inPoland from 1971 to 1985, and used mili-tary records to divide subjects into thoselikely to have been exposed to RFs ormicrowaves and those not exposed. It isestimated that in the great majority of theexposed situations, the fields were pulsemodulated emissions at 150 to 3500 MHzand were less than 2 pW/cm2, with aminority of exposures at higher levels.

An essential characteristic of a cohortstudy is that the information sources usedto document the exposure are identical forsubjects who develop the outcome of inter-est (cancer) and for those who do not. Thisinformation was collected from the recordsof central and regional Polish military hospi-tals and the central military medical board,and it is stated that information on exposureto possible carcinogenic factors was collectedfrom these sources as well as from the ser-vice occupational records (42). Thus a ser-viceman who developed cancer had moresources of information on possible RF expo-sures compared to a serviceman who did notdevelop cancer. This raises the possibility ofsystematic bias; such a bias would beexpected to produce an increased relativerisk for all types of cancers.

The results show an excess of all cancers(RR= 2.1, 95% confidence limits 1.1-3.6).For all leukemias and lymphomas, therewere 24 cases observed compared toapproximately 3.8 expected, giving a riskratio of 6.3 (95% confidence limits3.1-14.3). There were also considerableexcesses of cancers of the esophagus, stom-ach, colon, and rectum. These cancer types,unlike the leukemias and lymphomas, haverarely been implicated in connection withnonionizing radiation exposure.

The overall doubling of total cancerrates is inconsistent with other reports. Itraises the question of whether canceroccurrence is better recorded in those whowere exposed, perhaps because of variations

in their military rank, length of service, ortype of posting. Szmigielski (42) states that"it is not possible to offer a reasonableexplanation for the 3-fold increase in therate of stomach and colorectal adenocarci-noma" and that "surprisingly there was nodifference seen in the most commoncancer, lung cancer." No information ispresented on cancer mortality. Althoughmortality may be sometimes of less interestas it is affected by treatment, it has theadvantage of often being more reliablyrecorded than cancer incidence. In this sit-uation mortality information would be auseful addition to the study.

U.S. Navy Study. The Polish studyresults (42) contrast with those of an ear-lier cohort study of U.S. naval personnel,in which 20,000 men with maximumopportunity for exposure to radar emis-sions were identified and compared to asimilar number of 20,000 subjects with alower potential for exposure (29). Thosewith maximum exposure opportunity wereinvolved in electronic equipment repair;those with lower potential exposure wereinvolved in equipment operation. All sub-jects had graduated from U.S. Navy tech-nical schools between 1950 and 1954, andhad served on U.S. Navy ships at the timeof the Korean war. As with the Polishstudy, exposure was based on servicerecords. The outcome was cancer deaths,ascertained up to 1974. Mortality may be aless sensitive indicator of a hazard but ismore likely to be consistently recorded.The extent of potential exposure wasassessed in two ways: by job category andby hazard number-a measure of potentialexposure based on a review of individualrecords for all men who died from diseaseand for a 5% sample of the others.

Confidence limits and a test of trendover categories of exposure have beenapplied to the Robinette et al. (29) pub-lished data because the original did notpresent such measures (see "Methods").The total mortality rate and total cancermortality rate were nonsignificantlyincreased in the highest exposure group.For cancer of the respiratory tract, theadded analysis shows a significantlyincreased risk in the highest exposuregroup defined by hazard number (RR =2.2, 95% confidence limits 1. 1-4. 1),although a test of trend over the fourexposure groups specified is not signifi-cant (chi-square 3. 1, p> 0.07) and theanalysis by job category shows no signifi-cant excesses. No significant excesses wereseen for cancers of the digestive organs,

leukemias and lymphomas, or other can-cers. Other causes of death including heartdisease and other diseases were alsoassessed and no increases were seen. Totaladmission rates to Navy hospitals up to1959 were assessed in addition to deaths,which gives some measure of cancer inci-dence. Total cancer admissions wereslightly lower in the high exposure group.

Robinette et al. (29) condude that "theresults demonstrate that in a large group ofmen, many of whom may have receivedsubstantial exposures, any health effectswhich occurred were insufficient to beclearly perceptible at the level of mortalityor hospital morbidity at the time of expo-sure." One weakness of this study is that itcompares two groups with high and lowlevels of exposure, respectively; whether themortality rates in the low-exposure groupare different from those of an unexposedgroup is unknown.

Study of U.S. Amateur RadioOperators. In a study in the United States,Milham (43) identified radio operators inWashington State and California from1979 to 1984 and linked their names withdeath records up to 1984. Radio operatorswere identified from federal amateur radiooperator licenses. The study was restrictedto men, although women were excludedonly on the basis of name because no gen-der information was available. No informa-tion on other exposures was available,although the author noted that these sub-jects were likely to be exposed to electricshock, soldering fumes, and degreasingagents. The occupation (as given on thedeath certificate) was an electrically relatedjob for 31% of the Washington Stategroup, so exposures to electric power fre-quencies were also likely. The resultsshowed significantly lower death rates thanexpected from all causes and from all can-cers. There was a significantly increased riskof one of nine types of leukemia reviewed(acute myeloid leukemia) and also of can-cers of other lymphatic tissue. There werestatistically significant reductions in deathrates from cancer of the respiratory systemand of the pancreas. The lack of informa-tion on other relevant exposures limits anyfirm interpretation of this study.

Study of Breast Cancers inNorwean Feale Radio and TelegraphOperators. Cancer occurrence was studiedin a group of 2600 Norwegian femaleradio and telegraph operators who werecertified to work as radio and telegraphoperators between 1920 and 1980 andwho worked on merchant ships at sea (44).

Environmental Health Perspectives * Vol 107, Supplement 1 * February 1999160

RF AND HUMAN CANCERS: EPIDEMIOLOGY REVIEW

Table 3. Radiofrequency emissions and cancers: occupational studies and study of amateur radio operators.

Occupational/amateur radio operator studiesU.S. amateur Norwegian female ship Canada/France electric

Polish military (42) U.S. Navy (29) radio operators (43) radio operators (44) utility workers (46)Study characteristics

ExposureAscertainment

Exposed group

Frequency, MHzOutcome dataExposed group for RR values

OutcomeTotal deaths, all causesAll cancerLymphatic and hematopoeticAll leukemiaAcute myeloid leukemiaAcute lymphatic leukemiaAcute nonlymphoidChronic myeloid leukemiaChronic lymphatic leukemiaHodgkin diseaseNon-Hodgkin lymphomaLymphoma/lymphosarcomaOther lymphaticLungLarynx, lungRespiratory tractOther respiratoryOral cavityPharynxLip, oral cavity, pharynxEsophagusStomachEsophagus, stomachColonRectumColorectalDigestive organsLiverPancreasLiver, pancreasOther gastrointestinalProstateKidneyKidney, prostateBladderUrinary tractFemale breastCervixUterus (endometrium)OvaryBoneSkinMelanomaThyroidMultiple myelomaBrain/nervous systemBrain, astrocytomaBrain, glioblastomaOther cancers

RF/microwaveService records

Military, male

150-3500Incidence

2.07 (1 .12-3.58)*6.31 (3.12-14.32)*

8.62 (3.54-13.67)*5.75 (1.22-18.16)*

13.90 (6.72-22.12)*3.68 (1.45-5.18)*2.96 (1.32-4.37)*5.82 (2.1 1-9.74)*5.82 (2.1 1-9.74)*

1.06 (0.72-1.56)

0.71 (0.42-1.32)1.08 (0.82-1.24)

3.24 (1 .85-5.06)*

3.19 (1.54-6.18)*

1.47 (0.76-1.56)

0.86 (0.54-1.67)

0.67 (0.36-1.42)1.67 (0.92-4.13)

1.54 (0.82-2.59)0

1.91 (1.08-3.47)*

Microwave (radar)Service records

Military, male

MortalityHazard no. 5001 +

1.23 (0.98-1.52)a1.44 (0.96-2.07)a1.64 (0.70-3.25)a

2.20 (1.05-4.06)a

0.78 (0.15-2.31

1.17 (0.50-2.32)"

Radio operationRadio operatorslicense records

Amateur radiooperators, male

Mortality

0.71 (0.69-0.74)*0.89 (0.82-0.95)*1.23 (0.99-1.52)1.24 (0.87-1.72)1.76 (1.03-2.85)*1.20 (0.26-3.81)

0.86 (0.17-2.50)1.09 (0.40-2.38)1.23 (0.40-2.88)

0.47 (0.15-1.1)1.62 (1.17-2.18)*

0.66 (0.58-0.76)*

1.13 (0.71-1.72)1.02 (0.68-1.45)

1.11 (0.89-1.37)0.77 (0.42-1.29)

0.65 (0.33-1.17)0.64 (0.42-0.94)*

1.14 (0.90-1.42)0.94 (0.57-1.48)

0.66 (0.38-1.08)

1.39 (0.93-2.00)

Radio operationEmployment records

Radio operatorson ships, female

0.4-25Incidence

1.2 (1.0-1.4)*

1.1 (0.1-4.1)

1.3 (0.4-2.9)

1.2 (0.4-2.7)

0.4 (0.1-2.0)

1.3 (0.6-2.6)1.8 (0.7-3.9)

0.6 (0.0-3.5)

1.6 (0.3-4.8)

0.6 (0.0-3.6)

1.5 (1.1-2.0)*1.0 (0.6-1.7)1.9 (1.0-3.2)*0.8 (0.3-1.6)

0.9 (0.4-1.7)

1.0 (0.3-2.3)

0.7 (0.3-1.3)

PEMFEmployment records andfield measurements

Electric utility workers

5-20, possibly up to 300Incidence.90th percentile

1.39 (1.05-1.85)*0.96 (0.48-1.90)0.80 (0.19-3.36)1.02 (0.08-13.04)

1.91 (0.19-19.39)

2.98 (0.21-41.57)1.33 (0.23-7.68)1.80 (0.62-5.25)

3.11 (1.60-6.04)*

1.20 (0.13-11.28)

4.98 (0.73-34.00)

2.16 (0.22-20.88)

1.35 (0.43-4.19)1.54 (0.42-5.61)

1.47 (0.47-4.54)0.78 (0.23-2.56)

1.02 (0.34-3.07)

0.31 (0.03-2.82)

0.20 (0.03-1.39)1.90 (0.48-7.58)6.26 (0.30-132.2)0.57 (0.08-3.91)

Environmental Health Perspectives * Vol 107, Supplement 1 * February 1999

-, No information was given in the study. aRobinette et al. (29): data are observed/exposed ratios for the highest exposure group (hazard function 5001 +) compared to allsubjects, with 95% confidence limits; test of trend also applied, all results nonsignificant. *Statistically significant, p<0.05.

161

J.M. ELWOOD

The incidence of all cancers was modestlyincreased, although this was significant(RR= 1.2, 95% confidence limits 1.0-1.4).An excess risk was seen for breast cancer(RR= 1.5, 95% confidence limits 1.1-2.0)and also for uterine cancer (RR= 1.9, 95%confidence limits 1.0-3.2). There was nosignificant excess of leukemias, lymphoma,brain tumors, or of several other types ofcancer, although the results were based onsmall numbers.

The authors discussed exposures toshift work, time changes, and to light-at-night, as well as to RF emissions (405kHz-25 MHz) and ELF fields (50 Hz).The breast cancer association was furtherexplored in a nested case-control study,showing an association with shift work inwomen over age 50. There are many estab-lished risk factors for breast cancer, includ-ing several aspects of reproductive history,alcohol consumption, obesity, and a fam-ily history of breast of other cancers.Cancer of the uterus shares several of theseknown associations with reproductive anddietary factors (45). The increased risksfor both breast and uterine cancer, with noexcesses in leukemia or similar cancers, aremore suggestive of a relationship to repro-ductive or other lifestyle factors than to anassociation with RF emissions. An analysisadjusted for age at first birth (a major riskfactor for breast cancer), however, stillgave an RR value of 1.5 for the associationbetween breast cancer and work as a radiooperator. Tynes et al. (44) performed somespot measurements of fields in ships stillequipped with older equipment and foundthat "measured values at the operator'sdesk were below the detection level forboth electric and magnetic fields"; how-ever, measurements 0.5 m in front of theradio tuner showed magnetic fields at fre-quencies above 8 MHz, which exceededoccupational exposure limits. This studypresents interesting observations thatrequire further research.

Study ofElectric Utility Workers:Short-Duration Pulsed ElectromagneticField. Armstrong et al. (46) carried out anested case-control study within cohorts ofelectric utility workers in Quebec, Canada,and in France, in whom 2679 incidentcases of all types of cancer were identified.The exposure assessed was short-durationpulsed EMFs (PEMFs) generated mainly bydielectric switching operations. Exposuresin cancer cases and in comparison subjectswere assessed by a job exposure matrixbased on electric field measurements car-ried out on approximately 1300 workers

for 1-week periods in 1991 to 1992 toassess the proportion of subjects in eachjob who had a weekly mean exposure of> 100 ppb. That is, for > 100 ppb of time,the electric field was > 200 VIm in the 5 to20 MHz frequency band; this is equivalentto 14.4 milliseconds/40-hr week. However,the meters also responded to RF fields ofapproximately 150 to 300 MHz and toradio transmissions; thus, some of the highrecorded exposures could be due to RFfields (46).

The study has extensive data on otherrelated factors such as smoking and otheroccupational exposures. There were noassociations seen between exposure tothese EMFs and cancers that had beensuggested to be associated with EMF expo-sure, including leukemia, lymphomas,brain cancer, and melanoma. The resultsfor cancers of the lip, mouth, and pharynx(together) showed a nonsignificantincrease based on small numbers. Stomachcancer results showed a nonsignificantincrease, restricted to France and with noclear dose-response relationship; for allcancer there was a significant increase forworkers at or above the 90% percentile ofexposure (RR= 1.39, 95% confidence lim-its 1.05-1.85) but no significant increase(RR = 1.03) based on the median dose. Asignificant association was seen with lungcancer: RR =3.11, 95% confidence limits1.60-6.04 based on the 90% percentile,and RR= 1.27 (95% confidence limits0.96-1.68) based on the median. This wasseen almost exclusively in the Quebeccohort; the levels of exposure to PEMFswere higher in Quebec. The risk increasedwith duration of exposure, and was seenafter adjustment for smoking and otheroccupational exposures. The authors notethat the findings were unexpected and aroseafter assessment of 30 cancer sites and threetypes of exposure-approximately 100comparisons. They note: "However, severalfactors limit the strength of the evidencefor a causal relation; lack of precision inwhat the meters measured, little previousevidence for this association; and no ele-vated risk for lung cancer in the utilityworkers overall in comparison with thegeneral population" (46).

This study is methodologically thestrongest of the studies reviewed here. Thework histories were detailed and the resultswere based partially on field measurementsof exposures, with good information onother relevant factors. The fields, however,may not be relevant to the major sources ofRFR because they relate to a specific type

of emission in a lower frequency range of 5to 20 MHz. The study was negative inregard to leukemias and similar cancers,which were elevated in the Polish militarystudy (42) and in the first Sydney study(36). The Armstrong et al. (46) studyfound an unexpected association with lungcancer, which is one of the few sites notelevated in the Polish military study andhas not been suggested in any other studyofEMF exposures.

Case-Control StudiesSeveral case-control studies with specificmention of RF emissions have been pub-lished (Table 4).

Brain Cancer in U.S. Air ForcePersonnel. In this study (47), U.S. AirForce servicemen who served between1970 and 1989 and who developed braincancer were identified from service records,and each was compared to four comparisonsubjects (unaffected Air Force personnel ofthe same age and ethnic group). Occu-pational histories were taken from person-nel records and reviewed to assess likelyexposure to both low frequency and RFmicrowave fields. This assessment used a

job-exposure matrix developed by an

expert group including a radiation physi-cist, an occupational health specialist, andan industrial hygienist.

The authors found that the risk ofdeveloping a brain tumor increased con-

siderably with increasing military rank,and they could not suggest a good expla-nation of that relationship. In addition,they found a small association with elec-tromagnetic radiation, which was statisti-cally significant for RF microwaveradiation (RR= 1.39, 95% confidence lim-its 1.01-1.90) and was somewhat lower forELF radiation (RR= 1.28, 95% confidencelimits 0.95-1.74). This study did notinclude men who had left the service; a biascould arise if any subjects left the servicebecause of health problems that were laterdiagnosed as brain cancer.

Brain Cancer in U.S. Civilians withOccupational Exposures to Radio-frequency Emissions. This study (48)shows the complexity of exposures experi-enced through occupation. Cases were menwho died of brain or other related cancersin three areas of the United States; controlswere men who died from causes exdudingbrain cancer, epilepsy, stroke, suicide, or

homicide, matched on age, year of death,and area of residence. Information on

occupation was collected by interviewswith the next of kin, with response rates of

Environmental Health Perspectives * Vol 107, Supplement 1 * February 1999162

RF AND HUMAN CANCERS: EPIDEMIOLOGY REVIEW

Table 4. Radiofrequency emissions and cancers: case-control studies.

Case-control studiesBreast cancer

Brain tumors Brain tumors, fatal Testicular tumors Breast cancer in men deaths, women Eye melanoma

Study characteristicsReference Grayson, 1996 (47) Thomas et al., 1987 (48) Hayes et al., 1990 (49) Demers et al., 1991 (50) Cantor et al., 1995 (51) Holly et al., 1996 (52)Exposure RF/microwave Occupations with RF/microwave Radio communication RF fields Microwaves, radar

RF exposure workAscertainment Service records: Job title: two exposure Job title and self Job title Job, on death Interview

job-exposure matrix assessments report certificateExposed group U.S., military U.S., men > age 30, U.S., mainly military U.S., civilian U.S., women U.S., general

white populationCancer type, adult

Brain cancers 1.39(1.01-1.90)* 1.0 ( 0.51.9)aMale breast 2.9 (0.8-10.0)Testis - 3.1 (1.46.9)b,*Testis 1.1 (0.6-2.1)cFemale breast, white - 1.14 NSdFemale breast, black - 1.34 NSEye melanoma - - 2.1 (1.1-4.0)*

-, No information given in the study. &For RF exposure, not in electrical jobs. bBy self-reported exposure. cBy assessment from job titles; any exposure. dMaximum level ofexposure. *Statistically significant, p<0.05.

74% for the cases and 63% for the con-trols. Occupations were classified in termsof likely exposures by two methods: onebased on a predesigned list of occupationslikely to involve RF radiation, and onebased on a review of each job by a certifiedindustrial hygienist.

The single-factor analysis showed anexcess risk among men who had ever had ajob in which they were likely to be exposedto RF radiation (RR= 1.6, 95% confidencelimits 1.0-2.4). However, this risk was onlyseen in those whose potential exposure toRFR was in an electrical or electronic occu-pation, where the relative risk was 2.3. Bycomparison, those exposed to RFR withoutworking in such a job had a relative risk of1.0, and the risk was increased in electronicsworkers with no exposure to RFR.

Thus the increased risk was due tosome other aspect of work in an electricalor electronics occupation. Further analysisshowed increases in risk with jobs involvingexposure to soldering fumes, lead, andorganic solvents. The authors conclude that"that pattern of excess brain tumor riskamong electrical and electronics workers,and not among others exposed to radio-frequency radiation, suggests that simpleexposure to radiofrequency radiation is notthe responsible agent" (48).

Study ofTestiular Cancer. A study inthe United States (49) assessed 271 men18 to 42 years of age with testicular cancerand 259 controls, identified at three med-ical centers, two of which were militaryhospitals. Exposure to microwave andother radio waves was assessed in two ways:by an analysis based on job title and by

self-report. An excess risk was seen with theself-reported exposure (RR= 3.1, 95% con-fidence limits 1.4-6.9). However, no asso-ciation was seen with the analysis based onjob title (RR= 1.1); job categories classifiedas likely to have had the heaviest exposurehad RR values = 0.8. Therefore, the resultsare inconsistent.

Male Breast Cancer Study. Occupa-tional exposure to EMFs and breast cancerin men (a rare disease) has been studied(50) based on 227 cases from 10 areas ofthe United States, as compared to 300 con-trols. Exposure status was defined as everhaving been employed in a job that hadbeen classified as involving potential expo-sure to EMFs. The risks seen were highestamong electricians, telephone linemen,and electric power workers. Radio andcommunications workers had a nonsignifi-cant increase in risk (RR= 2.9, 95% confi-dence limits 0.8-10.0) based on sevencases. Risk did not vary with duration ofexposed employment. In this study theparticipation rates were low, especially inthe controls. The results can be regardedonly as a preliminary observation requiringfurther research.

Case-Control Study ofBreast CancerDeaths. Cantor et al. (51) performed acase-control study of women who diedfrom breast cancer between 1984 and 1989in 24 states in the United States, matchingeach with controls randomly selected fromnoncancer deaths and matched for age,gender, and race; over 33,000 cases and117,000 controls. Occupational data werelimited to that recorded on the death cer-tificate. A job-exposure matrix was used to

estimate the probability and level of 31workplace exposures, and the analyses werecarried out after adjusting for socioeco-nomic status. The authors conclude that"suggestive associations for probability andlevel of exposure were found for styrene,several organic solvents,...and several met-als/metal oxides and acid mists. Because ofthe methodologic limitations of this study,its primary value is in suggesting hypothesesfor further evaluation" (51).

RF EMF exposure was one of theexposures studied. It was classified in fivegroups ranging from nonexposed throughfour increasing levels of estimated expo-sure probability. For white subjects, a sig-nificant OR of 1.15 (95% confidencelimits 1.1-1.2) was reported for thefourth of the five levels, but there was noexcess in the highest level of exposureprobability (OR 0.99) and no regulartrend. An analysis using exposure level,that is, estimated intensity, classified asnonexposed and three levels, showed sig-nificant increases at level 1 and level 3(OR 1.15 and 1.14, respectively) but areduced risk at level 2 (OR 0.95, 95%confidence limits 0.9-1.0), and thus againno regular trend. Results for African-American women were somewhat similar,with some significant elevations in indi-vidual categories but no regular trend oneither analysis.

The authors' discussion takes intoaccount the issues of multiple testing andthe lack of a dose-response trend, and theirconclusion was "in this investigation, wefound no association with either ionizingor non-ionizing radiation" (51).

Environmental Health Perspectives * Vol 107, Supplement 1 * February 1999 163

J.M. ELWOOD

Case-Control Study ofIntraocularMelanoma: Microwaves and Radar.Holly et al. (52) conducted a case-controlstudy of intraocular melanoma based on221 male white patients residing in thewestern United States and seen at onecenter in San Francisco, compared to 447controls of similar age in the same geo-graphical area as identified through ran-dom digit dialing. A large number ofoccupational groups were assessed, withsignificant increased risks found inchemists, sailors or fishermen, welders, andhealth-related occupations. In terms of spe-cific exposures, significant excess risks wereseen with exposures to formaldehyde, pes-ticides, carbon tetrachloride, asbestos, anda marginally significant excess risk seenwith exposure to antifreeze. Only resultsfor the exposures for which there was anincreased risk are presented, and these arepresumably a subset of a large number ofdifferent exposures that were assessed, ofwhich the total number is not given.

The finding of an elevated risk for everexposure to microwaves or radar, with anOR of 2.1, 95% confidence limits 1.1-4.0,based on 21 of the 221 patients beingexposed, must be taken in this context.The authors do not mention this result intheir summary, and they point out that thedata on recall of specific exposures "aremore subject to recall bias than the majorfindings based on the occupational groups`(52). They state that the association withhealth-related occupations, which might bethe one most readily associated withmicrowaves, "could have resulted fromreferral bias" (52), given that this studywas carried out in only one referral center.

Other Human Studies Relevant toCancer Causation. Garson et al. (53) car-ried out chromosomal studies on 38 radiolinemen employed by Telecom Australiaand who had exposure to 400 kHz to 20GHz radio frequencies at or below occupa-tional limits, and 38 age-matched controlswho were derical staff with no exposure toRFs. Some 200 metaphases from each sub-ject were studied and scored by an observerwho was blind as to exposure status. Amonitoring committee including represen-tatives of Telecom Australia and the rele-vant union monitored the study and alldata were coded blind until the study wasfinalized. The results showed virtuallyidentical frequencies in a range of specifiedaspects of chromosomal damage, with theoverall risk ratio of aberrant cells being 1.0(95% confidence limit 0.8-1.3). Theauthors state that this was the first study of

chromosome damage in workers exposedto RFs.

In an in vitro experiment using highdosages of high frequency microwave radi-ation, samples of human whole blood wereexposed to 7.7 GHz radiation at powerdensities from 0.5 to 30 mW/cm2. Thisproduced an increase in chromosomalaberrations and micronuclei (54).

A few other studies assessed bloodcounts, which could be relevant to cancerbut are also influenced by many other fac-tors. Two studies found no effects in work-ers with occasional high exposures to radar(55,56).

DiscussionMethodsFor this review, a systematic search proce-dure has been used to identify all majorpublished studies dealing specifically withRFs and human cancer. Many otherstudies relate to exposures to EMFs withno further specification. These may indudeRF exposures but, in the absence of anyspecification, the major exposures are morelikely to be ELFs. Such studies have beenexamined in major reviews of ELF expo-sure. They include cohort studies oftelecommunication workers (57,58), work-ers in electrically related jobs (59), andstudies assessing exposures to visual displayterminals (60). Studies based only onnational routine data sets of incidence ormortality with occupation as recorded onthe death certificate or hospital record wereexcluded because they suffer from severalproblems: the quality of the data is limited;multiple testing is a major difficulty; andsuch reports, because of their size, are oftenunpublished or are published only as gov-ernment documents. Selected items fromthem published in the scientific literatureare particularly open to publication bias.

Although only studies published inscientific journals have been included inthis review, the report by Lilienfeld et al.(61) deserves attention because it has beengiven publicity by others. According toBergqvist (27), this study comparedemployees at the U.S. Embassy in Moscowwith employees of U.S. embassies in othereastern European countries because RF at0.5 and 10 GHz of up to 5 pW/cm2 for9 hr/day was detected at the MoscowEmbassy between 1963 and 1975, but suchradiation was not detected at otherembassies. The study of embassy employeesreviewed medical records for 71%, andused a questionnaire with a response rate of

42%, which is low. Bergqvist (27) quotes arelative risk for all cancers of 0.9 (95%confidence limits 0.5-1.4) and increasedrisks of leukemia and of breast cancers,each nonsignificant and based on only twocases, comparing the Moscow Embassystaff with U.S. national rates. For bothleukemia and breast cancer, the rates werelower for Moscow Embassy staff than forthe staff of other eastern Europeanembassies with no recorded RF exposure.Goldsmith (62) also comments on thisstudy and presents data showing excess risksof leukemia, brain tumors, and breast can-cers in embassy staff compared to expectedvalues, which are not specified but are pre-sumably based on U.S. rates, but the datashow lower risks in Moscow staff than inother embassy staff. Goldsmith also notesthat studies of chromosome changes andhemologic measures were also conducted,which he claims shows changes in theembassy staff as compared to foreign serviceexaminations carried out in Washington,although he quotes an expert panel thatconcluded that "no valid conclusions couldbe drawn from this study" (62). Goldsmithimplies that the results suggest there musthave been similar radiation exposures at theother eastern European embassies, butthere is no evidence for this. Overall, theresults of this study appear negative.

Types ofEpidemiological Study. Therelative importance of epidemiologicstudies can be referred to as a hierarchy ofstudies (63,64). The strongest evidence toassess a cause-and-effect relationship comesfrom an experimental study, but obviouslythis cannot be applied to potential hazardsin a general population. The best studies toassess such potential hazards are studiesbased on individuals with good informa-tion on the suspected causal factor, the dis-ease outcome, and on other relevant factorsrelated to the outcome. Such cohort andcase-control studies are the method bywhich most recognized causes of humancancer have been identified, such astobacco smoking, exposure to asbestos, orexposure to ionizing radiation. Usuallymany such studies need to be done before aconsensus can be reached (63).

Limitations of Studies ofRadio-frequencies and Health Effects: TheEcological Falacy. The type of evidenceavailable on health effects ofRF is not com-parable to the types of evidence available forsuch well-established links as those betweensmoking and lung cancer, asbestos andmesothelioma, or thalidomide and birthdefects. For RF, the studies of individuals

Environmental Health Perspectives * Vol 107, Supplement 1 * February 1999164

RF AND HUMAN CANCERS: EPIDEMIOLOGY REVIEW

are limited to small case-control studieswith poor estimation of RF exposure and tocohort studies of certain groups (such asmilitary personnel) whose exposure levelsare likely to be different from those of thegeneral population. Even these individuallybased studies do not show consistent results.

The studies of general populations arenot based on individuals but are based oncomparing community groups withhypothesized different levels of exposuredetermined, for example, by residence.These are ecological studies. It is falla-cious to assume that an association seen inan ecological study indicates a cause-and-effect relationship between individuals(65); this error has been referred to as theecological fallacy. Thus, many ecologicalstudies demonstrated relationships betweenheart disease frequency in different geo-graphic areas and various characteristics indrinking water, but further studies have notshown any important relationship at theindividual level (66). The explanation isthat geographical areas with different watersupplies have many other differences thatare more directly related to heart disease.

The ecological fallacy can only bedemonstrated when hypotheses, createdon the basis of comparisons betweengroups, are tested in high-quality studiesof individuals, with good measurement ofindividual exposure and individual healthoutcome. A recent example is a majorcase-control study to test the hypothesisthat childhood cancer was related toradon gas concentrations (67). Radon lev-els in the current and previous homes ofchildren who later developed leukemia,and in comparison children, were directlymeasured in a study involving nearly 1000children. The mean radon levels werelower for children who developedleukemia than for controls; the authorsconclude: "In contrast to prior ecologicalstudies, the results from this analyticstudy provide no evidence for an associa-tion between indoor radon exposure andchildhood leukemia" (67).

Criteria Used in Assessing Causality.Criteria have been developed that are gen-erally and internationally accepted for theassessment of epidemiologic evidence froman individual study, and from the totalityof evidence derived from a number ofstudies. There are two sets of criteria. Thefirst set comprises three factors that couldexplain an apparent association between anexposure and a disease, apart from a cause-and-effect relationship (63). These factorsare as follows:

* Bias in the observations that are made.For example, in a study based on aninterview recall of exposures, peopleaffected with cancer may more readilyrecall and report a previous exposurethan people who have not had cancer.

* The effect of other relevant factors,known as confounding. Thus theremay be an association between icecream sales and drownings in coastalareas due to a third related factor, thatof good weather.

* Apparent associations may be due tochance variation. This is assessed bystatistical methods.The first process in assessing whether a

particular study gives a valid cause-and-effectassessment is therefore to see if these alterna-tive explanations can be reasonably exduded.

The next process is to look for the spe-cific features expected if a biologic cause-and-effect relationship applies. Thesecriteria, often referred to as the BradfordHill criteria (68), are accepted by manymultidisciplinary international groups inthe assessment of cause and effect in cancerand in other major diseases (69). They areuseful but are not prescriptive or absolutecriteria, and their application must takeinto account measurement errors andprobable heterogeneity of both exposureand outcome. The criteria are as follows:The time relationship must be clear; thesuspected cause must clearly precede thedevelopment of the disease. The relation-ship should be strong, that is, there shouldbe a substantially higher rate of disease insubjects exposed to the potential causal fac-tor than in those not exposed. Thereshould be a dose-response relationship,that is, subjects with greater exposureshould have greater disease outcome rates.Specificity may be helpful, although it isnot always applicable. This means that aparticular causal agent causes a particulardisease rather than a whole range of dis-eases. Plausibility may be a useful criterionbut is often open to varied interpretations.This means that a mechanism for the rela-tionship between the causal agent and thedisease outcome can be recognized basedon knowledge of the underlying biology ofthe situation. A related criterion that issometimes added is analogy-that the rela-tionship is similar to some other estab-lished relationship-but this is merely anaspect of plausibility. Coherence impliesthat there should be an overall associationseen between the general distribution ofthe causal agent and the distribution of thedisease that it causes. It is a less useful

criterion because it holds only if the causalrelationship is strong. Consistency is themost important criterion. Consistency isassessed first as consistency within a study.For example, if radiation causes a particularcancer, we would expect to see that rela-tionship in men and women, in differentsocial groups, geographical areas, and soon. The most important criterion is consis-tency among various studies (63). In thegreat majority of situations the devel-opment of a consensus on whether a par-ticular agent causes (for example) cancer isbased on a consideration of the consistencyof evidence from a large number of studies,of different designs and in different popu-lations that overall give a substantial bodyof evidence.

Application ofCriteriafor Causality.The available studies on RF and humancancer have been assessed in regard tothese criteria.

The study designs used in general areweak. This applies particularly to theassessment of likely exposure. In all thegeneral population studies, the measure ofpotential exposure used has been the placeof residence at the time of diagnosis of thecancer or at the time of death. This is onlyan approximate indicator of the level ofexposure to the relevant wavelengths of RFtransmissions at the critical time for thecausation of the cancer, which will bemonths or years before diagnosis. Also,information on other relevant factors atthe individual level has not been collectedand cannot be adjusted for. Where area ofresidence is defined down to a small unit,as in the U.K. studies (34,35), there issome ability to control for other factors byusing variables related to that small area ofresidence, but where the area is larger, asin the first Sydney study (36), this is lesseasily done.

The occupational cohort studies aresomewhat stronger. However, althoughthese are studies of individuals in terms ofcancer occurrence, data on exposure to RFsare indirect, being based on job title andother details such as military posting, witheither assumptions or limited field measure-ments available, not for all the individualsin the study, but for other samples of thosein particular jobs or of the environment ofparticular jobs.

In the case-control studies, estimates oflikely exposure in the past have been basedeither on job titles or on self-report; noneof these studies are convincing. Thereforeall of these studies are relatively weak,which means that issues of observation bias

Environmental Health Perspectives * Vol 107, Supplement 1 * February 1999 165

J.M. ELWOOD

and the influence of other relevant factorscannot be easily discounted.

In terms of criteria for positive aspectsof causality, we can accept the time crite-rion that exposure was likely to precede theeffect. None of these studies give enoughdetail to assess the likely time intervalbetween exposure and an increase in risk.

In regard to strength, the strongestassociations were seen in the Polish mili-tary study (42). The difficulty is that thisstudy showed an excess of all cancer and ofmany types of cancer, including gastroin-testinal tract cancer, which has not beenrelated to radiation in any other majorstudy. It seems likely that there is a bias inthat study toward a general excess of diseasebeing recorded in association with higherexposure. A strong association with lungcancer was seen in the cohort study of elec-trical utility workers exposed to PEMFs,but the authors themselves regarded thiswith considerable doubt. The associationswere not strong in any of the other studies.In the Sydney study (36) the associationwith the incidence of childhood leukemiahad an odds ratio of 1.6. In the moreextensive and powerful UK study, even theequivocal results on total leukemia arebased on a maximum excess of only 15%seen only in some localities (35).

Few of these studies showed any consis-tent dose-response assessment. The utilityworkers (46) and the U.S. military study(29) had some measures of different likelyexposure dosages but did not show consis-tent results. The Sutton Coldfield and UKstudies (34,35) were based largely onassessing a measure of dose response, thatis, a gradation of risk with distance of resi-dence from the transmitter, and did notgive consistent positive results for anycancer site.

The criteria on which these studiesclearly break down is in specificity tocancer types and in consistency. If there isa real effect of RF on cancer, we wouldexpect to see it consistently on certaintypes of cancer in various studies.However, this is not found. Figure 1 sum-marizes the results for adult and childhoodleukemias. The strongest individual resultsfor RF in a general population were thefirst Sydney results for childhood leukemia(36). However, the Sutton Coldfield andU.K. 20 transmitter studies showed noclear excess of childhood leukemia(34,35), nor did the San Francisco study(41). For adult leukemias of all types,there was a weaker but still positive rela-tionship in the Sydney study (36). The

166

results for the U.K. 20 transmitter studiesgave some evidence of a decreasing trendwith residence, although the effect wassmall and there was no excess in thosewho live closest to the transmitter (35).The results for all adult leukemias wereweak and equivocal in the U.S. Navystudy (29). The Polish military studyshowed an excess of leukemia but also anexcess of several other cancers that werenot shown in any other study (42).

In adults, cancers of the central ner-vous system and the brain gave positiveresults in the Polish military study (42)and in the small case-control study ofbrain tumors in U.S. military (47) butshowed no association in the study of elec-tric utility workers (46) or the Sydneystudy (36), and no clear association in theSutton Coldfield study (35). There was noassociation with cancers of the brain inchildhood in any of the three studies thatassessed it (35,36,41).

The criterion of coherence, that is,whether the rates of these diseases vary intime and place with RF emissions, is onlyrelevant if it is claimed that these associa-tions are strong and if this exposure is themajor cause of these diseases. This criterionis not helpful.

The final criterion is that of plausibil-ity, which relates to whether an establishedmechanism relating in biologic terms thisexposure to cancer production is accepted.The experimental evidence, both inhuman cell systems and in the nonhuman

Type Study, referenceAdult leukemia E United Kingdom (35)

E Sydney 1 (36)C Polanda (42)C U.S. Navyb(29)C U.S. radio (43)C Norway(44)C Canada/France (46)

Child leukemia EEEEE

United Kingdom (35)Sydney 1 (36)Sydney 2 (40)Sydney 2b (40)

San Franciscoc (41)

0.10

situations, is in itself controversial. There isno consensus from major interdisciplinaryreview groups that the evidence dearly sug-gests a potential biologic mechanism forcancer causation, although a minority ofresults suggest potential effects in terms ofinitiation or promotion actions; therefore,this criterion is also unhelpful.ConclusionsSeveral studies have assessed associationsbetween likely exposure to RF and varioustypes ofhuman cancer.

The studies individually are weak and,as a consequence, the results cannot be eas-

ily interpreted in terms of cause and effect.The major impression from these studies istheir inconsistency. There is no type ofcancer that has been consistently associatedwith RF exposures.

The epidemiologic evidence falls shortof the strength and consistency of evidencethat is required to come to a reasonableconclusion that RF emissions are a likelycause of one or more types of humancancer. The evidence is weak in regard toits inconsistency, the weak design of thestudies, the lack of detail on actual expo-sures, the limitations of the studies in theirability to deal with other likely factors, andin some studies there may be biases in thedata used. 'Whereas the current epidemio-logic evidence justifies further research toclarify the situation, there is no consistentevidence of any substantial effect onhuman cancer causation.

1.00 10 100

RR, log scale

Figure 1. Relative risks and 95% confidence limits for studies of leukemia in adults and in children. Type of study:C, occupational cohort; E, ecological. 'All lymphatic and hematopoietic-total leukemia not given. bExcluding LaneCove area. cNo confidence limits given; nonsignificant.

Environmental Health Perspectives * Vol 107, Supplement l * February 1999

RF AND HUMAN CANCERS: EPIDEMIOLOGY REVIEW

ACKNOWLEDGMENTS: The author is sup-ported by the Cancer Research Trust (Dunedin).This work was stimulated by a request fromTelecom New Zealand to review this topic, andthe author has prepared other reports forTelecom; however, this review has not be seen orinfluenced by any staff or associates of Telecom,and represents only the author's views.

REFERENCES AND NOTES

1. International Commission on Non-ionizingRadiation Protection. Health issues related tothe use of hand-held radiotelephones and basetransmitters. Health Phys 70:587-593(1996).

2. Rothman KJ, Chou C, Morgan R, Balzano 0, GuyAW, Funch DP, Preston-Martin S, Mandel J,Steffens R, Carlo G. Assessment of cellulartelephone and other radio frequency exposurefor epidemiologic research. Epidemiology7:291-298 (1996).

3. International Commission on Non-IonizingRadiation Protection. Guidelines for limitingexposure to time-varying electric, magnetic,and electromagnetic fields (up to 300 GHz).Health Phys 74:494-522 (1998).

4. United Nations Environment Programme/WorldHealth Organization and International RadiationProtection Association. Electromagnetic Fields(300 Hz to 300 GHz). Environmental HealthCriteria Series 137. Geneva:World HealthOrganization, 1993.

5. National Radiological Protection Board.Biological Effects of Exposure to Non-lonisingElectromagnetic Fields and Radiation. Ill:Radiofrequency and Microwave Radiation.R-240. Chilton, UK:National RadiologicalProtection Board; 1991.

6. McKinlay AF, Andersen JB, Bernhardt JH,Grandolfo M, Hossmann K, Mild KH, SwerdlowAJ, Van leeuwen M, Verschaeve L, Veyret B.Radiotelephones and Human Health - Proposalfor a European Research Programme. Brussels:European Commission Directorate General Xil,1996.

7. Cridland NA. Electromagnetic fields and cancer:a review of relevant cellular studies. NRPB-R256. Chilton, UK:National RadiologicalProtection Board, 1993.

8. Polk C, Postow E. Biological Effects ofElectromagnetic Fields. 2nd ed. Boca Raton,FL:CRC Press, 1996.

9. Repacholi MH. Low-level exposure to radiofre-quency electromagnetic fields: health effectsand research needs. Bioelectromagnetics19:1-19 (1998).

10. Sarkar S, Ali S, Behari J. Effect of low powermicrowave on the mouse genome: a direct DNAanalysis. Mutat Res 320:141-147 (1994).

11. Lai H, Singh NP. Acute low-intensity microwaveexposure increases DNA single-strand breaksin rat brain cells. Bioelectromagnetics16:207-210 (1995).

12. Lai H, Singh NP. Single- and double-strand DNAbreaks in rat brain cells after acute exposure toradiofrequency electromagnetic radiation. Int JRadiat Biol 69:513-521 (1996).

13. Williams GM. Comment on "Acute low-inten-sity microwave exposure increases DNA single-strand breaks in rat brain cells" by Henry Laiand Narendra P. Singh. [Letter]. Bioelectro-magnetics 17:165 (1996).

14. Chou C, Guy A, Kunz LI, Johnson RB, CrowleyJJ, Krupp JH. Long-term, low-level microwaveirradiation of rats. Bioelectromagnetics13:469-496 (1992).

15. Szmigielski S, Szudinski A, Pietraszek A, BielecM, Wrembel JK. Accelerated development ofspontaneous and benzopyrene-induced skincancer in mice exposed to 2450-MHz microwaveradiation. Bioelectromagnetics 3:179-191 (1982).

16. Santini R, Hosni M, Deschaux P, Packeco H.B16 melanoma development in black miceexposed to low-level microwave radiation.Bioelectromagnetics 9:105-107 (1988).

17. Salford LG, Brun A, Eberhardt JL. Experimentalstudies of brain tumour development duringexposure to continuous and pulsed 915 MHzradiofrequency radiation. BioelectrochemBioenerg 30:313-318 (1993).

18. Repacholi MH, Basten A, Gebski V, Noonan D,Finnie J, Harris AW. Lymphomas in Ep-Pimltransgenic mice exposed to pulsed 900 MHzelectromagnetic fields. Radiat Res 147:631-640(1997).

19. Bawin SM, Kaczmarek LK, Adey WR. Effects ofmodulated VHF fields on the central nervoussystem. Ann NY Acad Sci 274:74-81 (1975).

20. Blackman CF, Elder JA, Weil CM, Benane SG,Eichinger DC, House DE. Induction of calcium-ion efflux from brain tissue by radiofrequencyradiation: effects of modulation frequency andfield strength. Radio Sci 14:93-98 (1979).

21. Albert EN, Slaby F, Roche J, Loftus J. Effect ofamplitude modulated 147 MHz radiofrequencyradiation on calcium ion efflux from avian braintissue. Radiat Res 109:19-27 (1987).

22. Lyle DB, Schechter P, Adey WR, Lundak RL.Suppression of T-lymphocyte cytotoxicity fol-lowing exposure to sinusoidally amplitude-modulated fields. Bioelectromagnetics4:281-292 (1998).

23. Byus CV, Lundak RL, Fletcher RM, Adey WR.Alterations in protein kinase activity followingexposure of cultured human lymphocytes tomodulated microwave fields. Bioelectro-magnetics 5:341-351 (1984).

24. Byus CV, Kartun K, Pieper S, Adey WR.Increased orthinine decarboxylase activity incultured cells exposed to low energy modulatedmicrowave fields and phorbol ester tumour pro-moters. Cancer Res 48:4222-4226 (1988).

25. Litovitz TA, Krause D, Penafiel M, Elson EC,Mullins JM. The role of coherence time in theeffect of microwaves on orthinine decarboxylaseactivity. Bioelectromagnetics 14:395-403 (1993).

26. Balcer-Kubiczek EK, Harrison GH. Neoplastictransformation of C3H/1 0T1/2 cells followingexposure to 120 Hz modulated 2.45 GHzmicrowaves and phorbol ester tumour pro-moter. Radiat Res 126:65-72 (1991).

27. Bergqvist U. Review of epidemiological studies.In: Mobile Communications Safety (Kuster N,Balzano 0, Lin JC, eds). London:Chapman &Hall, 1997;147-170.

28. Ovid: Medline [database on CD-ROM]. Version3.0. Bethesda, MD: National Library ofMedicine, 1966-. Updated monthly (Optional).

29. Robinette CD, Silverman C, Jablon S. Effectsupon health of occupational exposure to micro-wave radiation (radar). Am J Epidemiol112:39-53 (1980).

30. Breslow NE, Day NE. Statistical Methods inCancer Research. Volume 2: The Design andAnalysis of Cohort Studies. IARC Sci Publ No.82. Lyon:lnternational Agency for Research onCancer, 1987.

31. Rothman KJ. A sobering start for the clusterbusters' conference. Am J Epidemiol132(Suppl.1):S6-S13 (1990).

32. Maskarinec G, Cooper J, Swygert L. Investi-gation of increased incidence in childhoodleukemia near radio towers in Hawaii: prelimi-nary observations. J Environ Pathol ToxicolOncol 13:33-37 (1994).

33. Davis RL, Mostofi FK. Cluster of testicularcancer in police officers exposed to hand-heldradar. Am J Ind Med 24:231-233 (1993).

34. Dolk H, Shaddick G, Walls P, Grundy C, ThakrarB, Kleinschmidt L, Elliott P. Cancer incidencenear radio and television transmitters in GreatBritain. 1: Sutton Coldfield transmitter. Am JEpidemiol 145:1-9 (1997).

35. Dolk H, Elliott P, Shaddick G, Walls P, ThakrarB. Cancer incidence near radio and televisiontransmitters in Great Britain. 2: All high powertransmitters. Am J Epidemiol 145:10-17(1997).

36. Hocking B, Gordon IR, Grain HL, Hatfield GE.Cancer incidence and mortality and proximity toTV towers. Med J Aust 165:601-605 (1996).

37. Hocking B, Gordon I, Hatfield G, Grain H. Re:Cancer incidence near radio and televisiontransmitters in Great Britain. I: Sutton Coldfieldtransmitter. Il: All high power transmitters.[Letter]. Am J Epidemiol 147:90-91 (1998).

38. Greaves MF. Speculations on the cause ofchildhood acute lymphoblastic leukemia.Leukemia 2(2):120-125 (1988).

39. Dolk H. Reply: Re: Cancer incidence near radioand television transmitters in Great Britain.I: Sutton Coldfield transmitter. Il: All high powertransmitters. [Letter]. Am J Epidemiol 147:91(1998).

40. McKenzie DR, Yin Y, Morrell S. Childhood inci-dence of acute lymphoblastic leukaemia andexposure to broadcast radiation in Sydney-asecond look. Aust NZ J Public Health22:360-367 (1998).

41. Selvin S, Schulman J, Merrill DW. Distance andrisk measures for the analysis of spatial data:a study of childhood cancers. Soc Sci Med34:769-777 (1992).

42. Szmigielski S. Cancer morbidity in subjects occu-pationally exposed to high frequency (radiofre-quency and microwave) electromagneticradiation. Sci Total Environ 180:-17 (1996).

43. Milham S. Increased mortality in amateur radiooperators due to lymphatic and hematopoieticmalignancies. Am J Epidemiol 127:50-54(1988).

44. Tynes T, Hannevik M, Andersen A, Vistnes Al,Haldorsen T. Incidence of breast cancer inNorwegian female radio and telegraph opera-tors. Cancer Causes Control 7:197-204 (1996).

-45. Kvale G, Heuch I, Nilssen S. Reproductive factorsand cancers of the breast and genital organs-are the different cancer sites similarly affected?Cancer Detect Prev 15:369-377 (1991).

Environmental Health Perspectives * Vol 107, Supplement 1 * February 1999 167

J.M. ELWOOD

46. Armstrong B, Theriault G, Guenel P, DeadmanJ, Goldberg M, Heroux P. Association betweenexposure to pulsed electromagnetic fields andcancer in electric utility workers in Quebec,Canada, and France. Am J Epidemiol140:805-820 (1994).

47. Grayson JK. Radiation exposure, socioeconomicstatus, and brain tumour risk in the US AirForce: a nested case-control study. Am JEpidemiol 143:480-486 (1996).

48. Thomas TL, Stolley PD, Stemhagen A, FonthamETH, Bleeker ML, Stewart PA, Hoover RN. Braintumour mortality risk among men with electri-cal and electronic jobs: a case-control study.J NatI Cancer Inst 79:233-238(1987).

49. Hayes RB, Brown LM, Pottern LM, Gomez M,Kardaun JWPF, Hoover RN, O'Connell KJ,Sutzman RE, Javadpour N. Occupation and riskof testicular cancer: a case-control study. IntJ Epidemiol 19:825-831 (1990).

50. Demers PA, Thomas DB, Rosenblatt KA,Jimenez LM, McTiernan A, Stalsberg H,Sternhagen A, Thompson WD, McCrea CurnenMG, Satariano W, et al. Occupational exposureto electromagnetic fields and breast cancer inmen. Am J Epidemiol 134:340-347 (1991).

51. Cantor K, Stewart P, Brinton L, Dosemeci M.Occupational exposures and female breastcancer mortality in the United States. J OccupEnviron Med 37:336-348 (1995).

52. Holly EA, Aston DA, Ahn DK, Smith AH.Intraocular melanoma linked to occupationsand chemical exposures. Epidemiology 7:55-61(1996).

53. Garson OM, McRobert TL, Campbell LJ,Hocking B, Gordon I. A chromosomal study ofworkers with long term exposure to radiofre-quency radiation. Med J Aust 155:289-292(1991).

54. Garaj-Vrhovac V, Fucic A, Horvat D. The correla-tion between the frequency of micronuclei andspecific aberrations in human lymphocytesexposed to microwave radiation in vitro. MutatRes 281:181-186 (1992).

55. Barron Cl, Baraff AA. Medical considerations ofexposure to microwaves (radar). JAMA168:1194-1199 (1958).

56. Djordjevic Z, Kolak A, Stojkovic M, Rankovic N,Ristic P. A study of the health status of radarworkers. Aviat Space Environ Med 50:396-398(1979).

57. De Guire L, Theriault G, Iturra H, Provencher S,Cyr D, Case BW. Increased incidence of malig-nant melanoma of the skin in workers in atelecommunications industry. Br J Ind Med45:824-828 (1988).

58. Vagero D, Ahlbom A, Olin R, Sahisten S. Cancermorbidity among workers in the telecommunica-tions industry. BrJ Ind Med 42:191-195 (1985).

59. Tornqvist S, Knave B, Ahlbom A, Persson T.Incidence of leukaemia and brain tumours insome "electrical occupations." Br J Ind Med48:597-603 (1991).

60. Beall C, Deizell E, Cole P, Brill I. Brain tumoursamong electronics industry workers. Epidemi-ology 7:125-130 (1996).

61. Lilienfeld AM, Tonascia J, Tonascia S, LibauerCA, Cauthen GM. Foreign Service Health Status

Study-Evaluation of Health Status of ForeignService and Other Employees from SelectedEastern Posts. NTIS PB-288163. Washington:State Department, 1978.

62. Goldsmith JR. Epidemiologic evidence ofradiofrequency radiation (microwave) effects onhealth in military, broadcasting, and occupa-tional studies. Int J Occup Med Environ Health1:47-57 (1995).

63. Elwood JM. Critical Appraisal of Epidemio-logical Studies and Clinical Trials. 2nd ed.Oxford:Oxford University Press, 1998.

64. U.S. Preventive Services Task Force. Guide toClinical Preventive Services: An Assessment ofthe Effectiveness of 169 Interventions.Baltimore:Williams & Wilkins, 1989.

65. Morgenstern H. Uses of ecologic analysis inepidemiologic research. Am J Public Health72:1336-1344 (1982).

66. Last JM, ed. A Dictionary of Epidemiology. 3rded. New York:Oxford University Press, 1995.

67. Lubin JH, Linet MS, Boice JD, Buckley J,Conrath SM, Hatch EE, Kleinerman RA, TaroneRE, Wacholder S, Robison [L. Case-controlstudy of childhood acute lymphoblasticleukemia and residential radon exposure.J NatI Cancer Inst 90:294-300 (1998).

68. Hill AB. The environment and disease: associa-tion or causation? Proc R Soc Med 58:295-300(1965).

69. Weed DL. Causal and preventive inference. In:Cancer Prevention and Control (Greenwald P,Kramer BS, Weed DL, eds). New York:MarcelDekker, 1995;285-302.

168 Environmental Health Perspectives * Vol 107, Supplement * February 1999