American Journal of Epidemiology Copyright © 2003 by the Johns Hopkins Bloomberg School of Public Health All rights reserved

Vol. 158, No. 6 Printed in U.S.A. DOI: 10.1093/aje/kwg197

Occupational Exposure to Extremely Low Frequency Magnetic Fields and Mortality from Cardiovascular Disease

Niclas Håkansson1,2, Per Gustavsson3,4, Antonio Sastre5, and Birgitta Floderus1,2 1

Institute of Environmental Medicine, Karolinska Institutet, Stockholm, Sweden. National Institute for Working Life, Stockholm, Sweden. 3 Department of Occupational and Environmental Health, Stockholm Public Health Center, Stockholm, Sweden. 4 Division of Occupational Health, Department of Public Health Sciences, Karolinska Institutet, Stockholm, Sweden. 5 Health Assessment and Research Center, Midwest Research Institute, Kansas City, MO. 2

Received for publication November 5, 2002; accepted for publication April 7, 2003.

cardiovascular diseases; electromagnetic fields; myocardial infarction; occupations

Abbreviations: AMI, acute myocardial infarction; ELF, extremely low frequency.

In 1999, an epidemiologic study by Savitz et al. (1) on cardiovascular disease mortality in a cohort of electric utility workers showed an association between exposure to extremely low frequency (ELF) magnetic fields and cardiovascular disease mortality. The association was attributable to arrhythmia-related death and acute myocardial infarction (AMI), while ischemic heart disease other than AMI and atherosclerosis did not show the same pattern. In another study based on utility company records, Sahl et al. (2) analyzed mortality from AMI and chronic coronary heart disease. They concluded that no association was observed either for AMI or for chronic coronary heart disease in relation to cumulative exposure to ELF magnetic fields,

although the study also included results for AMI that suggested an association. Previous work in laboratory studies on human volunteers showed a reduction in heart rate variability upon intermittent magnetic field exposure (3). Heart rate variability is a marker of autonomic cardiac control, and reductions in heart rate variability have been shown to predict sudden death (4), all-cause mortality, and heart disease in prospective epidemiologic studies (5–9). Acute effects on heart rate variability from ELF magnetic field exposure have been found in some later laboratory studies, but not all, and the presence of these effects appears to be correlated with the level of nervous system arousal (10).

Correspondence to Niclas Håkansson, Institute of Environmental Medicine, Karolinska Institutet, Box 210, S-171 77 Stockholm, Sweden (email: [email protected]).

534

Am J Epidemiol 2003;158:534–542

Downloaded from aje.oxfordjournals.org by guest on October 26, 2010

The authors evaluated the relation between occupational exposure to extremely low frequency (ELF) magnetic fields and mortality from cardiovascular diseases. The study was based on 27,790 subjects from the Swedish Twin Registry. Exposure to ELF magnetic fields was assessed by linking occupation reported in 1967 or 1973 to a job exposure matrix. Four levels of exposure were related to cause-specific mortality through 1996, and primary and contributing causes of death were considered. The authors estimated relative risks by Cox regression, with adjustment for several cardiovascular disease risk indicators. The authors calculated the synergy index to evaluate potential interaction between exposure to ELF magnetic fields (>0.2 µT) and genetic susceptibility to acute myocardial infarction (AMI). Arrhythmia-related death, ischemic heart disease other than AMI, and atherosclerosis showed no association with ELF magnetic fields. The authors found a low-level increase in AMI risk in the highest exposure group (relative risk = 1.3, 95% confidence interval: 0.9, 1.9) and suggestions of an exposure-response relation (p = 0.02). A synergy index of 2.7 (95% confidence interval: 1.1, 6.6) in monozygotic twins indicated that the risk of AMI was strengthened among ELF magnetic field-exposed subjects with genetic susceptibility to the disease. The results for AMI corroborate previous findings from the United States. The unusual opportunity to include genetic susceptibility in the present analyses showed that evaluations of effect modification in vulnerable subjects are warranted in ELF magnetic field research.

ELF Magnetic Fields and Cardiovascular Disease 535

TABLE 1. Mortality distribution of a cohort selected from the Swedish Twin Registry, by category of occupational exposure to extremely low frequency magnetic fields, Sweden, 1967/1973–1996 Exposure* (µT) Total no. 0–<0.1

0.1–<0.2

0.2–<0.3

≥0.3

No. of subjects Men

16,680

621

11,005

3,148

Women

11,110

3,152

6,040

1,583

1,906 335

All subjects

27,790

3,773

17,045

4,731

2,241

Men

362,566

13,458

238,092

70,064

40,952

Women

250,567

71,819

136,599

34,567

7,582

All subjects

613,133

85,277

374,691

104,631

48,534

Men

4,646

159

3,146

727

614

Women

2,025

481

1,096

392

56

All subjects

6,671

640

4,242

1,119

670

Person-years of follow-up

No. of deaths

The aim of the present study was to investigate whether long-term exposure to ELF magnetic fields is associated with cardiovascular disease mortality, particularly AMI and arrhythmia-related death. An advantage over the previous studies was the substantial ELF magnetic field exposure diversity of a general population and the ability to control for several known risk factors for cardiovascular disease, including smoking, alcohol ingestion habits, physical activity, and body mass index (weight (kg)/height (m)2). We also aimed to evaluate whether genetic susceptibility to cardiovascular disease modifies the relation between exposure and disease. Specifically, we wished to examine whether individuals with increased potential for genetic susceptibility to AMI are more vulnerable to long-term ELF magnetic field exposure. It was possible to examine this because we used a study population comprising a large series of twin pairs. MATERIALS AND METHODS

The study was based on two subsamples from the Swedish Twin Registry, a nationwide registry including (in principle) all same-sex twins born in Sweden. We used the first cohort established in 1961, comprising 21,884 twins born in 1886– 1925. They answered a questionnaire in 1967, and we coded their answers regarding their main occupation according to the Nordic version of the International Standard Classification of Occupations (produced by the International Labour Organization). In addition, we used data from the second cohort, established in 1971, comprising 36,854 twins born in 1926–1958. For this group, the questionnaire data go back to 1973. Assessment of ELF magnetic field exposure was based on the occupation recorded in 1967 or 1973. Levels of exposure were obtained by linking the occupations to a previously elaborated job exposure matrix comprising the 100 most Am J Epidemiol 2003;158:534–542

common occupations (among men) in Sweden. The exposure estimates of the occupations were based on at least four dosimetric full-shift measurements from different workplaces. In all, more than 1,000 measurements were carried out, as described in detail previously (11). To increase the number of subjects, we used additional exposure information for some rare occupations (12). Students, housewives, other subjects without an occupation, and subjects with occupations not included in the job exposure matrix were excluded from the analyses. The exposure metric used in the analysis was the geometric mean of average workday mean values. The exposures were divided into four groups: <0.10 µT (low exposure, reference group), 0.10–0.19 µT (medium exposure), 0.20–0.29 µT (high exposure), and ≥0.3 µT (very high exposure). The cutoff value for the reference group was set low in order to minimize misclassification in the reference group. Information on cause-specific mortality was obtained through linkage with the causes-of-death registry, and both underlying and contributing causes of death were considered. Causes of death were coded according to the International Classification of Diseases. For the years 1967 and 1968, the Seventh Revision of the International Classification of Diseases was used; for 1969–1986, the Eighth Revision was used; and for 1987–1996, the Ninth Revision was used. We analyzed AMI as a primary cause of death and AMI as a primary or contributing cause. Furthermore, we collectively analyzed ischemic heart disease other than AMI, arrhythmia, and atherosclerosis as primary or contributing causes of death. We calculated relative risk estimates based on Cox regression (13, 14). All risk estimates were computed using SAS software (15). In the Cox regression analysis, the basic time dimension was calendar years, and the subjects were considered to be at risk from study entry (1967 or 1973) to the end of follow-up (1996), the year of death, or age 75 years,

Downloaded from aje.oxfordjournals.org by guest on October 26, 2010

* Geometric mean of average workday mean.

536 Håkansson et al.

TABLE 2. Relative risk of death from cardiovascular disease in relation to occupational exposure to extremely low frequency magnetic fields, Sweden, 1967/1973–1996 All subjects Cause of death and exposure level (µT)

No. of deaths

RR*,†

Men 95% CI*

No. of deaths

RR‡

Women 95% CI

No. of deaths

RR‡

95% CI

Acute myocardial infarction, primary cause§ Low (0–0.09)

44

1.00

14

1.00

30

1.00

Medium (0.10–0.19)

466

1.11

0.81, 1.53

406

1.35

0.79, 2.29

60

1.00

0.64, 1.56

High (0.20–0.29)

133

1.26

0.89, 1.79

109

1.51

0.86, 2.64

24

1.08

0.62, 1.87

Very high (≥0.30)

94

1.31

0.90, 1.90

90

1.57

0.89, 2.77

4

1.21

0.42, 3.48

p for trend

0.05

0.06

0.70

Acute myocardial infarction, primary or contributing cause Low (0–0.09)

49

1.00

15

1.00

34

1.00

Medium (0.10–0.19)

501

1.10

0.81, 1.49

436

1.35

0.81, 2.26

65

0.94

0.61, 1.43

High (0.20–0.29)

150

1.31

0.94, 1.82

120

1.55

0.90, 2.66

30

1.19

0.72, 1.98

Very high (≥0.30)

102

1.31

0.92, 1.87

97

1.59

0.92, 2.74

5

1.33

0.51, 3.45

0.02

0.04

0.40

Ischemic heart disease, other than acute myocardial infarction¶ Low (0–0.09)

29

1.00

15

1.00

14

1.00

232

0.87

0.57, 1.34

193

0.57

0.34, 0.97

39

1.19

0.64, 2.21

High (0.20–0.29)

52

0.79

0.48, 1.28

37

0.47

0.26, 0.85

15

1.25

0.59, 2.62

Very high (≥0.30)

31

0.70

0.40, 1.21

30

0.47

0.25, 0.88

1

0.60

0.08, 4.66

Medium (0.10–0.19)

p for trend

0.13

0.05

0.82

Arrhythmia, primary or contributing cause# Low (0–0.09)

9

1.00

104

1.68

0.85, 3.30

High (0.20–0.29)

25

1.54

0.72, 3.27

Very high (≥0.30)

18

1.80

0.81, 4.03

Medium (0.10–0.19)

p for trend

0.46

Atherosclerosis, primary or contributing cause** Low (0–0.09)

18

1.00

119

0.90

0.54, 1.50

High (0.20–0.29)

42

1.25

0.71, 2.19

Very high (≥0.30)

23

1.13

0.60, 2.13

Medium (0.10–0.19)

p for trend

0.18

* RR, relative risk; CI, confidence interval; ICD-7, -8, or -9, International Classification of Diseases, Seventh, Eighth, or Ninth Revision. † Adjusted for sex, age, smoking, and body mass index. ‡ Adjusted for age, smoking, and body mass index. § ICD-7* code 420.1; ICD-8* and ICD-9* code 410. ¶ ICD-7 codes 420.0 and 422.1; ICD-8 codes 412 and 413; ICD-9 codes 411–414. # ICD-8 code 427; ICD-9 codes 426 and 427. ** ICD-8 and ICD-9 code 440.

whichever came first (age 85 years was used in the agestratified analysis). The numbers of study subjects, personyears of follow-up, and deaths occurring during the study period are shown in table 1. The four exposure levels were entered into the regression model as dummy variables. Age was entered as age at the start of follow-up. We used six categories of smoking habits: never smoker, previous smoker, and four groups of current smokers (depending on the number of cigarettes smoked per day). The other factors were entered as follows: body mass index (in quintiles), alcohol drinking habits (excessive

consumption vs. lower consumption), and physical activity during leisure time (regular or hard vs. light). To control for socioeconomic factors, we restricted the study population to manual workers in additional analyses. Alcohol drinking habits and physical activity showed no effect on the relation and thus were deleted from the final model. By means of Cox regression, we evaluated potential exposure-response associations using trend tests with the assumption of equal distance between the four exposure levels. If a twin in a pair had died from AMI before age 75 years, we used this as a marker of potential genetic susceptibility to Am J Epidemiol 2003;158:534–542

Downloaded from aje.oxfordjournals.org by guest on October 26, 2010

p for trend

ELF Magnetic Fields and Cardiovascular Disease 537

TABLE 3. Relative risk of death from cardiovascular disease among manual workers in relation to occupational exposure to extremely low frequency magnetic fields, Sweden, 1967/1973–1996 All subjects Cause of death and exposure level (µT)

No. of deaths

RR*,†

Men 95% CI*

No. of deaths

RR‡

Women 95% CI

No. of deaths

RR‡

95% CI

Acute myocardial infarction, primary cause§ Low (0–0.09)

25

1.00

8

1.00

17

1.00

243

1.15

0.74, 1.77

222

1.40

0.69, 2.85

21

0.93

0.49, 1.78

High (0.20–0.29)

95

1.19

0.76, 1.86

74

1.44

0.69, 3.01

21

1.06

0.55, 2.05

Very high (≥0.30)

64

1.31

0.80, 2.14

60

1.53

0.73, 3.22

4

2.35

0.76, 7.25

Medium (0.10–0.19)

p for trend

0.24

0.37

0.40

Acute myocardial infarction, primary or contributing cause Low (0–0.09)

27

1.00

8

1.00

19

1.00

Medium (0.10–0.19)

264

1.22

0.80, 1.85

240

1.52

0.75, 3.09

24

0.94

0.51, 1.72

High (0.20–0.29)

105

1.27

0.82, 1.95

79

1.54

0.74, 3.20

26

1.17

0.63, 2.15

Very high (≥0.30)

71

1.43

0.90, 2.29

66

1.69

0.81, 3.53

5

2.57

0.93, 7.13

0.13

0.28

0.20

Ischemic heart disease, other than acute myocardial infarction¶ Low (0–0.09)

13

1.00

8

1.00

5

1.00

125

1.18

0.62, 2.25

111

0.68

0.33, 1.40

14

1.85

0.69, 4.93

High (0.20–0.29)

43

1.06

0.54, 2.07

28

0.53

0.24, 1.17

15

2.25

0.82, 6.16

Very high (≥0.30)

18

0.75

0.34, 1.66

17

0.42

0.18, 0.97

1

2.03

0.22, 18.44

Medium (0.10–0.19)

p for trend

0.16

0.02

0.14

Arrhythmia, primary or contributing cause# Low (0–0.09)

7

1.00

Medium (0.10–0.19)

60

1.31

0.60, 2.85

High (0.20–0.29)

23

1.24

0.54, 2.84

Very high (≥0.30)

10

1.00

0.38, 2.60

p for trend

0.69

Atherosclerosis, primary or contributing cause** Low (0–0.09)

9

1.00

Medium (0.10–0.19)

59

1.11

0.53, 2.32

High (0.20–0.29)

33

1.50

0.71, 3.20

Very high (≥0.30)

19

1.68

0.73, 3.88

p for trend

0.05

* RR, relative risk; CI, confidence interval; ICD-7, -8, or -9, International Classification of Diseases, Seventh, Eighth, or Ninth Revision. † Adjusted for sex, age, smoking, and body mass index. ‡ Adjusted for age, smoking, and body mass index. § ICD-7* code 420.1; ICD-8* and ICD-9* code 410. ¶ ICD-7 codes 420.0 and 422.1; ICD-8 codes 412 and 413; ICD-9 codes 411–414. # ICD-8 code 427; ICD-9 codes 426 and 427. ** ICD-8 and ICD-9 code 440.

the disease in the other twin. Below, we use the expression “genetic susceptibility” even if the variable is a proxy marker. AMI mortality among persons with ELF magnetic field exposure only, persons with genetic susceptibility only, or persons with both factors was compared with the AMI mortality of a reference group comprising individuals unexposed to ELF magnetic fields and without any indication of genetic susceptibility. A dichotomized measure of ELF magnetic field exposure was used, with a cutoff at 0.20 µT. We analyzed the interaction between the two factors by Am J Epidemiol 2003;158:534–542

testing whether the joint effect was greater than the sum of the independent effects of the single factors by calculating the synergy index (16). Confidence intervals for the synergy index were calculated according to the method of Hosmer and Lemeshow (17). A computer program developed by Lundberg et al. (18) that calculates the confidence limits was modified in order to fit the Cox regression analysis. In all of the analyses described above, the twin subjects were treated as a sample of individuals from the general population, disregarding twinship (19). To ensure that confi-

Downloaded from aje.oxfordjournals.org by guest on October 26, 2010

p for trend

538 Håkansson et al.

TABLE 4. Age-stratified relative risk of death from cardiovascular disease in relation to occupational exposure to extremely low frequency magnetic fields, Sweden, 1967/1973–1996 Age (years) ≤65

Cause of death and exposure level (µT) No. of deaths

RR*,†

66–75 95% CI*

No. of deaths

RR†

76–85 95% CI

No. of deaths

RR†

95% CI

Acute myocardial infarction, primary cause‡ Low (0–0.09)

17

1.00

27

1.00

52

1.00

183

1.10

0.64, 1.88

283

1.09

0.73, 1.63

306

0.68

0.49, 0.92

High (0.20–0.29)

51

1.15

0.64, 2.07

82

1.30

0.84, 2.00

70

0.68

0.47, 0.99

Very high (≥0.30)

37

1.31

0.70, 2.44

57

1.26

0.79, 2.00

47

0.63

0.42, 0.95

Medium (0.10–0.19)

p for trend

0.29

0.14

0.13

Acute myocardial infarction, primary or contributing cause Low (0–0.09)

19

1.00

30

1.00

57

1.00

194

1.08

0.65, 1.78

307

1.09

0.74, 1.59

327

0.67

0.49, 0.90

High (0.20–0.29)

56

1.16

0.67, 2.03

94

1.36

0.90, 2.05

75

0.67

0.47, 0.96

Very high (≥0.30)

42

1.39

0.78, 2.48

60

1.21

0.77, 1.91

49

0.61

0.41, 0.91

Medium (0.10–0.19)

0.13

0.11

0.08

Ischemic heart disease, other than acute myocardial infarction§ Low (0–0.09)

12

1.00

17

1.00

28

1.00

Medium (0.10–0.19)

84

0.77

0.38, 1.57

148

0.93

0.54, 1.60

185

0.82

0.54, 1.25

High (0.20–0.29)

12

0.42

0.17, 1.00

40

1.04

0.58, 1.88

43

0.88

0.54, 1.43

Very high (≥0.30)

12

0.65

0.27, 1.60

19

0.70

0.34, 1.40

32

0.89

0.53, 1.50

p for trend

0.10

0.44

0.96

Arrhythmia, primary or contributing cause¶ Low (0–0.09)

3

1.00

6

1.00

12

1.00

39

1.93

0.61, 6.07

65

1.55

0.67, 3.58

101

1.03

0.56, 1.93

High (0.20–0.29)

4

0.72

0.16, 3.16

21

1.94

0.79, 4.76

30

1.33

0.68, 2.60

Very high (≥0.30)

8

2.50

0.65, 9.55

10

1.45

0.53, 3.97

13

0.83

0.37, 1.86

Medium (0.10–0.19)

p for trend

0.82

0.47

0.98

Atherosclerosis, primary or contributing cause# Low (0–0.09)

5

1.00

13

1.00

33

1.00

26

0.75

0.27, 2.10

93

0.96

0.54, 1.71

179

0.72

0.48, 1.08

High (0.20–0.29)

9

0.95

0.29, 3.11

33

1.34

0.71, 2.54

34

0.59

0.36, 0.96

Very high (≥0.30)

5

0.95

0.26, 3.52

18

1.17

0.57, 2.40

28

0.72

0.42, 1.24

Medium (0.10–0.19)

p for trend

0.71

0.21

0.20

* RR, relative risk; CI, confidence interval; ICD-7, -8, or -9, International Classification of Diseases, Seventh, Eighth, or Ninth Revision. † Adjusted for sex, age, smoking, and body mass index. ‡ ICD-7* code 420.1; ICD-8* and ICD-9* code 410. § ICD-7 codes 420.0 and 422.1; ICD-8 codes 412 and 413; ICD-9 codes 411–414. ¶ ICD-8 code 427; ICD-9 codes 426 and 427. # ICD-8 and ICD-9 code 440.

dence intervals were not erroneously narrowed because of similarities within pairs, we performed Cox regression analyses that adjusted variance estimates for correlated outcomes (20–22). We accomplished this through the use of a macro in SAS that stems from the same theoretical background and yields the same results as does the published FORTRAN program of Lin (23). RESULTS

There was no overall increase in mortality across the exposure groups. The relative risks in the medium, high, and very

high exposure groups were 0.95 (95 percent confidence interval: 0.87, 1.04), 0.98 (95 percent confidence interval: 0.89, 1.09), and 0.98 (95 percent confidence interval: 0.87, 1.10), respectively. For AMI, we found low-level increases in the relative risks both for primary cause of death and for primary or contributing cause of death (table 2). The confidence intervals for the risk estimates were wide, but there was a suggestion of an exposure-response relation, and the test for trend yielded a p value of 0.05 for primary cause and a p value of 0.02 for primary or contributing cause. In the analysis of manual Am J Epidemiol 2003;158:534–542

Downloaded from aje.oxfordjournals.org by guest on October 26, 2010

p for trend

ELF Magnetic Fields and Cardiovascular Disease 539

TABLE 5. Relative risk of death and synergy index for acute myocardial infarction in relation to genetic susceptibility and occupational exposure to extremely low frequency magnetic fields, Sweden, 1967/1973–1996 Twin group and sex

Genetic susceptibility

ELF* magnetic field exposure (µT)

No. of deaths

Relative risk†

No

<0.20

463

1.00

No

≥0.20

211

1.19

1.01, 1.40

Yes

<0.20

87

4.50

3.35, 6.05

Yes

≥0.20

41

6.48

4.47, 9.39

No

<0.20

371

1.00

No

≥0.20

180

1.17

0.98, 1.39

Yes

<0.20

80

4.45

3.27, 6.05

Yes

≥0.20

37

6.07

4.11, 8.97

No

<0.20

92

1.00

No

≥0.20

31

1.21

0.80, 1.85

Yes

<0.20

7

5.04

1.79, 14.18

Yes

≥0.20

4

14.00

5.11, 38.39

No

<0.20

131

1.00

No

≥0.20

64

1.29

0.96, 1.75

Yes

<0.20

20

4.54

2.39, 8.62

Yes

≥0.20

15

11.25

6.17, 20.51

No

<0.20

102

1.00

No

≥0.20

54

1.27

0.91, 1.77

Yes

<0.20

19

4.77

2.44, 9.31

Yes

≥0.20

15

12.14

6.59, 22.37

No

<0.20

29

1.00

No

≥0.20

10

1.23

0.59, 2.58

Yes

<0.20

1

1.38

0.08, 23.73

Yes

≥0.20

0

No

<0.20

295

1.00

No

≥0.20

134

1.16

0.94, 1.42

Yes

<0.20

57

4.27

2.98, 6.12

Yes

≥0.20

25

5.23

3.25, 8.44

No

<0.20

239

1.00

No

≥0.20

114

1.13

0.91, 1.41

Yes

<0.20

52

4.08

2.81, 5.94

Yes

≥0.20

21

4.45

2.65, 7.48

No

<0.20

56

1.00

No

≥0.20

20

1.25

0.74, 2.11

Yes

<0.20

5

5.86

1.79, 19.15

Yes

≥0.20

4

19.07

6.48, 56.07

95% CI*

Synergy index‡

95% CI

All twins§ All subjects

Men

Women

0.91, 2.43

1.40

0.84, 2.36

3.06

0.66, 14.16

2.68

1.09, 6.58

2.76

1.10, 6.91

1.23

0.66, 2.32

1.07

0.54, 2.15

3.54

0.69, 18.14

Monozygotic twins All subjects

Men

Women

Dizygotic twins All subjects

Men

Women

* ELF, extremely low frequency; CI, confidence interval. † Adjusted for sex, age, smoking, and body mass index. ‡ Synergy index: 1 = no interaction; 2 = an effect among persons with both ELF magnetic field exposure and genetic susceptibility that is twice what would be expected from additivity of effects. § Includes monozygotic and dizygotic twins and twins of unknown zygosity.

workers only, we obtained similar point estimates but wider confidence intervals for the relative risks (table 3). Am J Epidemiol 2003;158:534–542

The relative risks for arrhythmia-related death were slightly higher than those for AMI, but the outcome was not

Downloaded from aje.oxfordjournals.org by guest on October 26, 2010

1.48

540 Håkansson et al.

DISCUSSION

Although we could observe only a weak association for AMI, the result pointed in the same direction as the result from the study by Savitz et al. (1) (figures 1 and 2). The outcome of an inverse association for ischemic heart disease other than AMI was also consistent with the findings from the Savitz et al. study (1). For subjects between 66 and 75 years of age, that is, up to 10 years after retirement, we observed a low-level increase in the relative risk, while subjects between 76 and 85 years of age showed no such increase. The lack of association may have been due to survivor effects. The outcome also showed that an increased risk of dying from AMI may be retained up to 10 years after cessation of the occupational exposure and that a potential biologic interaction is not restricted to acute effects. This finding might also show that it may be essential to truncate follow-up at a certain age, since effects from exposure that ceased long ago may fade away with increasing age at death from AMI. In the study by Sahl et al. (2), the distribution of ages at death was not reported, but follow-up beyond the relevant ages might have influenced some results. There seemed to be an inverse relation in relative risk for AMI as compared with ischemic heart disease other than AMI. An explanation could be that persons who die from AMI may deplete the pool of persons who are susceptible to later chronic coronary heart disease. Statistical precision in this study, particularly among women and for specific diagnoses, was sometimes too limited for us to obtain conclusive results. The proportional distribution of the specific diseases in our study was not fully comparable to that of the Savitz et al. study. For example, the proportion of persons with arrhythmia as the primary cause of death (1.1 percent) was lower than in US data (3.1 percent). A possible reason could be variations in diagnostic criteria between the countries, differences in how death

FIGURE 1. Relative risk of cardiovascular disease mortality in relation to occupational magnetic field exposure among Swedish twins, 1967/1973–1996.

certificates are completed, or different age distributions or risk factor patterns. We did not have access to data on morbidity. It is difficult to speculate on potential effects introduced by mortality data. Mortality and morbidity data should be more comparable for AMI than for other ischemic heart disease, where the progress of disease is slower, with longer survival times. It seems reasonable to think that the relative risks would have been strengthened if time to first heart attack had been analyzed, considering the hypothesis of an acute effect on heart rate variability from ELF magnetic field exposure. The models including control for confounding factors yielded results almost identical to those of the analyses controlling for age only. Socioeconomic status is an established risk factor for AMI that we could not consider because of a lack of appropriate data. On the other hand, we took into account smoking, body mass index, alcohol drinking habits, and physical activity, that is, lifestyle factors that may summarize an important part of the risk carried by “socioeconomic status.” When the analysis was restricted to manual workers, the precision was reduced but the risk esti-

FIGURE 2. Relative risk of cardiovascular disease mortality in relation to occupational magnetic field exposure among US electric utility workers, 1950–1988. Data were obtained from Savitz et al. (1).

Am J Epidemiol 2003;158:534–542

Downloaded from aje.oxfordjournals.org by guest on October 26, 2010

compatible with an exposure-response relation. The result for arrhythmia among manual workers did not show an effect. For ischemic heart disease other than AMI, we did not find any increased risk for the exposure groups; rather, the risk tended to be significantly decreased. For atherosclerosis, the relative risks were close to unity, but in the analysis of manual workers, an exposure-response relation was suggested (p = 0.05). In the main analyses, we excluded subjects over 75 years of age, but in the age-stratified analysis we also analyzed subjects between 76 and 85 years of age. This analysis showed coherent results for persons under age 65 years and persons aged 66–75 years (table 4), while subjects aged 76– 85 years at baseline showed a reduced risk of AMI. A small effect of ELF magnetic fields (>0.20 µT) was seen in the absence of genetic susceptibility to the disease, and a strong effect was found for genetic susceptibility in the absence of ELF magnetic fields (table 5). The joint prevalence of the two factors showed the strongest increase in risk, and a synergistic effect above additivity was indicated, particularly among monozygotic twins.

ELF Magnetic Fields and Cardiovascular Disease 541

ACKNOWLEDGMENTS

The Swedish Twin Registry is supported by grants from the Swedish Council for the Planning and Coordination of Research, the Swedish Social Research Council, and the MacArthur Foundation Research Network on Successful Aging. This study was supported by grant 2001-0210 from the Swedish Council for Working Life and Social Research.

REFERENCES 1. Savitz DA, Liao D, Sastre A, et al. Magnetic field exposure and cardiovascular disease mortality among electric utility workers.

Am J Epidemiol 2003;158:534–542

Am J Epidemiol 1999;149:135–42. 2. Sahl J, Mezei G, Kavet R, et al. Occupational magnetic field exposure and cardiovascular mortality in a cohort of electric utility workers. Am J Epidemiol 2002;156:913–18. 3. Sastre A, Cook MR, Graham C. Nocturnal exposure to intermittent 60 Hz magnetic fields alters human cardiac rhythm. Bioelectromagnetics 1998;19:98–106. 4. Algra A, Tijsen JG, Roelandt JR, et al. Heart rate variability from 24-hour electrocardiography and the 2-year risk for sudden death. Circulation 1993;88:180–5. 5. Tsuji H, Venditti FJ Jr, Manders ES, et al. Reduced heart rate variability and mortality risk in an elderly cohort: The Framingham Heart Study. Circulation 1994;90:878–83. 6. Tsuji H, Martin G, Larson MG, et al. Impact of reduced heart rate variability on risk for cardiac events: The Framingham Heart Study. Circulation 1996;94:2850–5. 7. Liao D, Cai J, Rosamond WD, et al. Cardiac autonomic and incident heart disease: a population-based case-cohort study. The ARIC Study. Am J Epidemiol 1997;145:696–706. 8. Dekker JM, Schouten EG, Klootwijk P, et al. Heart rate variability from short electrocardiographic recordings predicts mortality from all causes in middle-aged and elderly men: The Zutphen Study. Am J Epidemiol 1997;145:899–908. 9. Dekker JM, Crow RS, Folsom AR, et al. Low heart rate variability in a 2-minute rhythm strip predicts risk of coronary heart disease and mortality from several causes: The ARIC Study. Circulation 2000;102:1239–44. 10. Graham C, Cook MR, Sastre A, et al. Cardiac autonomic control mechanisms in power-frequency magnetic fields: a multistudy analysis. Environ Health Perspect 2000;108:737–42. 11. Floderus B, Persson T, Stenlund C. Magnetic field exposures in the workplace: reference distribution and exposure in occupational groups. Int J Occup Environ Health 1996;2:226–38. 12. Floderus B, Persson T, Stenlund C, et al. Occupational exposure to electromagnetic fields in relation to leukaemia and brain tumours: a case-control study in Sweden. Cancer Causes Control 1993;4:465–76. 13. Breslow NE, Day NE, eds. Statistical methods in cancer research. Vol 2. The design and analysis of cohort studies. Lyon, France: International Agency for Research on Cancer, 1987. 14. Clayton D, Hills M. Statistical models in epidemiology. Oxford, United Kingdom: Oxford Science Publications, 1993. 15. SAS Institute, Inc. SAS software, version 8.1. Cary, NC: SAS Institute, Inc, 2000. 16. Rothman KJ. Modern epidemiology. Boston, MA: Little, Brown and Company, 1986. 17. Hosmer DW, Lemeshow S. Confidence interval estimation of interaction. Epidemiology 1992;3:452–6. 18. Lundberg M, Fredlund P, Hallqvist J, et al. A SAS program calculating three measures of interaction with confidence intervals. Epidemiology 1996;7:655–6. 19. Cederlöf R. The twin method in epidemiological studies on chronic disease. Stockholm, Sweden: Institute of Hygien, Karolinska Institutet, 1966. 20. Lin DY. Cox regression analysis of multivariate failure time data: the marginal approach. Stat Med 1994;13:2233–47. 21. Wei LJ, Lin DY, Weissfeld L. Regression analysis of multivariate incomplete failure time data by modeling marginal distributions. J Am Stat Assoc 1989;84:1065–73. 22. White H. Maximum likelihood estimation of misspecified models. Econometrica 1982;50:1–25. 23. Lin DY. MULCOX2: a general computer program for the Cox regression analysis of multivariate failure time data. Comput Methods Programs Biomed 1993;40:279–93. 24. Delpizzo V, Borghesi JL. Exposure measurement errors, risk

Downloaded from aje.oxfordjournals.org by guest on October 26, 2010

mates were largely unchanged. This suggests that the potential impact of socioeconomic status should be limited. We based exposure assessment on the occupation given in the questionnaire, without taking duration of exposure into account, which was done in the US studies. On the other hand, the subjects from the older cohort that generated most of the cases belonged to generations that did not often change their occupation during their lifetime, and it is hard to judge the potential effects of this shortcoming. Controlling for twin cohort did not change the results. We also analyzed the cohorts separately. The risk estimates for the younger cohort were somewhat higher than those for the older cohort, but the confidence intervals were wider. Exposure assessment was based on a job exposure matrix and not on individual measurements of exposure, an ideal but unrealistic approach. The method should mainly lead to nondifferential misclassification and to risk estimates closer to unity, particularly for the very high exposure group (24). Misclassification due to changes of exposure within occupations over time may also contribute to bias, presumably also diluting the results. Coronary heart disease has a strong genetic component (25), which was also apparent from the present study. In addition to other known precursors with genetic components, the genetic influence on disease occurrence might be mediated by a genetic component for reduced heart rate variability. Reduced heart rate variability is a predictor for death from heart diseases (6–9). Some studies have shown higher correlations of heart rate variability between relatives as compared with nonrelatives (26–28), and a study of twins also showed a genetic component in heart rate generation and heart rate variability (29). Our analyses of synergy indicated that the combined occurrence of ELF magnetic field exposure and prevalence of AMI in the twin sibling increased the risk of AMI above additivity, particularly among monozygotic twins. The analyses were based on small numbers, but the results indicated that the effect of ELF magnetic fields is strengthened if a subject has a genetic predisposition for AMI, possibly induced by reduced heart rate variability. The question of effect modification from individual sensitivity such as genetic predisposition deserves to be further explored in studies of ELF magnetic fields and chronic diseases.

542 Håkansson et al. estimates and statistical power in case-control studies using dichotomous analysis of a continuous exposure variable. Int J Epidemiol 1995;24:851–62. 25. Marenberg ME, Risch N, Berkman LF, et al. Genetic susceptibility to death from coronary heart disease in a study of twins. N Engl J Med 1994;330:1041–6. 26. Singh J, Larson M, O’Donnell C, et al. Heritability of heart rate variability: The Framingham Heart Study. Circulation 1999;99: 2251–4. 27. Sinnreich R, Friedlander Y, Sapoznikov D, et al. Familial

aggregation of heart rate variability based on short recordings—The Kibbutzim Family Study. Hum Genet 1998;103:34– 40. 28. Sinnreich R, Friedlander Y, Luria MH, et al. Inheritance of heart rate variability: The Kibbutzim Family Study. Hum Genet 1999;105:654–61. 29. Voss A, Busjahn A, Wessel N, et al. Familial and genetic influences on heart rate variability. J Electrocardiol 1996;29(suppl): 154–60.

Downloaded from aje.oxfordjournals.org by guest on October 26, 2010

Am J Epidemiol 2003;158:534–542