Fish consumption and myocardial infarction: a second prospective biomarker study from northern Sweden14

Maria Wennberg, Ingvar A Bergdahl, Go ran Hallmans, Margareta Norberg, Thomas Lundh, Staffan Skerfving, kan Jansson Ulf Stro mberg, Bengt Vessby, and Jan-Ha
ABSTRACT Background: A benecial role of sh consumption on the risk of myocardial infarction (MI) has been reported and is mostly ascribed to n23 (omega-3) fatty acids. However, sh also contains methylmercury, which may increase the risk of MI. Objective: The objective was to determine how sh consumption and erythrocyte concentrations of mercury (Ery-Hg) and selenium (Ery-Se) are related to the risk of MI and whether n23 fatty acids (eicosapentaenoic and docosahexaenoic acids) in plasma phospholipids (P-EPA+DHA) are protective. Design: This was a case-control study nested within the northern Sweden cohort, in which data and samples were collected prospectively. The study included 431 cases with an MI after data and sample collection, including 81 sudden cardiac deaths (SCDs) and 499 matched controls. Another 69 female cases with controls from a breast cancer screening registry were included in sex-specic analyses. Results: Odds ratios for the third compared with the rst tertile were 0.65 (95% CI: 0.46, 0.91) for Ery-Hg, 0.75 (95% CI: 0.53, 1.06) for Ery-Se, and 0.78 (95% CI: 0.54, 1.11) for P-EPA+DHA. Ery-Hg and P-EPA+DHA were intercorrelated (Spearmans R = 0.34). No association was seen for reported sh consumption. Multivariate modeling did not change these associations signicantly. Sex-specic analyses showed no differences in risk associations. High concentrations of Ery-Se were associated with an increased risk of SCD. Conclusions: The biomarker results indicate a protective effect of sh consumption. No harmful effect of mercury was indicated in this low-exposed population in whom Ery-Hg and P-EPA+DHA were intercorrelated. Am J Clin Nutr 2011;93:2736. INTRODUCTION

controls in northern Sweden reported an inverse association between risk of MI and mercury concentrations in erythrocytes (EryHg) and EPA and DHA in plasma phospholipids (P-EPA+DHA) (8). It was suggested that this unexpected benecial association could be explained by Ery-Hg being a marker of sh consumption. The exposure level of MeHg is relatively low in northern Sweden (median Ery-Hg: 3.6 lg/L) (9), particularly when compared with eastern Finland [mean hair-Hg: 1.9 lg/g (6), corresponding to an Ery-Hg of 13 lg/L] (7, 1012). The multicenter study showed a somewhat higher exposure [mean nail-Hg: 0.25 lg/g (4), corresponding to an Ery-Hg of 6 lg/L] (7, 1012). In this larger, prospective, nested case-control study with 431 cases of rst MI and 499 matched controls in northern Sweden, we aimed to examine the inuence of Ery-Hg on MI as well as the effects of sh consumption, Ery-Se, and P-EPA+DHA. To study sex differences, another 69 female cases with MI and controls were identied from a breast cancer screening registry and were included in the sex-specic analyses. We also examined the relation between the study variables and the risk of sudden cardiac death (SCD).

Downloaded from ajcn.nutrition.org by guest on January 24, 2013

SUBJECTS AND METHODS

Study population and study design The study population was derived from the Northern Sweden Health and Disease Study (NSHDS), which consists of 3 sub From the Department of Medicine, Skelleftea Hospital, Skelleftea, Sweden (MW and J-HJ); the Department of Public Health and Clinical Medicine, Umea University, Umea, Sweden (MW, IAB, GH, MN, and J-HJ); the Department of Occupational and Environmental Medicine, University Hospital, Lund, Sweden (TL, SS, and US); and the Department of Public Health and Caring Science, Uppsala University, Uppsala, Sweden (BV). 2 The contents reect only the authors views; the European Union is not liable for any use that may be made of the information. 3 Supported by grants from the Vasterbotten County Council, the Foun dation of Medical Research in Skelleftea, the Swedish Research Council for Environment, Agricultural Sciences and Spatial Planning (2007-2024), the Swedish Council for Working Life and Social Research, the Medical Faculty, Lund University, and the County Councils of Southern Sweden and the European Union (Sixth Framework Programme; PHIME; FOOD-CT-2006016253). 4 Address correspondence to M Wennberg, Occupational Medicine, Umea University, 901 87 Umea, Sweden. E-mail: maria.wennberg@envmed.umu.se. Received February 22, 2010. Accepted for publication October 4, 2010. First published online November 3, 2010; doi: 10.3945/ajcn.2010.29408.
1

Fish is considered a healthy food, especially because of its content of the long-chain omega-3 (n23) fatty acids eicosapentaenoic acid (EPA, 20:5n23) and docosahexaenoic acid (DHA, 22:6n23); however, other nutrients might also be benecial, such as selenium. The benecial role of sh consumption on the risk of cardiovascular disease (CVD) has been extensively studied (13). On the other hand, sh also contains environmental contaminants, such as methylmercury (MeHg) and persistent organic pollutants. Several epidemiologic studies [eg, a multicenter study (4) and studies from eastern Finland (5, 6)] have found an association between the risk of CVD or risk factors for CVD (7) and concentrations of mercury in hair or nails. In contrast with the Finnish studies, a prospective nested casecontrol study of 78 cases of myocardial infarction (MI) and of 156

Am J Clin Nutr 2011;93:2736. Printed in USA. 2011 American Society for Nutrition
Supplemental Material can be found at: http://ajcn.nutrition.org/content/suppl/2010/12/20/ajcn.2010. 29408.DC1.html

27

28

WENNBERG ET AL

cohorts: the Vasterbotten Intervention Program (VIP) (13), the WHOs Multinational Monitoring of Trends and Determinants in Cardiovascular Disease (MONICA) Study in northern Sweden (14), and the Mammography Screening Project (MSP) (15). Both VIP and MONICA are health examination programs for CVD and diabetes. In VIP, the mean participation rate was 59%; until 1995, some health centers discontinued VIP in periods, and thus the participation rate varied between health centers. In the northern Sweden MONICA Study, the mean participation rate was 77%. Data collection in VIP and MONICA regarding CVD risk markers were very similar. The designs of VIP and MONICA were described in detail previously (13, 14). To increase the number of female cases, participants in the MSP were included in the sex-specic analyses. The participation rate in the MSP was 85% in the screening phase, and 57% of the participants donated blood samples. By 31 December 1999, 73,000 unique subjects had been screened in these 3 subcohorts in the NSHDS. Participants in VIP or MONICA were asked to complete a questionnaire concerning social background, medical history, and various lifestyle factors, including diet and smoking habits. A medical examination of the participants was conducted. Participants consented to donate blood samples, which were stored at 280C in a biobank for future research. Health screening information was stored in a database. The participants in the MSP did not ll out a questionnaire, but blood samples for future research were collected and information on smoking habits was obtained. This study consists of subjects participating in any of the health examinations mentioned above from years 1987 to 1999. Because very limited multivariate adjustment was possible for the participants in the MSP, they were not included in the main analyses. They were included in additional sex-specic analyses, in which fewer variables were considered. In the main analyses, 91% of the

cases were participants from VIP and 9% were from the MONICA Study. A prospective, nested, case-control study design was used (Figure 1). Consecutive cases of MI occurring from 1 October 1994 to 31 December 1999 were identied through the northern Sweden MONICA incidence registry (16). Of these cases, SCD was dened as survival time from onset of symptoms to death of ,24 h. Controls (one control for men and 2 controls for women) were matched to the cases for sex, age (62 y), date of health survey (6 4 months), type of health survey, and geographic region (municipality). Because of a smaller number of female participants, 2 controls were matched to female cases to increase statistical power. Cases and controls were excluded if previous MI, stroke, or malignant disease could not be excluded according to the questionnaire or case records. In addition, subjects without biobanked erythrocyte samples were excluded, leaving 431 cases and 499 controls (at least one control per case) from VIP and MONICA. Of these cases, 81 were dened as SCD. Another 69 cases and 126 controls from the MSP were included in the sex-specic analyses. The study was approved by the Research Ethics Committee of Umea University. All participants gave informed consent. Baseline variables Smoking habits were classied into daily smoking or nonsmoking (including previous smokers and occasional smokers). Body mass index (BMI) was calculated as weight (kg)/ height squared (m2). Hypertension was dened as a systolic blood pressure !140 mm Hg, a diastolic blood pressure !90 mm Hg, or the reported use of antihypertensive medication during the past 14 d. Diabetes was dened as self-reported

Downloaded from ajcn.nutrition.org by guest on January 24, 2013

FIGURE 1. Selection and exclusion algorithm for study subjects from the Vasterbotten Intervention Program (VIP) and the Multinational Monitoring of Trends and Determinants in Cardiovascular Disease (MONICA) Study from northern Sweden. Data were sampled between 1987 and 1999. MI, myocardial infarction; Ery-Hg, erythrocyte mercury; Ery-Se, erythrocyte selenium; P-EPA+DHA, sum of proportions of eicosapentaenoic acid and docosahexaenoic acid in plasma phospholipids. *One referent for men; 2 referents for women.

FISH CONSUMPTION AND MYOCARDIAL INFARCTION

29

disease from the questionnaire or a fasting glucose !7 mmol/L and/or a 2-h postload plasma glucose !11.0 mmol/L (!12.2 mmol/L in the VIP, as capillary plasma was drawn). Total cholesterol was measured in serum, sampled after .4 h of fasting. Educational level was categorized into fullled academic education (university or college) or not.

Consumption of fruit and vegetables Because there were different versions of the FFQ, consumption of fruit and vegetables was only estimated as a dichotomized variable. First, all questions on fruit and vegetables were converted to intakes per day and then computed. With the same response alternatives as described for sh consumption, the possible categorization was intake of fruit and vegetables 1 time/d or .1 time/d. If data were missing on one or several questions, data were coded as insufcient if the available information did not allow categorization of the subject to the .1 time/d per group. Physical activity level The different versions of the questionnaire used over time complicated the estimation of physical activity level. It was possible to identify subjects reporting low levels of physical activity. Subjects who on questions regarding leisure-time physical activity chose the answer never, hardly ever, or mostly inactive were included in this category. Blood sampling Venous blood samples were drawn without stasis into evacuated tubes after a minimum of 4 h of fasting in the VIP and the MONICA Study. In the MSP, blood samples were collected throughout the day, not necessarily fasting. Erythrocytes, buffy coat, and plasma were separated by centrifugation at 1500 g for 15 min, and aliquots were stored at 280 until used (19). Storage time (time between sampling and analyses) ranged from 8 to 20 y. Measurement of apolipoprotein B and apolipoprotein A-I Apolipoprotein (apo) A-I and apo B were analyzed in 2007 by using blood samples stored in the biobank since screening; apo A-I and B were measured by immunoturbidimetry with reagents from Dako (Glostrup, Denmark) and a calibrator (0947) on a Hitachi 911 multianalyzer (Roche Diagnostics GmbH, Mannheim, Germany). The ratio of apo B to apo A-I (apo B/apo A-I) was calculated and used instead of serum cholesterol in the statistical analyses. Measurement of fatty acids Fatty acids were measured in 2007 by gas-liquid chromatography after separation of lipids by thin-layer chromatography and transmethylation (20) and were expressed as a percentage of all fatty acids in phospholipids (Unit for Clinical Nutrition Research, Department of Public Health and Caring Sciences, Uppsala University, Uppsala, Sweden). The method imprecision (calculated as the CV for duplicate preparations and measurements) has been reported to be ,15.5% (21). Measurement of mercury

Dietary data Information on average dietary intake over the past year was acquired from a food-frequency questionnaire (FFQ). There have been some changes over time in the FFQ. The version currently used in the MONICA study (84-items) has been validated with ten 24-h diet records (17) and with determination of fatty acid content in erythrocyte membranes as biomarkers (18). In the VIP, a somewhat shorter version (6466 items) of the FFQ has been used since 1996. Each of the FFQs used 9 predened responses and were answered by most of the participants (n = 705). Another 217 participants completed older versions of the FFQ with 6 predened responses. The participants from the MSP (n = 195) did not ll out any FFQ, whereas another 8 subjects also lacked dietary data.

Downloaded from ajcn.nutrition.org by guest on January 24, 2013

Fish consumption The questions regarding consumption of fatty and lean sh were identical in the different FFQ versions. The questions asked were as follows: How often do you eat lean sh (eg, perch and cod)? and How often do you eat fatty sh (eg, herring, lavaret, and salmon)?. In addition, the question How often do you eat salty sh (salt herring)? was asked in most of the FFQ versions. Because herring is a fatty sh, we assumed that this was reected in the respondents replies regarding fatty sh, but for those who reported a more frequent consumption of salty sh than of fatty sh (n = 54), the estimation of salty sh consumption was interpreted as fatty. Because it is uncertain how people report fatty and salty sh consumption, alternative analyses were carried out in instances in which higher consumption of salty sh was reported. The frequencies in the questionnaires with 9 predened responses were quantied to meals per week: never = 0, a few times per year = 0.05, 13 times/mo = 0.50, once per week = 1.00, 23 times/wk = 2.50, 46 times/wk = 5.00, once per day = 7.00, 23 times/d = 17.5, and !4 times/d = 28.0. In the versions of the FFQ with 6 alternatives, the frequencies were quantied to meals per week: never = 0, ,1 time/wk = 0.5, 12 times/wk = 1.5, 35 times/wk = 4.0, 67 times/wk = 6.5, and .1 time/d = 10.5. Consumption of lean and fatty sh was computed to total sh meals per week. Consumption of total sh was also categorized in 4 groups of sh consumption: ,1 time/mo (n = 32 cases and 28 controls), 1 time/mo to ,1 time/wk (n = 74 cases and 87 controls), 12 times/wk (n = 260 cases and 319 controls), and .2 times/wk (n = 26 cases and 40 controls).

Alcohol consumption Questions on consumption of wine, spirits, and strong beer were included in all versions of the FFQ and were encoded to intakes per week of each beverage.

Mercury was measured in duplicate in acid-digested erythrocytes by cold-vapor atomic uorescence spectrometry (22). The limit of detection (3 times the SD of the blank) was 0.14 lg/L, and the CV was 5.3%. The analytic accuracy was checked by using Seronorm Trace Elements Whole Blood [SERO AS,

30

WENNBERG ET AL

Billingstad, Norway; lots MR4206 and 404109y; obtained: 2.0 6 0.16 (mean 6 SD) and 13.8 6 0.60 lg/L, respectively; recommended: 2.02.4 and 13.516.4 lg/L, respectively] and human blood quality control from Centre de Toxicologie du Quebec, Canada (lot M-0401; obtained: 1.9 6 0.15 lg/L; certied: 2.0 lg/L). Measurement of selenium The concentration of selenium in the erythrocyte samples was measured in duplicate by inductively coupled plasma mass spectrometry (ICP-MS; Thermo 7; Thermo Elemental, Winsford, United Kingdom) (23). The detection limit was 0.55 lg/L. The method imprecision was 3.2%. The analytic accuracy was checked against reference material. For Seronorm Trace Elements Whole Blood (lot 0503109; SERO AS, Billingstad, Norway), the results (n = 126) obtained were 125 6 6.8 (mean 6 SD) compared with the Recommended concentration of 113133 lg/L. For human blood quality control with samples obtained from Centre de Toxicologie du Quebec, Canada (lot ICP-01B-03), the obtained values were 176 6 12 (n = 78) compared with the certied value of 191 6 60 lg/L. Statistics The crude effect of each potential risk factor was examined by univariate conditional logistic regression (ie, including a single potential risk variable) based on the matched case-control sets (24, 25) in the VIP and the MONICA Study. To proceed with multivariate analyses, we analyzed Spearmans correlations (26) between the candidate risk factors in controls from the VIP and the MONICA Study; these correlations provide a descriptive support for understanding potential confounding effects in the multivariate settings. Fairly extensive multivariate models were possible for the participants in the VIP and the MONICA Study. Because the crude effect estimates of traditional risk factors (except BMI) and other study variables were in the same direction for both sexes and the effect modication by sex was not evident (except for BMI, but BMI showed only weak correlations with the exposure variables and, consequently, no strong confounding effects in the multivariate analyses), we decided to analyze combined data from men and women. Multivariate models were rst built for each exposure variable (ie, sh consumption, Ery-Hg, P-EPA +DHA, and Ery-Se). The data-driven adjustment covariates were rst sorted out. These covariates changed the point estimate, reecting a crude exposure effect of !10% (27), and were as follows: 1) sh intake, apo B/apo A-I, smoking, diabetes, education, physical activity level, consumption of strong beer, and consumption of fruit and vegetables; 2) Ery-Hg, systolic blood pressure, smoking, education, consumption of strong beer, and consumption of fruit and vegetables; 3) P-EPA+DHA, apoB/ apoA1, smoking and consumption of fruit and vegetables; and 4) Ery-Se, apo B/apo A-I, smoking, education, consumption of strong beer, and consumption of wine. To facilitate directly comparable models for the exposure variables, each data-driven model included all adjustment covariates sorted out (ie, diabetes, smoking, systolic blood pressure, apo B/apo A-I, education, physical activity level, and consumption of strong beer, wine, and fruit and vegetables (referred to as model 1). Extended

multivariate models were planned considering the Framingham Study, in which cholesterol concentration, systolic blood pressure, smoking, and diabetes were established as important (28). However, instead of using total cholesterol and HDL cholesterol as in the Framingham Heart Study, we used apo B/apo A-I, which has proved to be a better risk marker for MI (29). However, because all of the Framingham covariates were included in model 1, no extended multivariate model was built. In the main analyses, the exposure biomarkers were categorized as low, medium, or high by the tertile limits based on all controls from the VIP and the MONICA Study. We report the P value for trend (Wald test) of each trichotomized biomarker variable. In addition, we considered ner categorizations (quintiles) of the biomarker variables and continuous (not categorized) Ery-Hg, P-EPA+DHA, and Ery-Se. Fish consumption was analyzed in 4 groups of consumption and as a continuous variable. Also, we report the results from a multivariate model including Ery-Hg and P-EPA+DHA together, along with all adjustment covariates (referred to as model 2). We also performed statistical tests of no (multiplicative) interaction between Ery-Hg and PEPA+DHA. We also analyzed the effects on SCD with the same multivariate models as for all MI and sex-specic effects using the covariates smoking and apo B/apo A-Ithe covariates available for the females in the MSP. The statistical computations were carried out by using SPSS for Windows (version 15 and 17; SPSS Inc, Chicago, IL).
RESULTS

Downloaded from ajcn.nutrition.org by guest on January 24, 2013

Baseline characteristics concerning background variables and study variables for 431 cases of MI with 499 matched controls in the VIP and the MONICA Study and 69 cases with 126 controls in the MSP are presented in Table 1. For the whole study group, the mean consumption (and range) of sh was 1.26 meals/wk (08 meals/wk). Median (and range) concentrations of mercury and selenium in erythrocytes were 3.54 lg/L (0.0187 lg/L) and 126 lg/L (72.5713 lg/L), respectively, and the median relative level of P-EPA+DHA was 5.84% (2.7814.5%). Average time between baseline (Table 1) and the event was 3 y and 11 mo (range: 44036 d), and the mean age at MI was 58.7 y (34.1 77.1 y). Established risk factors for MI were more common in cases (Table 1). Storage time for blood samples ranged from 8 to 20 y. No signicant difference in median P-EPA+DHA values was observed in groups with different blood sample storage times (8 11 y: 5.86%; 1720 y: 5.75%). Correlations between sh consumption variables were examined. Signicantly positive, although not strong (Spearmans R , 0.20), correlations were observed between reported sh consumption and Ery-Hg, P-EPA+DHA, and Ery-Se (Table 2). The strongest correlation among sh consumption-related variables was seen between Ery-Hg and P-EPA+DHA (Table 2). In univariate analyses for cases and controls from VIP and MONICA, academic education, physical activity, and consumption of fruit and vegetables were associated with a decreased risk of MI. Smoking, consumption of strong beer, diabetes, BMI, systolic blood pressure and apoB/apoA1 were associated with an increased risk of MI (Table 3). No notable protective effect of self-reported sh consumption was found. Ery-Hg was clearly

TABLE 1 Baseline characteristics of 500 cases from the Northern Sweden Health and Disease Study with myocardial infarction during follow-up until 31 December 1999 and for 625 matched controls1 The Vasterbotten Intervention Program and the MONICA Study from northern Sweden Controls n Values Tertiles2 n Values n Values n Values Tertiles2 Tertiles2 Cases Controls Cases Tertiles2 The Mammography Screening Project

350 149 126 1996 (19951999) 1995, 1996 69 1.00 (05.50) 1.05 (0.104.50) 3.95 (0.1581.4) 3.68 (0.1916.7) 126 3.16 (0.2026.8) 123 (75.5250) 132 (81.0400) 126 135 (99.0713) 1.37 (0.455.44) 1.37 (0.684.52) 122 4.56 (2.228.41) 4.70 (2.128.55) 122 5.91 (2.9413.5) 6.06 (3.1112.3) 5.28, 6.66 5.62, 6.70 306 68 5.72 (3.4711.5) 5.93 (4.1511.2) 5.28, 6.34 5.31, 6.60 4.11, 5.08 4.28, 5.24 306 68 4.41 (2.318.26) 4.68 (3.237.54) 4.06, 4.83 4.27, 5.07 1.19, 1.60 1.19, 1.53 306 68 1.38 (0.503.75) 1.25 (0.413.62) 1.24, 1.60 1.04, 1.48 116, 132 122, 140 350 81 120 (72.5302) 126 (86.0207) 112, 130 119, 136 2.89, 5.21 2.95, 4.72 350 81 3.47 (0.0145.9) 3.04 (0.4323.4) 2.64, 4.75 2.38, 4.18 1.00, 1.50 1.00, 1.50 317 75 1.00 (08.00) 1.05 (04.50) 1.00, 1.05 1.00, 1.50

1993 (19871999) 1993 (19881998)

1991, 1994 1991, 1994

350 81

1993 (19871999) 1992 (19881998)

1991, 1994 1991, 1994

1996 (19951999)

1995, 1996

331 143

350 149

2.48, 4.60

69

2.57 (0.1886.9)

1.90, 3.43

350 147

128, 146

69

135 (98.5198)

129, 146

308 126

1.40 (0.685.54)

1.23, 1.84

69

1.33 (0.494.81)

1.14, 1.64

308 126

4.37 (1.778.29)

4.00, 4.91

69

4.14 (2.199.72)

3.60, 4.90

308 126

122

5.77 (2.7813.3)

5.32, 6.69

69

5.43 (3.1914.5)

4.85, 6.21

FISH CONSUMPTION AND MYOCARDIAL INFARCTION

338 145 14.1 22.6 64.8 57.2 82.3 89.1 0.05 (01.00) 0.00 (01.00) 0.00, 0.05 0.00, 0.02 319 75 313 70 310 69 326 77 7.98 11.7 57.7 46.4 75.2 88.0 0.05 (02.50) 0.00 (01.00)

21.0 21.4

329 76

38.9 36.8

126

15.9

69

59.4

333 146

324 138

Study variables Year of sampling3 Men Women Fish consumption (meals/wk)4 Men Women Erythrocyte Hg (lg/L)4 Men Women Erythrocyte Se (lg/L)4 Men Women P-EPA (%)4 Men Women P-DHA (%)4 Men Women P-EPA+DHA (%)4 Men Women Background variables Smokers (%) Men Women Academic education (%) Men Women Physically active (%) Men Women Fruit and vegetables .1 time/d (%) Men Women Strong beer (times/wk)4 Men Women 0.05, 0.05 0.00, 0.05

333 147

324 135

(Continued)

31

Downloaded from ajcn.nutrition.org by guest on January 24, 2013

32

TABLE 1 (Continued ) The Vasterbotten Intervention Program and the MONICA Study from northern Sweden Controls n 0.05 (02.50) 0.05 (02.50) 0.05 (02.50) 0.05 (07.00) 53.2 (3077) 58.4 (3073) 126 61.4 (5070) 4.05 6.80 25.4 (19.042.1) 25.6 (18.249.4) 42.6 50.3 84.7 (62.7111) 84.2 (66.8109) 134 (100208) 135 (95.5216) 6.04 (2.5910.4) 6.47 (3.7010.6) 0.84 (0.342.07) 0.76 (0.311.42) 0.73, 0.91 0.66, 0.86 350 80 5.56, 6.48 5.80, 7.03 334 77 6.55 (3.3910.0) 6.83 (3.6012.9) 0.98 (0.212.05) 0.89 (0.441.33) 126, 140 126, 140 336 77 139 (101194) 145 (103209) 131, 145 131, 149 5.98, 7.10 6.30, 7.40 0.87, 1.10 0.77, 0.98 81.3, 87.9 82.0, 87.1 336 77 87.7 (65.0154) 88.9 (66.8114) 83.0, 91.1 83.0, 91.1 336 77 61.3 72.7 24.0, 27.1 23.9, 27.0 338 78 26.7 (18.757.4) 26.0 (18.746.5) 25.4, 28.1 24.5, 27.7 338 78 9.76 8.97 50.1, 59.9 57.4, 60.6 350 81 53.3 (30-76) 58.5 (3074) 50.0, 59.9 57.2, 60.3 0.05, 0.50 0.00, 0.05 320 72 0.05 (05.00) 0.05 (01.00) 0.05, 0.50 0.00, 0.05 0.03, 0.05 0.05, 0.50 316 74 0.05 (07.00) 0.05 (02.50) 0.00, 0.05 0.00, 0.05 Values Tertiles2 n Values n Values n Values Tertiles2 Tertiles2 Cases Controls Cases Tertiles2 The Mammography Screening Project

325 141

330 142

350 149

58.1, 66.6

69

61.5 (5070)

57.8, 66.6

346 147

338 144

WENNBERG ET AL

336 143

334 143

334 143

Wine (times/wk)4 Men Women Spirits (times/wk)4 Men Women Age (y)3 Men Women Diabetes (%) Men Women BMI (kg/m2)4 Men Women Hypertension (%)5 Men Women Diastolic BP (mm Hg)3 Men Women Systolic BP (mm Hg)3 Men Women Serum cholesterol (mmol/L)3 Men Women apo B/apo A-I3 Men Women 126 0.70 (0.141.41) 0.63, 0.82 69

333 142

350 147

0.86 (0.361.22)

0.73, 0.92

MONICA, Multinational Monitoring of Trends and Determinants in Cardiovascular Disease; EPA, eicosapentaenoic acid; DHA, docosahexaenoic acid; P-EPA+DHA, sum of proportions of EPA and DHA in plasma phospholipids; BP, blood pressure; apo, apolipoprotein. 2 Values correspond to the 33rd and 67th percentiles. 3 Values are medians; ranges in parentheses. 4 Values are means; ranges in parentheses. 5 Hypertension dened as systolic BP !140 mm Hg, diastolic BP !90 mm Hg, or reported use of antihypertensive medication during the past 14 d.

Downloaded from ajcn.nutrition.org by guest on January 24, 2013

FISH CONSUMPTION AND MYOCARDIAL INFARCTION

33

TABLE 2 Covariation between variables: Spearmans rank correlations (partial analyses adjusted for age and sex) between risk factors for myocardial infarction and study variables in 499 controls from the Northern Sweden Health and Disease Study1 n Erythrocyte Hg P-EPA+DHA Erythrocyte Se Fish Lean sh Fat sh Smoking apo B/apo A-I BMI Diabetes SBP Physical activity Fruit and vegetables Wine Beer Academic education 499 434 497 474 479 480 483 497 482 493 477 462 480 466 459 479 Erythrocyte Hg 0.342 0.152 0.182 0.103 0162 20.004 20.006 20.005 20.021 20.0993 20.031 20.053 0.011 0.027 20.035 P-EPA+DHA Erythrocyte Se Fish Lean sh Fatty sh

0.202 0.192 0.055 0.252 20.042 0.103 0.066 0.037 0.004 20.016 0.029 0.134 0.080 0.174

0.0953 0.048 0.086 20.052 0.0963 0.057 0.002 20.039 0.058 0.044 0.134 0.076 0.134

0.792 0.662 0.036 20.005 20.023 20.018 20.124 0.164 0.123 20.011 0.028 0.043

0.063 0.090 20.033 20.086 20.019 20.0963 0.144 0.152 20.026 20.025 0.021

20.029 0.020 0.0923 20.017 20.045 0.033 0.003 20.008 0.058 0.031

1 P-EPA+DHA, sum of proportions of eicosapentaenoic acid and docosahexaenoic acid in plasma phospholipids; apo, apolipoprotein; SBP, systolic blood pressure. 24 Signicant correlation: 2P 0.001, 3P 0.05, 4P 0.01.

Downloaded from ajcn.nutrition.org by guest on January 24, 2013

associated with a decreased risk of MI. Point estimates for PEPA+DHA and Ery-Se were also ,1, but they were not statistically signicant (Table 4). Ery-Hg remained related to a decreased risk of MI in the multivariate models (P for trend = 0.025 in model 1) (Table 4). When data were analyzed by quintile (Q), the crude odds ratios (ORs) with 95% CIs for Ery-Hg were as follows, with Q1 as comparison group; Q2 = 0.86 (0.56, 1.32), Q3 = 0.61 (0.39, 0.94), Q4 = 0.70 (0.45, 1.08), and Q5 = 0.58 (0.37, 0.91). After adjustment according to model 1 in Table 4, there was still a signicant association for Ery-Hg in Q5 (OR: 0.48; 95% CI: 0.23, 0.99). An investigation of how inclusion of the competing shconsumption biomarkers affects the relation to MI showed that the associations with MI for Ery-Hg were only moderately attenuated (Table 4, model 2). We found no rm evidence of interaction effects between these 2 biomarkers (P = 0.059). An attempt to further describe the potential interaction through stratied analyses did not clarify this issue. An attempt to further describe the potential interaction through stratied analyses did not clarify this issue (see Table S1 under Supplemental data in the online issue). Women from the MSP were included in sex-specic analyses. A total of 500 cases were studied. Crude analyses were made separately for women in the MSP and combined for women in the VIP and MONICA, and the results were found to be similar. Therefore, groups were combined in multivariate analyses, which compared the effects among men and women. Associations were found to be similar for men and women (see Table S2 under Supplemental data in the online issue). With SCD as an endpoint (81 cases), protective relations appeared for Ery-Hg and P-EPA+DHA, which were similar to those of MI (see Table S3 under Supplemental data in the online issue), but only in the crude models. In contrast, a signicant association was found with Ery-Se in the adjusted model, which indicated an increased risk at high levels.

DISCUSSION

Ery-Hg was associated with a decreased risk of MI, but selfreported sh consumption was not. Both P-EPA+DHA and ErySe showed the same tendencies as Ery-Hg, but their associations with risk of MI were not statistically signicant. A probable explanation is that all biomarkers in some way reect sh consumption. We do not know of any mechanism that would make mercury protective against MI. On the contrary, some mechanisms of mercury might increase the risk of MI (30). It may seem contradictory that an association appears for a marker of sh consumption, but not for reported sh consumption. However, dietary surveys are affected by subjective factors causing errors. In the present study, the variation in reported sh consumption was low. The predened frequency alternatives may have contributed to this lack of contrast, which also have another consequence: we could not test for benets at very low concentrations of EPA+DHA (3). With few subjects reporting no consumption of sh (n = 7), any possible association would be difcult to demonstrate. Another potential limitation was the differences between FFQs between the cohorts and over time. This, however, is of minor concern because the questions on sh consumption were the same throughout (except for the MSP, for which no questionnaire was used).The main limitation with the FFQ was the lack of contrast mentioned above. In the present study, we therefore used biomarkers to get a better contrast for the exposure variables. The higher correlation between Ery-Hg and P-EPA+DHA than between sh consumption and these biomarkers supports the presumption that biomarkers reect real sh consumption better than does self-reporting. It is probable that Ery-Hg reects the general sh consumption better than does P-EPA+DHA. The latter has been suggested to reect intake from only the last days, whereas it has been suggested that erythrocytes reect the past

34

WENNBERG ET AL
TABLE 3 Univariate conditional logistic regression analyses concerning the inuence of potential background risk factors on risk of myocardial infarction in 431 cases and 499 controls from the Northern Sweden Health and Disease Study1 Risk factor and unit or category Background variables Smoking No Yes Academic education No Yes Physical activity No Yes Fruit and vegetables 1 time/d .1 time/d Strong beer, times/wk Wine, times/wk Diabetes No Yes BMI ,25 kg/m2 2530 kg/m2 .30 kg/m2 Systolic BP, mm Hg apo B/apo A-I 20.69 20.88 .0.88
1 2 3 4

Cases/controls 395/456

OR

95% CI

P for sex2

1.0 2.50 391/456 1.0 0.48 373/430 1.0 0.72 349/441 1.0 0.67 360/411 368/426 412/477 2.16 0.71 1.0 2.40 1.06 1.0 1.51 2.01 1.02 1.0 1.85 3.67

0.84 1.81, 3.46 ,0.001 0.94 0.30, 0.76 0.001 0.64 0.53, 0.97 0.03 0.34 0.46, 0.97 1.21, 3.86 0.49, 1.03 0.03 0.009 0.06 0.46 0.40

Downloaded from ajcn.nutrition.org by guest on January 24, 2013

0.36 1.35, 4.26 1.02, 1.11 1.11, 2.06 1.30, 3.10 1.01, 1.04 0.003 0.001 0.0014 0.023 0.66

407/468

399/359 430/497

,0.001 ,0.0014

0.16 0.96

1.23, 2.78 2.48, 5.42

OR, odds ratio; BP, blood pressure; apo, apolipoprotein. P for effect modier for sex. We found no evidence that BMI affected the risk of myocardial infarction in women. P for trend.

month (31). It may thus be that EPA+DHA in erythrocyte membranes reects a longer time span than does EPA+DHA in plasma phospholipids (31); however, this has been questioned, suggesting a similar time course (12 wk, ie, longer than a few days) for EPA+DHA in both matrices (32). Furthermore, levels of EPA and DHA are expressed as a percentage of fatty acids analyzed; thus, it is affected by the amount of other fatty acids in plasma. Ery-Hg, on the other hand, is a concentration. Even though total mercury in erythrocytes was measured in our study, we believe that it reects exposure to MeHg reliably. Twentythree samples from the northern Sweden MONICA Study were previously analyzed for total and inorganic mercury (median inorganic mercury: 0.5 lg/L; ranging from a detection limit up to 7.5 lg/L). It was found that inorganic mercury made up only 8% (range: 257%) of the total mercury (33). Thus, as our data suggest, Ery-Hg may be a better marker of sh consumption than P-EPA+DHA, even though levels of MeHg vary within and between sh species. The weakest correlation was seen between sh consumption and selenium concentrations. This is not surprising because many other foods are sources of selenium. Swedes have a higher intake of selenium from meats and milk products than from sh (34), even though a portion of sh contains more selenium than does an equal portion of meat. It has been suggested that Ery-Hg may be a marker of a healthy lifestyle (35). In this study, no correlation was seen between Ery-

Hg and the consumption of fruit and vegetables, physical activity, or education. This suggests that the association between Ery-Hg and decreased MI risk is due to Ery-Hg being a marker of sh consumption rather than to a healthy lifestyle. Selection bias was assessed by comparison studies of participants and nonparticipants in VIP (36) and in the MONICA Study (37). Only small differences were found regarding social and health factors, which indicated that selection bias was small. A limitation of this study was that only single measurements were performed, which increased the risk of misclassication. Also, the information for some confounders from the questionnaire was coarse, especially for consumption of fruit and vegetables and level of physical activity. However, we saw no relevant change in point estimates after adjustment for these variables, which excluded this as a source of statistical variation. Individuals with missing data on one or several adjustment variables, and therefore excluded from the multivariate analyses, may have differed from those with complete data. Again, reasonable consistency of point estimates after adjustment does not support this as a cause of statistical variation. Another limitation was the limited range of Ery-Hg concentrations, which means that an effect of mercury on MI risk may have been obscured by the protective effect of sh. It would be of interest to study the interaction between Ery-Hg and P-EPA +DHA. We observed only a tendency (P = 0.059) for an interaction,

FISH CONSUMPTION AND MYOCARDIAL INFARCTION

35

TABLE 4 Crude and multivariate conditional logistic regression analyses concerning the inuence of study variables on risk of myocardial infarction in 431 strata from the Northern Sweden Health and Disease Study1 Crude Unit or category Fish consumption (meals/wk) n ,1 meal/mo 1 meal/mo to ,1 meal/wk 12 meals/wk .2 meals/wk Ery-Hg (lg/L) n 22.90 lg/L 24.98 lg/L .4.98 lg/L P-EPA+DHA (%) n 25.39% 26.67% .6.67% Ery-Se (lg/L) n 2117.0 lg/L 2134.5 lg/L .134.5 lg/L
1

Model 12 P for trend OR (95% CI) 1.08 (0.80, 263 1.0 0.88 (0.40, 1.09 (0.55, 1.21 (0.43, 0.94 (0.88, 274 1.0 0.78 (0.46, 0.55 (0.32, 0.95 (0.82, 254 1.0 1.34 (0.82, 0.80 (0.46, 0.99 (0.98, 274 1.0 0.73 (0.43, 1.00 (0.60, 1.44) 0.517 1.94) 2.15) 3.33) 1.00) 0.025 1.32) 0.93) 1.10) 0.457 2.21) 1.38) 1.01) 0.993 1.22) 1.69) P for trend

Model 23 OR (95% CI) P for trend

OR (95% CI) 0.98 (0.82, 374 1.0 0.77 (0.41, 0.77 (0.45, 0.65 (0.31, 0.97 (0.93, 431 1.0 0.61 (0.43, 0.65 (0.46, 0.92 (0.83, 374 1.0 0.97 (0.70, 0.78 (0.54, 0.99 (0.98, 431 1.0 0.71 (0.50, 0.75 (0.53, 1.18)

0.323 1.43) 1.32) 1.35) 1.00) 0.008 0.85) 0.91) 1.02) 0.166 1.36) 1.11) 1.00) 0.091 0.99) 1.06)

0.94 (0.87, 254 1.0 0.83 (0.47, 0.61 (0.33, 1.02 (0.86, 254 1.0 1.48 (0.88, 0.96 (0.53,

1.01) 0.094 1.46) 1.09) 1.21) 0.978 2.46) 1.74)

Downloaded from ajcn.nutrition.org by guest on January 24, 2013

OR, odds ratio; n, cases with at least one control in the analyses; Ery-Hg, erythrocyte mercury; P-EPA+DHA, sum of proportions of eicosapentaenoic acid and docosahexaenoic acid in plasma phospholipids; Ery-Se, erythrocyte selenium. 2 Adjusted for variables affecting the point estimate !10% for any of the study variables = adjusted for apolipoprotein B/apolipoprotein A-I, smoking, systolic blood pressure, diabetes, educational level, consumption of fruit and vegetables, consumption of wine, consumption of strong beer, and level of physical activity. 3 Adjusted as for model 1 and the other sh-related variable (either Ery-Hg or P-EPA+DHA).

and our data were not heterogeneous enough to evaluate this further. The storage time of blood samples must be considered, especially for fatty acids because they are sensitive to oxidation. We observed no signicant time-dependent differences when levels of fatty acids in groups with different durations of storage were compared. Therefore, we do not believe that storage time inuenced the results. Only limited adjustment (for apo B/apo A-I and smoking) was possible in sex-specic analyses (see Table S2 under Supplemental data in the online issue), because the participants in the MSP did not ll out the same questionnaires. Associations between markers of sh consumption and risk of MI were similar in men and women in these analyses. In a population-based cohort in eastern Finland, levels of mercury in hair were associated with increased risk of cardiovascular outcomes (6). Also, Guallar et al (4) found an increased risk of MI in a multicenter study using nail-Hg. In contrast, the study by Yoshizawa et al (38) from the United States (38) showed no association between toenail mercury and risk of coronary heart disease. However, when dentists were excluded, a nonsignicant association between toenail mercury and elevated risk of coronary heart disease was observed. In a study from the Faroe Islands, mercury exposure was associated with increased blood pressure and carotid intima-media thickness (7). The general population in northern Sweden has low mercury concentrations as compared with populations in whom an increased risk of cardiovascular morbidity or risk factors for CVD has been

observed with higher mercury concentrations (4, 6, 7). The differences in exposure level might explain these seemingly conicting results. The increase in risk of SCD at higher concentrations of Ery-Se is surprising, but may have been a chance nding because of the small number of cases (see Table S3 under Supplemental data in the online issue). No signicant association was found in the crude model, but was found in the adjusted one. Previous studies of selenium and risk of CVD have been inconclusive, but suggest a weak protective effect (39). However, Bleys et al (40) found indications for a decrease in cardiovascular mortality with increasing serum selenium concentrations up to 120 lg/L and increased mortality at higher concentrations. We found no reports on selenium and SCD specically. However, our data suggest an association that ought to be followed up in larger studies. In conclusion, overall, the biomarker results indicate a protective effect of sh consumption, even though data on selfreported sh consumption show no clear protective association. No harmful effect of mercury was indicated in this low-exposed population, in whom mercury and EPA+DHA were intercorrelated. Any harmful effect of mercury must therefore be overridden by protective nutrients from sh.
We acknowledge the Northern Sweden Diet database team and the funds supporting it, including the Swedish Research Council and the Vasterbotten County Council. The authors responsibilities were as followsMW: contributed to data preparation, designed and carried out the data analyses, and participated in the data interpretation and drafting and nalizing of the manuscript; IAB:

36

WENNBERG ET AL
ments in the Northern Sweden Health and Disease cohort. Public Health Nutr 2002;5:48796. Wennberg M, Vessby B, Johansson I. Evaluation of relative intake of fatty acids according to the Northern Sweden FFQ with fatty acid levels in erythrocyte membranes as biomarkers. Public Health Nutr 2009;12: 147784. Van Guelpen B, Hultdin J, Johansson I, et al. Low folate levels may protect against colorectal cancer. Gut 2006;55:14616. Boberg M, Croon LB, Gustafsson IB, Vessby B. Platelet fatty acid composition in relation to fatty acid composition in plasma and to serum lipoprotein lipids in healthy subjects with special reference to the linoleic acid pathway. Clin Sci (Lond) 1985;68:5817. Smedman AE, Gustafsson IB, Berglund LG, Vessby BO. Pentadecanoic acid in serum as a marker for intake of milk fat: relations between intake of milk fat and metabolic risk factors. Am J Clin Nutr 1999;69:229. Sandborgh-Englund G, Elinder CG, Langworth S, Schutz A, Ekstrand J. Mercury in biological uids after amalgam removal. J Dent Res 1998; 77:61524. Barany E, Bergdahl IA, Schutz A, Skerfving S, Oskarsson A. Inductively coupled plasma mass spectrometry for direct multi-element analysis of diluted human blood and serum. J Anal Atom Spectrom 1997;12:10059. Hosmer D. Applied logistic regression. New York, NY: Wiley, 1989. Clayton D. Statistical models in epidemiology. Oxford, United Kingdom: Oxford University, 1993. Altman D. Practical statistics for medical research. London, United Kingdom: Chapman & Hall, 1991. Greenland S. Modeling and variable selection in epidemiologic analysis. Am J Public Health 1989;79:3409. DAgostino RB Sr, Vasan RS, Pencina MJ, et al. General cardiovascular risk prole for use in primary care: the Framingham Heart Study. Circulation 2008;117:74353. McQueen MJ, Hawken S, Wang X, et al. Lipids, lipoproteins, and apolipoproteins as risk markers of myocardial infarction in 52 countries (the INTERHEART study): a case-control study. Lancet 2008;372:224 33. Mozaffarian D. Fish, mercury, selenium and cardiovascular risk: current evidence and unanswered questions. Int J Environ Res Public Health 2009;6:1894916. Arab L. Biomarkers of fat and fatty acid intake. J Nutr 2003;133(suppl 3):925S32S. Skeaff CM, Hodson L, McKenzie JE. Dietary-induced changes in fatty acid composition of human plasma, platelet, and erythrocyte lipids follow a similar time course. J Nutr 2006;136:5659. Wennberg M, Lundh T, Bergdahl IA, et al. Time trends in burdens of cadmium, lead, and mercury in the population of northern Sweden. Environ Res 2006;100:3308. The National Food Administration in Sweden. 2010. Available from: http:// www.slv.se/sv/grupp1/Mat-och-naring/Vad-innehaller-maten/Saltmineraler/ Selen/ (cited 11 October 2010). Cundiff DK, Lanou AJ, Nigg CR. Relation of omega-3 fatty acid intake to other dietary factors known to reduce coronary heart disease risk. Am J Cardiol 2007;99:12303. Weinehall L, Johnson O, Jansson JH, et al. Perceived health modies the effect of biomedical risk factors in the prediction of acute myocardial infarction. An incident case-control study from northern Sweden. J Intern Med 1998;243:99107. Peltonen M, Huhtasaari F, Stegmayr B, Lundberg V, Asplund K. Secular trends in social patterning of cardiovascular risk factor levels in Sweden. The Northern Sweden MONICA Study 1986-1994. Multinational monitoring of trends and determinants in cardiovascular disease. J Intern Med 1998;244:19. Yoshizawa K, Rimm EB, Morris JS, et al. Mercury and the risk of coronary heart disease in men. N Engl J Med 2002;347:175560. Navas-Acien A, Bleys J, Guallar E. Selenium intake and cardiovascular risk: what is new? Curr Opin Lipidol 2008;19:439. Bleys J, Navas-Acien A, Guallar E. Serum selenium levels and all-cause, cancer, and cardiovascular mortality among US adults. Arch Intern Med 2008;168:40410.

participated in the analysis and interpretation of the data and the drafting and nalizing of the manuscript; GH and MN: participated in the discussion initiating the study, collection and administration of data, and nalizing of the manuscript; TL: carried out the chemical analysis of mercury and selenium and participated in drafting of the manuscript; SS: participated in the discussion initiating the study, data analyses, interpretation of the data, and the drafting and nalizing of the manuscript; US: participated in designing and carrying out the data analysis, interpretation of the data, and the drafting and nalizing of the manuscript; BV: carried out the chemical analysis of fatty acids and participated in drafting of the manuscript; and J-HJ: participated in the discussion initiating the study, data analysis, interpretation of the data, and the drafting and nalizing of the manuscript. The authors had no conicts of interest to report.

18.

19. 20.

21. 22. 23.

REFERENCES
1. Psota TL, Gebauer SK, Kris-Etherton P. Dietary omega-3 fatty acid intake and cardiovascular risk. Am J Cardiol 2006;98(4A):3i18i. 2. He K. Fish, long-chain omega-3 polyunsaturated fatty acids and prevention of cardiovascular diseaseeat sh or take sh oil supplement? Prog Cardiovasc Dis 2009;52:95114. 3. Konig A, Bouzan C, Cohen JT, et al. A quantitative analysis of sh consumption and coronary heart disease mortality. Am J Prev Med 2005; 29:33546. 4. Guallar E, Sanz-Gallardo MI, vant Veer P, et al. Mercury, sh oils, and the risk of myocardial infarction. N Engl J Med 2002;347:174754. 5. Salonen JT, Seppanen K, Nyyssonen K, et al. Intake of mercury from sh, lipid peroxidation, and the risk of myocardial infarction and coronary, cardiovascular, and any death in eastern Finnish men. Circulation 1995;91: 64555. 6. Virtanen JK, Voutilainen S, Rissanen TH, et al. Mercury, sh oils, and risk of acute coronary events and cardiovascular disease, coronary heart disease, and all-cause mortality in men in eastern Finland. Arterioscler Thromb Vasc Biol 2005;25:22833. 7. Choi AL, Weihe P, Budtz-Jorgensen E, et al. Methylmercury exposure and adverse cardiovascular effects in Faroese whaling men. Environ Health Perspect 2009;117:36772. 8. Hallgren CG, Hallmans G, Jansson JH, et al. Markers of high sh intake are associated with decreased risk of a rst myocardial infarction. Br J Nutr 2001;86:397404. 9. Wennberg M, Bergdahl IA, Stegmayr B, et al. Fish intake, mercury, long-chain n23 polyunsaturated fatty acids and risk of stroke in northern Sweden. Br J Nutr 2007;98:103845. 10. Fernlund PFG, Hanson A, Steno J, Lundh B. Laurells clinical chemistry in practical medicine. 6th ed. Lund, Sweden: Studentlitteratur, 1991 (in Swedish). 11. Svensson BG, Schutz A, Nilsson A, Akesson I, Akesson B, Skerfving S. Fish as a source of exposure to mercury and selenium. Sci Total Environ 1992;126:6174. 12. Skerfving S. Methylmercury exposure, mercury levels in blood and hair, and health status in Swedes consuming contaminated sh. Toxicology 1974;2:323. 13. Norberg M, Wall S, Boman K, Weinehall L. The Vasterbotten Intervention Programme: background, design and implications. Glob Health Action (Epub ahead of print 22 March 2010). 14. Stegmayr B, Lundberg V, Asplund K. The events registration and survey procedures in the Northern Sweden MONICA Project. Scand J Public Health Suppl 2003;61:917. 15. Hallmans G, Agren A, Johansson G, et al. Cardiovascular disease and diabetes in the Northern Sweden Health and Disease Study Cohort evaluation of risk factors and their interactions. Scand J Public Health Suppl 2003;61:1824. 16. Tunstall-Pedoe H, Kuulasmaa K, Amouyel P, Arveiler D, Rajakangas AM, Pajak A. Myocardial infarction and coronary deaths in the World Health Organization MONICA Project. Registration procedures, event rates, and case-fatality rates in 38 populations from 21 countries in four continents. Circulation 1994;90:583612. 17. Johansson I, Hallmans G, Wikman A, Biessy C, Riboli E, Kaaks R. Validation and calibration of food-frequency questionnaire measure-

24. 25. 26. 27. 28. 29.

Downloaded from ajcn.nutrition.org by guest on January 24, 2013

30. 31. 32. 33. 34. 35. 36.

37.

38. 39. 40.