Arthur Villeneuve, peintre barbier / Arthur Villeneuve / Arthur ...
60Hz · 2020. 4. 7. · Name: Paul Joseph Villeneuve Degree: Ph. D., Epidemiology Year: 2000 Thesis...
Transcript of 60Hz · 2020. 4. 7. · Name: Paul Joseph Villeneuve Degree: Ph. D., Epidemiology Year: 2000 Thesis...
Cancer among Ontario electric utility workers: the evaluation of alternate
indices of exposure to 60Hz electric and magnetic fields
Paul Joseph Villeneuve
A thesis submitted in conformity with the requirements for the degree of Ph.D. (Epidemiology)
Graduate Department of Public Health Sciences The University of Toronto
O Copyright by Paul Joseph Villeneuve 2000
National Library B * 1 of Canada Bibliothèque nationale du Canada
Acquisitions and Acquisitions et Bibliographie Services services bibliographiques
395 Wellington Street 395. rue Wellington OttawaON K 1 A W Octawa ON K1A ON4 Canada Canade
The author has granted a non- exclusive licence allowing the National Library of Canada to reproduce, loan, distribute or sell copies of this thesis in microform, paper or electronic formats.
The author retains ownership of the copyright in this thesis. Neither the thesis .nor substantial extracts fiom it may be printed or othexwise reproduced without the author' s permission.
L'auteur a accorde une licence non exclusive permettant à la Bibliothèque nationale du Canada de reproduire, prêter, distribuer ou vendre des copies de cette thèse sous la forme de microfiche/film, de reproduction sur papier ou sur format électronique.
L'auteur conserve la propriété du droit d'auteur qui protège cette thèse. Ni la thèse ni des extraits substantiels de celle-ci ne doivent être imprimés ou autrement reproduits sans son autorisation.
Name: Paul Joseph Villeneuve Degree: Ph. D., Epidemiology Year: 2000 Thesis title: Cancer among Ontario electnc utility workers: the evaluation of altemate indices
of exposure to 60Hz electric and magnetic fields Department: Department of Public Health Sciences
The University of Toronto
Abstract
Three research objectives were addressed within this thesis. First, the relationships
between indices of 60 Hz electric and magnetic tield exposures measured in Ontario electric utility
workers were evaluated to identiQ a series of independent exposure metrics for cancer risk
assessment. Second, using these exposure indices, cancer nsks were estimated using data fiom a
nested case-control study of Ontario Hydro workers assembled as a component of a previous Tri-
Utility Study. Cancers examined included: leukemia, brain cancer, malignant melanoma and Non-
Hodgkin's lymphoma. Third, a simulation study was undertaken to investigate the gains in power
that could be achieved by perfurming additional sampling to reduce the variance of the mean
exposure estimate assigned to each job category, or alternatively. by increasing the number of
cases and controls.
The principal components and correlational analyses revealed that aspects of field
strength. other than those of a centrai tendency, should be examined when perîorrning cancer risk
assessment in this population. In particular, metrics that capture fluctuations in tield strength, and
exposures above threshold cut-points warrant consideration. Moreover, the poor correlations
observed between electric and magnetic field exposure indicate that risk assessment should be
undertaken separately for both field entities.
The percentage of time spent above an electiic field within the range of 10-39 V/m was
found to be associateci with an increased risk of leukernia and Non-Hodgkin's lymphoma over and
above associations explained by either the geometnc or atithmetic mean. Duration of employment
was associated with an elevated tisk of leukemia and malignant melanoma. Magnetic field
exposures were not significantly associated with any cancers exmined. however, anaiyses of brain
cancer cases were limited by small sample sizes. Taken together, the results are consistent with
the hypothesis that electric fields act as a promoting agent in the process of carcinogenesis of
adult leukemia and Non-Hodgkin's lymphoma.
The simulation study reveaIed that greater improvements in power were realised by
increasing the number of cases and controls rather than by conducting additional exposure
assessments on an equivalent number of workers. No appreciable gains in power were achieved
by sampling more than 15 workers in each occupational grouping.
Acknowledgements
I am indebted to my thesis committee for the expertise that each contributed to this research. My supervisor Paul Corey provided invaluable guidance and biostatistical expertise not only for this thesis research, but also for other projects that 1 worked on during the past five years. 1 am thankfiil that we were ofien able to discuss the intricacies of simulation analyses and the neutral-zone trap during the sarne meeting. 1 am gratefid for the enthusiasm that David Agnew brought to the project as well as his willingness to explain the physics of electric and magnetic field exposures many times over. Jim Purdham's knowledge of occupational epidemiology and his insightful suçgestions enhanced the quality of this research.
Dr. Anthony Miller, who was an investigator on the original Tri-Utility Study, was kind enough to provide detailed and constructive comments in an exceptionally timely fashion. I was also fortunate to receive input fiom Lois Green of the Ontario Power Generation Corporation. Stuart Kramer of the Ontario Power Generation Corporation extracted the raw data fiorn the personal monitoring devices.
Thank-you to my examiners (Jim Heller, Jacek Kopec, Julia Knight, Andrea Sass-Kortsak and Don Wigle) for providing helpfül cumments and criticisms that greatly improved the final version of this thesis.
1 have been fortunate to have had tremendous support fiom my parents and sisters throughout this journey. They amaze me. Lee, thanks for the sage advice from sorneone who has sone through the same process before me.
1 would like to thank my colleagues in Ottawa, particularly Yang Mao of Health Canada who introduced epidemiology to me many years ago and to Dan Krewski at the Department of Epidemiology and Community Medicine at the University of Ottawa who patiently accommodated me while 1 have been finishing up my doctoral studies.
A final thank-you to my many fnends in Ottawa and Toronto who have ensured that this epidemioloçist got out to play.
Table of Contents
Abstract ................................................................ ii Acknowledgements ........................................................ iv
ListofTables .......................................................... viii
ListofFigures ......................................................... Uii
.................................................. Chapter 1 . Introduction 1
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1 .1 . Overview 1
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.2. Electric and magnetic fields 3
1.3 . Expenmental midies of cancer and electric and magnetic fields . . . . . . . . . . . 6
1 .3.1. Multistage carcinogenesis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
1.3.2 . 111 vitro studies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1 -3 .3 . Ir1 vivo studies 14
. . . . . . . . . . . . . . . . . . . . . . . . . . . 1 3 .4. Summary of experimental findings 17
1.4. Epidemiologic studies of cancer and occupational exposure to electric
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . and magnetic fields 19
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . - - - . . 1.4.1. Leukemia --- . . A 21
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.4.2. Brain cancer 29
1.4.3. Non-Hodgkin's lyrnphoma . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
1.4.4. Malignant melanoma . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
. . . . . . . . . . . . . . . . . . . . . . . . . . 1 .4.5. Summary of epidemiologic findings 37
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.5. Study rationale 38
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.6. Research objectives 40
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Chapter 1 : References 41
Chapter 2 . Correlations between indices of electric and magnetic field esposurcs ............................. in Ontario electric utility worken .. 60
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.1. Abstract 60
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.2. Introduction ... ... 61
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.3. Methods 64
2.4. ResuIts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69
2.5. Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75
Chapter 2: References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84
Tables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 89-103
Chapter 3 . Risk of cancer amoag Ontario electric utility workers for selected indices OC exposure to 60- electric and magnetic fields ............ 104
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.1. Abstract 104
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.2. Introduction 107
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.3. Methods 109
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.3.1. Case ascertainment 109
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3 -3 .2 Control selection 1 10
. . . . . . . . . . . . . . . . . . . . . . . . . 3 -3 .3 Electric and magnetic field exposure 1 10
. . . . . . . . . . . . . . . . . . . . . 3 -3 .4 Construction of the job exposure matrices 1 12
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3 -3 .5 Occupational confounders 1 15
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3 .3 .6 . Statistical analyses 1 17
3.4. Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 120
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.4.1. Leukemia 121
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.4.2. Brain Cancer 123
. . . . . . . . . . . . . . . . . . . . . . . . . . . 3.4.3. Non-Hodgkin's lymphoma .. .. 125
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3 .4.4. Malignant melanoma 127
3 .5 . Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 128
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.6. Conclusions 140
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Chapter 3 : References 141
Tables: . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 147-172
Chapter 4 . Variations in exposure between and within workers: implication for power calculations within a matched casocontrol design .................. 173
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.1. Abstract 173
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.2. Introduction ... 174
4.3. Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 177
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.4. Results 183
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.5. Discussion 185
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Chapter 4: References 191
Figures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 192-194
Tables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 195-202
Appendix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 203-208
Chapter 5 . Conclusions and recommendations for further research ............. 209
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Chapter 5 : References 212
vii
List of tables
Page
Chapter 2.
Table 2-1 List of electric field exposure indices used in the first stage of principal components analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 89
Table 2-2 Mean daily exposure (in V/m) for traditional measures of electric fields, by
occupational group, Ontario electric utility workers . . . . . . . . . . . . . . .. . . . 90
Table 2-3 Mean daily exposure (in V/m) for traditional measures of electnc fields. by work- site location, Ontario electric utility workers . . . . . . . . . . . . . . . . . . . . . . . . . 9 1
Table 2-4 Indices of exposure selected to represent each principal component factor axis for
'non-traditional' electric field exposures . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92
Table 2-5 Correlation coefficients obtained fiom principal component analyses using selected electric field metrics and traditional measure of exposure . . . . . . . . . . 93
Table 2-6 Correlation matrix of selected electric field metrics and traditional measures of electric field exposures among Ontario electric utility workers . . . . . . . . . . . 94
Table 2-7 Summary of principal components analyses of electric field metrics. by occupational title . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95
Table 2-8 List of magnetic field exposure indices used in the first stage of principal . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . components analyses 96
Table 2-9 Average daily exposure (in PT) for traditional summaries of magnetic fields, by . . . . . . . . . . . . . . . . . . . occupational group, Ontario electric utility workers 97
Table 2- 1 O Mean daily exposure (in PT) for traditional measures of magnetic fields, by work
. . . . . . . . . . . . . . . . . . . . . . . . . site location, Ontario electric utility workers 98
Table 2- 1 1
Table 2- 12
Table 2- 13
Table 2- 14
Table 3- 15
Chapter 3.
Table 3- 1
Table 3-2
Table 3-4
Table 3-5
Table 3-6
Indices of exposure selected to represent each principal component axis for 'Non- traditional' magnetic field exposures among a sample of Ontario electric utility
workers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99
Correlation coefficients obtained fiom Principal component analyses of selected
magnetic field metrics and traditional measures of exposure among Ontario electric . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . utility workers 100
Correlation matrix of selected magnetic field metrics and traditional summaries of
mapetic field exposure, Ontario eiectric utility workers . . . . . . . . . . . . . . . 10 1
Summary of principal components analyses of magnetic field metrics, by . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . occupational titie 102
Pearson correlation coefficients between traditional electric and magnetic field
exposures, Ontario electric utility workers . . . . . . . . . . . . . . . . . . . . . . . . . 103
Characteristics of Ontario Hydro workers, by case control status . . . . . . .
Pearson correlation coefficients for selected electric field exposure indices,
Ontario Hydro electric utility workers . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Pearson correlation coefficients for selected magnetic field exposure indices,
Ontario Hydro electric utiIity workers . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . Risk of leukemia for selected indices of average electnc field exposures 150
Risk of leukemia for selected indices of average rnagnetic field exposures . . 15 1
Leukemia standardized coefficients obtained fiom the conditional logistic mode1
containing terms for duration of employment, percentage of time spent above
electric field threshoid and average arithetic mean field exposure . . . . . . . 152
Table 3-7
Table 3-8
Table 3-9
Table 3- 10
Table 3-1 1
TabIe 3 - 12
Table 3- 13
Table 3 - 14
Table 3-1 5
Table 3- 16
Leukemia standardized coefficients obtained fiom the conditional logistic model
containing ternis for duration of employment, percentage of time spent above . . . . . . . . . . . . magnetic field threshold and arithrnetic mean field exposure 153
Risk of leukemia for selected indices of electric field exposure, by length of
employrnent, Ontario electric utility workers . . . . . . . . . . . . . . . . . . . . . . . . 154
Risk of leukemia for selected indices of electric field exposure by petiod of
exposure among Ontario electric utility workers ernployed for at least 20 years . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 155
Risk of brain cancer for selected indices of average electric field exposures . 156
Risk of brain cancer for selected indices of average magnetic field exposures 157
Brain cancer standardized coefficients obtained fiom the conditional logistic modei
containing tenns for duration of employment, percentage of time spent above electric field threshold and average arithrnetic mean field exposure . . . . . . . 158
Brain cancer standardized coefficients obtained fiom the conditional logistic model
for brain cancer containing terrns for duration of employment, percentage of time spent above magnetic field threshold and arithmetic mean field exposure . . . 159
Risk of brain cancer for selected indices of electric field exposure by period of
exposure among Ontario electric utility workers employed for at least 20years . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 160
Risk of Non-Hodgkin's lyrnphoma for selected indices of average electric field
exposures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 161
Risk of Non-Hodgkin's lymphoma for selected indices of average magnetic field exposures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 162
Table 3- 17
Table 3- 18
Table 3- 19
Table 3-20
Table 3-2 1
Table 3-22
Table 3-23
Table 3-24
Table 3-25
Table 3-26
NHL standardized coefficients obtained fiom the conditional logistic model
containing terms for duration of employment, percentage of time spent above
electric field threshold and average arithmetic mean field exposure . . . . . . . 163
NHL standardized coefficients obtained fkom the conditionai logistic model
containing terms for duration of employment, percentage of time spent above
magnetic field threshold and anthmetic mean field exposure . . . . . . . . . . . . 164
Risk of NHL for selected threshold indices of electric field exposure, by length of . . . . . . . . . . . . . . . . . . . . . . . . employment, Ontario electric utility workers 165
Risk of NHL for selected indices of electric field exposure by period of exposure
among Ontario electric utility workers employed for at least 20 years .. . . . 166
Risk of malignant melanoma for selected indices of average electric field exposures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 167
Risk of maiignant melanoma for selected indices of average magnetic field
exposures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 168
Malignant melanoma standardized coefficients obtained fiom the conditional
logistic model containing terms for duration of employrnent, percentage of time
spent above electric field threshold and average aritiunetic mean field
exposure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 169
Malignant melanoma standardized coefficients obtained from the conditional
logistic model containing terms for duration of employment, percentage of time
spent above electric field threshold and average arithmetic rnean field
exposure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 170
Risk of malignant melanoma by duration of ernployment, Ontario electric utility
workers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 171
. . . . . . . . . . Effect of historical corrections on risk estimates, by cancer site 172
Chapter 1.
Table 4- 1 :
Table 4-2A:
Table 4-2B:
Table 4-3 :
Table 4-4:
Table 4-5 :
TabIe 4-6:
Table 4-7:
Table 4A- 1 :
Table 4A-2:
Model assumptions for simulated populations of workers . . . . . . . . . . . . . 195
Exposure estimates, variance components and the Intraclass Correlation
Coefficients for the arithmetic mean electric field intensity, by job-title, Ontario Hydro workers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . - . . - . . . - 196
Exposure estimates, variance components and the Intraclass Correlation
Coefficients for the arithmetic mean electric field intensity, by job-title, Ontano
Hydro workers (logged exposures) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 197
The mean dope and standard error of the dope obtained fiom simulating 10,000
matched case control studies with P={O, 0.263, 0.526) . . . . . . . . . . . . . . . 198
Power of detecting an effect with an Intraclass correlation coefficient of p O . 5 and
for sampling workers daily over a 5-day work week. . . . . . . . . . . . . . . . . . . 199
Power of detecting an effect with an Intraclass correlation coefficient of p=0.2 and for sampling workers daily over a 5-day work week . . . . . . . . . . . . . . . . . . 200
Power of detecting an effect with an lntraclass correlation coefficient of p=0.8 and for sampling workers daily over a 5-day work week . . . . . . . . . . . . . . . . . . . 20 1
Power of detecting an effect with an lntraclass correlation coefficient of p=OS and for sampling workers daily over a 5-day work week . . . . . . . . . . . . . . . . . . . 202
Mean exposure estimates, variance components and the Intraclass Correlation
Coefficients for the percentage of time above 20 V/m, by job-title, Ontario Hydro
workers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 204
SAS program used to generate simulated data . . . . . . . . . . . . . . . . . . 205-208
xii
List of figures
. . . . . . . . . Figure 4- 1 : A flowchart of the simulation program used to estimate power 192
Figure 4-2: Sample size as a fûnction of required study power and standard deviation of the
mean exposure to electnc fields within each of the two cells of the job exposure
matrix (50 exposure samples taken) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 193
Figure 4-3 : Sample size as a function of required study power and standard deviation of the
mean exposure to electric fields within each of the two cells of the job exposure
matnx ( 100 exposure samples taken) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 194
Chapter 1.
Introduction
1.1. Ovewiew
The possibility that exposure to power fiequency electric and magnetic fields may cause
adverse hedth effects was first raised by reports of various health probkrns among workers at
high voltage substations [ I l . In 1979, Wertheimer and Leeper [2] suggested that relatively weak
levels of 60 Hz magnetic fields may be associated with an increased risk of childhood leukemia
thereby spurring research in the area of electromagnetic fields and cancer. Though several
epidemiologic studies of electrk and magnetic field exposures in workers have since been
undertaken, the findings have been inconsistent and whether such exposures cause cancer in
humans rernains controversial.
Magnetic and eiectric fields are compiex quantities that can be characterised by their
fiequency, waveform, polarization and amplitude 131. As a consequence, exposure can be
assigned in rnany ways. Initially, occupational studies that investigated the rektionship between
cancer and exposure to electric and magnetic fields were bas& on job titles that lacked direct
measures of exposures [4-71. More recently, the availabiiity of monitoring devices that capture
field exposures in the persona1 environment has enabled studies to employ more renned methods
of exposure assessrnent [8- 101. These monitors have typidy been used to obtain field measures
in a sarnple of the current workforce which are in tum used to constnict job exposure matrices
Chapter 1 Introduction 1
(JEMs) to derive cumulative time weighted average (TWA) magnetic or electric field exposures
for the subjects under study. These cumulative exposure indices have invariably been based on
the geometnc or arit hmetic mean field strengt h calculated from daily exposure surnrnanes
obtained From the monitoring devices. However. as mentioned previously, field exposures can be
measured in many different ways and examples of alternative exposure indices include: the
percentaçe of time spent above a specified field threshold. the standard deviation of the field
strength. and fluctuations in the field strength over successive time intervals. Although if1 v~tro
and i i ~ vivo studies support a role for several of these alternate definitions. they have not
established which. if any. is the most biologically relevant metnc. Indeed, the inability to
adequately characterise relevant exposure may have contributed to the absence of an association
between electric and magnetic fields with cancer in some studies. Few studies have examined the
relation between these alternate indices of rnagnetic field exposure and cancer [ 10- 131. and to
date. there has been no published work that has evaluated the relevance of alternate indices of
electrk fields.
The above points underscore the need to perform cancer risk assessrnent using altemate
measures of electric and magnetic field exposure in a population having heterogeneous exposures.
This task forms the primary objective of this thesis. The initial thesis chapter serves to provide an
overview of electric and magnetic fields and additionally, summarizes the pertinent experimentai
and epidemiological research of these exposures as they relate to cancer. Thereafter. the thesis is
composed of three distinct. yet interrelated, sets of analyses. The first, examines the relationships
Chapter 1 Introduction
3
between a variety of electnc and magnetic field indices measured in a sample of Ontario electric
utility workers in order to identie a suitable series of indices to mode1 cancer risk (Chapter 2) .
This cancer risk assessment forms the basis of the second analysis stage (Chapter 3). The third
and final analysis consists of a simulation study designed to evaluate how power can be improved
by either obtaining a more reliable estimate of mean exposure by sampling from a greater number
of workers, or by increasing the number of cases and controls (Chapter 4). These three sets of
analyses are presented within this thesis as separate manuscripts in successive chapters.
1.2. Electric and magnetic fields
The generation, transmission or use of electricity results in the creation of electric and
maçnetic fields due to the presence and motion of electric charges. These fields are vector
quantities characterised by a magnitude, direction and fiequency of the source [3]. The frequency
is the rate at which the field changes direction and is usually given in Henz (Hz) where 1 Hz is
one cycle per second. The frequency (f) and the wavelength (A) are related by the following
equation:
A = (speed of light) 1 f
Therefore. as the frequency increases. the wavelength shonens. Electnc power systems in the
United States and Canada operate at 60 Hz while 50 Hz is used elsewhere including Europe. The
term "power frequency fields" is commonly used to refer to the sinusoidal electric and magnetic
fields produced by 50160 Hz lines and devices. Similarly. the term "extremely low frequency
fields" refer to fields in the range >O - 300 Hz.
C hapter 1 Introduction
The electromagnetic field can be divided into the near-field and far-field regions. For
power frequency fields generated by antennas (e.g., transmission lines), the near-field region is
reactive and, in this region, electric and magnetic field components store energy without
contributing to radiation. The energy is periodically transferred between the antenna and the near
field. The reactive near-field eaends from the antenna up to the distance D, = / 2 ~ . Power
frequency fields operating at 60 Hz have a wave length of 5000 km. Beyond this distance of the
near-field (D,), electric and magnetic fields are interrelated with each other and by measuring
either the electric or magnetic field the other is readily calculable. However, within the near-field,
power frequency electric and magnetic fields Vary independently and they must be assessed
separately [ 141.
Electric fields are created by the presence of an electric charge. The electric charge on a
conductor creates an electnc field in the space around the conductor. The electric field intensity
E (a vector quantity) is defined as the force per unit charge exerted on a test charge at a point in
space. The electric potential at a point (P) in space (V) is defined by the following integral:
Electric field measures are expressed in volts per metre (V/m).
Chapter 1 Introduction
Magnetic fields are created by the motion of electric charges. Magnetic field currents
within an electrical system or used by equipment Vary widely and fiequently experience
instantaneous, hourly, daily or seasonal variability. Magnetic fields are measured in units of
amperes per metre (A/m), or more commonly in units of microTesla (CrT) or milligauss (mG) ( 1
PT= 10 mG). Unlike electric fields, magnetic fields are only present when an electrical device is
operating and are able to easily penetrate buildings. trees. tissue and other objects [ 151.
For both electric and magnetic fields, the magnitude of the field varies inversely with the
n" power of the distance to the source (D,). Specifically, E = Eo 1 D: (or M = Mo 1 D: ) where
E, (or Mo) is the magnitude of the field at the source. The actual rate of the drop-off (n) depends
on the geometry of the source. For a single long conductor, or at a long distance fiom the
niultiple conductor line, rl equals one. Alternatively, when the source is near a transmission or
distribution line rl equals two. For small sources such as appliances, n is assumed to be three.
In addition to electric and magnetic fields produced through the generation of electricity
there are several natural sources. Unlike power fiequency fields, which consist of 50/60 HZ
alternating current (AC) fiequencies, the Earth produces mainly static (or DC) fields. EIectric
fields are produced by thunderstorm activity whereas magnetic fields are thought to be generated
by electric currents within the Earth's core. Magnetic fields are also associated with deposits of
maçnetic ore. living cells and nerve impulses [Ml. Alternating current fiequencies are able to
induce currents within the human body.
Chapter 1 Introduction
6
There are certain occupations that have higher levels of exposure to magnetic and electnc
fields. These include flame cutters/welden, powerline maintainers, power station operators. sheet
metal workers. sewing machine operators, motion picture projectionists. telephone cable splicers
and electric power electricians. Workers in the electric utility industry can have electric and
magnetic field exposures that are, on average, 5- 1 O times higher than most other occupations or
residential levels [ 171. Further. exposures between workers employed in electric utilities are
typically heterogeneous with a subset of workers that receive background levels of exposure [15] .
1.3. Experimental studies of cancer and electric and magnetic fields
A large number of experimental studies of electric and magnetic fields have been camed
out to determine which, if any, stages of carcinogenesis rnight be affected by these exposures.
The current knowledge indicates that the combination of the geomagnetic field, AC magnetic
fields. and electric fields and currents induced or in the body may affect biological processes [18].
This section provides a briefoverview of the multistage mode1 of carcinogenesis.
Thereafter, the results of iri vitro and ;ri vivo studies that examined the bioeffects of magnetic and
eiectric fields as they relate to cancer are summarized. i r r vitro studies encompass those
experiments undertaken in cells or explanted tissue exposed and assayed outside the orçanism. In
contrast. ii, vivo studies involve experiments done in living animals or humans. Ir] vitro studies
are often used to provide the basis for the design of iri vivo studies by defining critical endpoint
and electromagnetic field parameters.
Chapter 1 Introduction
1.3.1. Multistage carcinogenesis
Carcinogenesis has been viewed as a process that c m be muttifactorial in nature and
encornpass multiple steps [19]. The traditional multistage model consists of an initiation and
tumour promotion stage [20,2 11, though some foms of this model have been extended to include
the processes of conversion and progression [22]. This section briefly describes the stages of this
model.
In the traditional multistage model, initiation represents the first stage and is considered to
involve cytogenetic damage that results in mutations to the genetic information carried in the
DNA [22]. Initiation can occur as a result of a single brief exposure to an initiating agent and
once this event has occurred. genetic damage is passed on to daughter cells during subsequent ce11
divisions. This stage includes the actions of inherited mutations, cancer causing viral genes and
altered cellular genes (oncogenes) [19]. Tumour initiators are generally thought to have a linear
dose-response relationship and an ail-or-none effect 1231.
The second stage, promotion, is considered to be a process that involves the application of
a further stimulus to an initiated cell to cause its progression to a fùlly developed malignancy [22].
Typically, promotion is an extended process that requires prolonged or repeated exposure to the
promoting agent [223. The exact length of this phase can be difficult to assess as it requires
knowledçe of when exposure to the initiating agent first occurred. Cancer promoters are also
characterized by the existence of a threshoId level or dose and reversibitity of effects [24]. These
Chapter 1 Introduction
agents can be systemic such as hormones or other growth enhancing factors or exogenous.
There are many processes that can perturb normal cell growth mechanisms. Detectable
effects of promoting agents include increased ce11 proliferation and effects on processes required
for cell growt h. In particular, certain enzyme activities may be altered. metabolic pat hways
stimulated, and sugar transport increased [22]. Co-promoting factors rnay also play a role in this
stage of carcinogenesis by enhancing the performance of cancer promoters. Continued exposure
to promoting or CO-promoting agents after turnour development rnay cause the tumour to evolve
with increased invasive and metastatic properties [19]. Therefore, tumour promoters may play a
role in the progression phase of the multistage model. The progression stage is characterised by
the "accumulation of hrther cellular changes, such as loss of growth control and the acquisition
of invasive behaviour" [22].
Recently, it has been suggested that the initiation-promotion-progression model does not
adequately describe the compfex interaction of protoxic and )toit-getto~oxic carcinogens [25 1.
Some have suggested that dividing the carcinogenic process into these genotoxic and non-
genotoxic phases represents a suitable alternative to the traditional forrn of the model [23, 261.
These models assume that initial DNA damage occurs and is followed by non-genotoxic
promotion and progression phases where each step involves a mutation [27]. In this context.
initiation is concerned with largely genotoxic events. while turnour promotion is associated with
molecular events resulting in an enhanced proliferation and selective growth of initiated cells.
Chapter 1 Introduction
9
This increased rate of proliferation may lead to an increased probability of a tumour developing
spontaneously [ B I .
The traditional multistage model has been used in the design of most irr vivo and irt virr.0
experiments of electric and magnetic fields. As a result, the following section interprets the
findings of these studies within the context of the traditional multistage model of carcinogenesis.
Nonetheless, it is done with the understanding that the biological mechanism(s) of action can Vary
across types of cancer and that the underiying models that descnbe the process of carcinogenesis
continue to be widely debated.
1.3.2. In vitro studies
Several investigators have examined changes in the fiinction of isolated cells in order to
investigate whether electric andor magnetic fields can cause cancer. Although it is difficult to
relate changes in cell fûnction directly to an increased incidence of cancer, such changes may
provide information on mechanisms thought to be reiated to the development of cancer. For
theoretical reasons, and because of findings from early studies. irt vitro research has focussed on
the tumour promoting potential of ELFs [18, 29, 301. Therefore, cellular studies have
concentrated on responses such as cell prolifération and other processes required for cell growth
and division. As before, these responses are represented by changes in enzyme activities,
activation of cellular signalling. synthesis of DNA and sugar transport [22]. In this section. the
results of i ! ~ viiro studies are surnmarized according to four main categories: genotoxicity, signal
Chapter 1 Introduction
tra~isduction. cell proliferation and impact on the immune response of cells.
(;et~oroxicity
Lt is widely acknowledged that the energy associated with ELFs is too low to cause direct
mutagenic damage to the DNA [3 1-33]. Nonetheless, exposure to ELFs could affect the
production of agents such as fkee radicals that may indirectly affect the structure of DNA. A few
studies have reported a higher incidence of chromosomal aberrations foIlowing exposure to
electric fields [34] or intermittent magnetic fields [35], and increased mutation rates in melanoma
cells have been observed after exposure to high magnetic field flux densities [36]. However, most
studies have not found evidence for DNA damage, chromosome aberrations, sister chromatid
exchange or the fidelity of chromosome segregation or micro nucleus formation 1371. For these
reasons. it is considered unlikely that ELFs are associated with the initiation or conversion stages
of carcinogenesis. Instead, it is generally accepted that if ELFs do affect carcinogenesis, it is
Iikely at the levei of promotion [ 18, 22, 381.
Chernical agents that fùnction as promoters do not themselves cause mutations but may
alter çene expressions or other processes that modi@ cellular metabolism and growth rates [29].
MYC', a gene that plays a central role in cell proliferation and apoptosis, is anomalously regulated
in the majority of cancers [39] and MYC transcription may be affected by magnetic field exposures
[40-421. The effects on gene expression appear to be dependent on fiequency. flux density and
exposure time [4 1, 421. However, other studies that have tried to replicate these findings have
Chapter 1 Introduction
been unsuccessfùl [4346]
L';g?tal rrm~sdlrct iorr
Different effects of ELFs on calcium ion homeostasis have been reported with the effects
differinç for specific frequencies and intensities [29]. Calcium ions are strictly regulated within ail
ce11 types and their flow across the ceIl membrane governs many physiologic processes such as
muscle contraction, egg fertilisation, hormone release and ceIl division [ 181. Initial studies that
have examined intracellular responses to ELF exposure in brain tissue observed changes in
calcium ion efflux between 10 and 30% [47,483. Both 60 Hz, 22 mT magnetic 1491 and 170 V/m
electnc field [50] exposures, in the presence of mitogens, have been show to increase the influx
of calcium in stimulated rat thyrnocytes. Other work has also found associations between ELFs
and calcium ion movements [5 1, 521 which could not be replicated elsewhere [53-551. Changes in
calcium ion homeostasis are important as they produce oxidative stress in the cell [56] and
prolonged stress has been linked to intracellular carcinogenic processes [57, 581. However. the
causal nature of these possible linkages is uncertain. Typically, irl vitro studies have appiied
exposures considerably higher than those that would be received by humans. Furthemore,
causality may not hold at the whole animai level because carcinogenesis is an irl vivo process that
depends of a vanety of physiologic factors, immune responses, phannacokinetic considerations
and metabolic effects [59].
Chapter 1 Introduction
12
Several studies have examined the relationship between magnetic fields and protein kinase
C (PKC) activity. Protein kinases fiequently fùnction as part of gene activity by transducing
signals from the ce11 surface to the cytoplasm and nucleus. PKC is believed to be the receptor for
turnour-promoting esters [60]. Monti and colleagues [6 1 1 found that exposing lymphocytes to a
50-Hz 8mT magnetic field increased the binding of PKC specific esters; this suggests that these
field exposures may be capable of modifjmg cellular responses to tumour promoters. Similar
findings have also been observed with electric field exposures [62].
Cd1 prolifer-atiott
One of the most important cellular responses elicited by known tumour promoting agents
is a sustained increase in the rate of proliferation. A 10- 15% increase in the cell cycle progression
of human lymphocytes was observed after exposure to a 5 mT. 50Hz magnetic field [63] which
was later confirmed in an independent replication [64]. Colony growth was also observed
following 10- 14 days of exposure to a 1 . 1 mT 60 Hz magnetic field [65] . As previously
rnentioned. 60 Hz electric fields have also been shown to increase DNA synthesis rates by 20%
and therefore, rnay affect cell proliferation [66]. Support for the proliferative capabilities of
pulsed electric fields comes with the observation of a 32% increase in fibroblast ce11 growth with
exposure to fields consisting of 1 -Hz, 1 -ms duration pulses, with a time-averaged current density
of 7 mAlm2 (peak current density 7 Nm2) [67]. Altered proliferation of cells in vitro has been
observed in a nurnber of studies, but in none were sham controls used 1391.
Chapter 1 Introduction
13
Some experiments using cancer cells have shown an enhanced effect on ceil growth
potential as measured by the activity of ornithine decarboxylase (ODC). ODC is an enzyme
essential to cell growth and is ofien activated in response to promoting agents [22]. For example.
ODC activity is markedly increased by phorbol esters which are tumour promoting compounds
[68] . Exposure to 60 Hz electric fields at levels as low as O. 1 mV/cm has been shown to increase
ODC activity in a varie5 of ceIl types [68]. Similarly, basal ODC activity has been stimulaîed in
mouse fibroblasts exposed to 55-65 Hz magnetic fields at flux densities between 1 - I OOpT [69].
Another study found that exposure to 50 Hz 0.1 mT magnetic fields increased ODC activity in
leukemic cell lines, but not in the non-leukernic lymphocyte line [70]. Elsewhere, increased
activity of ODC was optimised when the coherence of time varying magnetic field exposure was
maintained for a minimum period of time [69]. In contrast, other studies did not observe any
effects of magnetic field exposure on ODC activity [7 1. 721.
hpacr or1 imrrt~rrre resporrse of ceils
Phillips et al. 1731 observed an association between electnc fields and transferrin receptors
in human colon carcinoma cells. This receptor is correlated with the proliferation of normal and
maliçnant cells and associated with the receptor of natural killer cells (cytotoxic lymphocytes)
[37] . Experiments on canine and human leukocytes have not demonstrated significant effects of
ELF exposures on the immunologic tùnctions of normal or specifically immunised cells [ 161. On
the other hand, a dose-response relationship was observed between 60 Hz electnc field intensities
(0.1 - 1 O mV/m) and the suppression of cytotoxicity in mice T-lymphocytes [74]. This could
Chapter 1 Introduction
14
indicate one possible mechanism whereby an electrk field could inhibit an organism's resistance to
the carcinogenesis.
1.3.3. I n vivo studies
11, vitro studies have consistently shown that electric and magnetic fields are not
rnutagenic [3 1. 75-79]. As a result, animal studies of extremely low tiequency fields have been
desiçned to determine whether exposures may be cancer promoters. In this brief review. the
findings of these studies are summarised under three headings: animal studies of carcinogenesis,
effects on the immune system. and the influence of exposure to ELFs on circulating levels of
melatonin. For more detailed summaries of this body of research. the reader is referred elsewhere
[ l S , 291.
A l~imal sfudies of camNtogemsis
The results obtained in animal studies that exarnined the tumour promoting abilities of
electric and magnetic fields are equivocal. Rannug and CO-workers used two different chemically
induced turnour systems in rats and mice to examine the long-term effects of magnetic field
exposure on tumour development [SOI. No effects were noted with continuous exposure to 50 Hz
fields at 50 or 500 PT for 20 Wday over 2 years. However, a twofold increase in the incidence of
skin tumours was obsewed in the two species when they were exposed to intermittent fields; this
occurred when fields were applied 15s on and 15s off for 20 hours per day over 103 weeks.
Support for a promoting role of magnetic fields was found in a study of female rats [8 11 where
Chapter 1 Introduction
the animals that were exposed to either static or altemating magnetic fields accompanied by the
cancer initiat or nitrosomethylurea, showed both increased numbers and earlier apvearance of
mammary tumours. There was no such increase with exposure to magnetic fields alone. In later
studies that used a skin mode1 where mice were treated with the tumour initiator DMBA,
exposure to 2 mT magnetic fields at 60 Hz had no statistically significant effect on tumour
development [82, 831. On the other hand. magnetic fields (30mT) at 50 Hz were found to increase
the number of mammary tumours induced in rats by approximately 30% [84]. Leung and
colleagues exposed rats to 60 Hz electric fields for 180 days. The exposed and sham animals were
treated with a single dose of the initiator dimethylbenz(a)antracene. An increase in the number of
tumours per tumour bearing animal was observed [85]. Some studies support the theory that
ELFs serve as a tumour promoter [68, 84, 86, 871 whereas some do not [43, 88-90].
Itnmrurology
There is inconsistent evidence that ELFs affect the immune system. In an investigation of
the cellular changes of the immune system after exposure to 60 Hz electric fields between 150-
250 Vlm. no effects were observed [91]. However, animal studies have noted significant
decreases in the cytolytic capacity of radiofiequency fields [92] and the suppression of T-
lymphocyte activity afler exposure to 60 Hz electric fields [74].
Chapter 1 Introduction
16
Effects of EL F on cirada~ing rne/atorrirl
It has been suggested that ELFs may enhance carcinogenesis by reducing circulatory levels
of melatonin [58, 931. The mechanisms through which decreased levels of circulating melatonin
may enhance carcinogenesis include: the increased proliferation of stem cells at risk, the
stimulation of cancer cell proliferation and the impairment of immune fiinction [94]. Melatonin has
been shown to inhibit the growth of a wide range of cancers [95]. and to suppress tumour growth
in humans and animals [93, 96-98]. Melatonin also inhibits the secretion of estrogen and other
tumour promoting hormones [99- 10 1 j, enhances the immune system [97? 1021 and influences the
scavenging of tiee radicals [ 1 031. Melatonin is a powerfùl antioxidant that provides significant
protection for DNA against oxidative damage [58, 1041. Therefore, the reduction of melatonin as
a result of either electric or magnetic fields may increase the vulnerability of DNA to cancer
initiators or promoters [105]. Animal studies have demonstrated the ability of magnetic [ 1051 and
electric fields [ 106. 1071 to decrease levels of melatonin.
In humans, a recent analysis of 142 male Colorado electric utility workers found reduced
escretions of a melatonin metaboIite among those in the highest quartile of magnetic field
exposure relative to the lowest; light exposure was found to modiQ the magnetic field effect
[ 1081. Similar results were found among a group of Swiss railway workers [ 1091. The effects of
electric field exposures on melatonin levels were evaluated indirectly among individuals with
hiçher exposures due to the use of electric blankets [1 101. Overall, no significant differences in
melatonin concentrations were observed between those who used electric blankets and those who
C hapter 1 Introduction
did not.
1.3.4. Summary o f experimental findings
III ilitro strrdies
As a whole. the results fiom cellular studies indicate that biological effects are caused by
exposure to electric and magnetic fields. These effects include changes to ce11 hnctions
responsible for growt h, and perhaps, turnour promotion. However, t here remains uncenainty
regarding different links in the chain of signal transduction events and their interrelationships. In
addition. evaluating the effects of ELFs using cellular studies is difficult because commercial ceil
incubators are a known source of magnetic field exposure. Specifically, cultured celis may be pre-
esposed to magnetic or electric fields while in culture thereby biasing the results [29]. Attempts
to replicate cellular studies in independent laboratories has for the most part been unsuccessful
and exposures have typically been at much higher levels than humans would receive. Cellular
effects have been observed only at certain fiequencies, amplitudes or time periods suggesting that
effects may be observed only within certain exposure windows. At this time, NI vitro studies have
not demonstrated any coherent pattern of exposure-effect relationships. Therefore. the exposure
parameters that may be responsible for inducing the biological effects that lead to ce11 proliferation
and tumour promotion in initiated cells remain to be defined.
Chapter 1 Introduction
/t r vivo st tidies
The results obtained fiom animal studies of ELF and cancer are inconclusive. It shouId be
noted that most animal experiments were not undertaken to evaluate threshold doses of 50/60 Hz
electric and magnetic fields, and as a resutt, few have utilized exposure patterns at levels
commonly experienced by humans [ 191. Moreover, dosimetric patterns are dependent on body
shape and size, and therefore, the application of scaling factors between animais and humans is
not a straightforward procedure. A fiirther complicating factor in cancer studies of electric and
magnetic field exposure in anirnals is the possibility that dose-response features of field exposure
do not follow a Iinear fùnction, but rather exhibit windows in which the system displays enhanced
sensitivity [ I l Il. If such a window does exist, then the traditional way of using high doses to
identie carcinoçenic effects of a specific agent [112] would not be appropriate for studies of
ELFs. Juutilainen et al [26] in their review of animal studies and magnetic field exposures
indicated that the evidence suggests that field exposures may potentiate the effects of known
carcinogens only when both exposures are chronic. That is. effects are apparent when magnetic
fields and the known carcinogen interact repeatedly during a long-term experiment. I f true. the
power of animal studies to detect biological effects fiom electric and magnetic field exposures
would be low unless very large studies were undertaken.
Chapter 1 Introduction
19
(O~iclz~s~orr
In surnmary, the experimental evidence suggests that electric and /or magnetic fields may
exen a promoting role in the carcinogenic process. Epidemiologic studies that investigate the
relationship between ELFs and cancer should therefore incorporate features of promoting agents
into their exposure assessment efforts. Again, cancer promoters are characterised by the
existence of a threshold and require repeated exposure over a prolonged penod for tumour
promotion [24, 321.
1.4. Epidemiologic studies of cancer and occupational exposure to eelectric and magnetic fields
Since the initial report by Wertheimer and Leeper in 1979 [2], several epidemiologic
studies have been undenaken to investigate the relation between cancer and occupational
exposure to magnetic, and to a much lesser exTent, electric fields. The first generation of such
studies used crude methods to assess exposure. Exposure assessment in these studies did not use
direct measurements of field strength, but instead classified individuals according to job titles in an
intuitive manner [ I 131. It is possible that non-differential exposure measurement error was
introduced because the vanability of exposures were not adequately taken into account by
classiQing individuals within broadly defined occupational groups. In many instances, analyses
were perfonned using studies that were designed for some other purpose. Further. many of these
early studies did not collect information on potential confounding variables [ I 131.
Chapter 1 Introduction
More recently, a number of studies have been published that relied on performing
measurements for different categories of occupations which have then been applied to the work
histories of individual workers [8- 10, 12, 1 14- 1 161. Many of these studies sampled actual work
locations whereas others made use of personal monitoring devices using a representative sampie
of workers to estimate exposure for different job categories. The effect of historical changes in
exposure over tirne was made for but a few of these studies [9. 1 15. 1 171. These changes would
have occurred due to changes in work practices and the manner in which electncity was generated
and supplied by the utility. The majority of the studies that examined the relation between cancer
and electric and magnetic fields focussed on leukernia and brain cancer. A limited number of
studies have âiso examined the association between other cancers and exposure to electric and
magnetic fields. These cancers inciude breast cancer. lymphoma and malignant melanoma [30].
The relation between breast cancer and field exposures could not be assessed within the cohort of
Ontario Hydro workers due to the rarity of the disease in men and the size of the cohort.
Therefore. a review of breast cancer studies of ELF is not provided. However, the occupationa1
studies of malignant melanoma and Non-Hodgkin's lymphoma are summarised. As this section
details, the possibility that electric and magnetic fields are associated with these two cancers has
some support based on epidemiologic findings and expenmental evidence (see section 1.3). Non-
Hodgkin's lymphoma is also closely related to certain cases of chronic lymphocytic leukemia, and
a number of lymphomas terminate in a leukemic phase regardless of initial bone marrow
involvement [118]. The review focuses on those studies where exposure was assessed using
persona1 monitoring devices.
Chapter 1 Introduction
1.4.1. Leukemia
The first generation of occupational studies that examined the relationship between
leukemia and magnetic fields inferred exposures using job titles. The findings of these studies
were rnixed. Approximately one half of twenty studies published before 199 1 reported a modest
increase in leukemia risk among exposed populations that ranged fiom between 1.2 and 1.5
relative to those with background levels of exposure [113]; the remainder of the studies found no
association. Several of these studies reported positive findings for subjects with suspected
elevated magnetic field exposure; these included weak associations found for electrical engineers,
but virtually none for welders [113]. These studies cm be regarded as inconclusive because
direct measures of assessment were not performed and risk estimates were not adjusted for
several potential confounding variables. Other exposures thought to be related to leukemia that
may be present in the workplace include benzene, ionizing radiation and exposure to pesticides
[ l 191.
More detailed efforts to assign exposure were incorporated into the design of a Swedish
population-based study that examined associations between occupational exposure to magnetic
fields and leukemia [IO]. Analyses were performed on 250 leukemia cases identified between
1983 and 1987 and 500 individually matched controls. Exposure assessment was performed by
obtaining field measures for a sample of subjects based on the job they had held the longest during
the ten year penod preceding cancer diagnosis. A total of 1 ,O 15 measurements were obtained for
169 different job categories. A limited number of exposure metics were calculated that included
Chapter 1 Introduction
the arithrnetic mean, the median, the standard deviation and the proportion of t h e spent above
0.2 PT. A dose-response relationship between magnetic field exposure and chronic lymphocytic
leukernia (CLL) was observed. Subjects in the highest quartile of exposure had an odds ratio of
3 .O (95% CI= 1 -6-5 -8) relative to those in the lowest. Elevated risks of CLL were also observed
among subjects in the highest quartile of magnetic field exposure based on the tirne spent above
0.2 pT (OR=2.4 95% CI=1.3-4.3) and the standard deviation (OR=2.2. 95% CI=1.2-3.8) relative
to lowest quanile. No significant associations were observed between any of the indices of
magnetic field exposure and the occurrence of acute myeloid leukemia. The odds ratios were
adjusted for a variety of occupational confounders that included solvents, ionizing radiation and
urbadrural residence. Nonetheless, limitations of the study include exposure assessment that was
based only on recent occupational groupings and case information that was largely collected from
proxies (64% of al1 cases).
In 1993. Matanoski and colleagues reported on the relation between magnetic field
exposures and leukernia mortality in a cohort of telephone workers [ I l ] . Cases and controls were
identified frorn within 1.3 million active workers and 200,000 retirees of the American Telephone
& Telegraph Company. Analyses were based on 75 matched that consisted of one case that was
matched to three controls. Exposure assessment was performed by using persona1 monitoring
samples taken fiorn present-day workers in four categories of telephone line-jobs: cable splicers,
central office technicians, outside plant technicians and installers/maintenance/repair technicians.
Engineers were included amongst aH other workers not classified as lineman. The sampled
Chapter 1 Introduction
23
exposures were used to construct job exposure matrices using a TWA exposure as well as metrics
based the standard deviation, autocorrelations and peak exposures. A non statistically signiticant
increased risk was observed among those in the upper quartite of TWA arithmetic mean
cumdative exposure (OR=2.5, 95% CI=0.7-8.6) relative to those in the lowest quartile. Funher,
non-statistically significant associations between peak exposure scores above the median,
compared to below median scores, were observed among those that died fiom leukemia.
Specifically, the odds ratios were 2.4 (95% CI=0.7-9.0) and 6.6 (95% CI=0.7-58) for latency
periods of 10 and 15 years respectively. Acute myeloid leukemia accounted for the major@ of
the Ieukemia deaths. The study's drawbacks included scant information on occupational
confounders, most notably benzene, and a sizeable loss of subjects due to missing job records.
Leukernia risk was also examined using a case control study nested within a cohort of
3 6 , Z 1 electric utility workers in California [13]. Between 1960 and 1988. a total of 44 cases of
leukemia were identified using death certificates. A variety of magnetic field exposure metrics
were created based on persona1 monitoring data obtained fiom a sample of workers representing
3 5 occupational categories. The exposure metrics that were modelled inciuded the arithmetic
rnean. geometric mean, 9SLh and 99h percentiles. and the Fraction of time spent above I .O and 5.0
PT. Cumulative index scores derived using these metrics were not related to leukemia mortality.
Moreover, positive associations were not observed across a range of latency windows.
Unfortunately, the study lacked information on potential occupational confounders, relied on
decedent rather than incident cases, and could not identie deaths occumng outside of California
C hapter 1 Introduction
London et al. [12] examined the relation between the incidence of leukemia and magnetic
field exposures among male electrical workers aged 20-64 years in Los Angeles County. A total
of 2,355 cases and 67,2 12 controls were identified from the County's cancer registry between
1 972 and 1990. Of the 2,3 55 leukemia cases, 12 1 were employed in the electrical industry.
Subjects that had been diagnosed with a central nervous system tumour were not eligible as
controls. The last occupation held by the subject was used as a basis for exposure and was
determined by using the medical records from the cancer registry. An annual estimate of exposure
to magnetic tields was then calculated for electric occupations and a sample of non-electrical jobs
using a variety of metrics. Exposure indices that were examined included the arithmetic mean, the
time spent above 2.5 mG and the time spent above 25 mG. A weak positive relationship with
leukemia was observed with the arithmetic mean magnetic field exposure, specifically, for each 1
pT increase in exposure, the odds ratio was 1 -2 (95% CI= 1 .O- 1.5). Similarly, for each 10%
increase of time spent above 25 mG the risk of leukemia increased by 40% (OR= 1.4 95% CI= 1 .O-
2.1 ). These tisk estimates were adjusted for occupational exposure to known or suspected
leukemogens that included ionizing radiation, benzene, chlorinated hydrocarbons, other solvents
and pesticides. A notable weakness of this study was the reliance on a single job to estimate
exposure. In addition, to the extent that exposure to magnetic fields is related to other forms of
cancer that were diagnosed within the control population, the risks were underestimated. Finally,
as pointed out by the authors, if smoking was jointly related to cancer and the likelihood of
Chapter 1 Introduction
working in the electric industry, the results may be biased in either direction.
Many of the limitations of previous work were addressed with the nested case-control
study of workers fiom three large electnc utilities in France. Quebec and Ontario [9]. Cases were
ascertained between 1970 and 1989. Information was collected for several cancer sites that
included 140 leukemia cases and 546 matched controls. The job exposure matrices were
constructed separately for each o f the three utilities, though in each utility, field exposure was
estimated by measuring current exposures using personal dosimetry in a sample o f the current
workforce. Job exposure matrices were calculated for both the daily arithmetic and geometric
means. Further, these exposure metrics were corrected for hkto~ical changes in current loading
and changes in work practices. For each subject, a cumulative estimate o f exposure was
calculated by multiplying the mean exposure for each job by the length o f employment in each
job-titIe based on the work history of the cases and their corresponding controls. Incident cases
of cancer were identified from Company follow-up. hospitai records and registry data. An
elevated risk of nonlymphoid leukemia was observed among workers who had more than the
median cumulative exposure to magnetic fields (OR=2.4, 95% CI=l. 1-5.4). A similar increase in
risk was observed in the same exposure grouping for those diagnosed with acute myeloid
leukemia (OR=3.2, 95% CI= 1 -2-8.3). However, no clear dose-response patterns emerged and
risk estimates were dissimilar across the three utilities. It is difficult to interpret the differences in
the risk estimates across the three utilities because, unlike the two other utilities, the French
component o f the study excluded retired workers h m follow-up and the Quebec component
Chapter 1 Introduction
contributed a small number of leukemia cases (n=24).
A retrospective cohon mortality study was conducted using 138,905 workers employed
for at least six months at five electric utility companies in the United States [a]. Mortality follow-
up between 1950 and 1986 identified 20,733 deaths during 2.7 million person-years of follow-up.
A total of 164 deaths fiom leukemia were observed during this period. When death fiom leukemia
among these workers was compared relative to the general population, the standardized mortality
ratio was 0.76 (95%=0.64-0.88). When cornparisons were made within the cohort of utility
workers, duration of employment was associated with increased mortality fiom leukemia.
Magnetic field exposures were estimated for 28 occupational categories using personal
monitoring data obtained from a sample of workers using the AMEX personal monitoring device.
N o significant association was observed between the cumulative 'IWA magnetic field exposure
and death fiom leukemia as the odds ratio per PT-year was 1.03 (95% CI=0.95- 1 . I 1). The risk
estimates were adjusted for occupationai exposure tu polychlorinated biphenyls and solvents.
while exposure to ionizing radiation was believed to be insufficient to produce confounding. The
study was limited by the reliance on mortality rather than incident cases and data on electric field
exposures were not collected.
A Swedish population-based study was camed out to examine the relation between
leukemia and magnetic fields [120]. Cases and controis were identified fiom a cohort of
individuals, who were at least 16 years of age and had lived for at least one year on a property
Chapter 1 Introduction
Iocated within 300 m from any of the 200 and 400 kV power lines in Sweden. Approximately
400,000 such individuals were identified and a total of 325 cases of leukemia were identified
between 1960 and 1985 through a record linkage to the Swedish cancer registry. The occupations
of the cases and controls were detennined using data fiom the censuses performed by Statistics
Sweden every five years. Exposure assessment was based on the occupation heid during the last
census year before diagnosis and was based on a job exposure matrix (JEM) derived from another
population of working males [12 11. Residential exposure was defined by the magnetic fields
çenerated by power lines close to the house and estimated exposures took into account height
between towers, distance between towers, mean load on the power line and the distance between
the home and power line. An odds ratio of 1.7 (95% CI=1.1-2.7) was observed among those
having occupational exposures exceeding 0.2 PT. Arnong subjects having high residential and
occupational exposures, the risk was much more pronounced (OR=3.7, 95% CI=1.5-9.4). The
study was lirnited in that there were few subjects having information on both occupational and
residential exposures, electric fields were not measured, exposure assessment did not take into
account changes in jobs, and fùrther, for a large percentage of subjects occupational exposure
data were lacking.
Most recently. the findings from a retrospective cohort study among Danish electric utility
workers were reported [122]. A total of 32,006 men and wornen who had been employed at any
of the 99 electric utilities in Denmark for at least three months forrned the cohort. Employrnent
history dating back to 1909 was obtained fiom work records in the electric utilities. Vital status
Chapter 1 Introduction
28
was ascertained through record linkage to the National Death Certificate files for the years 1968-
1993. A total of 60 deaths fiom leukemia were observed. Exposure was assigned using a job
exposure matrix and was calculated using 127 magnetic field measurements taken from an earlier
study [ I Z ] to represent 475 combinations of job titles and work areas. Standardized leukemia
mortality ratios (SMRs) for these workers relative to the Danish male population were calculated.
No association between leukemia and exposure to magnetic fields was observed (SMR=0.9. 95%
CI=O.7- 1.2). Again, the findings were limited by the lack of data collected for potential
confounding variables and a less comprehensive exposure assessment than that used in rnost other
studies.
Secondary data analyses were undertaken by some of the original investigators of the Tri-
Utility Study [9] to examine the relation between electric fields and leukemia within the Ontario
[ 1 171 and French [l 153 utility workers. Five additional leukemia cases were identified during the
additional follow-up interval in the Ontario study population, and data on occupational exposures
to leukemoçens, that had not previously been available, were assembled [9]. Among Ontario
Hydro cases and controls, exposure to electric and magnetic field exposures were assessed
according to both job title and work-site locations. This perrnitted the creation of a JEM that was
defined by both these variables. Ontario subjects having a cumulative arithmetic mean electric
field exposure of greater than 345 V/m-years had an odds ratio of 4.45 (95% CI=l .O 1-19.7)
relative to those with less than 171 V/m-years [117]. No association between etectric fields and
leukemia was found in workers employed at Electricité de France, though again, these analyses
Chapter 1 Introduction
29
excluded follow-up in retired workers where the incidence of this disease is most comrnon [115].
Using exposure data collected in the original Los Angeles County study [12], the relationship
between electric fields and leukemia was later analysed and published in 1997 [124]. No
association between electric field exposures and leukemia was observed, however, field
measurements were available for only 28% of those workers monitored in the original study.
Further. electric field measures were not available for power line workers in Los Angeles county
and were inferred fiom measures taken elsewhere. The authors acknowledged difficulties in
assessing electric field exposures due to grounding and posturing. Moreover, as indicated earlier
in this section, this study was prone to several potential biases as a result of drawing controls fiom
a population of individuals diagnosed with selected cancers.
1-42. Brain cancer
As with leukemia, the relationship between brain cancer and exposure ta magnetic fields
was first examined by using occupational title as a sunogate for exposure [ 125- 1281. One such
study reported an excess number of deaths fiom glioma and astrocytoma in Maryland residents
t hat had formerly been employed as electrical workers [ 1281. Eisewhere, a nonconcunent cohort
study did not find any significant increases in brain cancer for electricians, power line workers or
telecommunications workers. but did for welders and rnetal cutters [129]. The above studies that
assessed the risk of brain cancer by using job title did not adjust for exposure to ionizing radiation,
a known risk factor for brain cancer.
Chapter 1 Introduction
30
More elaborate exposure assessrnent was perFormed in several subsequent studies. In the
Swedish case control study conducted by Floderus et al. (page 2 1 ) a total of 346 brain cancer
cases were identified [10]. Approximately 75% of these patients responded to the questionnaire
and analysis was based on 26 1 cases. The arithmetic mean exposure to magnetic fields for the job
held the longest within the ten years before cancer diagnosis was weakly associated with brain
cancer. Specifically, those in the upper quartile had an odds ratio of 1.4 (95% CI=0.9-3.1)
relative to those in the lowest exposure grouping. When exposure was categorized according to
the median daily magnetic field exposure, those in the upper quartile had an odds ratio of 1.5
(95% CI= 1.1 -2.0) relative to those in the lowest quartile. A similar increase in risk was observed
for the corresponding categorization of exposures above 0.2 PT (OR= 1.5. 95% CI= f -0-2.2). The
elevation in risk was apparent only for astrocytoma grades III-IV which accounted for 1 94 of the
26 1 cases of brain cancer. This finding is consistent with the notion that field exposures act on
the promotional phase as evidenced by experimental studies that indicate that promoting agents
can lead to cancers with increased metastatic properties [22, 241.
The nested case control study of electric utility workers ernployed by Southern California
Edison also considered brain cancer as an endpoint [ 131. A variety of indices of cumulative
exposure to magnetic fields were not found to be related to brain cancer. Further, risk estirnates
were not significantly different fiom the nuIl when various time windows of exposure were
analysed- The study was somewhat limited by a small number of cases (n=32). In addition, cases
were identified using death certificates which are unable to distinguish primary cancers fiorn
Chapter 1 Introduction
metastases [ 181
A non-statistically significant increased risk of brain cancer due to exposure to magnetic
fields was observed in the Tri-Utility study 191. Workers having a cumulative exposure to
magnetic fields that was greater than the median (3.15 PT-years) had a twofold increase in brain
cancer risk (OR=2.0, 95% CI=] .O-3.9). A re-analysis of the Ontario component of the Tri-Utility
Study was undertaken to examine the relation between brain cancer and exposure to electric
fields. Cumulative exposure to electric fields based on either the geometric and arithrnetic means
was not associated with brain cancer [117]. A similar re-analysis was perforrned for the French
cornponent of the Tri-Utility study [ I 151. Subjects having exposures in the 90" percentile had an
odds ratio of 3.1 (95% CI=l.I-8.7) relative to the b a d i n e group. This association persisted afier
adjusting for magnetic field exposure and was strongest among workers who had been employed
for at least 25 years.
A mortality study of the Hydro Québec workers conducted by Bans et al. [130], found an
elevated risk of brain cancer death among those with higher exposures to electric and magnetic
fields. Exposure was assigned using a job exposure matrix that had been designed for the Tri-
Utility Study [9]. However. because full work histones were not available, the JEM was applied
to the last job held in order to classi@ workers into those with background and above background
exposures. Background exposures for electric fields were < 5.6 V/m while for magnetic <O. 16
PT. The relative risk o f death fkom brain cancer among blue collar workers exposed to above
Chapter 1 Introduction
32
background levels of magnetic fields was 2.5 (95% CI=0.6-9.6) while for electric fields the
corresponding relative risk was 1.5 (95% CI=0.4-5 -2). Of note, t hese results are based on only
I O cases.
In a historïcal cohort study of U.S. electric utility workers, Savitz and Loomis, analysed
the association between magnetic field exposure and brain cancer nionality [8]. A total of 144
brain cancer deaths were observed between 1950 and 1986. As outlined in the previous section,
magnetic field exposures were ascenained by sampling fiom current workers using personal
dosimetry. Workers in the highest exposure category ( 2 4.3 PT-years) had an increased nsk of
dying fi-om brain cancer (RR-2.3, 95% CI=1.2-4.6) relative to those with exposures less than
<0.6 PT-years. The risk was more pronounced for cumulative exposure received d~t ing the past
2- 10 years (RR per PT-yeal-1 -9, 95% C H .3-2.8).
Brain cancer was also evaluated in the Swedish case-control study of occupational and
residential exposure to magnetic fields [120]. A totaI of 223 cases of central nervous system
(CNS) tumours were identified fiom the Swedish Cancer Registry. Occupational exposure to
magnetic fields was not associated with CNS tumours in patients with little, or no residential
exposure. A non-significantly increased risk for astrocytoma grades III-IV was observed among
those having both residential and occupational exposure that exceeded 0.2 p T (OR=2.2, 95%
CI4.6-8.5).
Chapter 1 Introduction
33
Harrington et al [13 11 evaiuated the relation between exposure to magnetic fields and
death fiom bain cancer in a cohort of 84,O 18 male and female electric utility workers who had
been employed for at lest six months by the Central Electricity Generating Board in the United
Kingdom. A total of 1 12 diagnoses of brain cancer were identified between 1972 and 1984
through death certificates and verified through the National Cancer Registry. Six controls were
individually matched to cases by age and gender. Exposure assessrnent was perfonned by
measuring exposures for 1 1 job groups within a sampte of 258 current employees of the utility.
Job exposure matrices were calculated using the time weighted arithmetic and geometric rneans.
No association was observed between cumulative magnetic field exposure and brain cancer was
observed for either of these two metncs over the total career or over the five years preceding
death. The study was limited in that exposure could not be determined for 16% of the cases and
10% of the controls.
More recently, Rodvall and colleagues [132] examined occupational exposure to magnetic
fields and brain cancer in a population-based case control study in Swedish males. The study
population included 84 cases of glioma, 20 cases of meningioma and 155 controls. The workers'
exposures were assigned by 1) categotizing subjects into electrical/non-electrical occupations, 2)
through expert review by an electrical engineer and 3) by taking magnetic field measurements at
work sites. When analyses were based on the measured fields, an odds ratio of 1.9 (95% Ck0.8-
5.0) for glioma was observed among those with exposures greater than 0.4 PT.
Chapter 1 Introduction
1.4.3. Non-Hodgkin's lymphoma
Elevated rkks of Non-Hodgkin's lymphoma ('MU) have been observed among workers
thought to have higher exposures to magnetic fields on the basis of job-titles [ 13 3- 1361. On the
other hand, a series of similar studies found no evidence to support this association [ 137- 1391.
Again, many of these studies lacked information for other workplace exposures thought to be
related to M L . These potential confounders include benzene. herbicides and insecticides [140].
In their analyses of 3 6 , Z 1 electric utility workers, Sahl et ai [ 131, examined the
association between exposure to magnetic fields and mortality fiom NHL, multiple myeloma and
Hodgkin's disease (grouped together). As described earlier in section 1.4.1, exposures were
assigned based on job titles and magnetic field measures in the work environment. Death fiom
lymphoma was not related to a variety of indices of magnetic field exposure.
Magnetic field exposure was not significantly associated with the incidence of NHL within
the Tri-Utility Study [9]. Specifically, electric utility workers in Ontario, Quebec and France that
had a cumulative exposure to magnetic fields greater than the median had an odds ratios of 1.2
195% CI=O. 8- 1 -9) relative to those with exposures less than the median. A non-significant risk
elevation was observed for subjects within the Ontario component of the study that had high
cumulative electric field exposures based on the arithmetic mean [117]. In particular, those in
upper tertile had an odds ratio of 2.4 (95% Ck0.7-8.3) relative to those in the lowest. No such
association was observed with employees fiom the French electric utiiity [115]. However,
Chapter 1 Introduction
35
follow-up for the workers of Electricité de France-Gaz terminated at retirement (between 55 and
60) before the age at which most NHL cases are identified [14 11.
Schroeder and Savitz exarnined the relation between magnetic field exposure and death
frorn lymphoma and multiple myelornas using a retrospective cohort study of workers employed
at five U. S. electric utilities [ 1 141. As in previous analyses of this study population [a], exposure
assessrnent was performed using personal monitoring devices. The design of the study only
permitted cumulative exposure to magnetic fields, based on the TWA arithmetic rnean, to be
calculated. Workers having a cumulative exposure greater than 4.3 PT-years did not have a
statistically significant increased risk of dying from NHL (OR=1.3, 95% CI=0.7-2.8) relative to
those with less than 0.6 PT-years. More pronounced risks, aibeit not statisticatly significant, were
observed for intermediate and higher grade lymphomas. Within this subset of cases, subjects
having exposures exceeding 2.0 PT-years had an odds ratio of 2.3 (95% CI=0.8-6.9) relative to
those having exposures less than 0.6 PT-years. Unfortunately. this study did not collect electric
field exposure data and therefore confounding cannot be ruled out. Furthemore, the study was
lacking information on occupational codounders.
1.4.4. Malignant melanoma
Increased risks of melanoma have been observed among workers suspected of having
elevated exposures to electric and magnetic fields that include workers in the welding [142, 1431,
the electronics industry [144, 1451, and the telecommunications industry [146, 1471. On the other
Chapter 1 Introduction
36
hand. neither intermittent or continuous exposure to magnetic fields were associated with
melanorna in an occupational cohort study involving 2.8 miIIion Danes [148], nor in a case control
study of electrical workers in New Zealand [149]. A recent proportional mortality study of
Amencan electrical workers employed in the construction industry found an excess number of
deaths from rnalignant melanoma (PMR=1.23) [150].
In the study by Tynes et al. [15 1 1, an excess risk of malignant melanoma was observed in a
cohort of5,088 subjects employed by Norwegian utilities that were exposed to higher levels of
magnetic fields. The cohort consisted of those male workers employed for at least one year
between January 1, 1920 and December 3 1, 198 5. Cancers were identified by linking data
contained in occupational information to databases maintained by the Norwegian cancer registry.
E'cposures to electric and magnetic fields were not based on persona1 monitoring, but rather, spot
measures taken at current work-sites by expenenced technicians. These measures were assumed
to represent typical patterns of exposures received in the work environment. Those with greater
than 35 PT-years of exposure had a standardized incidence ratio of 2.24 relative to those with less
than 5 PT-years (p<O.O5). However, this increased risk was confined to those with exposure to
PCBs. Analysis was limited by a small number of cases (n=19).
In the Tri-Utility Study [9], no association between cumulative magnetic field exposures
and the incidence of malignant melanoma risk. Those workers having cumulative magnetic field
exposures greater than the median had an odds ratio of 0.9 (95% CI=0.5-1.5) relative to those
Chapter 1 Introduction
with exposures less than the rnedian. Separate anaiyses within the Ontario cohort were
performed to examine the relation between cumulative electric field exposures and cancer; no
association was found between malignant melanoma based and cumulative exposure based on
the arithmetic and geometric means 11171. Within the French utility, a non-significant
elevation in melanoma risk was observed in the 9ûa percentile of cumulative electric field
exposures estimated using the geometric (OR=6.8 95 % CI=0.7-63.0) but not the arithmetic
mean (OR= 0.2, 95% CI=0.0-1.9) [115].
1 -4.5. Summary of epidemiologic findings
At this rime, there is no conclusive evidence that either magnetic or electric fields are
associated with an increased incidence of brain cancer, malignant melanoma, leukemia or Non-
Hodgkin's lymphoma. To date, only three studies have published results based on direct
measures of electric fields [Ils-1 171. In light of the observed relation between cumulative
electric field exposure and leukemia risk in Ontario workers [117], more work is needed in
this area. To date, the relation between cancer risk and electric field exposure has only been
assessed using metrics based on the daily geometric and arithmetic mean field strength. A
l i m i ted num ber of studies have performed cancer risk assessrnent using al ternate summaries of
magnetic fields [13]. though none of these have been able to adjust for exposures to electric
fields as a potential confounding variable.
Chapter 1 Introduction
38
1 S. Study rationale
Although the results frorn in vivo and in vitro studies of electric and magnetic fields
suggest that these exposures may be related to the promotional stage of carcinogenesis, they
have been unable to identify which aspect of exposure is most relevant. The vast majority of
studies of electric utility workers have defined field exposure using a cumulative TWA based
on the geometric or arithmetic rnean. Temporal characteristics of ELFs may be important for
evaluating biologic effects [ 1521.
To date, few occupational studies of power frequency fields and cancer have evaluated
alcernate indices of magnetic fields as they relate to cancer. The findings of these studies have
been inconsistent. and additionally, have not controlled for the potential confounding influence
of electric fields. Other studies have indirectly evaluated the potentially relevant aspect of
field exposures by modelling cumulative indices of magnetic fields that are based on the
geometric and arithmetic mean. The arithmetic mean is more sensitive to skewed data and is
better suited for modelling threshold effects. On the other hand, the geometric rneari serves to
minimise the influence of outliers and is closely related to the median of the exposure
distribution. Within some study populations, the arithmetic mean has been shown to be highly
correiated with threshoid indices, and could therefore be considered a suitable surrogate
measure. However, as will be seen in Chapter 2, the use of the geometric and arithmetic
means does not adequately capture the variability in either electric and magnetic fields within
Chapter t Introduction
39
the Ontario Hydro workers. and therefore, other metrics should be evaluated. TO date. no
study has evaluated indices of electric fields other chan those based on the arithmetic or
geometric means.
A clearer understanding of the relationship between power frequency electric and
magnetic fields and cancer is needed given the fact that these exposures are ubiquitous and steps
can be taken to reduce harmfùl exposures. particularly for electric fields which are easily shielded-
It follows that by identifjing the relevant exposure metric. risk can be more accurately determined
and cost-effective strategies to reduce worker exposures can be developed.
The simulation study presented in Chapter 4 examines how increases in study power can
be achieved by either increasing the study size or by expending additional resources to produce a
more precise estimate of the mean exposure within the cells of the job exposure matrix. The
simulation methods can easily be adapted to other studies where exposure data are collected
independentiy €rom case control data.
C hapter 1 Introduction
1.6. Research objectives
The primary research objective was to investigate the relation between various definitions
of exposure to 60 Hz electric and magnetic fields and cancer within a group of male Ontano
Hydro workers. Risk assessment was performed for leukernia. brain cancer, Non- Hodgkin's
lymp homa and malignant melanoma.
The specific objectives of the study were:
To evaluate the relation between indices of electric and magnetic fields so as to identiQ an
appropriate series of metrics to perform cancer risk assessment (Chapter 2).
To determine which, if any, aspect of magnetic or electric fields are associated with the
incidence of the cancers listed above (Chapter 3). The series of exposure indices identified
using correlational analyses (Chapter 2) were used to perform this cancer risk assessment
To explore variations in cancer risk according to various lag intervals (Le., exposure
windows) and duration of employment (Chapter 3).
Where applicable, to estimate the value of the critical threshold required to increase risk of
site-specific cancer (C hapter 3)
To evaluate, by using cornputer simulation. the effectiveness of different strategies to
augment power by reducing the variance of mean exposure estimates within a job
exposure matrix (Chapter 4).
C hapter 1 Introduction
41
Cha pter 1 : References
Asanova, T. P. and A. N. Rakov, [fiealth condi~iorrs of workers exposed to elecrricjkldr
of ope11 witchboard ir~stailatiorrs of 400400 kv. (Preliminary report)/. Gig Tr Prof
Zabol, 1966. lO(5): p. 50-2.
Wen heimer. N . and E. Leeper, Electrical wirhg corrfigrrations ard childhood carrcer.
Am J Epidemiol, 1979. 109(3): p. 273-84.
Ten for de, T .S. and W .T . Kaune, hremtiorr of extremely Iow freqzrerrcy electric arrd
magneticfields with hrma~rs. Health Phys, 1 987. 53(6): p. 585-606.
Mi I ham, S., Jr., Mortality from leirkemia iri workers exposed to eleclrical alrd mugrtetic
fields- [letfers/. N Engl J Med, 1982. 307(4): p. 249.
Howe, G. R. and J. P. Lindsay, A follow-rrp stzrdy of a tert-percerlt sarnple of the Carradiarr
labor force. I. Catrcer mortnlity in males, 1965-73. J Nat1 Cancer 1 nst, 1 983. 70( 1 ): p. 3 7-
44.
Wright, W.E., J.M. Peters, and T.M. Mack, Lettkaernta irr ivorkers exposed to electrical
mrd magnetic fields [letter/. Lancet, 1982. t(8308): p. 1 160- 1 .
Morton, W. and D. Ma janovic, Lerrkemia hciderrce by occ~rpation irr the Portlar~d-
~~2ulcozrver. metropoliturt area. Am J Ind Med, 1984. 6(3): p. 185-205.
Savitz, D. A. and D.P. Loomis, Magmticfield exposrrre irt relation to leirkemia arrd brairr
carrcer- mortality amorrg electric zrtility workers (prrblished ermîurn appears in Am J
Epidemiol1996Jlrl 15;144(2):205]. Am J Epidemiol. 1995. 141(2): p. 123-34.
Chapter 1 Introduction
42
9. T hériault, G.. et al., Car~cer risks associated with occzrpiiorial exposrrre to mcrgr~etic
fields amotrg electric trtility workrrs I I I Oritario alrd Qirebec. Catiada. arid Fraricx: 1970-
1989 [pirblished errattrm appears 111 Am J Epiderniol199-1 May 15; l39(f 0): IOfi3 / [sce
cornmerlfsj. Am J Epiderniol, 1994. lN(6): p. 550-72.
1 0. Floderus. B .. et al. , Occupational expomre to elrctromagrretic fields b r relut ION to
le~rkernia a d brairi tzcmors: a case-corrtrol stzrdv in Swedetr. Cancer Causes Control,
1993. 4(5): p. 465-76.
1 1 . Matanoski, G.M., et al., Leukemia ;II telephom lirmnen. Am J Epidemiol, 1993. 137(6):
p. 609-19.
12. London, S.J., et al., &postrre to ntagneticfldds amorrg electrical rvorkers I I I re/atiorr to
leikentia ri& in Los Angeles Coirmy. Am J Ind Med, 1994. 26(1): p. 47-60.
13. Sahl, J.D., M.A. Kelsh, and S. Greenland, Cohort arrd rrested case-coritrol stirdies of
hematopoietic carrcers utid brai11 cancer among electr~c rrtility workers. E pidemio 1 ogy ,
1993. 4(2): p. 104- 14.
1 4. Sagan, L. A., Ep~demiological and laborator- strrdies of powr freqrrerrcy dectric ar~d
rnclgrretic ficilds [see comments/. Jama, 1 992. 268(5): p. 625-9.
1 5 . B rac ken, T. D., fiiposam assessmerrt for porver freqrerrcy electric arld maprelicfie/ds.
Am Ind Hyg Assoc J, 1993. 54(4): p. 165-77.
16. Hendee, W.R. and J-C. Boteler, The questiori of hedth eflects from exposzcre to
clec~romagneticflelds fsee commetrts]. Health Phys. 1 994. 66(2): p. 1 27-3 6 .
Chapter 1 Introduction
43
Deadman. I. E.. et al., Occupariorcal arrd residetrtial60-Hz electromagrteticfie/dc atrd
high- freqzrency electric tratrsietrts: expomre assessmetrr rtsirrg a r lmt dosime ter
fprtblished erratum appears itr Am btd Hyg Assoc J 1996 Jrm;5 7/6/ :58O-3/. Am Ind Hyg
ASSOC J , 1988. 49(8): p. 409-19.
N IEHS, Assessrneta of health efjrctsfrom expomre to power-litx f r e q e ~ c y electric atrd
rnapeticfields, ed. C.J. Portier and M. Wolfe. 1998. Research Triangle Park, NC: U.S.
National Institute of Health.
Loscher, W. and M. Mevissen, A~rimal stttdies orr the role of 50/60-Hertz rnagrreticfields
ir~ carcirrogeriesis- Life Sci, 1994. 54(2 1 ): p. 1 53 1-43.
Berenblu k, 1. and P. Shubik, fie role of crotorr oil appiicatiotrs, associated with a sirrgle
paitrtit~gof a carcitroge~~, itr tirmorrr irrdmtiori of the mottse's skir~. Br I Cancer, 1947.
l (3 79-82).
Mottram, J.C., A developing factor irr expertmeritai blas~ogetresis- J Pat ho1 Bacteriol.
1944. 56: p. 181-7.
Cridland, N., Egects of power freqzrerrcy ~ /=e . rpos~r re s at the cellidar levd Radiation
Protection Dosimetry, 1997. 72(3-4): p. 279-290.
Hardell, L., et al. , f ipo~'~tre to extremely /ow fi-eqiret~cy electroniapetic fields arld rhr
ris& of mal ipratrt diseases-ur~ e valuutiorr of epidemiologica/ a t~d experimerrtal jiirdirrgs.1 rdir rgs-
Eur J Cancer Prev, 1995. 4 Suppl 1: p. 3- 107.
Pitot, B. C. and Y. P. Dragan, Facts urrd theories corrcewritrg the mcchanisms of
carcirrogerresis. Faseb J , 1991. 5(9): p. 2280-6.
Chapter 1 Introduction
Cohen, S. M., Ceil proliferatiorr itr the evah~atio~r of carci)rogmic risk atld the
irradequacies of the ir~itiatiot-promotion mode!. Int J Toxicol, 1998. 17(Suppl3): p. 129-
142.
Juutilainen, J . , S. Lang, and T. Rytomaa, Possible cocarcinogerric effects of ELF
e l e c t r m a e t i c e may repire repeated lotlg-term itrteracriot~ with howtr
carcirroger~ic factors. Bioelectrornagnetics, 2000. 21(2): p. 1 22-8-
Barrett, J-C., Mechrisms of multistep carcinogerresis urd carcinogen risk assessrnerit.
Environ Health Perspect, 1993. 100: p. 9-20.
Shaw, 1 .C. and H. B. Jones, Mechanisms of rtorr-gerrotoxic carcinogenesis fsee
commerrtsj. Trends Pharmacol Sci. 1994. 15(3): p. 89-93.
Loscher, W . and R.P. Liburdy, Animal arrd ceiidar studies OH carcirrogetric effects of low
ft-equency (50/60-Hz) magrreticfields. Mutat Res, 1998. 410(2): p. 185-220.
Mou lder, J . E. and K. R. Foster, Biologicaf effects ojpolver-freqztency fiel& as the_)? relate
IO curcirrogerresis. Proc Soc Exp Bi01 Med, 1995. 209(4): p. 309-24.
Murp hy, J .C ., et al., Ittternatiotral Commissiotr for Protection Agaiïrst Err virottmetr fal
Mirtagerrs urrd Carctr~oger~s. Power frequerrcy eiectric and magneficfields: a review of
gerretic toxicology. Mutat Res, 1993. 296(3): p. 22 1-40.
Adey, W . R., Ce// membrarres: the electromagnetic environmerif arrd camer promo tiori.
Neurochem Res, 1988. 13(7): p. 671-7.
Frey, A. H., Electromagr~etic fieki irrteractiotrs with biological Aystems. Faseb J, 1 993.
7(2): p. 272-8 1 .
Chapter 1 Introduction
45
3 4. d' Ambrosio, G., et ai., Chromosomal aberratious irrdttced &y EL K electric f~elds. J
Bioelectr, 1985. 4: p. 237-247.
3 5 . Nordenson, I., et al., Chromosomal aberrations irr hmarr amrriotic cells afier
ii~fermitter~t exposrre to fifty hertz mugneticfie/ds. Bioelectromagnetics, 1 994. 1 S(4): p.
293-30 1.
3 6. Mi yakoshi, J . , et al., Irrcrease î i ~ hypoxarr~hit~e-pdarrirre phosphorihosyf tram$kruse ger~e
matatiorrs by expo.smre ro high-der~sity 50-Hz magnetic fields [see commer~ts/. Mutat Res.
1996. 349(1): p. 109-14.
3 7. CIeary, S. F., A review of il1 vitro studies: lorv-freqttertcy electromagrretic fields. Am Ind
Hyg Assoc J, 1993. 54(4): p. 178-85.
3 8. Hest er, G. L . . E/ectric arrd ma~teticfie/ds:ma~ra~~~rg an trrrcertair~ err virorrmerrt.
Environment, 1992. 34: p. 7-3 1.
3 9. Lacy-HuIbert, A., J. C. Metcalfe, and R. Hesket h. Biuiogica/ respottses to electromagrretic
JTelds. Faseb J , 19%. 12(6): p. 395420 .
40. Goodman, R., et al., hpo~wre to electric and magnetic fields (M) irrcrcinses trarrscript
irr HI,-60 ceifs: does udaption to W f i e l d s ocau- ? Bioelectrochern Bioenerg, 1 992. 29:
p. 185-1 92.
4 1 . Goodman, R. and A. Shirley-Henderson, Transcripfiorr und trurrslation in ce//s exposed to
exfremely low freqtencyfie/ds Bioelectrochem Bioenerg, 199 1 . 220: p. 1383 - 1285.
Chapter 1 Introduction
46
Wei. L.X.. R. Goodman, and A. Henderson, Changes in levels of c-nryc and hisrome H2R
fo//owirrg expusure of cells f O low-fequetrcy simsoidal elec fromagrre f ic f i e : e viderice
for a rvirldow efJecî. Bioelectromagnetics, 1990. 1 l(4): p. 269-72.
Lacy-Hulbert, A., et al._ No ef/ecf of 60 fi elec~romagrieticJields or? MYC or hefa-ucfi~r
expressiort (11 human leirkemic celk Radiat Res, 1995. 144( 1 ): p. 9- 1 7.
Lacy-Hulbert, A., et crf. , Camrr risk atrd r l e c t r o m a ~ ~ e ~ i c r d s f leirer /. Nature. 1 99 5 .
375(6526): p. 23.
Saffer. J. D. and S .J. Thurston, Short exposwes to 60 Hr ntagneticfieids do trot alfer MYC
expressiotr in HL60 or Datrdi cells. Radiat Res, 1 995. 144( 1 ): p. 18-25.
Saffer, J . D. and S. J . Thurston, Carrcer ris& atrd electrornagneficfields [/effer/. Nature,
1995. 375(6526): p. 22-3.
B lac kman, C. F., et al., Effects of ELFfieIds on calciwn-ion efym from brairr [issue in
vitro. Radiat Res, 1982. 92(3): p. 5 10-20.
Bawin. S.M. and W.R. Adey. Serlsitivity of calcitm birtdiirg iti cerrhrd tissre [O wak
erwit-ortmerrtal elec fric fields oscillafi~rg al low freq~fetrcy. Proc Nat l Acad Sci U S A,
1976. 73(6): p. 1999-2003.
Walleczek, J - and R.P. Liburdy, Norrfherrnal60 Hz sirlrrsoidal mapretic-field expomre
enhances JjCa.? - trpfake in rat fhymocyfes: dependence otr mirogen activation. FEBS
Lett, 1990. 271(1-2): p. 157-60.
Chapter 1 Introduction
47
Li bu rdy. R. P.. Calcium sip~alling irt lymphocytes arid ELF$eldF. Eviderice for art
rkctric field rnetric arld site of irrteracriort irwolvirg the cafcium io~? chamel. FEBS Lett.
1992. 271: p. 53-59.
Lindstrom, E., et al., I~~tracrllrrlar calcium oscillatior~s inditced in a T-ce// lirw hy a weak
50 fi rnagneticjirld. J Cell Physiol, 1993. 156(2): p. 395-8.
Mc Leod, K. J. and C.T. Rubin, The eflecf of low-freqz(ericy ekctricalfleldî 011
osteogemsis [pzcblished rrratzcm appears 111 J Boite Joirif Swg Am 1992 Sep; 74(8): 12 74/.
J Bone Joint Surg Am, 1992. 74(6): p. 920-9.
Parkinson, W.C. and C.T. Hanks, Search for cyclotrott reso~runce h celfs hl vitro.
Bioelectromagnetics, 1989. lO(2): p. 129-45.
Sontag, W., Actiorr of extremely low freqrrericy electric fields ott the cyrosolic calcium
cotzcerrtratiorr of d@ereritiated HL-60 cells: no~tactivafed cells. Bioelectromagnetics.
1998. 19(1): p. 32-40.
Lyle, D.B., et a/., I~~tracrlldar calcizim signalitig by Jurkai T-fyrnphocytes exposed to a
60 Hz mapietic field Bioef ectromagnetics, 1997. 18(6): p. 439-45.
Reed, D. J . and M. W. Fariss, Glutathiorte depletiori and srisceptibility- Pharmacol Rev,
1984. 36(2 Suppl): p. 25s-33s .
Pascoe, G.A., Calcilcm homeostasis alid oxidative stress, in &trernely lorv freqrrerlcy
elecft~omap~etic jirlds: the quesf/or~ of cancer., B. W . Wilson, R.G. Stevens, and L. E.
Anderson, Editors. 1989, Battelle Press: Columbus, OH.
C hapter 1 Introduction
48
Poeggeler, B., et al., Melatonir,, hydro~vl radical-mediated oxidative damage. attd aging:
a hypothesis. J Pineal Res, 1993. 14(4): p. 15 1-68.
Pool, R., Electromagnetic fie/ds: the biological evidetlce fttmw/. Science, 1 990.
249(4975): p. 13 78-8 1 .
Kikkawa, U., A. Kishimoto, and Y. Nishkuka, The protein kirlase C famify: heteroge~teity
arrd ils implications. Annu Rev Biochem, 1989. 58: p. 3 1-44.
Monti, M.G., et al., Effect of ELFpiclsed electroma~teticfields: o t ~ prote111 kirrase C'
activatiot~ process 111 HL-60 leirkemia cefls. Journal of Bioelectncity, 1 99 1. 10: p. 1 19-
130.
Holian, O., et af., Proteitt kiwse C activity is altered I I I HL60 cells exposed to 60 Hr AC'
electricjields. Bioelectromagnetics, 1996. 17(6): p. 504-9.
Rosenthal, M . and G. Obe, Effects of 50-hertz electromagrteficfields ou profiJeratiott uttd
011 chromosomaf alteratiotts in htmart peripheral fymphocytes zmtreaf ed or pretreated
wifh chemical mutagem. Mutat Res, 1989. 210(2): p. 329-35.
Antonopoulos, A., et al., Cytological effects of 50 Hz electronragnetic fields ut? htrmatt
lymphocytes in vitro. Mutat Res, 1995. 346(3): p. 1 5 1-7.
West. R.W., et al., Etthancemettf of artchorage-irtdepettdem growth itt JB6 cells exposed
to 60 Hz mapteticfields. Bioelectrochemistry and Bioenergetics, 1994. 34: p. 39-43.
Noda, N ., el al., Eflect of eleclric crrrrenls orr DNA synthesis itt rat osleosarcorna cells:
cieper1dertce ott conditions thal itrfluence ceIl growth. J Orthop Res, 1 987. 3: p. 253-260.
-.
Chapter 1 Introduction
49
Cleary, S.F., et al., Mzxizrfario~i of terldorifibroplasia by exogerroils electric mrrerrfs.
Bioelectromagnetics, 1988. 9(2): p. 183-94.
Byus, C.V., S.E. Pieper, and W.R. Adey, The eflects of low-t.,~ergv 60-fi er~viror~metital
electrornapwtic e s irpor~ rhe growth-related ert-yme onrithine decarboxylase.
Carcinogenesis, 1987. 8( 1 O): p. 1385-9.
Litovitz. T . A-, D. Krause, and J. M. Mullins, Effect of coherertce rime of the applied
tnapreticfield or1 on~ithirie decarboxylase activig. Biochem Biophys Res Commun,
199 1 . 178(3): p. 862-5.
Valtersson, U., K.H. Mild, and M.O. Mattson, Orriirhirre decarboxylase activity ard
polyumir~e fevels are different irr Jtcrkat and CEAI-CM3 ce//s after expomre to a 50 Hz
mugreticfield. Bioelectrochemistry and Bioenergetics, 1997. 43: p. 169- 172.
Cress. L.W., R.D. Owen, and A.B. Desta, Orr~ithirte decnrboxylase activity iri L929 crlls
fohwir~g exposirre to 60 Hz magneticfields. Carcinogenesis, 1 999. 20(6): p. 1025-3 0.
Azadniv, M., et al., A test of Ihe hypothesis that a 60-Hz ntagrwtic jkld aficts orr~ithitre
decarho.ry1use activity in mouse L929 cells I ~ I vitro. Biochem Biop hys Res Commun,
1 995. 2 l4(S): p. 627-3 1.
Phillips, J . L., L. Rutledge, and W.D. Winters, Trar~sjierriri bit~dirtg to IWO humart co/ori
carci,ioma ce// /zrtes: characterization arrd effecr of 60-H,- cl~ctroniagr~elicfield.~. Cancer
Res, 1986. 46(1): p. 239-44.
L y1 e, D . B., et al., Stippressior~ of T- fymphocyre cytotoxiciîy followir~g exposire to 60-Hz
sitrt~soidal electric fields. Bioeiectromagnetics, 1988. 9(3): p. 303- 13.
Chapter 1 Introduction
50
Cohen, M.M., et al.. ïhe eyect of low-level60-Hz electromagr~eticfields or, hrrmari
fynphoid cells. 11. Sister-chrornatfd exchmges N I peripheral fyrnphocytes arid
iymphoblastoid ce// lines. Mutat Res, 1986. 172(2): p. 1 77-84.
Cohen, M.M., et al., Effecf of low-ievel, 60-Hz electrurnagrreticfields orr ht~marr
ly»phoid cells: 1. M~tofic rate arrd chromosome breahge b~ hirmarr peripheral
lymphocyfes. Bioelectromagnetics, 1986. 7(4): p. 4 1 5-23.
Reese, J.A., R.F. Jostes, and M.E. Frazier, fiposrrrci of mammaharr cells to 6 0 4 :
rnapjetic or electrkjields: anaiysis for DNA sit~gle-strand breaks. Bioelectromagnetics,
1988. 9(3): p. 23747.
Livinçston, G.K., et al., Reproductive iritegrity of mammaliurr cd... exposed tu pubver
fieqirericy electromagrreticfields. Environ Mol Mutagen, 199 1. 17( 1 ): p. 49-58.
McCann, J., et al. , A critical revtew of the gerrotoxic poteritial of electric ar~d magnetic
fields. Mutat Res, 1993. 297(1): p. 61-95.
Rannug, A.. et al., A stzrdy or1 skin trcrnozcr for?natiotr br mice with 50 Hz mag>ieticfie/d
expomre. Carcinogenesis, 1 993. 14(4): p. 573-8.
Beniashvili, D.S., V.G. Bilanishvili, and M.Z. Menabde, Law-freqlrerrc): electromu~retic
radiariorr erlhames the irductior~ of rat marnmary tumors b j ~ nirmsornethyi irrea. Cancer
Lett, 199 1 . 61(1): p. 75-9.
McLean, J.R., et al., Carrcer promotion in a mozrse-ski11 rnodèl by a 60-Hz mapeticfield:
ii. ï h o r de velopment at~d immio~e respmse. Bioelectromagnetics, 1 99 1 . l2(5): p. 273 -
87.
Chapter 1 Introduction
5 1
Stuc hl y, M . A., et al., Modif7catiori of tumor promotio~z irr the niouse skin by e x p i r e to
arr ahenratirig magneticfield Cancer Lett, 1992. 65( 1 ): p. 1-7.
Mevissen, M., et al., Effects of magnetic fiel& or> mammary ttimor developme~î iriduced
by 7,/2- dirnethylberiz(a)ar,rhracerle in rats. Bioelectromagnetics. 1 993. 14(2): p. 1 3 1-43.
Leung, F . C ., et al. Eflects of eleciric jields or, rat ntammary trcmor development Urd~iced
by 7,12-demythlherc(a)arrthracerre [A bstractj. in 10th Ar uncal meeting
Rioeiectromagnetics Society. 1 986. Gaitherburg MD.
Baum, A., et al., A hisropathologicaf strrdy on alteratioris in MA-irzd~cced mammary
carcirrogertesis in rats with 50 Hz. 100 mu T mupieticf7eld eruposrrre. Carcinogenesi s,
1995. 16(1): p. 119-25.
Mevissen, M., et al., Dponcre of DMBA-treated femafe rats i r ~ a 50-Hz, 50 micro Tesfa
magrreticjield: eficts on rnammary trtmor growth, melatoniri levels. arid T lymphocyte
activatiorr. Carcinogenesis, 1996. 17(5): p. 903- 10.
Dees, C., et al., Effects of 60-Hzflelds, estradioi arrd xenoestrogerrs ort hcmarr breast
came)- ce //S. Radiat Res, 1 996. 146(4): p. 444-52.
Loscher, W., et al. , Effects of weak aiteniating mu~retic/iefds orz noctrcrrd melatorrirr
prodrtctiorr arrd mamrnary carcirtogerresis I I I rats. Oncoiogy, 1 994. 5 l ( 3 ) : p. 388-95.
McLean, J. R., et al., The effect of 60-Hz mapeticjlelds ori CO-promotior~ of chernical&
irrdrrced skirr tzrmors orr SENCAX mice: a disclrssiorr of three sttidies. Environ Health
Perspect, 1997. 105(1): p. 94-6.
Chapter 1 Introduction
52
Moms, J.E. and R.D. Phillips, Eflects of 60 Hz-electricfirldv or1 spec~fic humoral arid
cellrilar comporients of the imnrtrm system. Bioelectromagnetics, 1982. 3(3): p. 43 1-7.
L y l e, D. B ., et al., Stppressiort of T-lymphocyte cy f otoxicity fo//owir~g expostire to
sirrt~midally amplitlide-md~ilatedfieIds. Bioelectromagnetics, 1983. 4(3): p. 28 1-92.
Reiter. R. J . , Pirieal glar~d, ce//~rlar proliferatiorr utrd neoplastic growth: art hisforical
accortrrt., in The pilleal glurid arid caricer-. D. Gupta A. Attanasio. and R. D. Reier.
Editors. 1988. Brain Research Promotion: Tubingen. Germany. p. 4 1-64.
Stevens, R. G.. et a/, , EIectric power, pinealjhncriori, artd the ris& of breast cartcer fsee
commerrtsj. Faseb J , 1992. 6(3): p. 853-60.
B las k, D. E., The ernergirlg role of ~ h e pir~eal glarrd alrd melator~iri in onccgertesis., in
Ertremely low frequemy e/ectromagneicfieI&: the qresfiorl of carrcw., B. W . Wilson,
R.G. Stevens, and L.E. Anderson, Editors. 1989, Batelle Press: Colombus, OH. p. 3 19-
335-
B lask, D . E . , Melu forrin il1 oricology, in Melatoriir r byosyrithesis, physiologieal effec fs, and
clirticalapplieatior~s.. H . Yu and R.J. Reiter, Editors. 1993, CRC Press: Boca Raton. FL.
p. 447-475.
Conti, A. and G.J. Maestroni, Thri clirtical rreuroimmt~rrotherapetctie role of melator~ir? irr
orrcology. J Pineal Res, 1995. 19(3): p. 103- 10.
Panzer, A. and M. Viijoen, fie validity of rnelalorriri us an oricostatic agertt. J Pineal Res,
1997. 22(4): p. 184-202.
Chapter 1 Introduction
53
Brzezinski, A., Mefatonitr t j r hmarrs. N Engl J Med, 1997. 336(3): p. 1 86-95.
Cohen, M., M. Lippman, and B. Chabrier, Roie of p~?reai gland 1rr aetioiogy and tre.rrime?rt
of breast cancer. Lancet, 1978. 2(8094): p. 8 14-6.
Cagnacci, A., Melatoriirr 1tr rrlatiorr to physiofogy in ad& hmarrs. J Pineal Res, 1996.
21(4): p. 200-13.
Nelson, R. J., et al., 7;he itrfrrierrcr of seasor,. ph0 toperid, and pitreal me fatorritr ort
immirr~e jïrrtctiorr. J Pineal Res, 1995. 19(4): p. 149-65.
Reiter, R.J., et ai., A review of the eviderm sirpp~irrg melalotrirr's roie as air
mrioxidnrrt. J Pineal Res, 1995. 18(1): p. 1 - 1 1 .
Tan, D.X., et al., The pitreal hormorre melatonin irlhibiis DNA uddi~ct fonnafiol~ irrdirced
by the chemicai curcitrogerî safroîe in vivo. Cancer Lett, 1 993. 70( 1 -2) : p. 65 -7 1 .
Reiter, R. J . , Melatorrin st~ppression by static artd extremeiy low freqtretrcy
efectromagrteticfie/ds: reiatiorrship to fhe reported ir~creased irrcider~ce of carrcer. Rev
Environ Health, 1994. tO(3-4): p. 17 1-86.
Wilson, B.W., et al., Chronic expomre to 60-H= ëkctricfiefds: effects orr pimm/fnlrctiotr
irr rhe rat. Bioelectromagnetics, 198 1 . 2(4): p. 37 1-80.
Wilson, B. W ., E.K. Chess, and L. E. Anderson, 60-Hz eiectric-fdd eflects orr pitreal
me/uror~iti rhythms: time cotwse for omet mrd recovery. Bioefectromagnetics, 1 986. 7(2):
p. 239-42.
Burch, J. B., et al., Reditced excret~on of a mefatonin metaboide in workers exposed fo 60
Hz ntagneticfieids. Am J Epidemiol, t 999. 150( 1 ): p. 27-3 6 .
Chapter 1 Introduction
54
Pfluger, D.H. and C.E. Minder, Ef/ects of expomre to 16.7 Hz magreticfields oti ririrtaty
6- hydroxymelatorrirr srilfare excretiort of Swiss ruihvay workers- J Pineal Res, 1 996.
21(2): p. 91-100.
Wilson, B.W., et al., Evidence for un egect of ELF electromngrtetic ficilds otr h r m t
pirreal g/arrd,fimctiort. J Pineal Res, 1990. 9(4): p. 259-69.
Li tovitz. T .A.- C. J. Montrose, and W. Wang, Dose-resportse imp/icatiorts of ihe trurtsiririt
natrire of electromap~etic- field-it~driced bioeffects: theorerical hypotheses and
predictior~s- Bioelectromagnetics, 1992. Suppl(1): p. 23 7-46.
HuK J . , Issrcrs and cotttroversies mrrorrndirtg qriulitative stt-u~egies for idetri fyi tg arid
forecast~tg carlcer catrsir~g agerits irt the hrrmarr envirorrmerrt. P hamac01 Toxico 1, 1 993.
72(Suppl 1): p. 12-27.
Fey chting, M., Occupational exposire CO electrotnagtretic fie fds arrd adrrlt le rikuemia: a
1-rview of the epidemiological evidence. Radiat Environ Biophys. 1996. 35(4): p. 23 7-42.
Schroeder, J.C. and D. A. Savitz, Lymphoma arrd multiple myelornu mortality irr relation
to magnetic field exposltre amotrg electric ritility workers. Am J Ind Med. 1997. 32(4): p.
392-403.
Guénel, P ., et al., hpo~rrre to SU-Hz electric field atrd im%ierrce of le~ikernia. brairi
irrmors~ ar~d other carrcers amortg Fretich elccrric rrtiiity workers [sec cumments/. Am J
Epidemiol. 1996. 144(12): p. 1 107-2 1 .
Khei fet s, L. I., et a/. , Occtipational electric and magrtetic field expomre artd lerikem ia. A
mcta- artai'ysis. J Occup Environ Med, 1997. 39(11): p. 1074-9 1 .
Chapter 1 Introduction
55
Miller, A. B., et al., Lezrkemia following ocarpatiorral exposrrre to 60-Hz electric mrd
magrretic fields amorig Ontario electric rrtility workers fsee comments /. Am J E pidemiol,
1996. 144(2): p. 150-60.
Higginson J, C.S. Muir, and N. Munoz, The lezrkemias, in Hizrmarr cat~cer:epidemioIogy
arrd errvironmental causes, H. Baxter, Editor. 1 992, Cambridge University Press:
Cambridge, UK. p. 470-475.
Linet, M . S . and R.A. Cartwright, The lez~kemias. Second ed. Cancer Epidemiology and
Prevention, ed. D. Schottenfeld and J.F. Fraumeni. 1996, New York, N.Y.: Oxford
University Press.
Feychting, M.. U. Forssen, and B. Flodems, Occtrpatior~a/ a~ld residerriial rnagrretkfwld
expomre arrd leukemia arrd ctvrtral rrervozrs system tzrmors. E piderniology, 1 99 7. 8(4): p.
3 84-9.
Flodems, B., T. Persson, and C . S tenlund, Magrretic-field fipoonires N I the Workplace:
Re fereuce Distribtrtiorr a~td fiposlrres ;PI Occupatiorrai Groirps. I nt J Occup Environ
Health, 1996. t ( 3 ) : p. 226-238.
Johansen, C. and J.H. Olsen, Risk of carrcer umorrg Dar~ish utility workers-a rzatiornvide
cohorr sfzrdy. Am J Epidemio!, 1998. 147(6): p. 548-55.
S ko tte, J . H., fipomre to power-fi-equency e/ectromagrreticfieids in Dermark. S cand J
Work Environ Health, 1994. t O ( 2 ) : p. 132-8.
Kheifets, L. I . , S.J. London, and J.M. Peters, Lezrkemia ris& utid occtrpatior~al elrctricfield
expoonrre i r ~ Los Attgeles Cozrtlîy, Californi. Am I Epidemiol, 1997. 146(1): p. 87-90.
Chapter 1 Introduction
56
Preston-Martin, S., B.E. Henderson, and J.M. Peters, Descriptive epidemiology of cerrtral
tlervolrs gstem neoplasms irt Los Alrgeles Cozrrtty. Ann N Y Acad Sci, 1 982. 38 1 : p. 202-
8.
Gallagher, R. P., et al., h i r ~ caricer ami exposrre to electromagnetic fie lds [letter/. J
Occup Med, 199 1. 33(9): p. 944-5.
Speers, M..4., J.G. Dobbins, and V.S. Miller. Occzpatiorral expomres arrd braiil carlcer
rnortality: a prelhitray stzrdy of east Texas residents. Am J Ind Med, 1988. 13(6): p.
629-3 8.
Lin, R. S., et al., Ocarpatior~al expostrre to electrornagr~etic fields and the ~~~~~reme of
hrairr tzrmors. An crrralysis ofpossible associalior~s. J Occup Med, 1985. 27(6): p. 433-9.
Mc Laughlin, J .K., et al., Occrrpatiorral risks for irrtracrarrial g/iornas irr Swedert. J Natl
Cancer Inst, 1987. 78(2): p. 253-7.
Baris, D., et ai., A rnortaiiîy study of electrical rrtility workers irr @rebec. Occup Environ
Med, 1996. 53(I): p. 25-3 1.
Hamngton, J. M., et al., Ocarpatiorra l expomre to magnelic flelds i r l relatiorr to morta lit):
fiom hrairr cancer amorlg elecrriciîy generation ar~d trartmissrûr~ worhrs. Occup
Environ Med, 1997. 54(1): p. 7- 13.
Rodvall, Y ., et al., Occupational exposire ta magnetic fields arrd brairi trtmoirrs itr
cerrtrd Sweden. Eur J Epidemiol, 1998. 14(6): p. 563-9.
Linet, M. S., et al. , nowHodgkirr k lymphoma and occ~patiorr N I Sweden: a registry hased
arrafysis. Br J Ind Med, 1993. 50(1): p. 79-84.
Chapter 1 Introduction
57
Figgs, L. W ., M. Dosemeci, and A. Blair, Ollited States rion-Hodgkids lymphornn
strrveillance by occlrpatiorr 1984- 1989: a twerrîy-forir state dearh certijicate strrdy Am J
Ind Med, 1995. 27(6): p. 8 17-3 5.
Dubrow, R. and D.H. Wegman, Camer and oc or pari or^ iti Massach~set&- a dearh
certrjkate sttrdy- Am J Ind Med, 1984. 6(3): p. 207-30.
Mi 1 ham, S., Jr., Mortaiity in workers exposed to eiecfr~magmttcfieids~ Environ Heal th
Perspect, 1985. 62: p. 297-306.
S cherr, P. A., G. B. Hut c hi son, and R. S. Neiman, Non-Hodgkirt 's lymphoma arid
occrrpatiorral expowe. Cancer Res, 1992. 52( 19 Suppl): p. 5 503 s-5 509s.
Schumacher, M.C. and E. Delzell, A death-certrficate case-control stlrdy of rion-
ffodgkirl's Iymphoma and oca~patiorr in merl irl North Curolina. Am J Ind Med, 1 988.
13(3): p. 3 17-30.
Blair, A., et ai., Evaittation of risks for rmt-Hodgkirr 's lymphoma by ocmpatior~ at~d
indrlsity expomresfr.om a case-corrtrol strrdy. Am .J Jnd Med, 1 993. 23(2): p. 3 0 1 - 1 2.
Arnerican Cancer Society, The Non-Hodgkin's Lymphoma Resowce Cmter (website):
htfp: w v . cancer. org. 1 999.
Greiner, T.C ., L. J. Medeiros, and E. S. Jaffe, Non-ffdgkiri's iyrnphoma. Cancer, 1995.
75( 1 S~pp l ) : p. 370-80.
Becker, N., J. Claude, and R. Frentzel-Beyme, Cartcer risk of arc welders exposed to
frrnws corrtairritg chromirrm artd nickel. Scand J Work Environ Health, 1 98 5. 1 1(2): p.
75-82.
Chapter 1 Introduction
58
Danielsen, T.E., et al., Iricidertce of carzcer amorig welders of mild steel alid other
shipyardrvorkers. Br J Ind Med, 1993. SO(12): p. 1097-103.
Vagero, D. and R. Olin, Orcidence of carlcer ;ri the elrctronics itrdi~stry.' ~(sirrg rhe tIew
Swedish Caticer Etivirormetit Registry as a screertirrg itrstnrmerrt. Br J Ind Med, 1 98 3 .
40(2): p. 188-92.
Nelemans, P. J ., et a/. . Me/uirorna mzd ocarpatiorr: resdts of a case-cor1 frol study itl I'he
Nefherlartds. Br J Ind Med, 1993. 50(7): p. 642-6.
De Guire, L., et al., hcreased incider~ce of malignu~~t melunoma of the skirr in worket-s U r
n fe/ecommw~ications iridmtv. Br J Ind Med, 1 988. 45( 1 2): p. 824-8.
Vagero, D., et ai., Car~cer morbidity amor,g workers 111 the telecommuncaio~~s it~d24stIy.
Br J Ind Med, 1985. 42(3): p. 19 1-5.
Guénel, P ., rr al., inciderrce of cartcer in persorts with oca~patiotra l e x p o ~ ~ ~ r e to
elcicfromu~ret~c~e/ds in Denmark. Br J Ind Med, 1 993. 50(8): p. 75 8-64.
Pearce. N.. J . Reif, and J. Fraser, Case-control stttdies of cancer N I New Zealarrd
electrical rvorkers. Int J Epidemiol, 1989. 18(1): p. 55-9.
Robinson, C. F., M. Petersen, and S. Palu, Mortai~ty patterrrs amoflg e/ectrica/ workers
ernployed I I I the US . corrstnrctior~ ir~drtstry, 1982-1987. Am Am Ind Med, 1999. 36(6): p.
6 3 0-7.
Tynes. T.. J.B. Reitan, and A. Andersen, hiciderice of cartcer amorrg,vorkers itr
Nonvegiart hydroelectric porver compan~es. Scand J Work Environ Heal t h, 1 994. 20(5):
p. 339-44.
Chapter 1 Introduction
1 52. Litovitz, T. A.. et al.. The role of temporal setrsiig in b~oelecfromapetic effecls.
Bioelectromagnetics, 1997. 18(5): p. 388-95.
Chapter 1 Introduction
Correla tions between indices of electric and magnetic field exposures in Ontario electric utility workers
2.1. Abstract
The preponderance of epidemiologic studies that have investigated the relation between
cancer and occupational exposure to electric and magnetic fields have relied on traditional
surnrnaries of exposure such as the cumulative time weighted arithmetic or geometric mean
exposure. Both animal and ceilular studies support the consideration of alternative measures of
exposure that are capable of capturing threshold and intermittent masures of field strength. The
analyses presented in this chapter were undertaken to identiQ a series of suitable exposure rnetrics
for an ongoing cancer incidence study within a CO hort of Ontario electric utility workers.
Principai components analysis (PCA) and correlational analyses were used to explore the
relationships within and between series of electric and niagnetic field exposure indices. These
were performed usuig exposure data coliected using personal moniton wom by a sample of 820
Ontario Hydro workers which yielded 4,247 worker days of measw-ernent data.
For both electrk and magnetic fields, the fk t tactor axis of the PCA was associated with a
series of intercorrelatai indices that included the geometric mean, median and arithrnetic rnean. A
considerable portion of the variability in electric and magnetic field exposures was accounted for
6 1
by two other principal component axes. The standard deviation and other threshold related field
measures were highly correlated with the second principal component factor a i s . To a lesser
extent, the variability in the exposure variables was explained by time dependent indices which
consisted of autocorrelations at five minute Iags and average transitions in field strength. These
results suggest that the variability in exposure data can only be accounted for by using several
esposure indices, and consequently, within this cohort, a series of metrics should be used when
exploring the risk of cancer that results fiom exposure to electric and magnetic fields.
Furthermore, the poor correlations observed between indices of magnetic and electric fields
reinforce the need to take both fields into account when performing cancer risk assessment in
t hese workers.
2.2. Introduction
Epidemiological studies of workers with high exposure to magnetic fields have
inconsistently demonstrated an increased risk of brain cancer and leukemia [ 1-61. The association
between field exposures and the occurrence of cancers at other sites has been examined to a lesser
extent and are inconclusive [7-IO]. For the most part, the study of eiectric field exposures and
cancer has been neglected though recently contradictory findings have been observed in three
separate study populations [ 1 1-1 31.
As outIined extensively in earlier sections (1.3 and 1.4), no clear health or biologic effects
resulting fiom exposure to electric and magnetic fields have been established. In addition, there is
Chapter 2 Correlations between indices of elect~ic and magnetic fields
62
no widely accepted biophysical mode1 to predict relevant exposure [ 141. For these reasons, there
is uncertainty about the appropriate exposure metric to use to evaluate biological effects in
epidemiologic studies. In studies of electric utility workers, the association between electric and
magnetic fields and cancer has typically been evaluated with a cumulative time-weighted average
(TWA) measure of exposure. As stated before, several experimental studies have suggested that
other aspects of field exposure are relevant tu the study of cancer. These field exposures include
intermittent or pulsed fields [IS, 161, and exposures within a specified window for a minimum
duration of time [17]. Inconsistencies in the findings of irl vitro, in vivo and epidemiological
studies underscore the need to clan@ the relationships between various measures of exposures.
The goal of the analyses presented in this chapter, was to examine the correlations within
and between a series of electric and magnetic field exposure indices using principal components
analysis (PCA). Identifjing metrics that are highiy correiated with each other allow redundant
metrics to be eliminated. Further, PCA can identiQ poorly correlated metrics which can be usehl
in recognizing independent aspects of exposure. PCA represents a suitable and objective means to
reduce a larse number of exposure indices to a smaller and more manageable subset that captures
the variability in exposure data.
S imilar met hodology has previously been empioyed to examine summary measures of
maçnetic fields [18, 191. The analysis presented in this chapter builds upon these analyses by
examining electric fields in addition to magnetic field exposures. Furthemore, a greater variety of
Chapter 2 Correlations between indices of electric and magnetic fields
63
exposure metrics have been exarnined. Correlationai analyses have also been used elsewhere to
evaluate different possible metrics [19-241. However these studies were limited by either smatl
sarnple size [Z], incomplete ascertainment of electric field metrics [19-231, exposure assessrnent
in non electric utility workers [2 1-23] or sizeable data losses [24].
Analyses presented herein are based on electric and magnetic field measures that were
collected in a sample of 820 Ontario electric utility workers using the ~ o s i t r o n ~ ' monitor
(Positron Industries Inc, Montreai, PQ) [25]. This monitor records field measures according to a
series of 16 levels (or bins), every minute, over the course of a workday and therefore, a large
number of exposure metrics could conceivably be created. Metrics were developed d e r
reviewing the biological and epidemiological literature. They were created based on the results of
animal and cellular studies and for exposure surnrnaries for which research information is lacking.
The bioiogical database relevant to "other metrics" is limited. Most studies have used continuous
exposures at one or more field intensities, and most have used only magnetic field exposure
systerns [14]. Where effects have been reported, there has generally been little success in
replicating them in other independent laboratories [14, 26, 271. To Our knowledge, no
expenmental study has reported a significant positive finding and then proceeded to investigate
different categorizations of exposure nor has any research been undertaken with the objective of
eliminatinç specific exposure indices fiom contention. Previous analyses of the Ontario Hydro
workforce revealed positive associations between the cumulative geometric and arithmetic mean
exposure to electric fields and certain cancers [12]. The present study was undertaken to evaluate
Chapter 2 Correlations between indices of electric and magnetic fields
64
which aspects of field exposure were the most important predictors of cancer risk, and further. to
determine whether alternate definitions of field exposure were related to cancers for which
positive findings have been reported elsewhere.
Exposure metrics that have been previously used in environmental epidemiological studies
are herein referred to as traditional exposure indices. Of these, my analyses included the
following summaries of daily exposure: the geometric mean, arithrnetic mean, standard deviation,
ninety-fifth percentile and median. Alternate measures of exposure were constructed fiom the
recorded monitor readings and are referred to in this chapter as 'rrorr-tradifiorral' indices. These
indices included the percentage of time spent above a threshold, the average transition in field
strength, the arithrnetic\geometrîc mean field exposure at or above a threshold, the time spent
above a threshold for a minimum duration, and autocorrelations at various lags.
2.3, Methods
Electric and Magnetic Field Exposure Assessrnent
Direct measurements of worker exposure under usual working conditions were obtained
using the Positron mode1 378 100 personal exposure monitor (Positron Industries, Montreal,
Quebec, Canada). The Positron is a portable pocket sized, battery operated electronic instrument
designed to monitor inunediate personal environmental exposure to 50/60 Hz magnetic and
electric fields. The Positron monitor filters the electric and magnetic field signal and was set up to
limit the measurement to 60 Hz fields. The devices were used to record electric and magnetic
Chapter 2 Correlations between indices of eIectric and magnetic fields
65
fields in the environment each minute. Each reading was assigned according to 16 predefined
exposure intervals or bins. The exposure intervals were 0-0.6 1, 0.6 1 - 1 2 2 . 1.22-2.44 . . . -5,000-
1 0,000, > 10,000 V/m for electric fields and 0-0.0 12, 0.0 12-0.024, 0.024-0.048, . . . ., 100-200,
>ZOO pT for rnagnetic fields. Each measure was assigned the value of the midpoint of the interval.
The monitors were tested and calibrated before use and at regular intervals during the study. A
more detailed description of this monitor as well as its' abitity to differentiate exposures by
occupational group and to obtain high compliance in workers has been documented [25, 281.
Measurements were originally performed on 895 workers that were sampled by job title
and work location. These measures were taken over the course of a five-day work week and
were defined by person, occupational group, work site and day. My analyses are based on the
daily sumrnary measures of electric and magnetic fields of 820 workers from 17 occupational
groups. Those with only electric or magnetic field exposure were not included, and this accounts
for most of the 75 workers removed fiom the 895 used in the Tri-Utility Study [Z]. The missing
data from these 75 was due to instrument failure, or strong evidence that the monitor was kept
close by (and so suitable for magnetic field exposure estimates) but not worn (and therefore not
valid electric field data). There were two worker records that could not be retrieved, due to
defects that appeared in the discs containing the raw data required to calculate the new metrics.
Chapter 2 Correlations between indices of electric and magnetic fields
Although estimates of home exposures were made for a sample of workers, these
exposures were not included in these analyses as the primary objective was to evaluate the risk of
cancer associated with occupational exposures. Residential measurements taken as part of the
Ontario Hydro sîudy of childhood leukernia indicate that arithmetic mean exposures to magnetic
field are on average 5-6 times higher among employees of Ontario Hydro when compared to
residential exposures [29]. Similariy, electric field exposures are approximately 2.5 times higher in
Ontario Hydro workers than in residences [29]. In general, the lack of association between home
and workplace exposures reduces the likelihood that our results will be confounded [30].
Exposure metrics
Only thresholds levels that were above the average exposure level of al1 workers were
considered in these analyses. It was also decided to exclude those metrics where positive
exposures occurred in less than 5 percent of the worker days. For example, for only 16 of 4247
worker days (0.4%) was a worker exposed to a threshold of 200 pT for a continuous interval of
15 minutes, and therefore, this metnc was dropped. It was felt that these metrics, which were
typicalIy high threshold exposures for a minimum period of time, would have little power to
discriminate disease status in subsequent case-control analyses. A total of 33 and 34 'non-
traditional' metrics of electric and magnetic field exposures, respectively, were included in our
analysis (Tables 2- 1,2-8).
Chapter 2 Correlations between indices of electric and magnetic fields
Statistical Analyses
Principal components analysis (PCA) may be used to analyse a set of interrelated
variables. The original variables are transfonned into a smaller set of uncorrelated variables that
are referred to as principal component axes. An a i s is a linear combination of a subset of
variables with which it has high correlation. One purpose of PCA is data reduction which is
achieved by explaining as much of the total variation in the complete set of variables with as few
principal components as possible. The principal component axes are uncorrelated, and thus
represent independent sources of variability. PCA is a frequently used tool in questionnaire design
and validation and is described in great detail in several statistical texts [3 1-33].
Tt follows that when PCA is applied to exposures of electric and magnetic fields, the axes
are summary measures which are representative of those exposure metrics that are most highly
correlated with them. These axes are ordered by the amount of the overall exposure variability
they explain, that is by their variance. The number of principal component axes is determined
based on the variances obtained from the correlation matrix of the PCA. In al1 our analyses,
principal component axes with a variance that exceeded one were retained.
PCA was performed on electric and magnetic fields separately. The principal component
axes were rotated so as to maximize correlations of a smaller number of metrics with each axis.
The Varirnax rotation method was used which maintained the orthogonality of the axes derived
from the PCA [33]. However, because the rotated axes correlate with fewer rnetrics, the ability to
Chapter 2 Correlations between indices of electric and magnetic fields
identiS, important metrics was improved.
PCA was performed in three stages. First, PCA was performed only on the 'non-
traditional' exposure summaries as a means of data reduction. Traditional measures were not
included in these analyses because their exposure patterns would be overwhelrned by the large
number of 'non-traditional' exposures. Indeed, preliminary analysis of our data demonstrated that
this was the case. The first stage of PCA yielded a series of axes. From these, the metric that was
most highly correlated with each axis was selected and retained for tùrther analyses. Selecting
metics in such a manner removed those metrics that were redundant.
In the second stage of analyses, PCA was performed on those metncs selected in the first
stage in combination with the traditional metrics of exposure. This allowed the relationships
between the selected non-traditional and the traditional measures of exposure to be examined.
Correlation matrices of the exposure indices that were strongly associated with the factor axes
derived in the second stage PCA were constnicted in order to fùrther evaluate the relationships
between our selected metrics, and make cornparisons to previous studies that have used this
technique [?O, 331.
Finally, PCA of the selected metrics and traditional exposure summanes was done within
occupational groupings and for combinations of occupational groupings and work site locations.
These analyses allowed us to detennine which metrics best accounted for the variability of
Chapter 2 Correlations between indices of electric and magnetic fields
69
exposure by occupational group and pennitted comparisons to results derived using al1 workers.
PCA for combinations of job title and work-site pennitted us to examine whether one metnc
could adequately represent the variability in field exposures within each such stratum.
2.4. Results
Electric fields
The 38 electric field exposure metrics used in these PCA analyses are outlined in Table 2-
1 . For the traditional exposure summaries, the mean daily exposure by occupational grouping is
shown in Table 2-2. The arithrnetic mean electric field exposure was markedly higher in
powerline maintainers (75.9 V/m), power maintenance electricians (60.8 Vlm) and foresters (28.1
V/m). Workers in these three groups also had higher mean exposures as defined by the other
t raditional metrics examined, though the range in values between job ti tles varîed considerably.
For example, the arithmetic mean for customer service representatives and powerline maintainers
differed by a factor of approximately seven while for the geometric mean, there was only a
twofold difference in exposure.
The overail mean electnc field for those workers with valid site exposures (10.4 V/m) was
lower than the mean based on measures taken from the entire sample of workers (22.1 Vlm)
(Table 2-2, 2-3). This can be attributed to the exclusion of those occupational groups with high
exposures (e.g., powerline maintainers, power maintenance electricians and foresters). These
workers spend most of their time away fiom their designated work locations, and hence, it was
Chapter 2 Correlations between indices of electric and magnetic fields
70
decided not to strati@ these exposures by work site. Electric field work-site exposures were
highest for measurements taken at the helipad (55.7 Vlm ), though this mean is based on only six
observations (Table 2-3). Relative to other work-sites, elevated electric field exposures were
observed at transformer stations, seMce centres, district offices and warehouses.
PCA on the 33 non-traditional electric field metrics yielded six factor axes. The electric
field metrics that were selected to represent each of these axes are displayed in Table 2-4. The
first axis accounted for 42% of the overall variance among electric field exposures. This axis
appeared to be representative of the percentage of time spent above a threshold and the metric
that was most highiy correlated with this mis was the percentage of time at or above 2500 Vlm
( ~ 0 . 9 5 ) . Similarly, the second axis was representative of the arithmetic mean of signals at or
above 156 V/m (r=0.94). Continued exposure to electric fields at or above 40 Vlm for five
minutes or longer was highly correlated with the third factor axis (r=0.94). As expected. this
metric was also highly correlated with the percentage of time spent above 40 Vlm. The average
electric field transition as measured by the difference in bin levels for adjacent reading was
correlated with the fifth factor axis, while the arithmetic mean of exposures at or above 2500 V/m
was hishly correlated with the last factor axis.
PCA was then performed on the above six exposure summaries and traditional measures
which resulted in three factor axes. The geometric mean (r=0.93), median (r=0.93), arithmetic
mean (0.88) and the percentage of time at or above 2500 Vlm ( ~ 0 . 8 8 ) were highly correlated
Chapter 2 Correlations between indices of eiectric and magnetic fields
7 1
with the first factor axis (Table 2-5). The arithmetic mean of electric field exposures at or above
1 56 Vlrn (r=O. 8 1 ) the arithmetic mean of electric fields at or above 2500 V/m (~0.86) and the
standard deviation (r=0.77) were highly correlated with the second factor axis. The average
electric field transition and autocorrelations at a five minute lag were highly correlated with the
third factor mis. However, the overall proportion of variability explained by this axis was only
16%. The percentage of time at or above 40 V/m was not strongly correlated ( 6 0 . 5 5 ) with any
of the three factor axes.
Pearson correlation coefficients between the metrics that were highly correlated with the
factor axes fiom the second stage PCA are presented in Table 2-6. The median and geornetric
means were highly correlated with each other (~0.92) but less so with the arithmetic mean
W0.80). Similarly, the percentage of time at or above 2500 Vlrn was highly correlated with the
geometric mean and the median. The anthmetic mean was correlated with the standard deviation
(r=0.8 1). The standard deviation and the arithmetic mean of electric field exposures at or above
2500 Vlm, which were both correlated with the second factor axis, were modestly correlated with
each other ( ~ 0 . 7 1 ). Finally, the metrics that were most highly correlated with the third factor
axis were weakly correlated with each other. These metrics consisted of the average transition in
the electric field and autocorrelations at a lag of five minutes (r=-0.29). Moreover, these metrics
were uncorrelated with al1 of the other exposure metrics included in the PCA (rc0.30). Similarly,
the percentage of time spent above 40 V/m was poorly correlated with al1 metrics used in the
second stage of the PCA (r(0.58).
Chapter 2 Correlations between indices of electric and magnetic fields
72
When analyses were performed separately for each of the 17 occupational groups, PCA
yielded between two and four axes (Table 2-7). The median and geometric mean were both
highly correlated with each other (P0.9) and with either the first or second factor axis (r>0.8) in
16 of 17 of the occupational groups. The standard deviation was typically correlated with a
factor axk that was not representative of the geometric mean or median. For each occupational
group, the metrics that were most highly correlated with the factor axes are tisted in Table 2-7.
For PCA that was performed within occupational group and work-site strata, findings were
Iimited by small numbers within many cells, specifically, the conditional variance for some metrics
was zero. Nonetheless, PCA yielded at least two factor axes in 1 O 8 of 109 strata.
Magnetic fields
A list of the 39 magnetic field exposure metrics examined using PCA is provided in Table
2-8. Elevated exposure to magnetic fields, as defined by the anthmetic mean were observed
arnonç operators ( 1.6 PT). protection and control technicians ( 1.2 gT) and power maintenance
electricians ( 1 . 1 PT) (Table 2-9). Work-site exposures were highest at hydroelectric generating
stations (3.1 PT), district offices (1 -3 PT) and transformer stations ( 1 . 1 PT) (Table 2- 10).
When PCA was applied to the 34 non-traditional measures of magnetic fields, six factor
axes were identified (Table 2-1 1). The first, accounted for approximately 42% of the variance of
the exposure indices. The average of exposures at or above 3pT for at least five minutes was
most highly correlated with the first axis ( ~ 0 . 9 7 ) . Similarly, the percentage of time spent above
Chapter 2 Correlations between indices of electric and magnetic fields
0.8 PT for at least five minutes was most correlated with the second factor axis (r=0.95).
Autocorrelations at a five minute lag, the geometnc mean of the magnetic tield at or above 0.8
pT, and the geometric mean of exposures at or above 0.8 pT were correlated with the fourth
through sixth axes, respectively. As with electnc fields, the six metrics that were the most highly
correlated with each of the factor axes were retained for further anaiysis.
The combined PCA of the previously selected magnetic field metrics and traditional
sumrnaries of exposure resulted in three factor axes (Table 2- 12). The geometric mean, median
and geometric mean of magnetic field exposures at or above 0.8 pT for at least five minutes were
strongly correlated with the first axis (r>0.90). To a lesser degree, the arithmetic rnean was also
correlated with the first factor axis (r=0.77). The highest correlations for the second factor a i s
were observed with the standard deviation (r=0.93) and the ninety-fifih percentile (r=0.78). The
correlation between the arithmetic mean of the magnetic field and the second factor axis was more
modest (r=0.67). Autocorrelations at a lag of 5 minutes (r=0.83) and the average transition in
magnetic field (r=-0.71) were correlated with the third factor a i s . The percentage of time spent
above 0.8 p T was poorly correlated with al1 factor axes (60.55).
The correlations between those indices evaluated dunng the second stage of the PCA are
provided in Table 2- 13. The geometric mean and median were highly correlated with each other
(r=0.92), but less so with the arithmetic mean (r=0.81, r=0.78). Similarly, the geometric mean of
magnetic field exposures at or above 0.8pT was highiy correlated with the geometnc mean and
Chapter 2 Correlations between indices of electric and magnetic fields
the median (r>0.85). As had been found for electric fields, the metrics correlated with the third
factor axis were weakly correlated with each other and with the other metrics (r<0.30). These
metncs consisted of autocorrelations at a five minute lag and the average transition in magnetic
fields. The percentage of time spent above 0.8 PT for at least five minutes was poorly correlated
with the anthmetic mean, arithmetic mean, median and threshold indices exceeding 3.0 PT.
When PCA was performed separately by occupational group, the number of factor axes
that resulted ranged from two to four (Table 2-14). The geometric mean and median were both
highly correlated (r>0.85) with the first or second factor axis in al1 17 occupational groups. In 15
of 17 groups, the standard deviation was most highly correlated with a factor a i s that was not
representative of the geometric mean (or median). As with electric fields, PCA performed within
strata ofjob title and work-site location yielded at least two factor axes for 108 of 109 celis. The
geometric and anthmetic means were correlated with the same factor axes in 59% of these strata.
Electric and magnetic fields were weakly correlated with each other (Table 2- 1 5). The
maximum correlation observed among the five traditional measures of electric and magnetic fields
was r=0.42. The correlation between traditional indices of electric and magnetic fields was even
weaker (60.10) when analyses were restricted to occupational groupings with high levels of
exposure to magnetic fields (results not shown).
Chapter 2 Correlations between indices of electnc and magnetic fields
2.5. Discussion
The analyses presented in this chapter reveal that the geometrk and arithmetic mean
exposures, by themselves, do not adequately capture the variability of either electric or magnetic
field exposures. indeed, the PCA suggests that field exposures are best characterised by at least
three factor components. These components which include meures of central tendency, the
standard deviation and indices that represent threshold exposures account for the majority of field
variations. To a lesser extent, the variability in field exposures are dependent on temporal metrics.
These rnetrics include autocorrelations at a five minute lag interval and the average transition in
field strength. A noteworthy finding is the poor correlations that were observed between indices
of electric and magnetic fields, which implies that both field exposures should be used when
assessing the risk of cancer (or other endpoint) within this population.
In the identification of a set of etectric and magnetic field exposure variables that predict
the probability of cancer in a multivariate model, an exposure variable is more likely to be selected
as an important contributor if it fùlfills three conditions. These conditions are that it be highly
variable, be poorly correlated with the other variables in the model, and along with these other
variables, be correlated to the probability of disease. The PCA and Pearson correlations analyses
identifies a series of metrics that satisfy the first two conditions.
The third condition was addressed by taking into account findings from biological studies
of extremely low frequency fields (ELFs). These studies suggest that potentially relevant aspects
Chapter 2 Correlations between indices of electric and magnetic fields
of exposure include ( 1 ) measures of central tendency, (2) threshold or peak exposures, (3)
duration of exposure above a threshold and (4) intermittent exposures. In the final selection of a
subset of metrics, electric and magnetic field exposures have been selected to address these four
aspects, and additionally, to satisfy the first two conditions ofa strong predictor variabie as
outIined in the previous paragraph.
Ideally, one would attempt to discriminate between case and control status using al1
possible metncs of exposure. For this study population, this would involve constructing
cumulative lifetime, and/or average exposure, estimates across working history records.
Performing such tabulations for 87 combined electric and magnetic field indices that are
dependent on both occupational group as well as site would clearly be quite onerous. Further,
such a tact would increase the likelihood of a spurious finding due to the dangers associated with
multiple testing. These analyses identi$ a manageable series of indices for which lifetime
esposures can be constnicted using a job exposure matrix. Further, the correlations between the
sampled workers can be compared to similar exposure assessrnent perforrned in other
occupational environments.
The results of our PCA suggest that the first, and most important, aspect of exposure that
needs to be captured is a rneasure of centra1 tendency. The selection of such a rnetric is somewhat
subjective and Our analysis suggests the consideration of three indices: the anthmetic mean, the
çeometric mean and the median. When PCA was performed on the entire data set, these three
Chapter 2 Correlations between indices of electric and magnetic fields
77
indices were correlated with the same factor axis and were highly intercorrelated. The median and
geometric mean were strongly correlated with the same factor axis for al1 the occupational groups
for magnetic exposures and al1 but one for electric field exposures. The effect of selecting one
over the other for subsequent risk estimation would be negligible. The arithmetic mean was less
correlated with the first factor axis than the geometric mean and median. Further, it did not
always fa11 on the same axis as the geometnc mean (or median) when analysis was performed by
occupational grouping. Because the arithmetic mean is more sensitive to skewed data (i-e., peaks
and thresholds) it is more proficient at modelling threshold effects than the geometric mean (or
median). In contrast, the median is dependent on continuously high field exposures. Modellinç
the arithmetic mean would be of value for comparative purposes as most epidemiologic studies of
ELFs have relied on this measure. However, given that other metrics are available to mode1
threshold effects, selecting the geometric mean to represent the aspect of central tendency would
be valued.
The standard deviation was highly correlated with the second principal component factor
axis for both electric and magnetic field exposures. Further, this metric was poorly correlated
with both the geometric mean and median. A knowledge of both the centra1 tendency and spread
of the data is essential in descnbing the distribution of exposure data. A large standard deviation is
also an indicator of fiequent andor large changes between high and low field exposures. For these
reasons, retaining the standard deviation as a metric in further risk estimation has merit.
Chapter 2 Correlations between indices of electric and magnetic fields
78
A number of indices that are suited for modelling threshold effects were included in the
analyses. These include the traditional exposure summaries represented by the 95' percentile and
the standard deviation. The anthmetic and geometnc mean for exposures received above a
threshold and the proportion of time spent above various thresholds represent two other series of
threshold type metrics. There are important differences between these two series that bear funher
elaboration. The arithmetic mean of field exposures received above a certain threshold may be
valuable for modelling high exposures of shon duration. For example, a worker may have a high
arithnietic mean for exposures received above 3 PT, even though such exposures were rarely
encountered. On the other hand, the percentage of time spent above selected threshold values is
more relevant for assessing effects resulting fiom chronic, or prolonged, elevated exposures.
Several threshold measures of electric fields were correlated with the second factor axis. The
correlation was strongest for the arithmetic mean of electric field exposures at or above 2500 V/m
( ~ 0 . 8 6 ) followed by the arithmetic mean of exposures above 156 Vlm (r=0.8 1) and the standard
deviation (r=0.77). The use of one of these metrics to model cancer risk would be recommended.
It should also be noted the percentage of time spent above the lower threshold value of 40 Vlm
for at Ieast five minutes was poorly correlated with al1 other metrics used in the PCA, and
therefore represents an independent aspect of field exposure. As before, experimental studies
suggest that if ELFs are involved in the carcinogenic process. they are likely to act as promoting
agents [I4, 26, 271 and such agents are characterized with the existence of a threshold, prolonged
exposure and reversibility of effects [34]. These aforementioned points suggest that it would be
worthwhile to model the effects associated with the percentage of time spent above lower electric
Chapter 2 Correlations between indices of electrïc and magnetic fields
field thresholds as well as the anthmetic rnean of exposures received above 2500 Vlm. Similar
reasoning could be applied to select magnetic field threshold indices for the purposes of cancer
nsk assessment.
As previously mentioned, expenmental studies have also found bioeffects resulting from
intermittent exposure to ELFs [15, 161. Possible candidate metrics to represent such exposures
are the average transition in electnc and magnetic fields as measures by the rnean change in bin
levels for adjacent samples. The PCA components analysis indicates that these metrics explain a
component of the exposure vanability as they are correlated with the third factor axis. Further
they were poorly correlated with the other metrics. For these reasons, one could also justiw
keeping autocorrelations at five minute lags for fùture cancer nsk assessment. However, it is
worth noting that the overall proportion of variance explained by the third factor axes is relatively
small for both electric and magnetic fields (<20%).
The findings for magnetic fields are consistent with other work that found high
intercorrelations between the geometric mean, anthmetic mean and fiactions of measurements
exceeding 0.5 and 1 .O pT [20]. The authors also found that the series of metrics consisting of
anthmetic mean, 95th percentile and fiactions of measures exceeding 5 and 10 pT accounted for a
significant portion of the vanability of exposure data as did a factor axis representative of the
standard deviation and fiaction of measures > 100pT.
Chapter 2 Correlations between indices of electric and magnetic fields
80
Similar analyses in Hydro Québec workers found that correlations between the arithmetic
and geometric means and alternative indices were strong (P0.8). These aiternative indices
included the 9oL percentile, and fraction of measurements exceeding 0.2, 0.39. 0.78. 1.56 and
6.25 p T 1241. However, within the Québec workers, the correlations between the arithmetic and
geometnc means were substantially weaker for lower thresholds of magnetic fields as represented
by the 20' percentile (60.45). The investigators observed similar patterns when electric field
measures were examined. The anthmetic and geometric mean were highly correlated with both
the fiaction of measures exceeding 20 and 78 V/m ( r X . 85). On the ather hand, these means
were weakly correlated with the 2oh percentile (r=0.29) . The results from the Ontario workers
also revealed weak correlations between the arithmetic and geometric mean and percentage of
tirne spent above lower threshold values in our study that increased with increasing threshold
values.
For both electric and magnetic fields, Armstrong and colleagues observed strong
correlations between the time weighted average ( W A ) and summaries of peak exposures [23 ] .
This Ied them to conclude that the use of the arithmetic mean as a summary measure would serve
as a reasonable proxy for assessing peak exposures. The arithmetic mean also has the desired
feature of avoiding the arbitrary choice associated with indices representing peaks or thresholds.
If cancer risk assessment could only be perfonned in our study population using one exposure
index, the arithmetic mean would be the preferred choice. This could be justified by the fact that
the arithmetic mean has modest correlations with measures that are highly correlated with the first
Chapter 2 Correlations between indices of electric and magnetic fields
two factor axes, specifically, the geometric mean (or median) and the standard deviation.
However, in a distribution that is not highly skewed, the arithmetic mean will be less effective in
identimng effects that are related to threshold or similar types of measure. Consequently. by only
using the anthmetic mean, potential biological effects associated with low threshold exposure
summaries rnay be missed.
The finding o f a weak correlation between indices o f electric and magnetic fields are
consistent with previous findings [19, 20, 231. Savitz and colleagues found higher correlations
between electric and magnetic fields among occupational groups than for different jobs held by
the same individual [20]. As a consequence, they concluded that a single measure of central
tendency appears to be adequate when exposures are assessed at the level o f job title. In contrast,
recent work has demonstrated that metrics other than the average magnetic field strength are
relevant in characterising occupational exposures in electric utility workers [19]. Specifically, the
study found that although average field strength was able to distinguish between high and low
exposed occupational groups it poorly discriminated exposures between highly exposed groups
[ 191. My PCA suggested that measures of central tendency explained most of the exposure
variability for the majority of the occupational groups. However, the number of principal
cornponent axes for the 17 occupational groups ranged fiom two to four suggesting that measures
of central tendency, by themselves, may not adequately account for the variability of either
magnetic or electric field exposures within occupational groups.
Chapter 2 Correlations between indices of electric and magnetic fields
82
The use of the series of metrics selected fiom these analyses to estimate cancer risk in the
cohort of Ontario electrical utiîity assumes the exposure mesures are representative of the much
larger cohort. Clearly, it is not feasible to address the potential associated bias by sampling al1
workers. However, it is important to note that occupational jobs that required exposure sampling
were identified by cornpiling a Iist ofjob titles observed in the case and control populations.
The measurement of electric fields is inherently more difficult than for magnetic fields. At
any point in time, the electric field measurement assessed with a personal monitor is influenced by
(i) the wearing location, (ii) the magnitude and direction of the local ambient electric field, the (iii)
posture of the body, and, to a lesser extent, (iv) the extent to which the worker is grounded. The
type of clothing generaily has little or no effect on the measurements (unless the monitor is worn
under wet clothing). In this study, workers were required to carry the monitor on a waist belt,
except when wearing a harness for work at elevation, in which case the monitor was attached to
the fiont strap of the harness. By having each worker Wear the monitor 7-8 hours each day for
five days, an average measurement could be made for the variety of body postures and electric
field environments normally encountered during the execution of the majority of routine tasks. To
account for the variation in job tasks that occur seasonally. members of relevant trade groups
were monitored throughout the year. Therefore, by specimng the wearing position of the monitor
and sampling many workers in each occupationai group for entire workdays, it is believed that the
electnc field average exposures can be used to quanti@ the relative exposures of these groups.
The range of electric and magnetic field mean exposure levels (Tables 2-6, 2- 10 ) suggests that
--
Chapter 2 Correlations between indices of electric and magnetic fields
the ability to separate occupational groups is at least as great for electnc fields as it is for
magnetic. However, stratification of specific occupational groups by work location revealed
fewer significant differences between sites for electric fields than for magnetic fields.
Further research shouid be undertaken to determine which metrics best differentiate
between workers who develop cancer compared to those who do not. Brain cancer and
haematological malignancies warrant particu1,r attention. This study identifies a senes of metncs
that serve as a starting point for such a discriminant analysis in this cohort. For electric fields, a
suitable series of metrics would consist of the geometric mean, the standard deviation,
autocorrelations at five minute lags, the arithmetic mean of exposure at or above 2500 V/m and
the average electric field transition in number of bins. Similariy for magnetic fields, an appropriate
series of metrics would consist of the geometnc mean, the standard deviation, autocorrelations at
five minute lags, the average mean of exposure at or above 3pT, and the average magnetic field
transition in number of bins. Cancer risk assessment using alternate definitions of electric and
magnetic field exposure is pursued in the next chapter.
Chapter 2 Correlations between indices of electric and magnetic fields
Chapter 2: References
Sahl, J . D . , M. A. Kelsh, and S. Greenland, Cohort and nested case-control studies of
hematopoietic cancers and brain cancer amung electnc utifity workers. Epidemiology,
1993. 4(2): p. 104-14.
Thér iau l t , G . , m al. , Cancer r iss usociated with occupational exposure to magnetic
fields among electric utility workers in Ontario and Quebec, Canada, and France:
1970- 1989 fpublished erratum appcars in Am J Epidemiol 1994 May 15; l39(1O):lO53/
fsee comments/. Am J Epidemiol, 1994. 139(6): p. 550-72.
H arr i ng ton, J . M. , et a 1. , Occupational eirposure tu magnetic fieus in relation to
morta!ity from brain cancer among electriciiy generation and transmission workers.
Occup Environ Med, 1997. 54(1): p. 7-13.
Sav i tz , D . A. and D . P. Loom is , Magnetic fietif erposure in relation to leukernia and
brain cancer mortaiiiy among electric utility workers [published erratum appears in Am
J Epidemiol 1996 Jul 15;144(2):205j. Am I Epidemiol, 1995. 141(2): p. 123-34.
Fe ycht i ng , M. and A. Ah1 bom , Magnetic fieus, leukemia, and central nervous system
tumors in Swedish adults residing near high-voltage power lines. Epidem iology , 1994.
S(5): p. 501-9.
Floderus, B. , et al. , Occupational txposure to electrornugnetic fields in relation to
leukemia and brain tumors: a case-control stiuiy in Sweden. Cancer Causes Control,
1993. 4(5): p. 465-76.
Chapter 2 Correlations between indices of electric and magnetic fields
85
Byus, C . V . , S.E. Pieper, and W. R. Adey, The @ec?s of low-energy 6û-Hz
environmental electromagnetic fields upon the growth-related enzyme ontirhine
decarboxylase. Carcinogenesis, 1987. 8(10): p. 1385-9.
Matanoski, G.M., P.N. Breysse, and E.A. Elliott. ElectromgneticfieId erposure and
male breast cancer [letter; commentj. Lincet, 199 1. 337(8743): p. 737.
Schroeder, J.C. and D. A. Saviu, Lymphorna and multiple myeloma mortai@ in
relation to magnetic field exposure arnong electric utility workers. Am J lnd Med,
1997. 32(4): p. 392-402.
Loomis, D.P., D.A. Savitz, and C.V. Ananth, Breast cancer mortaiity among femafe
elecrrical workers in the United States fsee commentsj. J Nat1 Cancer Inst, 1994.
86(12): p. 921-5.
Guénel, P., et al., Erposure to 50-Hz electn'cfeld and incidence of leukemia, brain
tumors, and other cancers among French elecrric utility workers fsee commentsl. Am J
Epidemiol, 1996. 144(12): p. 1 107-2 1.
M i l le r , A. B.. et al. , Leukemia following occupational aposure to 60-Hz electric and
magneric fields among Ontario electric u t i le workers fsee comrnentsj. Am J
Epidemiol, 1996. 144(2): p. 150-60.
K he i fets , L. 1. , S. J. London, and J. M. Peters, Leukemia risk and occupational elecf ric
f ieu erposure in Los Angeles County, California. Am J Epidemiol, 1997. 146(1): p.
87-90.
Chapter 2 Correlations between indices of electric and magnetic fields
86
NIEHS, Assessrnent of heakh Mects from erposure to power-line frequency electric and
mgneticfields, ed. C.J. Portier and M. Wolfe. 1998. Research Triangle Park, NC:
U.S. National Institute o f Health.
Beniashvili, D.S., V.G. Bilanishvili, and M.Z. Menabde, Low-frequency
electromagnetic radiation enhonces the induction of rat mmmary rumors by
nitrosomethyl urea. Cancer Lett, 199 1. 61( 1): p. 75-9.
Rannug , A., et al., A study on skin tumur formation in mice wirh 50 Hz mgnetic field
exposure. Carcinogenes is, 1 993. 14(4): p. 573-8.
Litovitz, T.A., D. Krause, and LM. Mullins, Effecr of coherence h e of the applied
magneric fieU on ornithine decarboxylase activity. Biochem Bioph ys Res Commun,
1991. 178(3): p. 862-5.
Sahl , J . D., et al. , Erposure to 60 Hz magnetic fields in the electric utility work
environment. Bioelectromagnetics, 1994. 15( 1): p. 2 1-32.
Zhang. J., 1. Nair, and I. Sahl. E m s function analysis of ELF magnetic field
aposure in the electric utility work environment. B ioelectromag net ics , 1 997. l8(5) : p.
365-75.
Sav i tz, D . A. , et al. , Correlations amng indices of electric and magnetic field txposure
in electric utiliîy workers. Bioelectromagnetics, 1994. 15(3): p. 193-204.
Wenzl , T. B., et al. , Magnetic fieid erposures in an automobile transmission plant.
Arch Environ Health, 1997. 52(3): p. 227-32.
Chapter 2 Correlations between indices of electnc and magnetic fields
87
Breysse , P. N. , et al. , 60 Hertz magnetic fieM erposure assessment for an investigation
of leukemia in telephone lineworkers. Am J Ind Med, 1994. L6(5) : p. 68 1 -9 1.
Armstrong, B.G., J . E. Deadman, and G. Theriault, Cornparison of indices of ambient
erposure fo WoQhertz electric and rnagneric fields. B ioelectromag ne t i f s . 1 990. 1 1 (4) : p.
337-47.
Dead man, J . E., B. G . Armstrong, and G. Theriaul t, Erposrtre to @Hz magnetic and
electric fiehi's ut a Canadian electric utifity. Scand J Work Environ Heal th. 19%.
22(6): p. 415-24.
Hé roux , P. , A dosimeter for assessment of exposures to ELF fields.
Bioelectromagnetics, 1991. 12(4): p. 24 1-57.
Lac y- H ul bert, A., I. C . Metcal fe, and R. Hesketh , Biological responses to
electromagnefic fields. Faseb J, 1 998. l2(6) : p. 395-420.
Loscher. W. and R. P. Liburd y, Animal and cellular studies on carcinogenic &ects 4
low frequency (50/6@Hz) magneticfiekis. Mutat Res, 1998. 410(2): p. 185-220.
Dead man, J . E. , et al. . Occupational and residential6û-Hz electromagne tic Jehi's and
high- frequency electric tranrients: exposure ossessrnent using a new dosimeter
fpublished erratum appears in Am Inâ Hyg Assoc J 19% Jun:57(6):58@3/. Am Ind
Hyg ASSOC J, 1988. 49(8): p. 409- 19.
Ontario H ydro, Summary of Electnç anà Magneric Field Measurements to J une 16th. .
1989, Ontario Hydro: Report ID: HSR-IR-89-2.Toronto.
Chapter 2 Correlations between indices of electric and magnetic fields
88
3 0. Knave, B . , Eïectric and magne tic fields and heafth oictcotnes-arr overview. Scand J Work
Environ Health, 1994. 2O(Spec No): p. 78-89.
3 1 . Rawlings, 1. O., Applied regression at~aiysis : a research fooï. 1 98 8, Pacific Grove,
California: Wadsworth and Brooks.
3 2 . Kleinbaum,D.G.,L.L.Kupper,andK.E.Muller,Appliedregre~sionat~a&sisarrdof~~er
muhivariate methods. 1988, Boston: PWS-Kent Publishing Co.
3 3 . Reyrnent, R. and K. G. Joreskog, Appliedjuctor uttaiysis i t ~ the rlatrrral scierices. 1 993,
New York, N.Y.: Cambridge University Press.
3 4. Pitot, H.C. and Y. P . Dragan, Facts amd theories corlcenrirlg the mechar~isms of
ccrrcirioger~esis. Faseb J, 1 99 1 . S(9): p. 3280-6.
Chapter 2 Correlations between indices of electric and magnetic fields
Table 2-1: List of electric f d d esposure indices used in the first stage o f principal componen ts anal ysis
-- - - --
Variable name Number of Description met rics
- -
ET-TH {s) 4
EAV-TH (s)
EGAV-TH {s)
EJAG
E-AVTRAN
EAV-TH 1 1
EGAV-TH 1 1
E-MEDIAN 1
E-STD 1
E-95 1
Percentage of time the electric field is at or above x={40. 156,625. or 2500 V/m).
Aritlunetic mean; for electric field exposures at or above s={40, 156- 625. or 2500 V/m) .
Geometric meani for clectric field exposures at or above x={40, 156,625 or 2500 V/m
Pcrccntage of electnc field masures that differ by at least 2 bins.
Averagei electric field transition in numbcr of bins for adjacent sarnples taken one minute apart
Percentage of Ume at or above s=40 V/m for at Ieast ~ ~ ( 5 . 1 5 minutes) and s= 156 V/m for at Icast y { 5 minutes).
Autocorrelations of s={ 1.5.15 or 30 minute lags).
Average of values at or abovc s=40 V/m for at least -{5. i 5 minutes) and s= 1 56 V/m for at Ieast y={5 minutes )
Geometric mean
Standard deviation of electric field measure (in V/m)
95th pcrcentile
' The average was calculated using (i) total time above the threshold and (ii) the total time for the period.
Chapter 2 Correlations between indices of electric and magnetic fields
Table 2-2: Mean daiiy exposure (in V/m) for traditional measures of electric fields, by occupational group, Ontario electric utility workers
Occupational - -
Daily Arithmetk Ceometric Median Standard 95"'
Croup measures mean mean deviation Penuntile
Clcrk 23 1 11.0 5.1 5.8 23.3 33. l
Controi maintainers 119 7.6 2.4 2.6 38.0 17.7
Customcr stvicc rt~re.sentative 13 1 1 1 . 1 5.8 6.8 21.7 29.9
Furcstcrs 2 10 28.1 3.9 4.0 90.3 129. I
Inspecter 6 3 5.5 1.7 1.7 14.4 23.1
Mcter rcader 93 7.6 3.1 3.2 18.2 26.6
Opcrator 353 12.8 3 .O 3.6 62.5 34.6
f'iitverlinc maintaincm 484 75.9 11.6 14.5 196.2 368 3
f'rol~ssional and manageriai 577 6.9 2.3 2.4 29.9 71.1
I'ot\.cr rnaintcnanct: eiatrician 235 60.8 6.5 9.2 188.2 27 1.2
S tockkccpers I l l 9.6 -.- 3 3 2.5 37.9 37.8
Supcn-isors (tcchnical and vade) 350 14.3 4.9 5.1 6 7 43.9
Truck drivtr 29 18.1 3.4 3 .O 44.1 74.2
Mriintcnance and scxurih 286 14.1 4.8 6.0 40.7 40.7
l'cchnical (othcr) 336 9.9 3.1 3.4 30.8 34.5
I'rotcction and control tc~hnicim 132 13.7 2.3 2.3 67.7 35.2
l'rade (gcncral) 397 8.5 2.9 3.9 30.1 24.4
Chapter 2 Correlations between indices of electric and magnetic fields
Table 2-3: Mean daily exposure (in V/m) for traditional measures of electric fields, by work-site location, Ontario electric utility workers
Job site Daily Arithmctic Gtomctric Median Standard 9 P measuw mcan mean deviation Penwntik
Administration
Arca otficc
Control cmtw
Constniction
I3istnct otficc
1 Iydroclcctric gtntrating station
I ielipad
lnspcction
Nuclcar gencrating station
Regional otlïcc
S c n i x centre
Technical facility
Thermal gcneration station
Transformer station
Warchausc
Al1 poups 3065 10.4 3.3 3.8 38.9 3 1.2
Chapter 2 Correlations between indices of electric and magnetic fields
Table 2-4: Indices of exposure selected to represent each principal component factor axis for 'non-traditional' electric field exposures
Axis Exposure Description met ric
Asis Correlation Variance with axis
1 ET-TH 14 Ptrcentage of time at or abovc 2500 V/m 12.1 0.95
2 EAV-TH 10 Rrithrnstic mean of elcctrïc ticld at or above 1 56 V/m 6.2 0.94
3 ETDS-5 Ptrctntage of h e at or above 40 Vlm for at lest 5 3.7 0.94
minutes
4 LAG-E-5 Autocorrelations at a lag of 5 minutes 3.3 0.92
5 E-AVTRAN Avtrage electric fIrId transition (in bins) for adjaccnt 2.0 0.95
readings
6 EAV-TH 14 Arithmctic mean of elcctric field at or above 2500 V/m 1 .G 0.6 1
Chapter 2 Correlations between indices of electric and magnetic fields
Table 2-5: Correlation coefficients obtained from principal component analyses* using selected electric field metrics and traditional measures o f exposure
Metric Description Asis #1 Axis #2 Axis #3
Non traditional uposure metrics
ETD8-5 Percentage of time at or above 40 V/m 0.54 0.16 0.34
LAG-E-5 Autocomlations at a 5 minute lag 0.09 0.08 0.83
ET-TH 14 Percentage of t h e at or above 2500 Vlm 0.88 0.3 1 -0.04
EAV-TH 14 Arithmetic mean of clectric field esposure at or O. 12 0.86 -0.0 1 above 2500 Vlm
EAV-TH 10 Arithmetic mcan of clectric field esposure at or O- 13 0.8 1 -0.08 above 156 V/m
E-AVTRAN Average electric field transition in nurnber of 0.00 O. 17 -0.73 buis for adjacent readings
Traditional exposure metrics
EAV - TH1 Arithmetic mean 0.88 0.4 1 0.00
EGAV-TH 1 Geomctric mcan 0.93 0.0 1 0.09
E-95 95& perccntilc O. 73 0.45 0.03
E-STD Standard dcviation 0.57 0.77 0.00
Variance explained by each factor mis 4.47 2.67 1.35
* the principal component axes were rotated using the Varimax rnethod; traditional and non- t raditional measures o f exposure were analysed simultaneously
Chapter 2 Correlations beGeen indices of e k n c and magnetic fields
Table 2-6: Correlation nintrix o f selectecl electric field metrics* and traditional numures of elcctric Field exposures smotig Oiitario etectric utility workers
LAO - E - 5 ET - TH14 EAV - TH1 EAV - TH14 EGAV_TH1 EAV-TH10 E-AV'PRAN E - MeDXAN E-STD ETDB-5 E-95
*A description of the variable naines used to represent tliese inetrics can be found in Table 2- 1.
Table 2-7: Summary o f principal components analyses o f electric field metrics, by occupational title
Job title Daüy Factor Metrics bighly corrclated with factor axes * summarks ases
Clerk 23 1 3 E-STD. EGAV-TH 1. LAG-E-5
Controi maintainas 119 3 EGAV-TH 1. EAV-TH 1. E-AVTRIW
Customrr s rn i ce rrpresentativr 13 1 3 EAV-TH 1. ET-TH i 4, E-AVTRAN
Forcstcrs 2 1 O 2 E-STD. E-MEDLW
Inspcctor 6 3 3 EGAV-TH 1. E-STD. LAG-E-5
Mcter readrr 9 3 2 EGAV-TFI 1. E - S m
Opcrator 3 53 4 EAV-TH 1. EGAV-TCIIO, EGAV-TH 1. E-AVTRAN
Po\verlinc maintainers 484 3 E-MEDIAN. EAV-TI-114. EAV-83
Profcssionüi and managtnaI 577 4 EAV-TH14. EGAV-THI. E-95, LAG-E-5
Poivcr maintenance eltxtrician 23 5 3 ET-TH 14. EAV-TH 1 4. E-AVTRAN
Stockecpers 1 1 1 3 EAV-TH 1. E A V T H 14. LAG-E-5
Supenisors (technical and iradc) 3 50 3 E-STD. EGAV-TI-II. LAG-E-5
Truck driver 2 9 4 EGAV-THI. E-95. EGAV-TH10. LAG-E-5
Maintenancc and security 286 3 E-STD. EGAV-TH 1. E-AVTMN
Tcchnicrtl (other) 336 3 E-STD. EGAVTH 1. LAG-E-5
Protcct~on and control tcchnician 132 3 E-STD. EGAV-TH 1. LAG-E-5
l'radc (gcncra1) 497 3 EAV-TH 1. ET-TI-1 1 4. E-AVTRAN
* those metrics that were most highly correlated with each of the factor axes are listed in succession
* a description of the metrics is found in Table 2- 1.
Chapter 2 Correlations between indices of electnc and magnetic fields
Table 2-8: List of magnetic field exposure indices used in the first stage of principal components analyses
Variable name Number of Description metrics
MT-TH{s)
MAV-TH{s)
MGAV-TH {s)
MJAG
M-AVTRAN
MTD{s) - { y )
LAG-M-{s}
NTD{s)-{y)
MAV-TH 1
MGAV-TH 1
M-MEDIAN
M-STD
M-93
Percentage of timc the magnetic field is at or abovc s={3, 12.5, or 50 pT)
Arithmetic mean: of magnetic field at or above s={3. 12.5. or 50 PT)
Geometric mean : of magnetic field at or abovc s={3. 12.5. or 5OpT)
Pcrcentage of magnctic field measures sarnplcs that diffcr by at lcast 2 bins.
Averageï magnctic field transition in nurnber of bins for adjacent samples taken one minute apart
Percentage ortirne at or above s=S pT for at least y={5.15.30 or 60 minutes) and s=12.5pT for at lcast y(5 .15 ) minutes
Autocorrclations of s={ 1-5-15 or 30 minute lags )
Average of values at or âbove s=3 PT for at lcast y(5.15.30 or 60 minutes), and s = 12.5 pT for !={5. 15 minutes).
Arithmctic mean
Standard deviation of magnctic field (in PT).
+ The average was calculated using (i) total time above the threshold and (ii) the total time for the perïod.
Chapter 2 Correlations between indices of electric and magnetic fields
Table 2-9: Average daily exposure (in PT) Cor traditional summaries of magnetic fields, by occupational group, Ontario electric utility workers
Occupational Daïiy Arithmetic Gcometric Median Standard 95" Croup measurcs mean mean deviation Percentik
Clcrk 23 1 0.27 0.19 0.2 l 0.48 0.58
Control maintainers I l9 0.57 0.20 0.20 1.32 1.75
Customer sc'rvicc r~prèsentativc
Forcstcrs
Inspxtor
Mctcr readtr
Opcrator
Powcrline maintainers
Prokssionril and managerial
Potvcr maintenance electncian
Stockkecpcrs
Suptn-isors (technical and vade)
Truck driver
Maintcnancè and secuity
Tcchnicai (other)
Protection and control technician
13 1
2 1
63
9 3
353
484
577
235
I I I
3 50
29
286
336
132
Tradc (gcncral) 497 0.60 O. 19 0.20 1-60 2.16
------------------------------------------------------------------------------------------ Al1 poups 4247 0.55 0.24 0.30 1 . 1 1 1 68
Chapter 2 Correlations between indices of electtic and magnetic fields
Table 2-10: Mean daily exposure (in PT) for traditional measures of magnetic fields, by work site location, Ontario electric utility workers
Job site Daily Arithmetic Gcometric Median Standard 95* mcasure~ mean mean deviation Perccntik
Administration
h a office
Control centrè
Construction
Distnct otXce
1-Iydroclcctric gcnerating station
I-Ielipad
Inspcction
Nuclcar gcnerating station
Regional otlice
Scn-icc ccntre
Tcchnical facilit!
Thcrmal generation station
Transformer station
Warchousc
Chapter 2 Correlations between indices of electric and magnetic fields
Table 2-1 1: Indices of exposure selected to represcnt each principal component axis for 'Non-traditional' magnetic field exposures among a sample OC Ontario electric utility workers
Axis Exposure metric Description Anis Correlation Variance with axis
1 NTDlO-j* Average of values at or above 3pT for at least 12.2 0.97
5 minutes
3 - MTD8-5 Percentage o f t h e at or abovc 0.8 pT for at 5.7 0.95
least 5 minutes
3 MAV-TH 10 Arithmetic mean of magnetic field at or above 5 .O 0.96
3 PT
4 LAG-M-5 Autocorrelations at a lag o f 5 minutes 3.1 0.9 1
5 M-AVTRAN Average magnetic field transition (in bins) for 2.0 0.96
adjacent readings
6 NGAV-THS* Gcometric mean of magnetic field at or above 1.3 0.59
0.8pT
* .4verage calculated with the denominator being the total time penod total time above a t hreshold
Chapter 2 Correlations between indices of electric and magnetic fields
Table 2-12: Correlation coeffkients obtained frorn Principal component analyses* of selected magnetic field metrics and traditional measures of exposure among Ontario electric utility workers
Metric - - --
Description Axis #1 Axis #2 Axis #3
Non traditional exposure met ries
MTD8-5 Percentage of time at or above 0.8 pT 0.54 0.20 0.38
LAG-M-5 Autocorrclations at a 5 minute iag 0.03 O. 10 0.83
NGAV-TH8 G Geometnc mcan of magnetic field at or above 0.9 1 0.2 1 -0.08 0.8 pT
NTD 10-5 a Average of cxposures at or abovc 3pT for ai 0.66 0.6 1 -0.09 least 5 minutes
MAV-TH 10 Anthmetic mean of magnetic field exposures at -0.03 0.67 O. 19 or above 3 pT.
M-AVTRAN Average magnctic field transition in number of -0.05 0.2 1 -0.71 bins for adjacent rcadings
Traditional exposure metrics
MAV-TH 1 Anthmetic mean 0.77 0.59 0.09
MGAVTH 1 Gcometric mean 0.96 O. 13 0.1 1
M-STD Standard dcviation 0.28 0.93 0.0 1
Variance esplained by each factor u i s 4.22 2.81 1.42
* the Principal component axes \vcrc rotated using the Varimau mcthod: traditional and non-traditional mcasurcs of cxposure were analyscd simultaneously
: average calculated with the dcnominator being the total timc period not thc total time above threshold
Chapter 2 Correlations between indices of electric and magnetic fields
Table 2-13: Correlatioii matrix of selected ningnetic field nwtiics* i i~id traditional sunimrries of mrgnetic field exposure, Ontario electric utility workers
LAG - M - 5 NGAV - TH0 MAV - TH1 MAV-TH10 MGAV - TH1 M - AVTRAN M - MEDIAN M - 95 M - STD MTD8 5 NTDlO-5 -
*A description of the variable names used to represent the metrics can be found in Table 2-8.
Table 2-14: Summary o f principal components analyses of magnetic field metrics, by occupational title
- - -ppp --
Job title Dail! Factor Mctrics highly correlated witb factor ases * summaries axes
Clerk 23 1 3 MGAV-TH1. MAV-TH10, NTD10-5, M-AVTRAN
Control rnaintain~xs 1 19 3 NTD 10-5. M-mDiAN, LAG-M-5
Customer sen-ice repr-niatire 13 1 3 MGAV-TH 1. M-95, LAG-M-5
I'orcsters
Inqxxtor
Meter rcadrr
Opcraror
210 4 M-MEDiAN. NTD10-5. MAV-TH10. M-AVTRAN
63 3 M A V T H 1. NGAV-TH8. LAG-M-5
9 3 4 MGAV-TH 1, M-STD, M-AVTRAN, LAG-M-5
353 2 MAV-TH 1 , M-AWWW
Poncrlinc maintainers 484 3 MGAV-TH 1. MAV-TH 1 0. M - A m N
I'rofcssional and managtrial 577 4 NTD 1 0-5, MGAV-TH 1. M - A V T W
1'on.t.r maintenance electrician 235 3 MGAV-TH 1. M-STD, LAG-M-5
StocUc~pt .rs 1 1 1 3 MGAV-TH 1 . MAV-TH 10, M-AVTRAN
Supcn.isors (technical and tradc) 3 50 3 M-STD, M-MEDIAN, M-AVTRAN
Truck drivcr
iMaintcnmcc and securih
Tcchnical (othcr)
29 3 MGAVTfl l , M-STD, LAG-M-5
286 3 M-STD, MGAV-TH 1. M - A V T W
336 3 MGAV-TH 1. M-STD, LAG-M-5
Protection and control tcchnician 142 3 M-STD, MGAV-TH 1
Tradc (general) 497 3 MGAV-TH 1. M-STD, LAG-M-5
* those metrics that were rnost highly corteiated with each of the factor axes are listed in succession
Chapter 2 Correlations between indices of electric and magnetic fields
Table 2-15: Pearson correlation coefficients between traditional electric and magnetic field exposures, Ontario electric utility workers
. -. -. - - - -
Electric field indices
Magnetic field Arithmetic Geometric Median 95th Standard indices Mean Mean Petcentile Deviation
Arithmetic mean 0.34
Geometric mean 0.25
Median O. 14
9 5 ~ percentile 0.38
Standard deviation 0.3 1
Chapter 2 Correlations between indices of electric and magnetic fields
Risk of cancer among Ontario electric utiiity workers: the evaluation of exposure indices of 60 Hz electric and magnetic fields
3.1. Summary
The association between various indices of eIectnc and magnetic field exposure and the
risk of cancer was evaluated using a nested case-control design within a cohort of 3 1,543 Ontario
Hydro workers. Detailed andyses were performed on individual sites of cancer of a priori interest
t hat included: leukemia, brain cancer, Non-Hodgkin's lymphoma and malignant melanoma. Four
controls were individuaiiy matched to each cancer case. There were 35,50,58 and 5 1 rnatched
sets for brain cancer and benign tumours, leukernia, rnalignant melanoma and Non-Hodgkin's
lymp ho ma, respect ively.
Job exposure matrices (JEMs) were constructed for a variety of electric and magnetic field
exposure indices that were chosen to represent the foilowing aspects of field exposure: central
tendency, standard deviation, transitions in field strength and the percentage of time spent above a
t hreshold. The cells of the JEM were defined by occupational title and work-site location.
Employrnent histories were used in conjunction with the E M S to estimate an average workiig
exposure for these different indices. Conditional logistic regression methods were used to
calculate odds ratios and their correspondhg 95% confidence limits.
105
The findings obtained from the analyses confimi the positive association between electric
field exposure and risk of leukemia observed in a previous analysis [l], and fùrther, support the
hypothesis that exposures above an electnc field threshold are important predictors of leukemia
risk. The percentage of time spent above electric field thresholds of 20 and 39 V/m was
associated with an increased leukernia risk after adjusting for duration of employrnent and the
anthmetic mean exposure to both electric and magnetic fields (p<0.05). Further. prolonged
exposure to electric fields was associated with a dramatic increased risk of Ieukemia. In
particular, among employees who had worked in the utility for at least 20 years, those in the
highest tertiles of percentage of time spent above 10 and 20 V/m had odds ratios of 10.2 (95%
CI=1.6-65.3) and 8.2 (95% CF1.2-54.4) respectively, when compared to those in the lowest
tertile. Non-significant elevations in risk were observed between indices of magnetic fields and
leukemia. These results support the hypothesis that electric fields act as a promoting agent in the
etiology of adult leukemia.
The assessment of the relation between brain cancer and benign brain tumours and
exposure to altemate indices of electric and magnetic fields was limited by a small number of
cases (n=35). Nonetheless, electric field exposures, including threshold indices, were not
associated with brain cancer. Non-significant elevations in risk were observed among subjects in
the upper tertile oFrnagnetic field exposure based on the arithmetic and geometric mean.
Duration of employment was not related to brain cancer.
Chapter 3 Risk of cancer among Ontario electric utility workers
Similar to leukemia, exposures above selected electric field threshold cut-points were
associated with an increased risk of Non-Hodgkin's lymphoma (NHL). In contrast, duration of
employment was not related to an increased risk of NHL. After adjusting for the arithmetic mean
electric and magnetic field exposures, the percentage of time spent above I O and 20 V/m
remained significant predictors of risk (p<0.05). Odds ratios in excess of three were observed in
the upper textiles of the percentage of time spent above 4.9. 10. 20 and 39 Vlm. Magnetic field
exposures were not significantly associated with NHL. Although these increased risks are far less
dramatic than observed for leukemia, they are consistent with the hypothesis that electric fields
may act as a promoter in the etiology of NHL.
Neither electnc nor magnetic field exposures were related to malignant melanoma. Non-
significantly elevated risks of malignant melanoma were observed with an increased length of
employmen: in the Ontario utility. Those who were employed for at least thirty years had a
twofold increase (OR=2.01, 95% CI=0.57-7.27) in risk relative to those that worked less than ten
years.
In summary, our analysis supports the hypothesis that electric fields may play a role in the
promotion of leukemia and Non-Hodgkin's lymphoma. Further work is needed to evaluate the
role of applying historical corrections to altemate exposure metrics, and subsequently, the impact
on risk estimates. In addition, exposure assessrnent based on alternate indices of electnc and
magnetic fields should be incorporated into îùture studies of both adult and childhood cancers.
Chapter 3 Risk of cancer among Ontario electric utility worken
3.2. Introduction
The possibility that occupational exposur e to electric and magnetic fields is associated
with an increased risk of leukemia was raised by a series of reports that appeared more than a
decade ago [2-51. A number of epidemiologic studies of workers have since investigated the
association between several cancers, primarily brain cancer and leukemia, and exposure to
magnetic fields [6, 73. More recently, the relationship between electric fields and cancer has corne
under greater scrutiny [l, 8, 91. As outlined in the first chapter (section 1.4), the findings of these
studies have been equivocal, and the hypothesis that magnetic and\or electric field exposures are
reiated to cancer remains unresolved.
Magnetic and electric fields are complex entities that can be characterised by their
frequency, waveform, polarization and amplitude [ 1 O]. As a result, there are potentially several
different parameters that can be used to characterize exposure. Studies of cancer and workplace
exposures to rnagnetic and electric fields have fiequently relied on a cumulative time weighted
average (TWA) exposure derived using the arithmetic or geometric mean exposure to estirnate
risk. Examples of alternate exposure metrics that could be used for risk assessrnent include the
proportion of time spent above a specified threshold, the standard deviation of field strength, and
fluctuations in field strength over successive time intervals.
Chapter 3 Risk of cancer among Ontario electric utility workers
108
As before, it is not known whether etectric and magnetic field exposures are related to the
incidence of cancer. However, as outlined in sections 1.3 and 1.4, it is generdly accepted that if
extremely low fiequency fields (ELFs) cause cancer it is through the promotion stage of the
carcinogenic process [7, 1 1- 131. Although vitro and in vivo studies support a role for several
of these alternate definitions of electric and magnetic fields, they have not established which, if
any, is the most biologically relevant metric. For the above reasons. there is uncertainty about
which aspect of electric and/or magnetic fields should be used to evaluate cancer risk in
epidemiologic studies.
The evaluation of altemate indices of magnetic field, as they relate to cancer, has been
done explicitly in some studies [ 14- 161. Other work has assessed cancer risk by using cumulative
electric and magnetic field exposures based on the geometric and arithmetic mean field strength
[ l , 8. 17- 191. The arithmetic mean is more sensitive to skewed data and is better suited for
modelling threshold effects whereas the geometric mean, which is closely related to the median of
the exposure distribut ion, minimises the influence of outliers.
Previous analyses of this study population examined the association between these
cumulative TWA magnetic and electric field indices and cancer [ 1, 171. However, correlational
anaIysis of exposure data fiom this workforce indicates that the use of the arithmetic and
seometnc means, by themselves, does not adequately characterise the variability of either electric
or magnetic field exposures (Chapter 2). Although a limited number of studies have evaluated the
--
Chapter 3 ~isk of cancer among Ontario electric utility workers
association between altemate indices of magnetic field exposures and risk of cancer in workers
[ 14. 15, 201, to date, analyses using comparable indices of electric fields have not been done.
3.3. Methods
This study is a nested case-control design within a cohort of 3 1,543 Ontario Hydro
workers. The cohort consisted of male employees who were either actively employed at some
point between January 1, 1973 and December 3 1 , 1988 or who were pensioners known to be
alive as of January 1, 1970. The period of observation for the detection of cases extended until
the earliest of diagnosis of cancer, death or the end of follow-up (January 1 , 1989). These data
were originally assembled as a component of the Tri-Utility study that also included workers fiom
Quebec and France 1171. The original investigators have described, in great detail, the design of
the initial study elsewhere 1 1 , 171.
3.3.1. Case ascertainment
Cases were identified using record linkage of the study cohort with records of the Ontario
Cancer Registry. The cases consisted of al1 cancers and benign brain tumours, (ICD(9): 140-
208,225) newly diagnosed in the relevant time penod among male employees with more than one
year of continuous service. The relevant time period for linkage was 1970- 1988 for the retirees
and 1973- 1988 for the active workers. To be eligible for inclusion in the study, the case had to be
found in the sampling fiarne used for the selection of the corresponding control. Only the first
prirnary tumour in an individual was eligible, except for previous basal or squamous cancer of the
Chapter 3 Risk of cancer among Ontario electric utility workers
skin
3.3.2. Control Selection
Controls were randomly selected from the employee files of Ontario Hydro. These
employees were individually matched t o cases using the following criteria: i) sarne years of birth.
ii) alive in the diagnosis year of the case and iii) no evidence of p i o r diagnosis of any cancer
(including ICD (9) 225) except basal or squamous cancer o f the skin. For each of the cancer sites
examined in this thesis, the aim was to select four controls per case. A worker could have been
validly selected as a control for a case, and subsequently, also included in the study as a case. The
number of cases and controls for each cancer site were as follows:
Cases and controls by cancer site
1 Cancer Cases Con trols
Brain cancer 35 Leukemia 50 Malignant melanorna 58 Non Hodgkin's lymphoma 5 1
3.3.3. Electric and magnetic field exposure
For the Tri-Utility Study, direct measurements o f worker exposure under usual working
conditions were obtained using the Positron mode1 378 100 persona1 exposure monitor (Positron
Industries, Montreal, Quebec, Canada) [2 11. The Positron is a portable pocket sized, battery
Chapter 3 Risk of cancer among Ontario elecuic utility workers
operated. electronic instrument that monitors immediate personal environmental exposure to
50/60Hz magnetic and electric fields. The device filters the electrk and magnetic field signal to
limit the measurement to 60 Hz fields and was used to record electric and magnetic field strength
every minute over the course of the work day. Each reading was assigned to one of 16 predefined
exposure intervals or bins depending on the magnitude of the exposure. The exposure intervals
were 0-0.6 I . 0.6 1 - 1 -22, 1 22-2-44. . . . 5,000- 10,000, > 10,000 V/m for electrk fields and 0-0.0 12.
0.0 12-0-024, 0.024-0.048, .. . 100-200, >200yT for magnetic fields. Every reading assigned to a
particular bin was given the value of the midpoint of the bin interval. The monitors were tested
and calibrated before use and at regular intervals during the study. A more detailed description of
this monitor as well as its ability to differentiate exposures by occupational group and to obtain
high compliance in workers has previously been published [2 1, 22 3.
As part of the exposure assessrnent in the Tri-Utility Study [ 173, field measurements were
perfomed on 895 Ontario Hydro employees that were sampled by job title and work location.
These measures were taken over the course of a five-day work week and were defined by person,
occupational group, work site and day. The analyses presented within this thesis use electric and
magnetic field measures obtained from 820 of the original 895 workers. Those workers that were
missing either an electric or magnetic field daily summary measure were not included in these
analyses; these account for most of the 75 workers that were dropped fiom the previous analysis
[ 171. The missing data fiom these workers was due to instrument failure, suggestive that the
monitor was kept close by (and so suitable for magnetic field exposure estimates) but not wom
Chapter 3 Risk of cancer among Ontario electric utility workers
(and therefore not valid electric field data). There were two worker records that could not be
retrieved due to defects that appeared in the discs containing the raw data required to calculate
the new exposure metrics.
Although estimates of home exposures were made for a sample of workers, these
exposures were not included in these analyses as the primary objective was to evaluate the risk of
cancer associated with occupational exposures. Residential measurements taken as part of the
Ontario Hydro study of childhood leukemia indicate that arithmetic mean exposures to rnagnetic
field are on average 5-6 times higher among employees of Ontario Hydro when compared to
residential exposures [23]. Similady, electric field exposures are âpproximately 2.5 times higher in
Ontario Hydro workers than in residences [23]. In general, the lack of association between home
and workplace exposures reduces the likelihood that Our results will be confounded [24].
3.3.4. Construction of the job exposure matrices
For the purposes of the risk assessment presented in this thesis, a job exposure matrix
(JEM) was constructed for each of the exposure indices used in our analyses. This involved
creating daily summaries of exposure based on the samples that were obtained each minute over
the course of the workers that were sampled. Each JEM was defined using the 17 job titles and
15 work site locations that had been used by the original investigators [1, 171. Ce11 entries of the
JEM were calculated using personal monitoring exposure data obtained fiom the sample of 820
workers (see above).
Chapter 3 Risk of cancer among Ontario electric utility workers
For six of the occupational groups (Powerline Maintainers, Meter Readers, Foresters,
Truck Divers and Customer SeMce Representatives) exposure was not stratified by work site
because these employees spend most of their time away from their designated work location. For
the remaining 1 1 groups, job title and work site specific means were calculated fiom the Positron
readings obtained fiom the sampled workers. Although efforts were made at the onset of the
study to obtain sufficiently large numbers to estimate exposure by job-title, in some instances, the
number of subjects within certain job-title and work site strata were small (Le., <5). In these
instances, the corresponding overall job title and work site means were averaged to produce the
estimated exposure for this job-title and work-site specific ceil of the JEM. The mean exposures
for these cells with small samples sites were weighted by the inverse of the variances ofjob title
and work-site exposures. 1 also considered Analysis of Variance (ANOVA) methods to estimate
exposures for these cells with small numbers. In the end, the ANOVA approach was not adopted
because the resulting predicted values that were derived were unstable, and not felt to be an
adequate representation of exposures within the relevant job-title and work-site strata.
Where work-site information could not be coded for a subject's work history record.
exposure was estimated using an overall mean calculated for each occupational group- To
calculate this overall mean, the number of person years spent in each work site, within each
occupational grouping, for al1 cases and controls with work-site information was tabulated. The
overall mean for each occupational group was then calculated as a mean of the work-site means.
within the occupational group. weighted by the number of person years spent in each work-site.
Chapter 3 Risk of cancer arnong Ontario electric utility workers
As before, site data fiom the six occupational groups were not used in these calculations.
The exposure for each subject was calculated by multipiying the duration of employment
in each work history record by the corresponding entry of the JEM and summing these products
up to obtain a cumulative estimate of exposure. Cumulative exposures to electric and rnagnetic
fields were calculated fiom date of hire until the date o f a cancer diagnosis. o r diagnosis date of
the matched case for controls. Average exposures were calculated by dividing the cumulative
exposure estirnate by the total length of employment. Because cumulative indices o f exposure are
dependent on length of employment, for each worker, these indices will be correlated with each
other. Therefore, in order to optimise our ability to discriminate risk patterns between exposure
metrics considered, average exposure indices were modelled while including a term for duration
of employment in the multivariate model. In other words, our modelling incorporates both
components of cumulative exposure, namely intensity and duration. The risk of cancer was
assessed for the following exposure metrics:
Electric fields Magnetic fwlds
3. GcomctTic mcm
3. Standard dcviation in field strcngth
2. Gcomeiric mean
3. Standard deviation in ficld strtngth
4. Autocorrelation at 5 minute lags 4. Autocorrelation at 5 minute lags
5 . Avcragc transition in field strength 5. Average transition in ficld strength
6. Ptrccntagc of tirne spent above: 2.4, 5, 10.20, 6. Pcrcentage of timc qxnt above 0.2,0.8, 3.0.
39.78. 156.625 and 2500 V/m 12.5 and 5OuT -- -- - - .- -.
Chapter 3 Risk of cancer among Ontario electric utility workers
3.3.5. Occupational confounders
Personal identifjmg information, occupational history and data relating to occupational
confounders were also collected by the original investigators of the Tri-Utility study [17].
Classification by socioeconomic status was accomplished by using a scheme thought to reflect
employee status when first hired. The classification scheme that was used follows: . - -- - - . -
SES category Description
Profmionais, esecutives
Managers, supervisors, technicians
Ernployrus in administration and commerce
Sirillcd worktxs and thcu foremen
Partly skillcd workers
There are certain workplace exposures that have been implicated as potential risk factors
in the etiology of those cancers under study. A site specific list of established and possible
occupational risk factors was compiled based on published reviews [25, 261 and is summarised in
the table below:
Cancer
Non f-Iodgkin 's lmphorna
Established risk factors
Ionizing radiation. benzcne. tobacco use
Ionizing radiation
--
Sunlight
Possihk risk factors
Pcsticidcs and herbicides
Tobacco LW, pctroleurn csposurcs I Bcnzcnc, pcsticidcs and herbicides, tobacco use
Ionizing radiation I
Chapter 3 Risk of cancer among Ontario electric util ity workers
Quantitative estimates of ioniting radiation exposure were obtained fiom the Ontario
Hydro Radiation Dose Information System which is maintained in accordance with the Atornic
Energy Control Act. As a result, for each subject an estimated cumulative workplace exposure to
ionizing radiation (in sieverts) was available. Data fiom non occupational exposures (i.e., medical
sources) were not collected. It is doubtful that occupational exposures to electric and magnetic
fields would be correlated with medical radiation exposure to radiation, and therefore,
confounding is unlikely to have occurred.
For the Tri-Utility study, exposure estimates were made for known and suspected
carcinogenic agents as defined by the International Agency for Research of Cancer categories 1,
2A, and 2B [27]. Estimates of exposure for these occupational agents took into account
historical changes over time and were developed by an industrial hygienist at Ontario Hydro and
by consulting longtime employees of the utility. This exposure assessment was made blind to case-
control status. Cumulative time weighted average estimates of exposures to these occupational
agents were derived by surnming up exposures received dunng the subjects employment history.
Due to the previously observed association between leukemia and magnetic fields within
the Tri-Utility study, more detailed efforts were undertaken to assemble exposure data for
leukernogens in a subsequent re-analysis [ I I . A list ofjob-titles and work-sites relevant to
leukemia cases and matched controls was created and subjects were categorized as having
probable or possible exposure to carcinogens. Again, these tasks were performed by an
Chapter 3 Risk of cancer among Ontario electric utility workers
occupational hygienist blinded to case-control statu [28]. A cumulative index of exposure to
benzene was calculated based on the product of confidence of exposure, as well as exposure
intensity, frequency, and duration. These exposure estimates were used in the recent analysis of
t his study population [ 11.
In sum, the occupational confouriders created in these databases that were used in this
present study induded exposure to ionizing radiation, benzene, sunlight, 2,4-diclorophenoxyacetic
acid (2,4-D). and 2,4,5-trichlorophenoxyacetic acid (2,4,5-T).
3.3.6. Statistical analyses
The data were analysed using methods appropriate to the matched design of the study.
Odds ratios and their corresponding 95% confidence intervals were estimated for each of the
indices of exposure. All descriptive statistics were generated using SAS [29]. Correlational
analysis was also performed to examine the relationships between duration of employrnent, and
cumulative exposure and average exposures during the work history of the subject. Specifically,
Pearson correlation coefficients were calculated for these aspects of electric and magnetic fields.
The constructed E M S were used to evaluate the relationship between exposure and the
incidence of cancer. As previously described, this assessment was performed separately for the
four cancer sites and by using several exposure indices that were identified as independent
exposures capable of explaining the variability in field measures (see Chapter 2). Average
Chapter 3 Risk of cancer among Ontario electric utility workers
exposures were categorised into tertiles according to fiequency distribution observed in al1
controls. The categorization of exposure in this manner minimires the impact of outliers and
makes no assumptions on the underlying dose response pattern. Moreover, it ensures a balanced
nurnber of controls across the strata, thereby increasing the efficiency of the regression model.
Potential occupational confounders (POCs) were used to adjust risk estimates in a
muitivariate model, and varied by cancer site based on published information on site-specific risk
factors. The cancer sites and the POCs for which adjustment was made were: -- -- -- - - . - -
A voihble occupational confounder daîa, by cancer site
Cancer Potential occupational confoundetfs)
1,eukcmia: loniung radiation btwene
Non-I-Iod&n's I_vmphorna: 2.4-D. 2.4.5-T and berurne
Brain: Ionking radiation
Although ionizing radiation and benzene are recognised risk factors for leukemia, the
roles of 2.4-D and 3,4,5-T in the etiology of this cancer are less clearly understood. These risk
factors were also examined to determine whether they fùrther confounded Our risk estimates. A
hlIy adjusted model was used to estimate risk for magnetic field exposures that included a term
for average exposure to electric fields. Similarly, risk estimates for electric field exposures were
adjusted for average exposure to magnetic fields.
-- -
Chapter 3 Risk of cancer among Ontario electric utility workers
119
A more in-depth analysis of potential threshold effects was examined by estimating the
percentage of time spent above specified bin values as defined by the Positron. The lower edges
in V l m of the nine bins used were: 2.4, 5, 10, 20, 39, 78, 156,625 and 2,500. Similarly, cut-
points used for magnetic fields (PT) were: 0.2, 0.8, 3.0, 12.5 and 50. As before. the percentage
of time spent above a specified threshold was defined by both job title and work site, and was
obtained From the appropriate JEM. These threshold exposure indices were entered into a mode1
that aiready included the anthmetic mean field strength in order to determine whether these
indices made a substantial improvement in predicting risk above an index conventionally used.
This was forrnally tested using the Wald Chi-Square statistic. Further, the relative roles of
du ration of employment, arithmetic mean electric and magnetic field exposures, and percentage
of time spent above selected thresholds were assessed by calculating standardized coefficients.
As outlined by Selvin [30], this method represents a suitable means to compare factors that have
been measured using different units. Standardized coefficients were calculated by dividing the
regression coefficients by their corresponding standard errors.
Chapter 3 Risk of cancer among Ontario electric utility workers
3.4. Results
Selected characteristics of al1 cases and controls within the nested cohort of Ontario
Hydro workers are presented in Table 3-1. As expected due to the matched design of the study,
the distribution of birth years between cases and control were similar and greater than half the
subjects were bom before 1920. Slight imbalances are due to different matching ratios for
different cancer sites. Less than 10°/o of cases and controls were exposed to each of ionizing
radiation, 3.4-5-T and 2,443. In contrast, 54 % of cases and 49% of controls had estirnated
exposures to benzene. As a whole, subjects were employed by the utility for a long period of
time as approximately 60% of cases and controls worked for at least 20 years.
A correlation matrix that included the variables duration of employrnent, arithmetic mean
electric field exposure, cumulative arithmetic mean electric field exposure and percentage of time
spent above different electric field thresholds is presented in Table 3-2. For the sake of brevity, I
have presented correlations for only 5 of the 9 threshold cut-points. Duration of employment was
poorIy correlated with al1 exposure indices (r<0.45), Correlations between the average electric
field arithmetic mean and the percentage of time spent above various thresholds increased with
higher threshold cut-points. For example, the Pearson correlation between the percentage of
time spent above 39 V/m and the average anthmetic rnean electric field strength was 0.87.
Similar correlation patterns were observed with the corresponding indices of magnetic field
exposure (Table 3-3).
Chapter 3 Risk o f cancer among Ontario electric utility workers
3.4.1. Leu kemia
The relationship between indices of average electric field exposures and leukemia is
presented in Table 3-4. Subjects in the highest tertile of electric field exposure, based on the
anthmetic mean, had an odds ratio of 4.38 (95% CI=1.33-14.43) when compared to those in the
lowest exposure category. Non-statistically significant elevations were observed with increased
electric field exposures based on the other metrics that were modelled. In particular. a threefold
increase (OR=3.08, 95% CI=0.97-9.80) in leukemia risk was observed among those subjects in
the highest tertile of electric field exposures based on daily standard deviations in field strength
relative to those in the lowest tertile.
Unlike electric fields, no statistically significant increases in leukemia risk were noted for
average magnetic field exposures (Table 3-5). Non significantly elevated nsks were observed in
the upper tertiles of the arithmetic (OR=2.3 1, 95% CI=0.66-8.12) and geometric mean magnetic
field exposures (OR=1.8 1, 95% CI=0.66-4.95) compared to the lowest tertile. A non-
statistically significant increased risk was also observed in the upper tertile of autocorrelations at
a five minute lag (0Rz2.46, 95% CI-0.89-6.78).
The results tiom modelling the percentage of time spent above selected electric field
thresholds in models that also contained the arithmetic mean electric and magnetic field strength
and duration of employment are presented in Table 3-6. In al1 five models, duration of
employment was positively associated with an increased risk of leukemia (p r 0.05). M e r
Chapter 3 Risk of cancer among Ontario electric utility workers
controlling for the arithmetic mean electric and magnetic fields, the percentage of time spent
above 20 and 39 V/m were significant predictors of leukemia rkk (p c0.05). Nthough
significant improvements in mode1 fit were not observed with other threshold values, the
magnitude of the standardized coefficients for these indices exceeded that of the average
arithmetic mean electric field strength. This suggests that leukernia nsk is more sensitive to
exposures above a threshold than to the arithmetic mean.
The percentage of time spent above selected magnetic field threshold values was not
associated with leukemia risk above and beyond that accounted for by duration of employrnent
and the arithmetic mean exposure to electric and magnetic fields (Table 3-7). As before, length
of empfoyment was positively associated with leukemia risk ( ~ ~ 0 . 0 5 ) .
Stratified analysis was perfonned to fürther evaluate the separate contributions of
duration of employrnent and exposure above electric field thresholds as they relate to leukemia
risk (Table 3-8). For individuals who were employed for less than 20 years, electric field
exposures were not associated with leukemia, however, the precision of the risk estimates was
quite poor as few leukemia cases worked less than 20 years. In contrast, dramatic increases in
risk were observed among those with higher electric field exposures who had been emptoyed for
at least 20 years. Those in the highest tertile of percentage of time spent above 10 and 20 V/m
had odds ratios of 10.17 (95% CI= 1.58-65.30) and 8.23 (95% CI=1.24-54.43) respectively,
when compared to those in the lowest tertile. Less pronounced risk elevations were observed
Chapter 3 Risk of cancer among Ontario elec& utility worken
123
for the arithmetic electrk field mean, and for thresholds up to 5 Vlm and above 39Vlm.
Analyses by period of exposure revealed more pronounced nsks of leukemia for exposures
received during the first ten years of employment (Table 3-9).
The slopes of exposure estimates with and without histoncal corrections are compared in
Tabte 3-26. As noted earlier, histotical correction factors were available for the arithmetic and
geometric mean electric and magnetic fields. For leukemia, the dope of the average electric field
exposure was steeper when historical corrections were taken into account (0.020 vs 0.0 15). The
Wald Chi Square test statistic indicated that the difference between these slopes was of
borderline significance (p=0.07). The difference between magnetic field exposures with and
without histoncal correction factors was significantly related to the odds of developing leukemia
(p=0.04) (Table 3-26). However, neither estimate of magnetic field exposure was significantly
related to case status.
3.4.2. Brain cancer
The evaluation of the relationship between brain cancer and benign tumours and indices
of magnetic and electric fields is based only on 35 cases, and therefore, the precision of the risk
estimates is somewhat poor. Average exposures to electnc fields were not related to the
incidence of brain cancer (Table 3- 10). Non-significant increases in brain cancer risk were
observed in the highest tertile of magnetic field exposure based on the average geometnc
(0R32.47, 95% Ct=0.76-8.04) and arithmetic mean (OR=2.62, 95% CI=0.70-9.74) relative to
Chapter 3 Risk of cancer among Ontario electric utility workers
those in the lowest grouping (Table 3-1 1).
The percentage of time spent above the electric field threshold values of 5, 10, 20, 3 9 and
156 V/m was not associated with brain cancer after adjusting for duration of employment and the
average arithmetic mean electric and magnetic field exposures (Table 3- 12). Similarly, threshold
indices of magnetic fields were not related to brain cancer (Table 3- 13)- Duration of employment
was not a significant predictor of brain cancer incidence in any of the models exarnined.
The relationship between brain cancer and the arithmetic and geometric mean electric and
magnetic field exposures, by penod of exposure, is presented in Table 3- 14. No significant
results were observed for exposures occumng either during the first ten years of employrnent, or
for exposures received after this period. However the width of the confidence intervals for these
odds ratios were wide. No significant differences in risk were noted when historical correction
factors for either the anthmetic or geometric mean electric and magnetic field exposures were
taken into account (Table 3-26).
Chapter 3 Risk of cancer among Ontario electric utility workers
125
3.4.3. Non-Hodgkin's lymphoma
Subjects in the upper tertile of electric field exposure based on the anthmetic mean had a
twofold increase (OR=1.99, 95% CI=0.71-5.53) in the risk of NHL relative to those in the
iowest tertile, though this risk was not significant at the 5% levei (Table 3-15). A similar risk of
NHL was observed among subjects in the highest tertile of electric field exposures based on the
geometric mean (OR=2.30, 95% CI=0.8 1-6.5 1).
The corresponding risk estimates for indices of average magnetic field exposures can be
found in Table 3-16. Those in the upper tertile of exposure of magnetic field exposure based on
5 minute autocorrelations had an odds ratio of 2.70 (95% CI-1 . O W . 18) relative to those in the
lowest tertile. A threefold increase in risk was observed within the middle tertile of the arithmetic
mean exposure to magnetic fields (OR=3.05, 95% Cï=1.00-2-24), however, this risk decreased
considerably in the highest exposure category (OR= 1 -4 1, 95% CI=O.4 14.8 1).
The risks of NHL obtained fiom modelling the percentage of time spent above selected
thresholds of electric field strength simultaneousIy with duration of employment, and the
arithmetic mean electric and magnetic field indices are presented in Table 3- 17. Du~ation of
employment was not related to the risk of M U in any of the five models. The standardized
coefficients representing the percentage of time spent above selected thresholds were positive
and larger than other terms entered into the model. In particuiar, the percentage of time spent
above 10 and 20 V/m were significant predictors of NHL risk over and above any association
Chapter 3 Risk of cancer among Ontario electric utility workers
126
accounted for by the arithmetic mean electric and magnetic field exposures (p<0.05). The
percentage of time spent above selected magnetic field thresholds was not associated with NHL
(Table 3-1 8).
To fùrther evaluate the risk of NHL according to the percentage of time spent above
selected thresholds of electric field, analyses were repeated using models that did not include
du ration of employrnent and arit hmetic mean exposures to electric and magnetic fields (Table 3-
19). This analysis was done as both duration of employment and the arithmetic mean electric
and magnetic field exposures were not significant predictors of case-control status. Stratified
analysis by duration of employment was not performed as length of employment was not related
to case-control status, nor was its interaction with any field indices significant. An approximate
threefold increased risk NHL was observed among subjects in the upper tertile of percentage of
time spent above 4.9, 10, 20 and 40 Vlm relative to those in the lowest tertiles. The respective
odds ratios were 3 -26 (95% CI=1.17-9-04), 3 .O6 (95% CI=] .06-8-86), 3.12 (95% CI= 1 .O6-
9.20) and 3.52 (95% CI= 1.30-9.58). The magnitude of these risk estimates exceeded those
previously observed for the geometric and arithmetic mean electric field exposures (Tables 3-
15,16).
Further analyses were camed out to examine the risk of NHL for electric field threshold
indices by period of exposure (Table 3-20). As a whole, risk estimates were more pronounced
for exposures received during the first 20 years of exposure compared to those received
Chapter 3 Risk of cancer among Ontario electric utility workers
The risk of MX, as indicated by the dope associated with arithmetic mean electric field
exposure, was more pronounced when historical correction factors (0.005 vs 0.003) were taken
into account, however, this difference was not significant (Table 3-26). Risk estimates did not
change appreciably when historical corrections for magnetic fields were considered.
3.4.4. Malignant Melanoma
Neither indices of average exposure to electric fields (Table 3-2 1 ) nor magnetic fields
(Table 3-22) were associated with malignant melanoma. In contrast, duration of employment was
positively associated with an increased incidence of melanoma (Tables 3-23 & 3-24). A dose
response relationship was observed with increasing length of employrnent, however, significant
differences with the referent category were not observed. Specifically, those who worked for at
least 30 years had an odds ratio of 2.01 (95% CI=0.56-7.25) relative to those working less than
ten years (Table 3-25). Historicai corrections to exposure did not result in significant changes to
slopes of the arithmetic mean field exposures (Table 3-26).
Chapter 3 Risk of cancer among Ontario electric utility workers
3.5. Discussion
A two-step approach was adopted to examine the relationship between alternate indices
of magnetic and electric field exposure and risk of cancer within Ontario electric utility workers.
Initially, we exarnined how different indices were related to each other using correlation and
principal components analysis (Chapter 2). These findings were then used to select a series of
indices that form the basis of the cancer risk assessment presented in this chapter.
Correiational analysis is an important first step to identi@ a senes of exposure indices that
can be used to perform cancer risk assessment. In particular, the use of indices that are highly
correlated with each other will be redundant in an epidemiologic analysis, while low correlations
are usetùl to identiQ independent components of exposure. Uniike Chapter 2, correlations were
performed using the work histories of the cases and controls. This perrnitted the relationship
between duration of emptoyment, cumulative and average estimates of exposure to be compared.
The poor correlations observed between the arithmetic mean electric field and the percentage of
time spent above 10, 20 and 39 V/m indicate that the arithmetic mean is a poor surrogate to
mode1 effects resulting from these threshold exposures. Indeed. the continuous variables
representing the percentage of time spent above 20 and 39 V/m were significant determinants of
leukemia risk over and above that accounted for by the arithmetic mean electric field exposure.
It is a property of lognormal (or positively skewed) data that the correlation between the mean
and the proportions above certain thresholds increases as the threshoid cut-point increases, and
this relationship between exposures in electnc utility workers has been observed elsewhere [3 1.
Chapter 3 Risk of cancer among Ontario electric utility workers
129
321. These points suggest that the arithmetic mean may be suitable to assess health effects that
result fiom exposure above high threshold values, but less proficient for lower field cut-points.
The arithmetic and geometric mean electric field were poorly correlated with metrics
representing time spent above 10 and 39 V/m, however, they were highly correlated with
exposures above 156 V/m. Therefore, within this data set. the use of the arithmetic and
çeornetric mean electric field strengths are unsuitable for assessing the relationship between NHL
and exposures above intermediary threshold values ( e g , 4 5 6 V/m).
The heterogeneity of exposures between different study populations may lead to the
selection of different indices to mode1 risk. For example, Savitz et al [33] observed high
correlations for job title groups with the time spent above 20 V/m (r=0.79), and 0.2 PT (r=0.87)
and the corresponding arithmetic mean electric and magnetic field strengths. In contrast,
Armstrong and colieagues [3 11 found that the percentage of time spent above an electric field
threshold of 20 V/m was poorly correlated with the arithmetic electric field mean (r=0.54), yet
the percentage of time spent above 0.2 pT was correlated with the arithmetic mean magnetic
field (r=0.79). It follows that for some populations, risk assessrnent using the anthmetic mean
exposure may be suitable to capture increased risks that resuit fiom exposures above certain
threshold cut-points.
Chapter 3 Risk of cancer among Ontario electric utility workers
Although misclassification of exposure by using job-exposure to infer individual
exposures matrices is possible, the manner in which exposure to electric and magnetic fields was
assessed is a major strength of this study. Few occupational studies have collected information
on electric field exposures. In addition, readings fiom the Positron monitor were collected fiom
workers each minute over the span of a 5-day work week. This ailowed for an extensive set of
exposure metrics to be evaluated when performing cancer rkk assessment. Further, our job
exposure matrices were defined according to both job title and work-site location. Work-site
location has previously been demonstrated to be an important determinant of leukemia risk
within this study population [28].
By using the same job and work-site categorisations that were employed in previous
analyses of this study population [l], it is assumed that these categorisations are suitable for the
alternate indices of exposure. In other words, these categorisations classi@ exposures into
homogeneous çroupings. Analysis of the tiequency distributions of exposure indices within each
job-title and work-site location revealed no bimodal distributions and provided support for this
assumption.
Few occupational studies have evaluated the relationship between electric fields and
cancer. Assessing exposure to electrk fields is much more difficult because, unlike magnetic
fields, electric fields are perturbed by the body of the worker. Efforts were made to rninimize
this error by requiring workers to carry the monitor on a waist belt, except when wearing a
Chapter 3 Risk of cancer among Ontario electric utility worken
131
harness for work at elevation, in which case the monitor was attached to the tiont strap of the
harness. Workers wore the monitor 7-8 hours each day for 5 days, and therefore, we obtained
an average measurement for the variety of body postures and electric field environrnents
normally encountered during the execution of the majority of routine tasks. Further. to account
for the variation in job tasks that occur seasonally, members of relevant trade groups were
monitored throughout the year. By specifying the wearing position of the monitor and sampling
many workers in each occupational group for entire workdays, it is believed that the electric field
average exposure captures the relative exposures of these groups. The range of electric and
magnetic field mean exposure levels (Table 2-6,2- 10 ) suggests that the ability to separate
occupational groups is at least as great for electric as it is for magnetic fields.
It would have been advantageous to have information on non occupational confounding
variables. However, with the exception of cigarette smoking, there are few non-occupational
risk factors. Permission was not granted to obtain data from subjects or their next of kin,
therefore this study had to rely on Company records. For some subjects, smoking and alcohol
data were available fiom medical records, however, it was incorsistently reported and more
complete for the case population. It is important to note that for lung cancer, which is strongfy
associated with tobacco use, smoking has rarely been found to be an important confounder [34,
3 51.
Chapter 3 Risk of cancer arnong Ontario electric utility workers
As noted earlier, analysis was limited by small numbers, particularly for brain cancer.
Cancer risk assessment should be undertaken in larger populations. This would enable risks to
be estimated for rarer forms of cancers, some of which have been shown to be associated with
ELFs. For example, occupational exposures to magnetic fields were found elsewhere to be
related to testicular cancer, particularly arnong men less than 40 years of age [36]. There has
also been some support for an association between occupational magnetic exposure and male
breast cancer [37, 381. Although these cancers are rare, their study is valuable in that it may
provide ches whether ELFs increase the risk of cancer through a hormonal mechanism.
Unfortunately, risk assessment for these cancers within the Ontario Hydro workforce was not
possible due to small numbers. There were only 26 and 10 cases of testicular and breast cancer,
respective1 y.
Leukemia
For several reasons. our results support an increased risk of leukemia owing to both
increased duration of exposure and exposure above a critical threshold. First. the risk of
leukemia was more pronounced when subjects were categorised according to the arithmetic
mean exposure rather than the geometric mean. The arithmetic mean, is more sensitive to skewed
data, or peak exposures than the geometric mean because the distribution of individual exposures
in the Ontario workers is skewed to the right. When average exposure to electric fields and
duration of employrnent were modelled simultaneously, elevations in risk were associated with
both increased exposure and duration of employment. Compelling evidence for the relevance of
Chapter 3 Risk of cancer among Ontario electric utility workers
133
exposures above thresholds cornes fiom modelling the percentage of time spent above selected
electnc field threshold values while controlling for duration o f employment, and the arithmetic
mean electric and magnetic field strengths. The percentage of time spent above 20 and 39 V/m
were significant predictors of leukemia risk over and above the arithmetic mean electric field
exposure which is conventionally used to model threshold effects. Sirnilar patterns of risk were
observed when threshold measures were modelled with the geometric rather than arithmetic
mean field exposure. Finally, when analyses were restricted to subjects who had worked with the
utility for at least 20 years, dramatic increases in leukemia risk were observed with the
percentage o f time spent above 10 and 20 V/m. Elsewhere, no association between electric fields
and leukemia was found in workers employed at Electricité de France [SI and in Los Angeles [9].
Unlike these other studies, my analyses exarnined exposure metrics other than geometric and
anthmetic mean, and the estimates o f exposure took into account variations in field strength not
only by job title, but also by work site. The contribution of work site exposures on risk
assessment has been shown to be very important in previous analyses of this study population [ 1.
381. In light of our findings for increased risks among those with long term employment, it is
also possible that discrepancies between Our findings and other studies may be due to a more
stable workforce. Historically, there was very little staff turnover at Ontario Hydro and it was
common for people to have stayed there over their entire working career. The cohon of workers
within the French electric utility was much younger, and as a result, few workers in their
cohort would have experience the long term exposures that were cornmonplace in the Ontario
cohort [171. Further, the ascertainment of cancers within the French utility workers did not
134
extend past the end of active employment. These three rasons may account for the discordant
f i nd ings between leukemia and electric field exposures with in the Ontario and French cohorts.
Few studies have investigated the relation between leukemia and magnetic field exposure
indices other than those based on the arithmetic and geometric mean field strength. Floderus and
coworkers 1\61 found that occupations with fiequent or large variations between low and high
magnetic field intensities were more common among subjects with chronic lymphocytic
leukemia. Similarly, an increased risk of leukemia was observed among workers with long term
exposure to intermittent fields and as well as to peak exposures [14]. Exposures above 2.5 pT
were found to be positively related to leukemia risk among Californian electric utility workers
11 51. In contrast, no association was observed between a variety of exposure summaries
(including peak exposures) and death fiom leukemia [20]. A Swedish study of magnetic fields
and adult leukemia found increased risks of cancer among subjects with high occupational
exposures ( > O 2 PT), with more pronounced risks observed among those who also had high
residential exposures [39]. The above studies did not measure electric fields, and were therefore
unable to control for their potential confounding effects. Again, to date there are no other
published studies that have perforrned cancer risk assessrnent using alternate indices of electric
fields.
Chapter 3 Risk of cancer among Ontario electric utility workers
135
This study population was initially assembled as one component of the larger Tri-Utility
Study [17] in order to provide sufficient power to identi@ an odds ratio of two for leukemia rkk
had that been the true value. Theoretically therefore, there is limited power to assess statistically
significant differences in cancer risk across indices of EMF exposures. However, given that
higher risk estimates of leukemia were found in previous analyses [1, 171 than had been predicted
in the planning of the study, the study puwer within the Ontario Hydro data set has proven to be
sufficient on its own. Further, the detail of exposure information captures several aspects of 50th
electric and magnetic fields thereby providing a unique opportunity to develop hypotheses for
fiiture studies of occupational cohons.
Another important strength of this dissertation is the detailed data on occupational
confounders that had been assembled by the original investigators. Although individual exposure
data were collected for workplace exposures to 2,4-D and 2,4,5-T, these variables were not
entered into the final models as they had a negligible impact on the leukemia risk estimates and
their inclusion diminished the power of the models by widening the confidence intervals.
The recent investigation of cancer risk within this cohort used job exposure matrices that
took into account historical changes in exposure [ I l . For those analyses, the development of
these job title and site specific historical adjustment factors was complex involving interviews
with relevant current and retired supervisory personnel, and a combination of data available fiom
Ontario Hydro archives. These data included peak power loads at a number of specific points in
Chapter 3 Risk of cancer among Ontario electric utility workers
136
the Ontario Hydro system and overall system data including total energy generated, operating
voltages and total lengths of transmission and distribution circuits. Because historical adjustment
factors were only available for the arithmetic and geometric mean ELF exposures, our job
exposure matrices were not dependent on calendar period. The inability to take into account
historical changes in exposure may result in increased exposure misclassification thereby limiting
the ability to detect associations with other metncs. Indeed, the siope of the arithmetic mean
electric field exposure and risk of leukemia was more pronounced for exposures with historical
corrections. Although this suggests that these analyses may underestimate the true risk, this does
not imply that the observed exposure index-specific patterns of risk may differ. Specifically,
there is no apparent reason why the convexity associated with risk estimates for leukemia across
increasing threshold values would no longer persist if historical corrections had been included.
Nonetheless, fùrther analyses that examine the role of historical correction factors for exposures
on risk estimates, both within this study population and elsewhere are needed.
For most exposure metrics considered, increased risks of leukemia were also observed
with both increasing cumulative exposure to electric and magnetic fields (results not shown).
When average exposure to magnetic fields was modelled rather than cumulative exposure, risks
were attenuated. This indicates that the increased risk associated with the cumulative measure of
magnetic field strength is in part attributable to its dependence on duration of employrnent.
Chapter 3 Risk of cancer among Ontario electric utility workers
137
In summary, the results fiom Our analyses contirm the positive association between
leukemia and exposure to electric fields previously observed in this study population [l]. This
study uniquely demonstrates that this association is explained by both exposure to electric fields
over a prolonged period and exposure above a threshold. M e r controlling for worker
differences in electnc field exposures, no significant associations between leukemia risk and
magnetic fields were observed. The absence of data relating to historical correction factors, may
have resulted in an attenuation of risk estimates, in particular, for the arithmetic mean electric
field exposures and risk of leukemia. Despite this, our findings support the hypothesis that
electric fields rnay act as promoters in the carcinogenic process. It is recommended that similar
analyses be pursued in other study populations that have data on exposure to electnc as well as
magnetic fields.
Brain cancer
Eiectric fields were not associated with risk of brain cancer. There was some suggestion
of an increased risk ofbrain cancer, as measured by annual average exposure to magnetic fields
based on arithmetic and geometric mean, and standard deviation. However, no significantly
elevated risk estimates were observed when the percentage of time spent above various
thresholds of electric or magnetic fields were considered. tt is possible that risk of brain cancer is
dependent on high exposure to both electnc and magnetic fields, unfmtunately in our study
analyses was limited by the srnall number of cases and the inclusion of benign tumours.
Chapter 3 Risk of cancer among Ontario efectric utility workers
138
Previous analyses revealed a positive relationship between brain cancer and cumulative
exposure to magnetic fields, but not electric fields [Il. Compared to these estimates, Our risks
are attenuated, which may be partIy attributed to the inability to account for historical changes
and our inclusion of benign turnours. In a study of workers fiom Électricité de France-Gaz de
France utility workers [SI, an odds ratio of 3 .O8 (95% CI= 1-08-8.74) was observed for al1 brain
tumours for exposure above the 90th percentile (2387 V/m years). Many of the workers were
exposed to petroleum products which have been implicated in some studies as a potential risk
factor for brain cancer [40-421. It is possible that some of the observed increased risk of brain
cancer in the French workers is due to incomplete control of petroleum exposures. Increased
risks of brain cancer resulting from exposure to magnetic fields have been observed in some
studies [ 1 8, 431, but not others 1441. Further analyses of Ontario workers were performed to
determine if exposure to diesel &mes confounded our results; the risk estimates were essentially
unchanged.
Non-Hodgkin's lymphoma
In Canada, the age adjusted incidence rate of Non-Hodgkin's lymphoma has increased
more dramatically than for any other male cancer site between 1970 and 1988 1451. The
etiology of this disease remains poorly understood, though recent investigations have suggested
that behavioural factors such as smoking, diet, and physical activity are not strongly related to
NHL 1461. For these reasons, increasing NHL rates over time may partly be explained by
increased exposure to environmental or occupational agents. Increases in individual exposure to
Chapter 3 Risk of cancer among Ontario electric utility workers
139
ELFs at the general population level during the last 30 years coincide with increased NHL
incidence rates. The results fiom Our study suggest that exposures to electric fields above
threshold cut points of 10 and 20 V/m increase the risk of NHL.
Recently, the relationship between magnetic field exposures and NHL mortality was
investigated in cohort of electric utility workers; analysis was perfomed separately by histologie
type [19]. It was found that the risk associated with cumulative exposure to magnetic field was
more pronounced for intermediate and high grade NHL relative to low grade. Unfortunately, the
small sample size in this population did not permit separate analyses by tiistologic type.
However, our study offers the advantage of being able to control for electric field exposures and
our findings are based on incident cases.
Our analysis extends the previous NHL cancer risk assessrnent within this cohort [ I l by
rnodelling a more extensive series of exposure metrics and by examinhg risk according to
exposures received at different time periods during employment. The results suggest that
exposure to magnetic fields is not associated with NHL. In contrast, electnc field exposures
above a threshold of 10 and 39 V/m were positively related to the incidence of this cancer.
Most important, after adjusting for the arithmetic mean electric field exposure and duration of
employment. the risk associated with the percentage of time spent above 10 and above 39Vlm
was more pronounced. Unlike Ieukemia, no positive relation was observed with duration of
employment.
Chapter 3 Risk of cancer among Ontario electric utility workers
140
Malignant melanoma
Neither magnetic nor electric fieId exposures were related to malignant melanoma. This
is consistent with the majority of studies that have examined this endpoint.
3.6. Conclusions
In summary, Our analyses suggest that electric fields may act as promoters for both
Ieukemia and Non-Hodgkin's Iymphoma. Neither rnagnetic fields nor electric fields were strong
predictors of brain cancer or malignant melanoma. Analyses of brain cancer were limited by a
small number of cases and the inclusion of benign tumours. The increased risk of leukemia and
Non-Hodgkins' lymphoma for electric fields, which are more easily shielded by the body, and not
more magnetic fields may in part be explained by the differences in induced currents produced by
field exposures at various sites in the human body. Dawson and colleagues observed that under
electric excitation, the maximum current density amplitude occurs in the blood which is a factor
of 46 times greater than the whole-body average value [47]. In contrast, for magnetic fields, the
highest current density amplitude occurs in the cerebrospinal fluid, which is a factor of 28 times
higher than the whole body average value. If increased rates of cancer occur as a result of
electric and magnetic fields influencing induced current within the human body, these results by
Dawson et al. suggest that electric fields may be a more important indicator of Ieukemia risk than
magnetic fields. In contrast, their findings suggest that magnetic fields may be more relevant in
the etiology of brain cancer. The above points reinforce the need for tiirther studies of ELF and
cancer that incorporate alternate indices of exposure to electri~ and magnetic fields.
Chapter 3 Risk of cancer among Ontario electric utility workers
141
Chapter 3: References
Miller. A. B., et al., Letrkernia following ocnrpatio~rai expomre to 60-Hi electric alid
magnetic field arnollg Ontario electric titiiity workers (sec comments/. Am J Epidemiol,
1996. 144(2): p. 150-60.
Mil ham, S ., Jr., Morta lity frorn leukem ia 111 workers exposcd to electrica/ and rnugneiic
fiel& fletters]. N Engl J Med, 1982. 307(4): p. 249.
Wright, W. E., J.M. Peters, and T . M . Mack, Lerrkaemiu in workers exposed to eiec~rkal
and map1eticfields [leterj. Lancet, 1982. 2(8308): p. 1 160- 1 .
Coleman, M., J. Bell, and R. Skeet, Lezrkaemia imidetrce in elecfrical workers fkfter'j.
Lancet, 1983. 1(833 1): p. 982-3.
Mc Do wall, M. E., Leukemia mortality irr elecrrical workers in Enplarrd ami Waies
/letter/. Lancet, 1983. l(83 18): p. 246.
Miller, R.D., J.S. Neuberger, and K.B. Gerald, Brah carlcer and lelrkernia and expomre
to power-freqzrency (50- to 60- Hz) e fecrric ami rnagnefic@elds. E pidemiol Rev. 1 997.
19(2): p. 273-93.
NLEHS, Assessrnent of heaith eflects from exposire to power-he freqverrcycy elecfric ami
magrieticfieids, ed. C.J. Portier and M. Wolfe. 1998, Research Triangle Park, NC: U.S.
National Institute of Health.
Guénel, P., et ai-, Exposure to 50-Hr eiectric field and incideme of ierikemia, brait1
tzrmors, arrd other cancers among Fre~lch electric lrtifity workers [see camments/. Am J
Epidemiol, 1996. IU( l2 ) : p. 1 107-21.
Chapter 3 Risk of cancer among Ontario electric utility workers
142
Kheifets, L. I., S. J . London, and J.M. Peters, Lrzrkemicr ris& at~d occz~patio~~al electric
field exposrrre itl Los Atlgeles Cmmty, Cal#!ornia. Am J Epidemiol, 1997. t46(1): p. 87-
90.
Tenforde, T. S. and W.T. Kaune, Interaction of extremeiy low fregirency elecfric arid
magrreticfielcis with hi~mar~s. Health Phys, 1 987. 53(6): p. 585-606.
Hendee, W. R. and J-C. Boteler, The qzrestiotl of health effecis from exposïrc. to
electromagnetic fields (see commentsj. Health Phys, 1 994. 66(2): p. 127-3 6.
Ade y, W. R., Cell membranes: the electromagrw tic environmetlt and catlcer promoriotl.
Neurochem Res, 1988. 13(7): p. 67 1-7.
Cridland, N., EfSecs of powerfreqtrency lrSMF expusures at the cellrriar level. Radiation
Protection Dosimetry, 1997. i 2 (3 -4): p. 279-290.
Matanoski, G.M., et al., Leukemia in telephorte litremen. Am J Epidemiol, 1993. 137(6):
p. 609-19.
London, S .J., et al., fipo~rre tu magnetic fields among electrical worker-s in relatiot~ to
Ieukemia risk iri Los Ailgeles Cotmty. Am J Ind Med, 1994. 26(1): p. 47-60.
Floderus, B., et al., Ocncpatiortal expusure to electromagt~etic fie lds 111 relatiotr to
leukernia utld braitt tumors: a case-control stlrdy in Sweden. Cancer Causes Control,
1993. 4(5): p. 465-76.
Chapter 3 Risk of cancer among Ontario electric utility workers
143
Thénault, G., et ai-, Catrcer r i sh associated with occtrptiorrai exposrre to magnetic
fields amorrg electric utiIity workers iri Ontario and Quebec. Carlada, and Fmtce:
19 70-1989 fprbIished ewattim appears irr Am J Epidemioi 1994 May 15; 139(10): 1 O53/
fsee commerttsJ Am J Epidemiol, 1994. 139(6): p. 550-72.
Savi tz, D. A. and D. P. Loomis, Magrtetic fie Id exposure in re Iation to Ieirkemio and
brait] cancer rnortality amortg elecrric ritility workers (ptrblished errattcm appears in Am
J Epidemiol1996 Jrcl fS;I 44(2):205j. Am J Epidemiol, 1995. 141(2): p. 123-34.
Schroeder, J.C. and D. A. Savitr, Lymphoma and mrritiple myeioma mortality in relatiotr
to magrtetic field expostrre among electric utiIity workers. Am J Ind Med, 1997. 32(4): p.
392-402.
Sahi, J. D., M. A. Kelsh, and S. Greenland, Cohort and rtested case-corrtrol sttrdies of
hernatopoietic cawers and brairt cancer amortg elecrric trtility workers. Epidemiology,
1993. 4(2): p. 104- 14.
Héroux, P., A dosirneter for assessment of expowes ro ELFflridx Bioelectromagnetics,
199 1 . lt(4): p. 241-57.
Deadman, J.E., et al., Ocatpational and residential60-Hz eiectromagneticfields arrd
high- frequency electric tramiertts: expomre assessment rrsirrg a new dosimeter
fptrblished erratum appears in Am I!rd Hyg Assoc J 1996 Jun;5 7(6):580-3/. Am Ind
Hyg ASSOC J, 1988. 49(8): p. 409- 19.
Ontario Hydro, Sttmmary of EIectric and Magnetic Fie Id Measzrremetrts to Jlme 16th.
1989, Ontario Hydro:Report ID:HSR-IR-89-2. Toronto.
Chapter 3 Risk of cancer among Ontario electric utility workers
1 44
Knave. B.. Electric and magrtetic fields and health otrtcornes-ott overview. Scand J
Work Environ Health, 1994. ZO(Spec No): p. 78-89.
Schottenfeld, D. and J. F. Fraumeni, Cancer Epiderniology arrd Preverttion. second ed.
1996. New York: Oxford University Press.
Amencan Cancer Society, Cancer Resozrrce Ceriter website http:/:3yww- cancer. org. 1 999.
International Agency for Research on Cancer and World Health Organization. Overu//
evahra~iorrs of carcinogenicity: an upItatig qf iA RC mot tographs vo ltmes I to 4 2 , .
1987, IARC: Lyon.
Miller, A.. T. To, and C . Wall, Techical report orr the epidemiologica/ strdy or7 the
hg-term ef/ects of exposicre to 60 Hz electric and magnetic field Orrtario Hydro
Sectord Ar~afysis, . 1996, Department of Preventive Medicine and Biostatistics,
University of Toronto: Toronto. p. 43.
SAS Institute, Statistical A~~aiysis System (version 6-12). 1 998, Raleigh, NC : SAS
Institute.
Selvin, S., Statistical artafysis of epidemiologic data. 1996, New York: Oxford
University Press. 260-263.
Armstrong, B.G., J.E. Deadman, and G. TheriauIt, Comparisorr of irtdices of ambierrt
expositre to 60-hertz electric and magnetic fields. Bioelectroma_enetics. 1990. 1 l(4): p.
3 3 7-47.
Sahl, J.D., et al., fiposure to 60 Hz magnetic fiel& itt the electric irtility work
erivirottment- Bioelectromagnetics, 1 994. 15( 1): p. 2 1-32.
Chapter 3 Risk of cancer among Ontario electric utility workers
145
Savitz, D. A., et ai-, Correiations amo~g itid~ces of rlectric arid magnetic field exp0~7cre
in electric trtility workers. Bioelectromagnetics, 1994. 15(3): p. 193-204.
Blair, A., S. K. Hoar, and J. Walrat h, Cornparison of crude utid smokitig-u~zcsted
statldard~zed rnortaiiiy ratios. J Occu p Med, 1 98 5 . 27( 1 2) : p. 8 8 1 -4.
Siemiatycki, I., et al., Degree of corifoz~rrdir~g bim related to srnoking. ethtiic grotcp, atid
sociot.coriomic stat1cs in estimates of the associatiotrs betweetr ocacpatiorr n ~ d caricer. J
Occup Med, 1988. 30(8): p. 6 17-25.
Stenlund, C. and B. Floderus, Ocapatioriai exposcre to magnetic fields II I relatioti to
male breast cartcer and testictdar cancer: a Swedish cuse-coritroi stzrdy. Cancer Causes
Control, 1997. S(2): p. 184-9 1.
Demers, P.A., et ai., Ocmpational expomre tu e l e c t r o a r e i c e d arid breast catlcet-
in merl [see cornmentsj. Am J Epidemiol, 1 9 9 1 . 134(4): p. 3 40-7.
Loomis, D. P ., Caricer of breast arnorig men in electrical oceupatiotis f /etter]. Lancet,
1 992. 339(8807): p. 1482-3.
Feychting, M., U. Forssen, and B. Flodenis, Ocncpatio~iai aiid residetitiai rnagrieticfield
cxpo'iltre utid ielrkernia arld central rrervoics systern tumors. Epidemiology, 1997. 8(4): p.
3 84-9.
Lagorio, S., et al., Mortality offi/lingstatiotr attetidar~ts. Scand J Work Environ Health,
1994. 20(5): p. 33 1-8.
Delzell, E., et al., Case-series itivestigation of intracra~tial neopiasms at a petrochemical
research facility. Am J Ind Med, 1999. 36(4): p. 450-8.
Chapter 3 Risk of cancer among Ontario electric utility workers
146
42. Divine, B. J . , C .M. Hartman, and J. K. Wendt, I/pdde of the Texaco rnortality str@
19-17-93: Part I. Arialyss of overall patterns of mortalify a m i g reficritrg. research. ar~d
petrochemical workers. Occup Environ Med, 1999. 56(3): p. 167-73.
43. Floderus, B., S. Tornqvist, and C . Stenlund, Incidettce of seiectedcarrcers 111 Swedish
railway workers. 1961 - 79 [see commerrtsj. Cancer Causes Control, 1 994. S(2): p. 1 89-
93.
44. Harrington, J. M., et al., Ocaipatior~al expostrre to magrletic fields N I relariorr to
rnorrality from brain cancer among electricity gerreratjori arld transmissiorr workers.
Occup Environ Med, 1997. 54( 1): p. 7- 13.
35. National Cancer Institute of Canada, Cariadiari Cartcer Statistics 1998, . 1998, National
Cancer Institute of Canada: Toronto.
46. American Cancer Society, ïïte Norr-Hodgkitr's Lymphoma Resoicrce Cerrter, . 1 999.
47. Dawson,T.W.,K.Caputa,andM.A.Stuchly,Acomparisorrof60Hzt~riifonnmagrieîic
mrd ekctric indrrctiori Ni the htlman body. Phys Med Biol, 1997. 42(12): p. 23 19-29.
- - -- - -- p.
Chapter 3 Risk of cancer among Ontario electric utility workers
Table 3-1: Charactenstics of Ontario Hydro workers, by case control status
Variable Percentage Percentage of o t cases ' controls '
Birth year < 1920 1920-< 193 5 193 5-< 1950 1 %O+
Date of hire < 1940 1 940-(1960 1960+
Exposed to potential occupational confounders Ionizing radiation Benzene 2,4,5-T 3,4-D
Socioeconomic status Professionals, executives Managers, supervisors, technicians Employees in administration and commerce Skilled workers and their foreman Partly skilled workers
l u ration of employment < 10 years 10-<20 years 20-<30 years 30 + years
' Data were collected fiom 1,484 cancer cases
' Data were collected from 2,183 controls
Chapter 3 Risk of cancer amo@ Ontario electric utility workers
Table 3-2: Pearson correlation coeflicients for selected electric field exposure iiidices, Oiitnrio Hydro electric utility workers
Duration of Employment (A)
Cumulative electric field arithmetic mean (B)
Average electric field arithmetic mean ( C )
Percentage of time above2.4V/m (D)
Percentage of time above 5 V/m ( E )
Percentage of time above 10 V/m (F)
Percentage of time above 20 V / m (G)
Percentage of time above 39 V / m (Hl
Table 3-4: Risk of ieukemia for selected indices of average electric field exposures
Average of electric field Cases Controls Odds ratio ' and Odds ratio * and exposure (in V/m) t 95% CI 95%
Arithmetic mean O - c 8.58 8.58 - c 13.67 13.67 +
Ceometric mean O-< 1.60 1.60 -c l -96 1.96 +
Standard deviation 0-Q6.22 26.22 -42 .5 1 52.5 1 -t
Autocorrelations at 5 minute lags
0-CO. 14 O. 14-~0.17 0.17+
Average transition in field strength
0-<0.87 0.87-~0.96 0.96 +
f - Categorised into tertiles according to the distribution of esposurc of al1 controls; thc average was calculatcd by dividing the cumulative csposure for the relevant indes by the numbcr of ycars employed
' Odds ratios wcrc adjusted for duration of cmployment. socioeconornic status, ycar of hirc. csposurc to benzene and ionizing radiation.
' Odds ratios wcrc adjustcd for duration of ernplopcnt. sociocconornic status. year of hire. csposurc to benzcnc. ionizing radiation and arithmetic mean exposure to magnetic fields.
Chapter 3 Risk of cancer among ebctnc utiIity workers Leukemia
Table 3-5: Risk of leukemia for sdected indices of average magnetic field exposures
Average magnetic field Cases Controls Odds ratio ' and Odds ratio ' and exposure (in PT) 95% CI 95% CI
Arithmetic mean O - c 0.22 I l 72 1.0 - 1.0 - 0.22 - c 0.47 13 62 1.14 (0.29-4.45) 1-10 (0.28-4.28) 0.47 + 26 65 2.67 (0.79-8.99) 2.3 1 (0.66-8.12)
Geometric mean 0-<0.074 i 4 65 1.0 - 1.0 - 0.074 -4. IO0 16 73 0-92 (0.37-2.28) 1.12 (0.43-2.93) 0.100 + 20 6 1 1.25 (0.52-3.0 1) 1.8 1 (0.66-4.95)
Standard deviation 0-<0.43 12 72 1.0 - 1.0 - 0.43-c 1.12 16 59 1.24 (0.37-4.16) 1.20 (0.36-4.06) 1.12 + 22 68 1.80 (0.57-5.67) 1.45 (0.444.85)
Autocorrelations at 5 minute lags
O-<0.24 15 69 1.0 - 1.0 - 0.24-~0.26 6 72 0.35 (0.1 1- 1.09) 0.35 (O. 1 1 - 1.09) 0.26+ 29 58 2.30 (0.99-5.35) 2.46 (0.89-6.78)
Average transition in field strength
O-< 0.53 16 66 1.0 - 1.0 - 0.53 -~0 .62 20 66 0.67 (0.27- 1.68) 0.53 (0.20- 1.4 1) 0.62+ 14 67 0.38 (O. 13- 1.15) 0.35 (0.1 1-1.1 1)
+ Categorised into tertiles according to the distribution o f exposure o f al1 controls; the average was calculated by dividing the cumulative exposure for the relevant index by the number of years employed 1 Odds ratios were adjusted for duration o f employment. socioeconomic status at time of hire, year of hire, exposure to benzene and ionizing radiation.
Odds ratios were adjusted for duration of employment, socioeconomic status at time of hire. year of hire, exposure to benzene, ionizing radiation, and arithmetic mean exposure to electric fields.
Chapter 3 Risk of cancer among electric utility workers Leu kemia
Table 3-6: Leukeniia strndardized coeificients* obtained from the conditional logistic niodel containiiig terins for durrtion of employment, percentrge of tirne spciit above electric field thresholdand average arithmetie inern field exposure
-- --
Exposure index$
Lower bound of electric field threshold
1 O Vlm 20 Vlm 39 Vlm
S.C. S.C. S.C. S.C. -- --
Average arithmctic mcan (Mabmctic field)
Avcragc arithmctic mcan (Elcctric field)
Duration of cmplqmcnt
Pcrccntagc of timc spcnt abovc clcctric ficld t hrcshold
* S.C.= standardizcd cocfficicnt; thc standardizcd cocfficicnt \vas obtaincd by dividing thc rcgrcssion cocnicicnt by its standard crror and \vas adjusteci by sociocconomic staius and csposurc to bcrizciic and ionizing radiation.
$ Esposurc indiccs wcrc modcllcd siniultaiicously,
KI p-value bascd on thc Wald Chi-sqiiarc tcst statistic
Table 3-8: R i s k o f leukemia for selecteâ indices of electric field exposure, by length of employmtnt, Ontario electric utility workers
- -~ - -
Electric field exposure index Employed s 20 years Employed >20 years Odds ratio and 9SoA CI Odds ratio and 95% CI
Arithmetic mean (V/m) 0 - < 8-58 8.58 - < 13.67 13.67 +
Percentage of aime above 2.4 V/m O - < 33.32 33.32 - < 38.17 38.17 +
Percentage of time above 5 V/m O - < 18.20 18.20 - < 22.29 22.29 +
Percentage of time above 10 V/m O - ~ 1 1 . 1 9 11.19 - (15.27 15.27 +
Percentage of time above 20 V/m O - c 5.88 5.88 - c 8.82 8.82 +
Percentage of time above 39 V/m O - < 2.80 2.80 - c 4.78 4.78 +
Percentage of time above 78 V/m 0 - < 1.26 1.26 - e 2.25 7 3 j + -.-
' Odds ratios w r e adjustcd for socioccanomic statu. duration o f cmploqment and csposurc to bervenc and ionizing radiation. 4- Catcgoriscd into tcrtilcs according to the distribution o f esposurc o f all controls
Chapter 3 Risk of cancer arnong electric utility workers Leukemia
Table 3-9: Risk of leukemia for selected indices of eltctric field exposure by pe"od of exposure among Ontario electric utility workers employed for at least 20 years
Electric field exposun index First 10 years of Ycan 10 - (20 of After 20 yean of employmcnt cmpbymcnt employmtnt
Arithmetic mean (Vlm) O - < 8.58 8.58 - C 13.67 13.67 +
Pcrcentage of time above 2.4 V/m O - < 33.32 1 .O 1 .O 1 .O 33.33 - < 38.17 5.56 0.98-32.53 12.82 2.57-63.8 1 3.1 1 1.14-13.28 38.17 + 5.95 1 .O 1-34.96 5.75 1 -26-26.18 4.28 1.16- 13-68
Perccntage of time above 5 V/m O - < 18.20 1 .O 1 .O 1 .O 18.20 - < 22.29 9.2 1 1 10-76.52 2.04 0.62-6.75 3.47 0.95-9.9 1 22.29 i- 10.94 1.53-78.12 3 2 1 1.07-9.67 3.03 1 .O 1-9.52
Percentage of time above 10 V/m O-eI l .19 1 .O 1 .O I .O 11.19 - <13.27 10.48 1.20-90.98 1.8 1 0.58-5.65 2.27 0.83-7.22 15.27 + 14.79 1.76-124.10 3.66 1.14-11.76 3.05 1.19-10.18
Pcrcentage of time abovc 20 V/m O - c 5.88 1 .O 1 .O 1 .O 5.88 - < 8.82 9.34 1.10-79.43 2.14 0.66-6.95 1.33 0.33-1.33 8.82 + 13.45 1.58- 1 14.20 2.27 0.68-7.58 2.30 0.90-8.1 O
Percentage of time above 39 V/m O - < 2.80 1 .O 1 .O 1 .O 2.80 - < 4.78 4.86 0.90-26.43 26.90 2.68-270.10 4-19 1.16-11.81 3.78 + 6.32 1.15-34.74 17.67 1.76- 177. I G 4.59 1 .07- 14.0 1
Percentage of time above 78 V/m O - < 1.26 1 .O I .O 1 .O 1.36 - < 2.23 4.88 0.94-25.1 9 3.04 0.73-12.71 2.08 0.45-3.96 2.25+ 3.73 0.8 1-1 7.09 4.38 1.18-16.24 4.0 1 0.95-8.74
' Odds ratios wcre adjustcd for sociocconomic status and duration of employmcnt and csposurc to ionizing radiation and bcnzcne.
t Categoriscd into tcrtilcs according to the distribution of cxposure of al1 controis
Chapter 3 Risk ofcancer among electric utility workers Leukemia
Table 3-10: Risk of brain cancer for selected indices of average electric field exposures
Average of electric field Cases Controls Odds ratio ' and Odds ratio ' and exposure (in V/m)t 95% CI 95O/0 CI
Arithmetic mean O - < 8.58 8.58 - < 13.67 13.67 +
Geometric mean O-< 1.60 1.60 -< 1.96 1.96 +
Standard deviation 0-e6.22 26.22 - 4 2 . 5 1 52.5 1 +-
Autocorrelations at 5 minute lags
O-<O. 14 O. 1 +<O. 17 O. l7+
Average transition in field strength 0-43.87 0.87-~0.96 0.96 +
$ Catego~scd into tcrtiles according to the distribution of esposure of al1 controis: the average \vas calculated by dkiding the cumulative esposure for the relevant indcs by the nurnber of ycars çmploycd
' Odds ratios wcre adjusted for duration of emplo~ment. sociocconomic status. year of hirc. csposure to bcnzcnc and ionizing radiation.
Odds ratios kvcrc adjusted for duration of emplo~nçnt, socioeconomic status, ycar of hirc. exposure to bcnzcnc. ionizing radiation and arithmetic mcan cxposurc to magnctic ficlcis.
Chapter 3 Risk of cancer among electric utility workers Brain cancer
Table 3-1 1: Risk of brain cancer for selected indices of average magnetic field exposures
Average magnetic field Cases Controls Odds ratio ' and Odds ratio and exposure (in PT) * 95% CI 95% CI
Arithmetic mean O - -= 0.22 0.22 - c 0.47 0.47 +
Geometric mean O-<0.074 0.074 -<O. 100 0.100 +
Standard deviation O-~0.43 0.43-< 1.12 1.12 +
Autocorrelations at S minute lags
0-~0.24 0.24-<0.26 0.26+
Average transition in field strength
O-< 0.53 0.53 -<0.62 0.62+
-f- Categorized into tertiles according to the distribution of exposure of al1 controls; the average was calculated by dividing the cumulative exposure for the relevant index by the number of years employed 1 Odds ratios were adjusted for duration of employment. socioeconomic status at time of hire, year of hire, and exposure to ionizing radiation.
Odds ratios were adjusted for duration of employrnent, socioeconomic status at time of hire, year of hire, exposure to ionizing radiation, and arithmetic mean exposure to electric fields.
Chapter 3 Risk of cancer among electric utiIity workers Brain cancer
Table 3-12: Brain cancer standardized coefficients* obtaiiied froni the condition81 logistic niodel containing ternis for dura tiun of employmen t, percentage of time spent above electric field threshold and average arithmetic mean field exposure
Lower bound otelectric field threshold
10 Vlm 20 Vlm 39 Vlm 156 Vlm
S.C. S.C. S.C.
Avcragc arithmctic mcuri (Mngnctic field)
Average arithmctic meun (Electric field)
Durntion of cmplqmsnt
Perccnlagc of tinic spcnl nbow clcctric field tiircsliold
* S C = standardized coefficient; the standardized coefficient was obtained by dividing the regression coeffcient by its standard error and was adjusted by socioeconomic status and exposure to ionizing radiation.
$ Exposure indices were modelled simultaneously.
p-value based on the Wald Chi-square test statistic
Table 3-13: Brain cancer standardized coeflicicnts* obtained from the conditional logistic niodel containing terms for duration of employmcnt, percent~ge of tirne spent above magnetic field threshold and arithmetic mean field exposure
Exposure index$
huer hound of magnetic fwld t hreshold
S.C. S.C.
Avcrnge arithrnctic mean (Magnetic field)
Pcrccntngs of tinic spcni abo\rc tnagnetic field ihrcshold
* S C . = standardized coefficient; the standardized coefficient was obtained by dividiny the regression coefficient by its standard error and was adjusted by socioeconomic status and exposure to ionizing radiation
$ Exposure indices were modelled sinwltaneously.
a p-value based on the Wald Chi-square test statistic
Table 3-14: Risk of b ra in cancer for selected indices oîelectric field exposure by period o f exposun among Ontario electnc utility workers employed for at least 20 years
Period of evposure
Exposure index First 10 years o f Af ter f i rst 10 years o f employment employment
Electric field arithmetic mean (V/m) O - < 8.58 8.58 - < 13.67 13.67 +
Electric field geometric mean (V/m) O-<1.60 1.60 -<l.96 1.96 +
Magnetic field arithmetic mean (PT) O - < 0.22 0.22 - < 0.47 0.47 +
Magnetic field geometric mean (PT) O-(0.074 0.074 -<O. 100 0.100 -t
' Odds ratios wcrc adjustcd for sociocconomic status. duration o f employment and csposurc to ionizing radiation.
.f- Catcgonscd into tcrtilcs according to the distribution o f csposure of al1 controls
Chapter 3 Risk of cancer among electnc utility workers Brain cancer
Table 3-15: Risk o f Non-Hodgkin's lymphoma for selected indices o f average electrk field exposures
Average o f electric field Cases Controls Odds ratio ' and Odds ratio ' and exposure (in V/m)f 95% CI 95% CI
Anthmetic mean O - < 8.58 8.58 - < 13.67 13.67 +
Geometric mean O-< 1 -60 1.60 -< 1.96 1.96 +
Standard deviation O-c26.22 26.22 -<52.5 1 52.51 +
Autocorrelations a t 5 minute lags
0-<O. 14 O. 14-<O. 17 0.17+
Average transition in field strength
O-<0.87 0.87-<0.96 0.96 +
t Categorised into tertiles according to the distribution of exposure of al1 controls; the average was caiculated by dividing the cumulative exposure for the relevant index by the number of years employed 1 Odds ratios were adjusted for duration of employment, socioeconomic status, year of hire, exposure to benzene and 2,4,5-T. ' Odds ratios were adjusted for duration of employment, socioeconomic status, year of hire, exposure to benzene, 2,4,5-T and the arithmetic mean exposure to magnetic fields.
- -- - - -
Chapter 3 Risk of cancer arnong electric utility workers Non Hodgkin's lymphoma
Table 3-16: Risk of Non-Hodgkin's lymphoma for selected indices of average magnetic field exposures
- - - - - - - - -
Average magnetie field Cases Controls Odds ratio ' and Odds ratio ' and erposure (in PT)+ 95% CI 95% CI
Arithmetic mean 0 - < 0.22 0.22 - < 0.47 0.47 +
Standard deviation O-cO.43 O.43-<l.l2 1.12 +
Autocorrelations at 5 minute lags
O-cO.24 0.24-<0.26 0.26+
Average transition in field strength
O-< 0.53 0.53 -<0.62 0.62+
f Catcgoriscd into tertilcs according to thc distribution of exposure of al1 controis: the average was calculatcd by dividing the curnulativc csposure for the relevant index by the number of yçars cmployed
' Odds ratios werc adjusted for duration of cmploymcnt, sociocconomic status at timc of hirc. ycar of hirc. csposurc to bcwme and 2,4,5-T.
= Odds ratios were adjusted for duration of employment, sociocconomic status at timc of hire, yçar of hirc. csposure to benzenc, 2.4,s-T and arithmetic mcan exposurc to electnc ficIds.
Chapter 3 Risk of cancer among electric utility workers Non Hodgkin's lymphoma
Table 3-17: NHL standardized coeîiïcients* obtained froni tlie conditional logistic rnodel contrining ternis for duration of employment, percentage of time spent above electric field threshold and average rrithmetic mean field exposure
1 h iwer hound of electric field throshold
Average arithmciic mcan 0.42 (I~lcctric field)
Duration of cmploymçnt 0.12
Pcrcçii(agc of t inic spcnt 1.49 abovc clcctriç ficld ihrcshold
E~posure index% 5 Vlm IO Vlm
S.C. p- i il il^ S.C.
Avcrogc aritlinictiç mcari -0.62 0.53 -0.65 (Magnetic ficltl)
0,m -0.04
A
20 Vlm
S.C.
- S.C.
-0.53
-0.30
0.09
1.57
* S,C.= standardized coefficient; the standardized coefficient was obtained by dividing tlie regression coefficient by its standard error and was adjusted by socioeconomic status. and exposure to benzene and 2,4,5-T.
$ Exposure indices were modelled simultaneously
0 p-value based on the Wald Chi-square test statistic
Table 3-18: NHL standardized coeîricients* ohtnined from the coiiditional logistic niodel coiitaining tcrms for duration of employment, percentage of time spent rbove magiietic field threshold and arithmetic mean field exposure
Lower bound of magnetic field threshold
S.C. S.C.
--
p-value
0.86
0.38
0.75
0.95
Awragc criihmctic mean (Mrignctic ficld)
Average ariihnictic mcnn (Elcctnç ficld)
Pcrceniage of timc spcnt obovc magnctic field ihrcshold
* S.C.= standardizcd cocflicicnt; tlic standardi;ï..cd cocfficicnt was obtoincd by dividing ttic rcgrcssion cocfficicnt by its standard crror md was adjustcd by sociocconomic status and exposure to benzene and 2,4,5-T.
$ Esposurc indices wcrc rnodcllcd sitnultancoiisly.
o p-vûliic bascd on the Wald Chi-square tcst stotistic
Table 3-19: Risk of NHL for selected threshold indices of electric field exposure, by length of employment, Ontario ekctrie utüity workers
Electric field exposure index Odds ratio and 9SoA Standardized coeffîcient C.I. and p-value
Percentage of time above 2.4 V/m O - c 33.32 33.32 - < 38-17 38.17 +
Percentage of time above 4.9 V/m 0 - < 18.20 18.20 - < 22.29 22.29 +
Percentage of time above 10 V/m O - (1 1.19 1 1.19 - (15.27 15.27 +
Percentage of tirne above 20 Vlm O - c 5-88 5.88 - < 8.82 8.82 i
Percentage of time above 39 Vlm O - < 2.80 2.80 - < 4.78 4.78 +
Percentage of time above 78 Vlm O - c 1.26 1.26 - < 2.25 2.25+
f ercentage of time above 156 V/m 0 - < 0.43 0.43 - C 1 .O4 1 .O4 +
' Odds ratios were adjusted for socioeconomic status, exposure to benzene and 2,4,5-T and duration of employment t Categonsed into tertiles according to the distribution of exposure of al1 controls
Chapter 3 Risk of cancer among electric utility workers Non Hodgkin's lymphoma
Table 3-20: Risk * of NHL for selected indices of electric field exposure by period of exposure, Ontario electric utility workers employed for at least 20 years
Electric field exposure index ' First 10 ?cars of Ycan 10 - (20 of After 20 'tan of cmployment cmployment emplo~mcnt
Arithmetic mean (Vlm) O - < 8.58 8.58 - c 13.67 13.67 i
Percentage of time above 2.1 Vlrn 0 - < 33.32 9- - 3 32.2- - c 38.17 38.17 +
Percentagc of timc abovc 5 V/m O - < 18.20 18.20 - < 22.29 22.29 +
Percentage of time above 10 V/m 0 - < 1 1.19 11.19 - (15.27 15.27 +
Percen tage of time above 20 Vlrn O - < 5.88 5.88 - < 8.83 25.82 +
Percentage of time above 39 Vlm O - < 2.80 2.80 - < 4.78 4.78 +
Perccntage of time above 78 Vlrn O - < 1.26 1.26 - < 2.25 2.25-t
- - - - - - -
' Odds ratios were adjusted for socioeconomic status, exposure to benzene, 2,4,5-D and duration of employment f- Categorïsed into tertiles according to the distribution of exposure of al1 controls
Chapter 3 Risk of cancer among electric utility workers Non Hodgkin's lymphoma
Table 3-21: Risk of malignant mclanoma for selected indices of average electric field exposures
Average of electric field Cases Controls Odds ratio ' and Odds ratio ' and exposure (in V/m) 95% CI 95% CI
Arithmetic mean O - < 8.58 8.58 - < 13.67 13.67 +
Geometric mean O-<1.60 1.60 -<l.96 1.96 +
Standard deviation O-<26 -22 26.22 -<52.5 1 52.5 1 +
Autocorrelations at 5 minute lags
O-<o. 14 O. 14-<O. 17 O. l7+
Average transition in field strength
O-<0.87 0.87-<0.96 0.96 +-
-f Catcgorised into tcrtilcs according to the distribution of esposurc of al1 controIs: the average was calculatcd by dividing thc curndativc csposure for the relevant index by the nurnber of 'cars employai
' Odds ratios were adjusted for duration of ernplopent. s o c i a ~ n o m i c status at time of hire. S a r of hirc. and csposurc to sunlight.
' Odds ratios w r e adjusted for duration of cmployment, sociocconomic status ai timc of hirc. -car of hirc. csposurc to sunlight and average esposurc to magnetic ficlds.
Chapter 3 Risk of cancer among electnc utility workers Malignant melanoma
Table 3-22: Risk of malignant meIanoma for selected indic- of average magnetic field exposures
Average magnetic field Cases Controls Odds ratio ' and Odds ratio and exposure (in pT) 95% CI 95% CI
Arithmetic mean O - < 0-22 0.22 - < 0.47 0.47 +
Geometric mean O-<O. 074 0.074 -<O. 1 O0 0.100 +
Standard deviation O-<0.43 0.43-< 1.12 1.13 f
Autocorrelations at 5 minute lags
O-<O. 24 0.24-<0.26 0.26+
Average transition in field strength
O-< 0.53 0-53 -<0.62 O.62+
Catcgonsed into tcrtiles according to the distribution of exposure of al1 controls: the average was calculatcd by dividing the cumulative esposure for the relevant indes by the number of ycars cmployed
' Odds ratios wcre adjusteci for duration of employmcnt. socioeconornic status at timc of hirc. ycar of hirc. and csposurc to sunlight.
Odds ratios wcrc adjustcd for duration of employmen& socioeconomic status at time of hirc. year of hirc. csposurc to sunlight and average csposurc to clectric fields.
Chapter 3 Risk of cancer among electric utility workers Malignant melanoma
Table 3-25: Risk o f malignant melanoma by duration o f employment, Ontario electric utility workers
-
Years employed Cases -- -
Controls Odds ratio + and 95% confidence interval
'Odds ratios were adjusted for socioeconomic status, and exposure to sunlight and ionizing radiation
Chapter 3 Risk of cancer among electric utility workers Malignant melanoma
Table 3-26: Effcct of historical corrections on risk estimates, by cancer site
Exposures with Exposures without historical corrections historical corrections p-value *
Cancer Slope Standard Slope Standard error error
Leukemia Arithmetic mean EF Arithmctic mcan MF
Brain Cancer Arithmctic mean EF Arithmctic mcan MF
Malignant melanoma ' Arithmetic mean EF Anthmetic mean MF
Non-Hodgkin's lymphoma Arithmctic mcan EF Arithmctic rnean MF
EF= elcctric field: MF= magnctic field
* p-value was obtained Gom the Chi-square test statistic for the term that represented the average difTcrencc in field csposures in a conditional logistic mcuiel: the difference was defincd as the arithrnctic rncan csposurc with historical corrections lcss esposure ~vïthout historical corrections.
.' Odds ratios wcrc adjusted for: socioeconomic status. csposure to beucne and ionizïng radiation, and duration of cmployment
Odds ratios were adjusted for: sociocconomic status. exposure to ionizing radiation and duration of crnplo~ment
Odds ratios werc adjustcd for: socioçconomic status. csposure to sunlight. ioniilzng radiation and duration of employmcnt
" Odds ratios werc adjustcd for: socioeconomic status, esposure to bcnzene and ionixing radiation. 2.4.5-T and duration of employment
Chapter 3 Risk of cancer among electric utility workers
Chapter 4.
Variations in exposure between and within workers: implications for Power calculations within a matcheà case-control design
Job exposure matrices (JEMs) are a fiequently used tool in epidemiologic studies designed
to evaluate the relation between workpiace exposures and risk of disease. Investigations of this
genre, require that the working history of each subject be defïneà in a rrianner that aliows the
amount of t h spent within different possible job categories to be determhed. Furthermore,
estimates of exposure must be avaüable within each of these job categories. One approach
involves the calculation of the mean exposure leveI, for each job category, by sampling fkom some
subset of workers.
A simulation study was undertaken to evaluate the effet that the variabiiity of this rnean
exposure, that arises fiom sampüng variations both between and within sampied worken, has on
the power of a case-control study to detect a hypthesized efféct. The profiles of the simulated
populations were based on the rnatched case control data o b t d Fom the cohort of Ontario
electric utility workers which formed the basis of earlier cancer risk assemnent (Chapter 3).
Controls were individuaiiy matched to cases by year of b i h .
Overali, the simulation resuhs suggest that greater improvements in power can be reallled
by increasing the nurnber of cases and controls rather than by conducting additional exposure
assessrnent s on an equivaient nurnber of workers. Nonet Mess, whcn t hc relation between
exposure and disease was much stronger, the discrepancy between gains in power achieved by
increasing the number of sub@ts or sampiing fiom additionai workers was modest. These resuits
were consistently observed for exposures across a range of Intraclass correlation coefficients
Further work is needed to evaluate the extent that the variability of the mean exposure
estimate within a job exposure matrix (JEM) affects power according to the number of possible
job categorizations. Simulation studies provide guidance for developing strategies to improve
power by either expending greater efforts in collecting exposure data or increasing the size of the
study population. Deciding where to allocate such resources is dependent on several factors that
include: the hypothesized relation between exposure and disease and the cost, time and effon
required to collect data on either exposure or fiom cases and controls.
2 Introduction
The design of occupationai and environmental epidemiologic studies has advanced fiom an
era where some exposures that were crudely inferred using broadly based descriptions of job
titles, can now be assigned using sophisticated monitoring devices. For example, within the cohort
of Ontario electric utility workers, the Positronnf monitor measured exposure to both electnc and
magnetic fields each minute over a five-day work week to estimate workplace exposure in a series
of cases and controls [ 11. Elsewhere, monitoring of fine particulate and sulphate matter has been
performed within state and counties to evaluate the risk of mortality resulting fiom exposure to air
pollutants [ 2 ] . These are but two examples of a growing body of work in which exposure is
assessed independently of study participants that have been classified according to disease status.
Nonetheless, for most exposures, the assignment of individual exposures using sophisticated
monitoring devices, while ideal, is not an option due to either prohibitive costs or limitations of
current technology. Further, occupational studies that have relied on persona1 monitoring devices
generalty estimate exposure using a sample of current employees and, therefore, describing
exposures received by workers in the distant past may prove difficult. Unlike most other studies
of electnc utility workers, in the Ontario Hydro case-control study. exposures were assessed for
both electric and magnetic fields, and to some extent, efforts were made to adjust for historical
changes in exposure levels.
Power Calculations
In such nested case control designs, the goal of exposure assessment is to obtain a
representative sample of individuals (or locations) in order to accurately characterize the
exposures of the subjects who fonn the basis of later risk assessment. The two components of
such a study design, narnely, exposure assessment and ascertainment of disease status, cm
influence the power of the study to detect the hypothesized effect.
Occupational studies will fkquently incorporate monitoring data into job-exposure
matrices. One approach to creating such a matrix involves the estimation of a mean level of the
exposure variable for groups of workers in different job categories. Thereafter, these mean
exposures are applied to the working history of the study subjects to derive a summary exposure
variable for those who become diseased and for those that do not. The variance of the mean
exposure estimate within each cell of the job exposure matrix is dependent on three key elements:
1 ) the number ofworkers sampled, 2) the number of repeated measurements taken on each
subject and 3) the underlying exposure distribution. An increase in study power can be achieved
by reducing the variance of the mean exposure estimate.
The variance of a mean exposure within the job exposure matrix can be expressed in terms
of the exposure vanability between and within the workers that are sampled. Estimates of
between and within worker exposure variability are readily calculable when repeated measures
have been collected fiom a group of individuals. It is comrnon to obtain measures within the same
subjects over the course of several days. The variance between measurements on the same
individual will reflect changes in daily work routines and measurement error. The Intraclass
Correlation Coefficient (ICC) is a statistic that is used to describe the relation between the two
components of variance.
Power Calculations
Specifically, the ICC @) is defined as follows:
where ab ' = variance of exposure between workers
o, ' = variance of exposure within workers (e.g., day to day variations)
It is advantageous to constmct job exposure matrices using homogeneous groupings
whereby the variance between subjects is small (ab ' -@ O) [3]. However, the variance of the
mean exposure estimates within the JEM will also be infiuenced by within worker variability.
Within worker variability may be large due to sizeable variations in exposure that a worker will
typically experience over a day, and fùrther, by measurement error. For exarnple, the estimation
of electric field intensities using the Positron device is infiuenced by the wearing position of the
device. The relationship between the mean exposure estimate (Y) within the SEM and the
between and within components of variance can be expressed as follows:
var (Y) = + -
Where w represents the number of workers sampled and d represents the number of repeated
measures (days) that sampling took place. Altematively. using the relation between oc and p in
equation 1, the variance can be expressed in terms of the ICC @) as follows:
Power Calculations
The relationship between the variance of this exposure mean within the job exposure
matrix and study power was evaluated by simulating case control studies with similar
characteristics to the nested case-control study of Ontario Hydro workers. A notable design
feature of the Ontario study is the availability of exposure data that contains repeated daily
measures of electric and magnetic fields taken on a large sample of workers. Specifically, daily
measures were obtained for 820 workers for up to the span of a five-day work week. As a result.
between and within worker variability in exposures could be calculated for a variety of exposure
indices. The simulation study pays particular attention to assessing the improvements in power
that could be achieved by increasing exposure samphg using a greater number of workers to
create the JEM relative to increasing the number of cases and controls. The implications of the
simulated results as they apply to other observational studies are discussed.
4.3. Methods
An overview of the simulation and analysis system is provided in Figure 4- 1. As before,
simulation programs were developed to evaluate strategies to augment study power by either
increasing the number of cases and controls or perfoming exposure assessment by sarnpling fiom
a greater number of representative workers. The profiles of the simulated populations are based
on the demographic and exposure charactenstics of the nested case-control study of the Ontario
electnc utility workers [ I l . It was decided to model exposure to electric fields and the incidence
of leukemia as our exposure and outcome of interest, however, the simulated results are
generalizeable to other exposures and diseases. A description of the generation of the work and
exposure histories for the subjects within the simulated studies follows. A listing of the
assumptions used to create each case control study is found in Table 4- 1 .
Generation o f work histories
In al1 analyses, controls were individually matched to cases by year of birth. It was
assumed that birth year was normally distributed with the mean occumng at 19 19 and a standard
deviation of 14 years. These results are based on the observed distributions of subjects in Ontario
Power CaIculations
Hydro case-control study 111. The initial design of the Ontario utility study had matched cases to
controls by birth year in order to control for possible age, period and cohon effects on the
exposure disease relation [4]. In particular, within the Ontario Hydro workforce, the length of
employrnent was correlated with the period when the worker was first hired. The simulated
populations assurned that length of employrnent was normally distributed with a mean and
standard deviation of 25 and 10 years, respectively. Further, when generating the duration of
employment for each subject, the negative correlation observed with birth year was taken into
account. This Pearson correlation coefficient (r) was based on data collected fiom cases and
controls within the cohort of Ontario Hydro workers. It was found that birth year and duration of
employment were inversely related to each other (F-0.3, p<O.OS).
The simulation analysis assumed that workers could potentially have been employed in one
or two different job categories. The first job type was defined as one with lower background
exposure levels to electric fields and whose duties consisted largely of managerial and
administrative tasks. The other job category was representative of technical duties charactenzed
by higher exposures. The proportion of time spent in each job title was determined by sampling
from a randorn unifonn distribution (T - U(0,l)) and was based on empfoyment histories of
members of the Ontario Hydro workfixce. If T ~ 0 . 2 7 then the worker spent his entire working
career in the administrative job. If O.27< T 50.67 then the worker was employed in both job
positions. The proportion of time in each job categorization was binomidly distributed with
n= 100 and p=0.5. Therefore, for these subjects, on average, time was split between these two
jobs. Lastly, if T >O.67 then the worker spent his entire career in a technical position.
Generation of exposure histories
As previously mentioned, Our exposure of interest was the arithmetic mean electnc field
exposure. This was chosen largely due to the previously observed associations between electric
fields and incidence of Non-Hodgkin's lymphoma and leukemia within the Ontario Hydro
workforce (Chapter 3). Additionally, environmental epidemiologic studies fiequently use this
Power CalcuIations
rnetnc, and therefore, the development of this simulation program may have greater relevance to
power ca~culations for studies than had another metric been used. Less detailed analyses were
also performed for the exposure index "percentage of time above 20 V/m" to assess potential
differences in these findings and their corresponding power implications.
The daily arithmetic mean exposure to electric fields was calculated using measures taken
over the course of the work day. An inspecti~n of the daily arithmetic means revealed that they
were loçnomally distributed. Indeed, the use of a lognormal distribution to derive the arithmetic
mean ensures that each simulated electric field exposure wili be positive. As before, the exposure
summary variable for each study subject was created by assuming that employment could be held
in up to two job titles. Realistic values of distribution parameters were obtained using the
sampled measures obtained from the current Ontario Hydro workforce. The mean exposure for
the iow exposure group was 7.4 V/m and was represented by a lognormal distribution defined by
the parameters ~ ' 1 . 5 , a=1 .O. The second job category was the high exposure group with mean of
16.3 V/m represented by a lognormal distribution having p=2.3 and O= 1.0. For each simulated
worker, the cumulative working exposure to electric fields over the tenure of employment was
calculated by multiplying the time spent in each job by the corresponding job-specific mean
estimate of exposure. Exposure estimates were calculated first by assuming that the true
exposure was known. and then additionally, by assuming that the estimate of the mean exposure
in each of the two job categories was derived by using a sample of individuais followed over a
number of days.
An estimate of the anthmetic mean electric field exposure for each job-title that contained
no measurement error was calculated by converting the job specific lognormal parameters using
the following equation [ 5 ] :
Power Calculations
where CL*,, was the arithmetic mean electric field exposure
pL was the mean of the logged daily mean values
o' was the sarnple variance of the logged daily mean values
The electric field exposure estimates for the group means that incorporated error were
then calculated assuming that sampling was conducted in w workers repeated over d days. It was
assumed that the same number of workers were sampled over a 5-day work week in each of the
two job titles. The overall variability in the mean exposure for each entry in the JEM that
included a component of error, was cdculated across a range of ICC. The variance that was
estimated based on sarnpled measures is denoted by using sL and its relationship to the ICC @) is
described as follows:
Where s ' = overall sample variance, s , ' = between subject variance and s , ' = within subject
variance. The between subject variance was calculated using the sample variance and ICC.
The within subject variance was then calculated as follows:
2 2 s, = s 2 - Sb (&? 7,
This then perrnitted an estimate of the standard error of the mean to be calculated for the mean
exposure within each job category (Y,).
Power Calculations
As Y, represented the logged exposures (i.e., Y, = In (xJ) an estimate of mean exposure was
calculated by transforming back to the proper scale using Equation 4. However, because the
expected value of a lognormally distributed variable is dependent on its variance, a scaling factor
was incorporated in this transformation. This ensured that the expected vdue of the mean
exposure estirnate that was measured with error was equal to the tme exposure. Three values of
the ICC were considered in Our models: 0.2, 0.5, 0.8. It is important to note that exposure
assessrnent was not performed directly on the case and control population, but rather on a
representative sample of current Ontario Hydro workers.
Determining case-control status
Case status was determined by assuming that cumulative electric field exposure (with no
measurement error) was related to the probability of developing leukemia (case status). Subjects
were classified into cases and controls by using a slight variation of fonnulae presented elsewhere
[6. 71. Specifically. the variable L, was simulated in the following way
p, L,, = p E + In -
l -Py
where i denotes the matched set within the case-control study
j represents an individual worker
E represents the cumulative exposure to electric fields (in V/m)
p, - U(0,I) uniform distribution
If L, O then the subject was defined as a control, whereas, L, > O represented cases. Values of
L, were generated until the sufficient number of cases and control had been generated within each
matched set. Cases were matched to controls in a 1 : 1 ratio.
Power Calculations
Based on a previous risk assessrnent [ 11, it was assurned that cumulative exposure to
100 V/m-years increased the risk of leukernia (case status) by 30%. In other words, the value for
p was set at 0.263 (i.e., Q ~ - ' ~ ~ = 1.3) TO evaluate the eaent to which exposure sampling strategies
may have influenced power when the exposure was more strongly related to disease, analyses
were repeated using a P o f 0.526. The validity o f the simulation program, and the detemination
of case-control status therein was assessed by perfonning analyses assuming disease was not
related to exposure @=O). Under such a scenario, the anticipated power would be 5%. assuming
that tests were declared significant when pc0.05.
Analysis and power calculations
The variation in the ICCs across job titles for the anthmetic mean exposure to electric
fields was evaluated using variance o f components analysis. Analyses were repeated for the
logged exposures and for the exposure metric "percentage of time above 20 V/m" to explore how
the ICC's differed between the exposure metrics. These analyses were performed using 4,247
daily measures o f electric fields obtained in a sample of 820 electric utility workers which have
been described in detail elsewhere (Chapter 2).
Conditional logistic regression was performed on the simulated studies to generate nsk
coefficients and estimate power. Exposure to electric fields was the only predictor o f case-control
st atus. The presented results are based on 5,000 simulated studies. Wit hin each study, rnodels
were fit using the "tme" exposure and the estimated exposure based on daily sampling in a group
of workers. Again, the estimated exposure was obtained by sampling Y, fTom a lognormally
distributed population with mean of 1 .O or 1.5 and standard error as defined by Equation 8. The
power was defined by the proportion o f slopes that were statistically significant at ~ 0 . 0 5 .
Regression analyses were performed assuming there was no error in the estimate of
exposure. These analyses were repeated after incorporating error into the model. Power was
calculated assuming cases were matched to controls in a 1 : 1 ratio and for sample sizes ranging
Power Calculations
fiom 100 to 200 subjects. These estimates of power were derived assuming work site monitoring
was pedorrned on 5, 10, 15, 20 and 25 workers within each job category. This process was
repeated assuming that the relationship between exposure and disease was stronger (i-e.,
p=0.526).
As conditional logistic regression models will not converge when the maximum estimate
of exposure among the case population is less than the minimum exposure observed within the
entire control population, the simulated program checked to ensure that no such geeneerated studies
were produced. The inclusion o f even a small number o f such studies could bias the results as the
parameter coefficients fiom these individual studies would be severely inflated. Because our
analyses are based on larger sample sizes, that is at least 50 matched cases and controis, no
convergence difficulties were encountered.
1.1. Results
Estimates o f the ICC for the arithrnetic mean electric field are presented in Table 4-2A.
Overall, the between subject variability was smaller than that accounted for within workers
(ICC=0.28). As expected, the between and within sample variances were larger for those job
categories with high exposures. The corresponding table for the iogged exposure yielded similar
resuIts (Table 4-2B). For example, arnong truck dnvers, the majority o f the overall exposure
variance was accounted for by within subject variability (swL1 .23) while variability between
sampled workers was small (%'=O. 16). For both the arithmetic mean and logged exposures, the
values of the ICC varied across job categorizations. A similar table for the metric "percentage of
time above 20 V/m" can be found in the Appendix (Table A-1).
Power Calculations
Estimates of the parameter (P) that describe the exposure disease relation where "true"
exposure was assumed to be known found in Table 4-3. As expected, the mean value of the
fitted estimates that were obtained by perfomiing regression analyses on the simulated studies
were equal to the hypothesized values. When exposure was assumed to be u~elated to disease
@=O), the power was estimated to be 4.5 %, which is near the expected value of five percent
based on an alpha of 0.05. The results sumrnarized in this table support the fact that the
simulated program, including the methods to define cases and controls [6, 71, Iùnctions as
i ntended.
lncreases in power that could be achieved by increasing the number of subjects versus
increasing the number of workers on which exposure assessment will be pefiormed were
estimated and the results are contained in Table 4-4. With a P of 0.263 and a p 0 . 5 , a study of
100 cases individually matched to 100 controls could achieve a power of 79.6 % assuming "true"
esposure was known. As expected, the power of the study to detect the stated effect increased
with a greater number of cases and controls and dso when exposure monitoring was performed
using a larger number of subjects. In the situation where a total of 30 workers, split evenly
between the two job categories, were sampled to estimate the job specific mean exposure, there
was little to be gained in terms of increasing power by sarnpling additional workers.
These analyses were repeated using ICC values of 0.2 and 0.8 (Table 4-5,4-6). This
permitted the evaluation of whether sampling strategies to improve power might differ if between
subject variability accounted for a greater (or smaller) portion of exposure variability in the
exposures of those sampled. Modest improvements in power were observed with an ICC=0.2
cornpared to an ICC of 0.8 in this model where the total exposure variance remained fixed (Tables
4-5,4-6). Again, for both these values of the [CC, little improvement in power could be achieved
by sampling additional workers above and beyond 30.
Power Calculations
Improvements in power that could be gained by sampling additional workers or increasing
cases and controls when the exposure disease relation was more pronounced are presented in
Table 4-7. Again, the increase in power was negligible (< 0.3%) when 10 extra workers were
sampled given that assessments had already been performed in 30 workers. However, increasing
exposure assessments fiom 10 to 20 total workers resulted in approximately a 2% gain in power
which exceeds the corresponding increase in power when the effect size was smaller (P=0.263)
(Table 4-4).
The relationship between sample size and power assuming a exposure assessment was
based on 50 measurements is iliustrated in Figure 4-2. When the total number of measures was
heid constant, sampling fiom a greater number of workers rather than increasing the number of
repeated measures taken for the same worker, led to a greater gain in power. A similar pattern
was observed when exposure assessment was based on 100 measures, though the differences
were much smaller (Figure 4-3).
4.5 Discussion
This chapter has presented methods that can be adapted to perform power calculations for
other case control studies in which exposure assessment is performed within a sample of
individuals that do not form part of the case or control series. The power calculations that were
performed using the profiles of the Ontario Hydro workers indicate that for the most part,
improved power can be realized by increasing the number of cases and controls rather than by
sampling using a similar increased number of exposure measures in workers. Nonetheless, when
the relation between exposure and disease is more pronounced, it is possible that similar gains in
power can be attained within a range of case/control study sizes by sampling exposures from a
greater number of workers. The results also suggest that for the situation where the total possible
number of exposure measures is fixed power wiil be optimized by reducing the number of
repeated measures while maximizing the number of workers to be monitored.
Power Calculations
The simulation program was complex and to vetify the validity of the model several
checks were perfonned. This included repeating the simulations using different seeds to ensure
that the random generating routines within SAS were not producing anomalous results.
Additionally, the power was calculated for the situation where there was no relation between
exposure and disease. Simulations results revealed that 4.5% of slopes were significant, which as
expected. coincides with the alpha of 0.05. Repeating this analysis using a different seed for the
random number generator that determined used to produce case or control status resulted in 5.1
% of slopes being statistically significant at the same alpha level. Perhaps most importantly, the
point estimates of the simulated parameters were consistent with the hypothesized relation
between exposure and disease when tme exposure and case-control status were modeled using
conditional logistic regression. The possibility that the conditional logistic regression model
produced non-convergent results for the individual studies was also examined - no such studies
were generated within the simulation routines.
It would be desirable to have direct measures of electric field exposure on al1 cases and
controls within the study population. Clearly, this is not feasible as within many occupational
cohorts, employment spans several decades and disease fiequently occurs before study onset
where exposure can be measured. Nonetheless, in some studies, particularly those of a cross-
sectional nature, the use of individual exposure data rather than grouped data may be available to
evaluate the exposure disease relationship. Recently, Seixas and Sheppard investigated the effect
of random error in exposure data on the exposure-disease relation using exposure estimates such
as the individual mean, grouped mean and a hybnd estimator [8]. They found that the disease
exposure relation was attenuated for the individual mean exposure estimates particularly, when
the within individual variance was large and the between variance was small. In contrast. the use
of a group mean substantially eliminated bias. Seixas and Sheppard also found that increasing the
between individual vanability resulted in a substantially increased uncertainty while the standard
error was not affected by an increase in within subject variability. Similarly, Our simulations
revealed a somewhat lower study power when a higher value of the ICC was modeled while
Power Calculations
assuming the total variance of the mean exposure was fixed. A higher value of the ICC implies
that the between subject variance accounts for a larger proportion of the overall variance than the
variability observed within subjects. In sum, when the total number of possible measures is fixed,
a greater increase in power can be achieved by increasing the number of workers sampled rather
than increasing the number of days of measurements. As evidenced by equation #6, increasing the
number of workers will decrease both the wittiin and between subject components of variance.
As it was recognized that Ontario Hydro workers exhibit considerable variations in
exposure across job-tities, exposures were evaluated within work site. This was done t o
subdivide the workers into groups with more homogeneous exposure. The success of doing this
can be directly measured by comparing differences in the estimates o f mean exposures within
work-site. For example, within powerline maintenance electricians the arithmetic mean exposure
to electric fields was 60.8 V/m while the standard error of this mean was approximately 14.6 V/m.
The mean exposure for powerline maintenance electricians at hydro generating stations was 10.8
V/m while the accompanying standard error of this mean 2.1 V/m. In contrast, the variance o f the
mean exposure among powerline maintenance electricians who worked at transformer stations
was quite high with a mean of 92.1 V/m and a standard error of 25.2 V/m.
Analyses were repeated using the exposure metric "percentage of time spent above 20
V/m". Again, analysis of this metric was undenaken t o examine the extent to which our power
calculations would be influenced by the choice of metnc and additionally and fùrther, because the
relevance of using threshold related exposure metrics to assess cancer risk bas been established
(Chapter 3). In the simulations that used "the percentage o f time spent above 20 V/m". the
cumulative exposure index was scaled in order to make a meaningfùl cornparison to the power
estimated based on an increase of 100 V/m-years in the arithmetic mean. Specifically, scaling was
done so that the continuous measure o f the cumulative time spent above 20 V/m corresponded to
the same percentile increase for the anthmetic mean. Similar results were found for the
cumulative exposure based on the percentage of time above 20 V/m. Namely, that sampling more
Power Calculations
than a total of 30 workers in the two job categorizations resulted in negligible gains in power.
The estimate of power for the threshold metnc was on average t 5% lower than the arithmetic
mean which can be attributed to an increased standard error of the mean exposure for the
percentage of time above 20 V/m within the job exposure matrix. The job with the lower
exposures was defined by the lognormal distribution having parameters p= 1.9 and O= 1 .O whereas
for the lower exposure category had p= 1 .O and a= 1.3. Therefore, the coefficient of variation,
defined by the ratio o fu / p was larger in both job titles for the threshold metric than the
arithmetic mean and, as a result, the overall estimate of power was lower.
The figures contained within the chapter illustrate that when the total number of exposure
samplings are fixed, power can be optimized by sampling Fiom a greater number of workers rather
than increasing the number of repeated samples taken fiom a srnalier subset of workers. This
result had been based on a ICC of 0.5, in other words, that the total sample variance of the
exposures was evenly accounted for by between and within worker variability. It was also found
that for other values of the ICC ranging between 0.1 and 0.9, that power was maximized when
efforts were made to sample a greater number of workers rather than increasing the number of
repeated samples. However, the magnitude of the improvements in power was influenced by the
value of the ICC. For example, with an ICC=O. 1, increasing the number of sampled workers did
not increase power to the same extent as increasing the number of sarnpled workers when the ICC
was 0.9. In sum, when within worker exposure variability is proportionally larger than between
subject variability, sampling a greater number of workers will not increase power to the same
extent as when between worker variability is large.
The presented simulation analyses are based on the scenario where there are only two
possible job categorizations - those either exposed at high or low levels. The results indicate that
little gain in power can be achieved by sampling greater than 1 5 workers within each cell of this
job exposure matnx. In practice, it is likely that the job exposure matrix may be comprised of a
greater number of cells. For example, within the Ontario Hydro case control study the job
Power Calculations
exposure matrix was defined by 17 job categorizations and 15 different work-sites. Although
t hese resuits suggest that it is preferred to have at least 1 5 workers within each ce11 of the job
exposure matrix, the simulations do not address the question of how power may be afTected with
a much greater number ofjob categorizations. ültimately, this effect is determined by the
transitional probabilities associated with the workers moving between jobs. The use of a greater
number of exposure categories will on one hand better discriminate the mean exposure profile
between the workers, while on the other. the summing up of these exposures across employee's
work histories may result a s u m m q variable of exposure that is less precise. Assessing the net
effect on power under such a scenano is complicated. In particular, elaborate multi-state or
Markov based rnodels would be needed to develop job transitional matrices that define the
probability of moving from one ce11 of the job exposure matrix to another.
In a nested case-control study, the size of the case and control population is determined by
the size of the initial population, the tength ofetiologic period and the ability of record linkage
procedures to identifjr al1 diagnosed cancer cases. The cancer cases used in this study were
identified between January 1. 1973 through December 3 1, 1988 while the initial size of the cohort
was 3 1.543 male workers. In some studies, where the endpoint may be common, the collection of
data for al1 diseased individuals may not be necessary in order to attain sufficient power.
However, within the Ontario Hydro study, efforts were made to construct a cornpfete
occupational history of al1 identified cases and controls. This first required assembling a cohort of
workers by merging year end employment files and checking for overlap. Subsequently, records
were linked to files at the Ontario Cancer Registry to identie cancer cases and manual resolution
of questionable links was performed. Matched controls were then selected fiom the assembled
cohort and work histories were copied to determine changes (and associated dates) in either job
title and or work site location. Data entry was then the performed. The related cost of assembling
these data was approximately $200 per subject.
Power Calculations
The process of perfoming exposure assessrnent of electric and magnetic fields within the
Ontario Hydro workers consisted of several steps. First, a list of eligible subjects to be solicited as
volunteer subjects was assembled using the case and control job history file and the 199 1 Hydro
emptoyee database. Subjects were selected based on their job title and work location, and taken
in proportion to their occurrence in the job history file. However, additional measurements were
made for occupational groups that were known or suspected of having a large number of variable
tasks or working conditions, for example, jobs with seasonal variability such as powerline
maintainers. Approval was obtained fiom senior management and unions, and managers of al1
sites to be visited. Site contacts were identified and provided with a list of eligible subjects.
Thereafter, instructions were provided to groups of eligible subjects, and monitors and
instmctions to those who volunteer. Technicians retumed 5-7 days later to retrieve monitors.
readout and check data, and answer questions. The log sheet and monitoring information was
then entered into a database and programs were then mn to create job exposure databases and
checked for invalid records. As there were approximately 1000 volunteers (and coliected about
900 valid records) it is estimated that the average cost to identi9, monitored and assign exposure
to each monitored worker was estimated to be $300.
In conclusion, this chapter has exarnined several key factors that influence power in
studies that infer exposures using persona1 monitoring data. These factors include the number of
workers and repeated measures that are sarnpted within job categoties, the variability of exposures
between and within workers, and the size of the hypothesized effect. As the values of these
variables are uniquely determined by each study, simulation analyses represents a suitable means
to take these factors into account in order to both estimate and maximize study power.
Power Calculations
Chapter 4: References
1. Miller, AB., et ai., Lerrkemia foliowing occupational expomre to 60-fi elecfric and
magnetic fields among Orttario electric utiiiw workers [see c o m m e n ~ . Am J E pidemio 1,
1996. 144(2): p. 150-60.
7 - . Dockery, D. W., et ai., An association betweeri air po/lrrtion und mortality iti six U.S.
ci fias (see commentsj. N Eng! J Med, 1 993. 329(24): p. 1 7 5 3 -9.
3 . Rap paport, S. M., H. Kromhout, and E. S yrnans Ici, Variation of exposure between workers
hi homogetieozrs expomre grozrps. Am Ind Hyg Assoc J, 1 993. 54( 1 1 ): p. 654-62.
4. Thériault, G., et al., Cancer risùs associated with oca~patiotral ex-omre to magnetic
fields amolrg electric utiiity workers in O~ttario and Quebec. Cmradu, arld France: 1970-
1989 (plrbiished erratirm appears in Am J Epidemiol1994 May 15; 1390 0): 1053 J [see
commerrtsj. Am J Epidemiol, 1994. 139(6): p. 550-72.
5. Rap paport, S. M. and S. Selvin, A method for evalzratirtg the meair exposrrre fi-orn a
fopiormal distribution. Am Ind Hyg Assoc J, 1987. 48(4): p. 374-9.
6. Brown, C . C ., et al., Energy adjrrstmerir methods for rnr fritiotral epidemiology: the egect
of categorization. Am J Epidemiol, 1994. 139(3): p. 323 -3 8.
7. Bellach, B. and L. Kohimeier, Energy a@stmer~t does riof coritrol for differerltial recul/
hias irr )izrfritionai epidemioiogy. J Clin Epidemiol, 1998. 5 L(5): p. 3 93-8.
8. Seixas, N. S. and L. S hep pard, Mmrimi:irig acctrracy und precision zrsitig i~rdividtrd uttd
groirped exposrwe assessmants. Scand J Work Environ Health, 1996. 22(2): p. 94- 10 1 .
Power Calculations
Figure 4-1: A flowchart of the simulation prograrn used to estimate study power
Define "truen exposure to electric fields in 2 job categmzaions
h
. Simulation parameters of cohort are &fined (e-g., length of employment, birth year, proportion of tirne spent in each job category)
)
I Simulate case-control status using cumulative working exposure based on 'Me" exposure 1
Incorporate exposure measurernent erra (B ased on ICC, number of workers sampled, total niunber of days of sarnpling)
) and po&lahd exponire-disease relaGon 1
v
Perforrn ri& assessrnent using conditional logistic regression for exposures measured =%th and without errar
1 Power calculations (based on regression anaiyses) I I
Repeat loop for different values of the ICC and different ratios of cases: controls, workers sampled and days sampled
Power Calculations
Figure 4-2: Saniple size as a fmction of required study power and standard deviation* of the mean exposure to electric fields within each of the two cells of the job exposure matrix (P=0.263, ICC=0.5 AND 50 EXPOSURE SAMPLES TAKEN)
ML
100 2 0 3 0 400 500
Sample Size Exposure assessrnent sampling strategy
( + True Exposure -t 25 wakers per job o v a I day 1 -+- I worker over 25 days - *- 5 workers over 5 days
Figure 4-3: Sample size as u function of required study power and standard deviation* of the mean exposure to electric fields withiii each of the two cells of the job expasure mntrix (D=0.263, ICC=0.5 AND 100 EXPOSURE SAMPLES TAKEN)
300
Sample Sizc
Exposure assessrnent sampling st rategy
-+- True Exposure
--t- 1 worker per job over 50 days
--c- 10 workers per job over 5 days - - - =5 workers per job over 10 days
Table 4-1: Mode1 assumptions for simulateâ populations of workers
Factor Variabk Assumptioa name
1 . Birth ?car BY BY - N (p.a2) wherr p=1919. a=13
2. Duration of rmplojmtnt DUR DUR - N (p.a2) w h a e p=Z. a= 1 O and
r(E3Y.D)=- 0.3
3 . Esposure to elrrctnc fields in job # I EI log (El) - N (p. a ) w h a c p=2.3 and a=i .O
4. Esposurc to clecuic field in job #2 & Iog (L, - N (p. O) whcre p = 1 3 md a= 1-0
5. Proportion of tirnr sptrit in job # 1 PI Pl = for 27% of workers Pl=] .O
for 32% of worktrs P,=O
for 3 1 % worktxs PI--BiN ( 1.0.5)
6. Proportion of time sptnt in job categow #2 PZ PZ= 1-Pr
7. Cumulative esposure to e l t ~ t r i c fields E E = (Pl El + P,EJ x DUR / 100
8. Number of cases Variable
9. Nurnber of individually matched controis Variable
10. Number of workers sampled in each job W, Variai (5 or 10)
I 1. Number of days of sampling 4 Varicd (5 or 1 0)
1 2. A~.eragc= esposurc to rlcctric fields t' s = P I E l + PZ&
13 Probabilit~ of c m status ~ v e n eqosure (E) Pm) P(D) = P x E + logit(8)
w h a c 9 4 . 2 6 3 and 6 - U(0,l)
Power Calculations
Table 4-3: The mean dope and standard error of the dope obtained h m simuhting 10,000 matched case contml studies with fl ={O, 0.263,0.526) '
' ,8 describes the hypotheskd relation b e e n cumulative exposure to e l d c fields (per 100 V/m-years) and
disease occurrence; =0,263 corresponds to a 3û% inaeased risk of disease per 100 V/m-years h i l e f l =O
corresponds to no association benkreen exposure to elec fields and disease and #l =OS26 conesponds to a 70% increase per 1 00 V/m -years.
Po wer Calculations
Table 4-4: Power* of detecting an e f k t with an Intraclass correlation coefficient of p 0 . 5 and for sampling workers daily over a 5 day work week.
1 : 1 case-control ratio Beta=O.263
Number of Numbcr of workers used to estimate exposure (2w) subjccts (ZN)
* rc~ul ts arc bucd on 5,000 sïmulated studies
Power Calculations
Table 4-5: Power* of detecting an effect with an Intraclass correlation coefficient o f p 4 . 2 and by sampling workers daily over a 5 day work week.
1: 1 case-control ratio Beta=0.263
Number of subjects (ZN)
-
Number of workers used to estimate exposure (2w)
* I i c h rcsult is bascd on 5,000 sirnulated studies
- Cascs nvre individually matchsd to controls in a I : 1 ratio.
Power Calculations
Table 4-6: Power* of detecting an effect with am Intraclass correlation coemcient of p 0 . 8 and for sampling workers daily over a 5 day work week
1: 1 case-con trol ratio Beta=0.263
Numbrr of Number of workers used to utimate exposurc (2w) subjccts (2N)
* rcsults arc bascd on 5.000 sirnulated studies
Power Calculations
Table 4-7: Powerf of detecting an e f k t with an Intraclass correlation coeiricient of p 4 . 5 and Tor sampling workcrs daily over a 5 day work week
1: 1 case-control ratio Beta4.526
- - --
Number cf Numbcr ofworkcrs uscd to utimatc exposurc (2w) subjecl~ (2A9
1 O 20 3 0 40 50 No measurmmt trror
* rcsultx arc bascd on 5.000 sirnulated studies
Power calculations
Table A-2: SAS program used to generate simulated data
DATA SIMUL; E - S E E D ~ = ~ ~ ~ ~ ; E - S E E D ~ = ~ ~ S ~ ; B Y ~ S E E D = ~ ~ ~ ~ ~ ; J - S E E D = ~ ~ ~ ~ ; J ~ - ~ E E D = ~ ~ ~ ~ ; D_SEED=1548; CC_SEED=4196; '** MEAN AND SIGMA OF LOGNORMAL DISTN; M-1=2.3;~-l=1.0; M-2=1.5;~-2=1.0; **** CALCULATION OF MEAN VARIANCE FOR EACH JOB CATEGORY ; **** THE LOG OF EACH EXPOSURE WAS CAtCULATED Yi=LOG(Ei) ;
**** THEN PROC VARCOMP Y=ID W A S RUN t
1CC1=0.50;ICC2=0.50; VAR-Bl=ICCl* (S-ltS-1) ; VAR-W-l= (S-i*S-1) -VARARB1; VM-B2=ICC2* (S-2*S-2) ; VAR-W-2=(S-2*S-2)-VAR_B2; N_DAYSl=S;N-DAYS2=5; *** NUMBER OF DAYS SAMPLED -USER DEFINED; N-WORK1=5;N_WORK2=5; *** NUMBER OF WORKERS SAMPLED - USER DEFINED; ""' CONVERT LOG EXPOSURE TO REAL EXPOSURES; GOLD-Jl=exp (M-1+ ( ( S 1*S 1) /2) ) ;
GOLD-J2=exp (M-2+ ( (~-2*~-2) /2) 1 ; ***** CALCULATE STANDARD ERROR OF MEAN EXPOSURE FOR JI AND J2 ; SE-JI= ((VAR-Bl/N-WORK1) + (VAR-WW1/(N-DAYS1*NWWORKI) 1 ) " 0-5; SE-J2= ((VAR-B2/N-WORKS) + (VAR-WW2/(N-DAYS2*NWWORK2) ) ) * * 0.5; * * t ~ * * * * * * t * * * * * * * * * * * t * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * Y t ;
"*** LOOP TO GENERATE 5000 CASE CONTROL STUDIES 8
X * t * * * * * * * * * * * * * * * * * * * * * * * * * * t * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * ;
30 STUDY=1 TO 5000; %LET MLOGl = M-1; %LET STDLOGl = SE-JI; %LET MLOG2 = M-2; %LET STDLOG2 = SE-J2; MLOGl = O + &MLOGl ; STDLOGl = O + &STDLOGl ; MLOG2 = O + &MLOG2 ; STDLOG2 = O + &STDLOG2 ; CALL RANNOR(E-SEED1, LNEl-JI) ; ElJl =M-1 + STDLOG1 * LNE1-JI; El-JI=EXP (ElJ1) + (GOLD-J1 - EXP (m-1 + (STDLOGl*stdlogl/2) ) )
CALL RANNOR (E-SEEDS , LNE1-J2 ) ; ElJ2=M-2 + STDLOG2 * LNE1-J2; EI_JS=EXP(ElJ2) + (GOLD-J2 - EXP(M-2 + (STDLOG2*STDLOG2/2) 1 ) ; IF El-JI <O THEN El-J1=0,5; IF El-J2 <O THEN El-J2=O.S;
Power calculations
*****************************************************************; "*** LOOP TO GENERATE ONE CASE CONTROL STUDY O F 100 MATCHED SETS; rt***r*tt*t*t****tt*t*******ttttt*tt***t**********~*******w******;
DO CCSET=l TO 100; IF CCSET=l THEN DO;
MAXCASE1=O;MAXCASE2=O;MINCONT1=3OOOOO;MINCONT2=30000; END ; RETAIN MAXCASE1 MINCONTl MAXCASE2 MINCONTS; CALL RANNOR (BY-SEED, BYR) ; "* SIMULATE BIRTH Y=; BYEAR=1919 + 14*BYR; BYEAR=ROUND (BYEAR, 1) ; *~t**************t****œ**C******************************;
r t t LOOP TO GENERATE ONE MATCHED SET (1 CASE, 1 CONTROL); r * * * * * * * * * * * * * * * * * * * * * * * i t * * * * * * * * * * * * * * * * .
DO J=l TO 200; "* ASSIGN THE % OF TIME SPENT IN J1 AND J2; *** BASED ON ONTARIO HYDRO WORKERS EMPLOYMENT RECORDS; CALL RANUNI (J-SEED. JTRAN) ; **** 27% ofworkers stayed on high exposure job entire career; IF Oc=JTRANc=0.27 THEN DO; P-Jl=l;P-JS=O;JCAT=l;END; **** 32% OF WORKERS STAYED IN LOW EXPOSURE JOB OVER ENTIRE CAREER; IL 0.68<=JTRAN<=1.00 THEN D0;P-Jl=O;P-J2=l;JCAT=2;END; * * * * 41% worked i n low and high; I F 0.27~ JTRAN< 0.67 THEN DO;
*** those t h a t worked in both jobs ( i / 2 time spent in l o w ) ; CALL RANBIN(J2-SEED,lOO,O.SO.rr); P_Jl=RR/lOO; P-J2=1-P-JI; JCAT=3 ;
END ; ********************************************~ * '** KEEP TRACK OF TOTAL CASES AND CONTROLS ; *******~************************************ ; IF J=1 THEN DO;TOTCASE=O;TOTCON=O;FLAG=O;END; RETAIN TOTCAÇE TOTCON;
**********************************************~ * * * OH workers: mean ernployment dur=25 * * 8 - t t * standard dev dur=lO t t . I
**********************************************. CALL RANNOR (D-SEED, DUR) ; DUR= 25 - (4.2 * BYR) + (9.08 ' D U R ) ; IF DUR c 0.5 THEN DUR=0.5; DUR= ROWND (DUR, O -1) ; **** NUMBER OF MATCHED CONTROLS; N-CONT= 1 ;
Power calculations
**************************************************************; *** CALCULATE CUMULATIVE EXPOSURE * * * . t
* EXPO-O MEASURED WITH NO ERROR tt*.
*** EXPO-1 EXPOSURE WITH ERROR t t t . t
**************************************************************; EXPO-O = ( (P-Jl*COLD_Jl) + (P JS*GOLD-J2) ) *DUR/100 ; EXPO-1 = ( (PJl*El-JI) + (P-J~*E~-JZ) *dur/100; * * * DETERMINE CASE CONTROL STATUS; C U L RANUNI (CC-SEED, U) ; LOGITU = LOG(U/(1-U) ) ; **= PROBABILITY OF LEUKEMIA beta=0.2 per 100 V/m; "* INCRERSE RISK OF LEUKEMIA OF 25% PER 100/Vm; P-L = (0.263 * EXPO-O) + LOGITU ; IF P-L < O THEN CASE=O; IF P-L > O THEN CASE=l; **" WANT EACH MATCHED SET TO CONTAIN 1 CASE and 4 CONTROLS; IF CASE=l THEN TOTCASE=TOTCASE+CASE; IF CASE=O THEN TOTCON=TOTCON + 1; IF TOTCASE>l AND CASE=l THEN FLAG=-1; *** THROW AWAY AS ALREADY HAVE 1 CASE; IF TOTCON > N-CONT AND CASE=O THEN FLAG=-1; *** THROW AWAY AS ALREADY HAVE 1 CONTROL; IF TOTCON<= N-CONT AND CASE=O THEN FLAG=O; IF TOTCAÇEc=l AND CASE=l THEN FLAG=O; IF TOTCONwN-CONT AND TOTCASE>l THEN J=200; *** STOP LOOP AS HAVE 1 CASE, N-CONT CONTROLS; IF FLAG=O THEN DO; IF CASE=l THEN MAXCASEl=M.AX(MAXCASEl, EXPO-O) ; IF CASE=O THEN MINCONTl=MIN (MINCONTl, EXPO-O) ; IF CASE=1 THEN MAXCASE2=MAX (MAXCASE2 , EXPO-1) ; IF CASE=O T F T N MINCONTS=MIN (MINCONTS , EXPO-1) ; OUTPUT;
END ; END ; / * - - - - - - - - END OF MATCHED SET LOOP - - - - - - - - - * /
END ; /* - - - - - - - - END OF CREATING ONE MATCHED CASE CONTROL STUDY - - - - -* /
END ; "** END OF CREATING 500 MATCHED CASE CONTROL STUDIES; KEEP STUûY EXPO-O EXPO-1 CCSET STUDY CASE
SE-J1 SE-J2 GOLD-JI GOLD-J2 P-J1 P-J2 MINCONTl MINCONT2 El-J1 El-J2 DUR MAXCASE1 MAXCASE2;
PROC SORT; BY STUDY CCSET:
RUN;
Power calculations
/t - - - - - - - - CHECK FOR CONVERGENCE DIFFICULTIES - - - - - - - - - - - * / DATA C0NV;SET SIMUL;BY STCTDY;IF LAST-STUDY;
CONV-ERR=O ; IF MAXCASEl<MINCONTI OR MAXCASE2cMINCONT2 TREN CONV_ERR=l; TITLE ' NUMBER OF SIMULATED STWDIES WHERE MAX (El CASES c MIN (E) CONTROLS ' ; PROC FREQ;TABLES CONV-ERR;
RUN ; DATA NEW;
SET SIMUL; BY STUDY CCSET; TIME=L;IF CASE=O THEN TIME=2; * * * CONDITIONAL LOGISTIC REGRESSION FOR EXPOSURES WITH NO ERROR; PROC PHREG NOSUMMARY N0PRINT;MODEL TIME*CASE(O)=EXPO-O
/TIEÇ=DISCRETE RL;STRATA CCSET; BY STUDY; OUTPUT OUT=SCENO XBETA=SLOPPYO STDXBETA=STDSLOPO;
!?UN; DATA NEW;
SET SIMUL; BY STUDY CCSET; T'IME=I;IF CASC=O THEN TIME=S; w r t CONTIONAL LOGISTIC REGRESSION FOR EXPOSURES WITH ERROR;
PROC PHREG NOSUMMARY NOPR1NT;MODEL TIMEfCASE(0)=EXPO-1 /TIES=DISCRETE RL~STRATA CCSET; BY STUDY;
E L i ; DATA G0LD;SET SCEN0;BY STWDY;
IF FIRST-STUDY; SLOPE=SLOPPYO/EXPO-O; SE-SLOPE=STDSLOPO/EXPO-O; NsTAT=SLOPE/sE~SLOPE; ??VALUE=I- PROBNORM (NSTAT) ; TAG=O ; IF PVALUEcO.05 THEN TAG=l; PROC MEANS;VAR SLOPE SE-SLOPE; TIîLE1 'POWER CALCULATIONS FOR E PROC FREQ; TABLES TAG;
RUN; DATA G0LD;SET SCEN1;BY STUDY;
IF FIRST.STuDY; SLOPE=SLOPPY~/EXPO-1; SE~SLOPE=STDSLOP~/EXPO~~; IF SLOPPYl=. THEN PPPP=1; NSTAT=SLOPE/SE_SLOPE; PVALUE=l-PROBNORM(NSTAT); TAG=O ; IF PVALUE<O.OS THEN TAG=l;
:XPOSURE MEASURED WITHOUT ERROR ' ;
TITLEl 'POWER CALCULATIONS FOR EXPOSURE WITH SAMPLING & MEASUREMENT ERROR'; PROC MEANS;VAR SLOPE SE-SLOPE; FROC FREQ;TABLES TAG PPPP;
RUN ; /* c ~ c ~ < c ~ c c < < < c < ~ c c ~ ~ ~ ~ c ~ ~ END OF PROGRAM >>>>>>>>>>>>>>>>>>>>>>>9/ ;
Power calculations
Conclusions and recommendations for further research
The availability of detailed electric and magnetic field exposure assessment w i t h the
Ontario Hydro workforce permitted several key reseafch questions to be addressed within this
thesis that had not previously been evaluated by the original investigators of the study. Most
notably, the relation between various indices of electric and magnetic fields was exarnined and
additiondy, cancer risk assessment was performed using a selected series of exposure indices for
a variety of cancers.
1 found that several exposure indices are needed to capture the variabiiity of 60 Hz electric
and magnetic field exposures in the Ontario Hydro workers and therefore, a series of indices
should be considered when assessing ELF related health outcomes in this occupational
population. Moreove- the poor correlations observed between electric and magnetic fields
indicate that these exposures be modeled as separate variables when attempting to characterize
risk.
An increased risk of leukernia and Non-Hodgkin's lymphoma was observed among those
workers more likely to have exposures above thresholds in the range of 10-39 Vlm Perhaps most
important, for these cancers, these threshold metrics were predictive of rkk over and above the
associations noted with the time weighted average field strength This is noteworthy as the TWA
210
has typically been used to characterize risk in studies of electric utility workers. Magnetic fields
were not strongly associated with any of the cancers examined, though there were few brain
cancer cases.
At first glance, the finding of an increased risk of NHL and leukemia with electric, and not
magnetic field seems surprising, given that electric fields are more easily shielded by buildings and
the human body. However, as outlined in the surnmary of experimental studies, electric field
exposures can influence several cellular fiinctions, ornithine decarboxylase and levels of
melatonin. Furthermore, recent experimental work has demonstrated that exposure to 10 V/m
results in induced currents that are much greater in the blood than in other body structures [l].
In contrast, magnetic fields have been shown to affect induced currents to a greater extent in
cerebrospinal fluid [Il suggesting that magnetic fields may be more relevant than electric fields in
the etiology of brain cancer.
Further efforts are needed both in humans and at the cellular level to describe the
ditFerences in the eRects of magnetic and electric fields. Some recent work has assayed levels of
melatonin in electric uti1ity workers in relation to their exposure to magnetic fields [2, 31. Such
work could readily be extended to assess melatonin levels as they relate to electric field
exposures.
Conclusions and recornmendations for fbrther research
As before, my analysis was limited by a small number of brain cancer cases. It is
recommended that future record linkage be pursued in order to increase the power of the study to
evaluate the relation to between field exposures and the incidence of brain cancer and additionally,
to alIow for risks to be assessed for testicular cancer. Recent work has demonstrated an
association between magnetic fields and this cancer [4], and fùrther, levels of melatonin, which are
mediated by 60 Hz field electric and magnetic exposures, may influence the development of
testicular cancer [ 5 ] .
Finally, the Radiation Protection Branch of Health Canada is currently conducting animal
experiments to investigate the effects of electric and magnetic field exposures on processes related
to brain development [6]. Recently, a number of studies have come forward that suggest that
power fiequency field exposures rnay play a role in the development of neurodegenerative
disorders, including Alzheimers disease [7- 1 O]. If this is tme, tùrther study of these health
outcomes in utility workers are needed. The use of the Ontario Hydro cohort would be
particularly valuable given the quality of electric and magnetic field exposure and additionally, the
age profile of this cohort.
-
Conclusions and recommendations for krther research
Chapter 5: References
Dawson, T.W., K. Caputa, and M-A. Stuchly, A comprisorr of 60 Hz wriform magnetic
artd eiectric field i~tductiotr itr the humatr body. Phys Med Biol, 1 997. 42( 1 2) : p 23 1 9-29.
Pfluger, D. H. and C.E. Minder, Effects of expomre to 16.7 Hz mapteticfieldr on irri~rafy
6- hydroxymelatoriin sulfate excretion of Swiss raiIway workers. J Pineal Res, 1996.
ZL(2): p. 91-100.
Burch, J.B., et ai., Redrtced excretion of a melatonin metabolite 1tr workers exposed to 60
Hz magneticfiel& Am J Epiderniol, 1999. 150(1): p. 27-36.
Stenlund, C . and B. Floderus, Occtputional expomre to magneticfields in relatiotr to
)riale h a s t cancer and testiclrlar cancer: a Swedish case-control sttrcfy. Cancer Causes
Control, 1997. 8(2): p. 184-91.
B lask, D. E., me emerging role of the pineal gland atd melatonin in oncogetresis.. in
lht-emely low freqttency electromagnetic fieldsr the question of catrcer., B. W . Wilson,
R.G. Stevens, 2nd L.E. Anderson, Editors. 1989, Batelle Press: Colombus, OH. p. 3 19-
335.
MacLean J, Personal Commzrnication, . 2000.
Feychting, M., et al., Dementia artd occupational expomre to magrretic field. Scand J
Work Environ Health, 1998. 24(1): p. 46-53.
Savitz, D. A., D. P. Loomis, and C.K. Tse, Electrical ocarpations and netrrodegenerative
disease: arralysis of C/. S. mortality data. Arch Environ Healt h, 1 998. 53( 1 ): p. 7 1 -4.
Chapter 5 Conclusions and recommendations for fiirther research
9. SobeI, E., et al., Elevated risk of Alzheimer's diseuse among workers with likely
electromagneticfield exposure. Neurology, 1 996. 47(6): p. 1477-8 1 .
1 0. S obel, E. and 2. Davani pour, Electromagnetic field e x p u r e may cause increased
production of amyloid beta and eventuall) lead to Alzheimer's diseme. Neurology, 1996.
47(6): p. 1594-600.
Chapter 5 Conclusions and recomrnendations for fùrther research