CLINICAL RESEARCH
European Heart Journal (2011) 32, 1416–1423 doi:10.1093/eurheartj/ehr036
Prevention/epidemiology
The prognostic value of adipose tissue fatty acids for incident cardiovascular disease: results from 3944 subjects in the Scottish Heart Health Extended Cohort Study Mark Woodward1,2,3*, Hugh Tunstall-Pedoe 1, G. David Batty 4,5, Roger Tavendale 1, Frank B. Hu 6,7, and Se´bastien Czernichow8,9 1 Cardiovascular Epidemiology, Institute of Cardiovascular Research, University of Dundee, Dundee, UK; 2Department of Epidemiology, Johns Hopkins University, Baltimore, MD, USA; 3The George Institute, University of Sydney, P.O. Box M201, Missenden Road, Sydney, NSW 2050, Australia; 4Medical Research Council Social & Public Health Sciences Unit, Glasgow, UK; 5Department of Epidemiology and Public Health, University College London, London, UK; 6Department of Nutrition, Harvard School of Public Health, Boston, USA; 7 Department of Epidemiology, Harvard School of Public Health, Boston, USA; 8De´partement de Sante´ Publique, Hoˆpital Avicenne (AP-HP) & Universite´ Paris 13, Bobigny, France; and 9Unite´ de Recherche en Epide´miologie Nutritionnelle (UMR INSERMU557, INRA, CNAM, UP13), CRNH-IdF, Bobigny, France
Received 7 October 2010; revised 29 November 2010; accepted 10 January 2011; online publish-ahead-of-print 22 February 2011
Aims
Dietary fats are routinely considered key determinants of cardiovascular risk, yet the scientific basis of this association has never been demonstrated using objective measures of fat intakes in a large prospective study in a general population. ..................................................................................................................................................................................... Methods Adipose tissue was taken from 3944 participants, predominantly aged 40–59 years, in Scotland, 1984–87. Perand results centages of individual fatty acids were measured using gas chromatography. Over a median of 19.5 years, 870 incident cardiovascular disease (CVD) events occurred. Hazard ratios (HRs) were obtained from Cox models and the additional prognostic value, accounting for variables in the Framingham and ASSIGN CVD risk scores, were assessed using discrimination indices. Adjusting for age, sex, total and HDL-cholesterol, systolic blood pressure, smoking, hypertensive medication use, diabetes, socio-economic status and family history, the percentage of monounsaturated adipose tissue fatty acids had a positive log-linear relationship with incident CVD: the HR comparing risk between the fourth and first quartiles was 1.29 (95% confidence interval: 1.05, 1.59). n-3 polyunsaturated fat showed the reverse trend, the corresponding result being 0.77 (0.63, 0.94). These two composite variables improved the classification of incident CVD events by 1.0 and 6.4%, respectively, with only the latter being significant at the 5% level. ..................................................................................................................................................................................... Conclusions A diet which is proportionately rich in polyunsaturated fat, as opposed to other fats, is expected to decrease the risk of CVD independently of the effects of common CVD risk factors, including social deprivation. Taking account of such diets improves the classification of future CVD events.
----------------------------------------------------------------------------------------------------------------------------------------------------------Keywords
Fatty acids † Myocardial infarction † Stroke † Risk score
Introduction Although diet is widely recognized as an important determinant of cardiovascular disease (CVD), this health behaviour is difficult to quantify precisely due to variations in portion size, cooking methods, and the balance of ingredients in food items, particularly processed food. Usual methods of dietary assessment, such as
food frequency questionnaires, diet history or repeated 24-h recalls, are often used in large epidemiological studies, yet these suffer from all the above problems, as well as errors in recall.1 These problems are especially true when quantifying intake of individual fatty acids, which are thought to be of paramount importance in chronic disease.2 Accordingly, there is considerable interest in the utilization of biomarkers for fatty acid intake,
* Corresponding author. Tel: +61 2 9993 4588, Fax: +61 2 9993 4588, Email:
[email protected] Published on behalf of the European Society of Cardiology. All rights reserved. & The Author 2011. For permissions please email:
[email protected].
1417
The prognostic value of ATFAs for incident CVD
through blood samples or subcutaneous adipose tissue.3 An important strength of adipose tissue fatty acids (ATFAs) is that they represent long-term (months to years) exposure to intakes of fatty acids, compared with days or weeks of exposure for blood or red blood cells levels.4 The relatively few studies of ATFA conducted to date have been hampered by the small numbers of subjects,5 and all have used cross-sectional or case – control designs6 – 17 which are suboptimal for investigating potential causality.18 Since tissue concentrations of fatty acids may be influenced by pathophysiological changes induced by diseases, prospective cohort studies are more suitable for examining the relationship between these biomarkers and CVD risk. To our knowledge, the largest sample of adipose tissue used to quantify fatty acids was collected in the Scottish Heart Health Extended Cohort Study (SHHEC). Previous reports from the cross-sectional phase of SHHEC have shown cross-sectional associations between ATFAs and coronary heart disease (CHD) at a regional level19 and changes in ATFAs with ageing.20 In both cases, only the first tranche of baseline SHHEC data were utilized. Here, we use all the available baseline data and, for the first time, report analyses of ATFAs in the prospective phase of the SHHEC. Since the major risk factors for CVD are well established, the primary aim of the current work is to ascertain whether ATFAs have any extra prognostic value, over and above standard risk engines. Of these, the Framingham risk score21 is the most commonly used world-wide for estimating individuals’ 10-year risk of CVD, while the ASSIGN CVD risk score22 is being implemented in Scotland.
Methods The study population for the SHHEC includes participants in the Scottish Heart Health Study (SHHS), a national random population survey between 1984 and 1986, and six associated random population surveys in Scotland.22,23 Subjects were recruited through the lists of General Practitioners. At the time of inception, Scotland had particularly high levels of coronary mortality in both sexes and internationally low intakes of polyunsaturated fats. A pre-specified aim of the study was, thus, to investigate whether dietary risk factors would add to existing, known, risk factors in the prediction of CVD. Adipose tissue was extracted from subjects in the SHHS and in two of the associated surveys: the 1986 MONICA study in north Glasgow and Edinburgh and a 1987 survey in Aryshire and Arran—all cities or regions of Scotland. Both these surveys used exactly the same questionnaire and clinic procedures as in the SHHS. Each random sample included subjects aged 40 – 59 years, with a small number outside this range also included in the MONICA surveys. All participants received questions on demographics, medical history, and lifestyle, including tobacco use, and a food frequency questionnaire, from which daily intakes of macronutrients were estimated.24 Participants were invited to attend clinics where blood pressure, weight, and height were measured and a 12-lead electrocardiogram was applied and a specimen of adipose tissue was taken from the outer upper arm using a 3 mm skin biopsy punch and frozen, initially at 2208C for up to 5 days and then for a median of 2 years (minimum 16 months; maximum 35 months) at 2408C. The fatty acid methyl esters were separated using gas chromatography and individual fatty acids were expressed as a percentage of the total amount of fatty acids present in the sample. Further details of the assay method
were given previously.19,25 The specific ATFAs measured are shown in Table 1. From the three constituent surveys, 12 534 subjects were asked to provide adipose tissue samples; after removing those who refused (16%) and failed extractions (46%), there were 4763 (38%) from whom valid fatty acid analyses could be drawn. Failed extractions were generally due to the punch having extracted skin only—to minimize discomfort, a relatively non-invasive method of biopsy punching was used25 and repeat biopsies were not allowed in the study protocol. There were no important differences between those included (n ¼ 4763) and excluded (n ¼ 7771) in any of the variables used in this manuscript: mean age at baseline was 48.8 vs. 48.5 years; mean systolic blood pressure, 132.0 vs. 132.7 mmHg; mean serum total cholesterol, 6.41 vs. 6.45 mmol/L; HDL-cholesterol 1.51 vs. 1.52; 47 vs. 52% were women; 37 vs. 39% were smokers; 1.3 vs. 1.7% had a history of diabetes; and 24.5 vs. 24.5% had CVD events, 19.6 vs. 19.8% had CHD events, and 6.7 vs. 6.9% had cerebrovascular events, during follow-up. Subjects who gave written permission were followed up through death registrations and the national record linkage database,22 up to the end of 2005. The primary outcome studied here is incident CVD: deaths attributed to a cardiovascular cause [International Classification of Diseases (ICD) 9 codes 390 – 459, ICD 10 codes I00-I99]; any hospital discharge diagnosis post-recruitment of CHD (ICD 9 410-414, ICD 10 I20-I25) or cerebrovascular disease (ICD 9 430– 438, ICD10 G45, I60-I69); or surgical codes for coronary artery bypass graft or percutaneous coronary angioplasty. Secondary outcomes were CHD and cerebrovascular disease. So as to analyse incident events, anyone with evidence of CVD at baseline was excluded from all the analyses reported here. This was determined from questions on previous doctor diagnoses of angina, heart attack, or stroke at baseline; whether the ECG was suggestive of myocardial infarction using Minnesota codes26; and by extending the search of the national linkage database back to 1981 to look for hospitalizations for CHD or stroke or any coronary surgical procedures prior to recruitment.
Statistical methods Since several of the ATFAs had highly skewed distributions, quartiles, and Spearman rank correlations were used to describe the data. Cox proportional hazards models were used to estimate hazard ratios (HRs) and 95% confidence intervals (CIs) for CVD by the quarters of each ATFA and for some appropriate combined variables: saturated (SFA: C14:0 + C16:0 + C18:0), monounsaturated (MUFA: 16:1, n-7 + C18:1, n-9 + C20:1, n-9), n-3 polyunsaturated (C22:5, n-3 + C22:6, n-3), n-6 polyunsaturated (C18:2, n-6 + C18:3, n-6 + C20:3, n-6 + C20:4, n-6) and polyunsaturated (n-6 PUFA + n-3 PUFA) fatty acids, and the PUFA to SFA ratio. Associations with ATFAs were adjusted for two sets of potential confounding factors: those used in the Framingham CVD score21 (age, total and HDL-cholesterol, systolic blood pressure, diabetes, history of blood pressure medication, and current smoking status) and those used in either the Framingham or the Scottish ASSIGN score22 (adding family history of CHD and socioeconomic status to Framingham, and refining the variable used to measure current smoking as number of cigarettes smoked per day). Socio-economic status was measured by the Scottish index of multiple deprivation (SIMD),22 which increases with increasing deprivation. Anyone with missing values for any of these variables was excluded from all analyses. Further adjustment sets, not reported here, substituted highest level of education achieved for SIMD and added body mass index (weight divided by the square of height, in kg/m2)— neither changed the HRs in any important way. Since no interactions between ATFAs and sex were significant at the 1% level, results are
1418
M. Woodward et al.
Table 1
Baseline summary statistics for 3944 men and women in the Scottish Heart Health Extended Cohort Study Men (n 5 2095)
Variable
Women (n 5 1849)
............................................................................................................................................................................... Continuous variables: mean (standard deviation) or median (first to third quartile)a Age (years) Systolic blood pressure (mmHg)
49.0 (6.9) 133.2 (18.5)
48.9 (6.6) 130.0 (20.0)
Total serum cholesterol (mmol/L)
6.29 (1.13)
6.49 (1.31)
High-density lipoprotein cholesterol (mmol/L) Scottish index of multiple deprivation score
1.38 (0.37) 16.4 (8.52– 31.1)
1.68 (0.42) 16.0 (8.55–31.1)
Cigarettes per day (for smokers)
20 (15– 25)
18 (12– 20)
............................................................................................................................................................................... b
Nutrient intakes per day Fat (g)
88.0 (73.7– 106.7)
74.3 (61.1–90.3)
Saturated fat (g)
38.9 (31.6– 48.0)
33.5 (26.3–41.6)
Polyunsaturated fat (g) Protein (g)
10.6 (8.10– 14.1) 85.8 (73.9– 99.2)
8.50 (6.40–11.4) 73.3 (63.7–84.9)
Carbohydrates (g)
270.6 (222.1–329.2)
185.9 (149.8–229.7)
Alcohol (g) Energy (kcals)
14.9 (4.10– 31.4) 2316 (1953–2710)
3.70 (0.00–9.50) 1706 (1448– 2019)
............................................................................................................................................................................... Adipose tissue fatty acidsc Saturated fat (SFA) (%)
27.0 (24.9– 29.2)
25.2 (22.6–27.7)
C14:0 Myristic C16:0 Palmitic
2.31 (1.81– 2.83) 21.0 (19.7– 22.4)
2.28 (1.75–2.76) 19.4 (17.9–21.0)
C18:0 Stearic
3.64 (2.97– 4.41)
3.39 (2.66–4.29)
62.7 (60.2– 65.2) 8.66 (7.45– 9.98)
64.0 (61.3–67.0) 9.23 (7.96–10.56)
50.6 (49.0– 52.2)
51.3 (50.0–53.1)
3.08 (2.76– 3.48) 9.93 (8.45– 11.64)
3.05 (2.75–3.42) 10.4 (9.15–12.1)
Monounsaturated fat (MUFA) (%) C16:1, n-7 Palmitoleic C18:1, n-9 Oleic C20:1, n-9 Eicosaenoic Polyunsaturated fat (PUFA) (%) n-6 polyunsaturated fat
9.51 (8.04– 11.01)
9.94 (8.67–11.61)
C18:2, n-6 Linoleic C18:3, n-6 g-Linolenic
8.57 (7.13– 10.20) 0.20 (0.15– 0.26)
8.99 (7.75–10.62) 0.17 (0.12–0.22)
C20:3, n-6 Dihomo-g- linolenic
0.10 (0.06– 0.12)
0.15 (0.11–0.18)
0.63 (0.54– 0.72) 0.42 (0.33– 0.52)
0.60 (0.51–0.69) 0.46 (0.36–0.58)
0.25 (0.20– 0.30)
0.26 (0.21–0.32)
0.17 (0.13– 0.22) 0.37 (0.31– 0.44)
0.19 (0.15–0.26) 0.42 (0.35–0.51)
Smokers
765 (37)
640 (35)
Diabetes Family history of coronary disease
21 (1) 547 (26)
19 (1) 569 (31)
Blood pressure lowering medication
108 (5)
196 (11)
C20:4, n-6 Arachidonic n-3 polyunsaturated fat C22:5, n-3 Docosapentaenoic C22:6, n-3 Docosahexaenoic PUFA/SFA
............................................................................................................................................................................... Binary variables, n (%)
a
As appropriate (for skewed variables quartiles are shown). Estimated from a food frequency questionnaire. The relative amount of this fatty acid in relation to the total adipose fat, per person.
b c
presented across the sexes, with sex added to the two risk adjustment sets. Tests for linear and quadratic effects, on the logarithmic scale, were carried out using log likelihood tests.18 To ascertain whether ATFAs have any prognostic value for CVD, having allowed for the risk factors included in the Framingham and ASSIGN scores, areas under the receiver operating characteristic curves (c-statistics) and relative integrated discrimination indices (RIDIs),27 after adding ATFAs to the two adjustment sets considered,
were estimated using bootstrap resampling (500 times) to correct for self-testing.28 Approximate P-values and 95% CIs were obtained by assuming normal distributions for these parameters. In these analyses, only those ATFAs which were independently significant after multiple adjustments in the Cox models were included. Adipose tissue fatty acids were measured by continuous terms in these analyses, taking linear terms only unless quadratic terms were found significant in the Cox analyses, in which case these were added. Non-ATFA
1419
The prognostic value of ATFAs for incident CVD
continuous variables were treated as in the Framingham and ASSIGN scores—i.e. the Framingham variables were log transformed. Every such variable was taken together with its interaction with sex: that is, effect modification by sex was allowed for in every case, as in the Framingham21 and ASSIGN22 scores. All analyses were performed using SAS version 9.1.
Results Of the 4763 with adipose samples, 4433 had no evidence of preexisting CVD at baseline. Of these, 3944 (47% women) had no missing values for any of the covariates chosen for multivariable analyses. The mean age at baseline was 49 years in each sex and, by modern standards, the prevalence of smoking was high, and that of diabetes was low, in both sexes (Table 1). Almost two-thirds of adipose fat was monounsaturated and only 10% polyunsaturated, with virtually the same percentages for each sex. In the general context of observational epidemiology, there were high age/sex-adjusted correlations (of +0.18 or more) between both the SIMD score (increasing social deprivation) and cigarette dose and both MUFA (precisely, C16:1, n-7 and C18:1, n-9, in a positive direction) and PUFA (precisely, C18:2, in a negative direction) (Supplementary material online, Table S1). The only other correlations of this order of magnitude were between C18:3, n-6 SFA and HDL-cholesterol (positively) and between cigarette dose and the PUFA to SFA ratio (negatively). During follow-up (median of 19.5 years), 870 incident CVD events occurred; 651 of these 870 had a CHD event and 219 had a cerebrovascular event (44 had both). Many of the ATFAs were independently associated with CVD after adjustment for Framingham variables (Table 2). For the HRs comparing the extreme quarters, only C20:1, n-9, C18:3, n-6, C20:3, n-6, C20:4, n-6 and the PUFA to SFA ratio were non-significant at the 5% significance level. In most cases, there was evidence of a linear trend, sometimes together with curvature to suggest an arrested trend (threshold effect). For example, n-6 PUFA shows similarity between the risks in the first two-quarters and then a drop in risk to the third- and fourth-quarters, which have similar risk. As the percentage of SFA and PUFA (n-3 and n-6) increased, the risk of CVD tended to go down, while MUFA had a positive relationship with the risk of CVD. Further adjustment for family history, SIMD score and cigarette dose accounted for the effects of SFA and n-6 PUFA, but left a residual positive effect of MUFA and negative effect of n-3 PUFA. Figure 1 summarizes the associations for the major subtypes of ATFAs, showing HRs contrasting the extreme quarters, after adjustment for all the confounding variables in the two risk scores (ASSIGN and Framingham) combined. For additional insight, Supplementary material online, Table S2 shows the effects of the ATFAs on CVD after adjustment for only age and sex and adjusting for age, sex, SIMD score, and family history of CHD. Supplementary material online, Tables S3 and S4 show the effects of ATFAs on CHD and cerebrovascular disease, respectively, adjusting as in Table 2. In each case, the broad inferences drawn are the same as for Table 2, although the PUFA to SFA ratio showed some evidence of an effect on CVD after minimal adjustments (Supplementary material online, Table S2).
After correction for internal testing, many MUFA and PUFA (n-3 and n-6) individual ATFAs added significantly (i.e. approximate P-values, after bootstrapping, ,0.05) to Framingham variables, both for discrimination and correct classification of who will and will not develop CVD in SHHEC (Table 3; left side). When adding individual ATFAs to the prediction model with all variables considered here (Framingham + ASSIGN), none gave approximate P-values , 0.05 for discrimination, but PUFA variables (not others) appear to improve classification significantly. Total n-3 PUFA adds most to the classification of CVD, increasing the accuracy of distinguishing future cases from non-cases by 6.4% when added to the Framingham variables, and by 6.0% when added to the Framingham and ASSIGN variables together. Confidence intervals for these percentages, using bootstrap percentiles, were 2.5 – 10.6% and 2.0 –10.3%, respectively (normal-based equivalents were similar: 2.3 –10.6 and 1.7 –10.3%). Total n-6 PUFA (predominantly C18:2, n-6) improves classification of CVD by 4.5% when added to the Framingham variables, and by 3.4% when added to all variables.
Discussion This is the first long-term prospective cohort study to have investigated the associations between ATFAs and incident cardiovascular events. We have shown that relative intakes of monounsaturated and polyunsaturated fatty acids were independent and opposite predictors of CVD. Both for n-3 and n-6 PUFA, those with proportionate intakes in the upper 20% had 20% less CVD than those in the lowest 20%, after allowing for a wide range of known risk factors. Furthermore, we have demonstrated that adding adipose tissue PUFA increased the prognostic capacities of the Framingham and ASSIGN risk equations. We thus give robust evidence to support the common theory that the balance of fats in the diet is important for cardiovascular health, even after allowing for classical risk factors that they affect, such as dyslipidaemia and hypertension. We have found that some of the relative effects of fat intake are likely to be captured by a sensitive measure of socio-economic status. Even after additional allowance for this and family history of CHD, n-3 PUFA improved classification of CVD by over 6%, which compares favourably to 1 –3% for fibrinogen29,30 and 7% for CRP,29 found in similar analyses using the SHHEC. The protective independent predictive value of n-3 ATFAs for CVD observed in this study is consistent with results from numerous case –control studies investigating the associations between blood or red cell membrane levels of n-3 PUFA and CHD.3,31 Further, as observed in our prospective cohort, previous epidemiological studies and intervention trials have observed beneficial effects of n-6 PUFA on CHD.2 Our study is the first to investigate SFA and MUFA from adipose tissue and incident CVD, and thus no comparison is possible with other published data. There is no gold standard biomarker for the quantification of the usual dietary intake of total fat, often prone to underreporting.4 Intakes of several fatty acids can be estimated through different biomarkers such as blood (serum, plasma, or red cell) levels and adipose tissue obtained from a skin biopsy punch. Each of these methods have specificities related to their analytical methods and
1420
M. Woodward et al.
Table 2 Associations between fatty acids and incident cardiovascular disease after adjustment for Framingham variablesa and after adjustment for Framingham and ASSIGN (‘all’) variablesb Fatty acid/Adjustment
Hazard ratio (95% confidence interval) vs. first quarter
..........................................................................................................
P-value
Quarter
..........................................................................................................
Second
Third
Fourth
Linear
Quadratic
Framingham
0.86 (0.71, 1.04)
0.87 (0.71, 1.05)
0.74 (0.60, 0.90)
0.003
0.81
All C14:0 Myristic
0.87 (0.72, 1.06)
0.94 (0.77, 1.14)
0.81 (0.66, 1.00)
0.06
0.99
Framingham
0.91 (0.76, 1.10)
0.78 (0.64, 0.94)
0.77 (0.63, 0.93)
0.006
0.38
All C16:0 Palmitic
0.95 (0.79, 1.15)
0.84 (0.69, 1.01)
0.85 (0.70, 1.03)
0.10
0.31
Framingham
0.87 (0.72, 1.07)
0.81 (0.66, 0.99)
0.76 (0.62, 0.94)
0.01
0.45
0.88 (0.72, 1.07)
0.86 (0.70, 1.05)
0.83 (0.68, 1.02)
0.10
0.74
0.90 (0.74, 1.08)
0.85 (0.71, 1.03)
0.78 (0.64, 0.96)
0.01
0.14
0.90 (0.75, 1.08)
0.90 (0.74, 1.09)
0.85 (0.69, 1.04)
0.11
0.18
1.22 (0.99, 1.49)
1.22 (1.00, 1.50)
1.44 (1.18, 1.76)
,0.0001
0.29
1.19 (0.97, 1.46)
1.15 (0.93, 1.41)
1.29 (1.05, 1.59)
0.004
0.47
Framingham
1.02 (0.83, 1.24)
1.07 (0.87, 1.31)
1.30 (1.07, 1.58)
0.002
0.04
All C18:1, n-9 Oleic
1.00 (0.81, 1.22)
1.01 (0.82, 1.24)
1.21 (0.99, 1.47)
0.03
0.04
Framingham
1.12 (0.91, 1.37)
1.27 (1.04, 1.55)
1.39 (1.14, 1.71)
0.0005
0.70
1.07 (0.87, 1.32)
1.19 (0.97, 1.45)
1.25 (1.02, 1.54)
0.02
0.77
1.00 (0.82, 1.21)
0.93 (0.77, 1.14)
1.15 (0.95, 1.39)
0.19
0.56
1.02 (0.84, 1.23)
0.93 (0.76, 1.13)
1.11 (0.91, 1.34)
0.29
0.87
0.95 (0.80, 1.14)
0.79 (0.65, 0.96)
0.79 (0.65, 0.97)
0.02
0.002
0.97 (0.81, 1.16)
0.81 (0.66, 0.98)
0.84 (0.69, 1.03)
0.07
0.004
Framingham
0.96 (0.80, 1.15)
0.80 (0.66, 0.97)
0.80 (0.66, 0.98)
0.03
0.002
All C18:2, n-6 Linoleic
0.98 (0.82, 1.17)
0.83 (0.68, 1.01)
0.85 (0.69, 1.04)
0.10
0.007
Framingham
0.93 (0.78, 1.12)
0.79 (0.65, 0.95)
0.79 (0.64, 0.96)
0.03
0.001
0.95 (0.79, 1.14)
0.82 (0.67, 0.99)
0.83 (0.68, 1.03)
0.10
0.007
0.84 (0.70, 1.02)
0.95 (0.79, 1.16)
0.92 (0.75, 1.14)
0.38
0.52
0.85 (0.70, 1.03)
0.96 (0.79, 1.17)
0.95 (0.77, 1.17)
0.56
0.62
............................................................................................................................................................................... Saturated fat (SFA)
All C18:0 Stearic Framingham All Monounsaturated fat (MUFA) Framingham All C16:1, n-7 Palmitoleic
All C20:1, n-9 Eicosaenoic Framingham All Polyunsaturated fat (PUFA) Framingham All n-6 polyunsaturated fat
All C18:3, n-6 g-Linolenic Framingham All C20:3, n-6 Dihomo-g-linolenic Framingham
0.80 (0.66, 0.96)
0.79 (0.65, 0.97)
0.81 (0.65, 1.00)
0.04
0.07
0.80 (0.66, 0.96)
0.81 (0.66, 0.99)
0.83 (0.67, 1.03)
0.05
0.08
0.93 (0.76, 1.14)
0.96 (0.79, 1.17)
1.11 (0.91, 1.34)
0.72
0.59
0.93 (0.76, 1.13)
0.93 (0.76, 1.13)
1.03 (0.85, 1.25)
0.97
0.82
Framingham
0.92 (0.76, 1.12)
0.80 (0.65, 0.97)
0.79 (0.64, 0.96)
0.02
0.002
All
0.94 (0.78, 1.14)
0.78 (0.64, 0.95)
0.77 (0.63, 0.94)
0.01
0.003
0.89 (0.73, 1.09)
0.84 (0.69, 1.03)
0.80 (0.65, 0.98)
0.03
0.09
All C20:4, n-6 Arachidonic Framingham All n-3 polyunsaturated fat
C22:5, n-3 Docosapentaenoic Framingham
Continued
1421
The prognostic value of ATFAs for incident CVD
Table 2 Continued Fatty acid/Adjustment
Hazard ratio (95% confidence interval) vs. first quarter
..........................................................................................................
P-value
Quarter
..........................................................................................................
Second
Third
Fourth
Linear
Quadratic
0.91 (0.75, 1.11)
0.85 (0.70, 1.04)
0.77 (0.63, 0.95)
0.02
0.03
0.85 (0.71, 1.04)
0.78 (0.64, 0.95)
0.78 (0.64, 0.96)
0.05
0.002
0.86 (0.71, 1.04)
0.75 (0.62, 0.92)
0.76 (0.62, 0.93)
0.03
0.0006
Framingham
0.87 (0.73, 1.05)
0.88 (0.73, 1.06)
0.91 (0.75, 1.11)
0.56
0.16
All
0.84 (0.70, 1.01)
0.85 (0.70, 1.03)
0.91 (0.74, 1.11)
0.58
0.07
............................................................................................................................................................................... All C22:6, n-3 Docosahexaenoic Framingham All PUFA/SFA
a Framingham variables are sex, age, total serum cholesterol, HDL-cholesterol, systolic blood pressure, history of taking blood pressure medication, smoking (current vs. not), diabetes, and sex21. b ‘All’ variables are sex, age, total serum cholesterol, HDL-cholesterol, systolic blood pressure, history of taking blood pressure medication, cigarettes smoked per day, diabetes, sex, family history of coronary disease and socio-economic status22.
Figure 1 Hazard ratios for cardiovascular disease comparing highest to lowest quarters of individual adipose tissue fatty acids, after adjustment for sex, age, total serum cholesterol, HDL-cholesterol, systolic blood pressure, history of taking blood pressure medication, diabetes, cigarettes smoked per day, family history of coronary disease, and socio-economic status. Diamonds show results for the summation of all fatty acids within the four subtypes shown. Bars and diamonds extend to the limits of the 95% confidence intervals.
capacity to represent a different duration of exposure of fatty acid intakes (days for serum; weeks for red cells; years for adipose tissue) and level of correlation to fatty acid dietary intakes. Three main types of ATFA biomarkers have been related to dietary intakes: n-3 and n-6 PUFAs, trans unsaturated and some subgroups of SFA derived from dairy products. As they are not produced endogeneously, they represent good biomarkers for
validation studies. In a recent systematic review, it has been estimated that the level of correlation of n-3 PUFA intake and n-3 level in adipose tissue ranged from 0.40 to 0.60, according to the method of dietary assessment.5 Similar correlations were identified with n-3 blood levels; however, they represent a short-term exposure to n-3 intakes compared with adipose tissue levels. On the other hand, several factors can influence measured levels of
1422
M. Woodward et al.
Table 3 Marginal discrimination and reclassification statistics, with approximate P-values, for incident cardiovascular disease within 10 years when individual adipose fatty acid variablesa are added to only Framingham variablesb and when individual adipose fatty acid variablesa are added to Framingham and ASSIGN (‘all’) variablesb Added to Framingham variables
..................................................................... Discrimination
.....................................................................
Reclassification
Discrimination
P-value
R
P-value
Dc
................................ Dc
Added to all variables
.........................
................................ P-value
Reclassification
.........................
R
P-value
............................................................................................................................................................................... SFA
0.0011
0.10
1.1
0.08
0.0000
0.99
0.1
0.39
C14:0 C16:0
20.0006 0.0006
0.52 0.48
2.0 0.9
0.02 0.13
20.0004 20.0001
0.58 0.96
0.5 0.1
0.14 0.40
C18:0
0.0020
0.02
0.3
0.26
0.0006
0.75
0.0
0.46
............................................................................................................................................................................... MUFA C16:1 C18:1
0.0021
0.002
2.7
0.01
20.0002
0.87
1.0
0.11
0.0021 0.0008
0.02 0.36
1.5 2.6
0.05 0.008
0.0000 20.0002
0.91 0.90
1.0 1.0
0.11 0.09
0.006
............................................................................................................................................................................... PUFA PUFA-6 C18:2 C20:3
0.0024
0.02
4.9
0.002
0.0008
0.48
3.8
0.0023
0.02
4.5
0.003
0.0007
0.47
3.4
0.008
0.0022 20.0005
0.03 0.54
4.2 1.9
0.003 0.07
0.0006 20.0012
0.52 0.27
3.0 1.8
0.01 0.06
PUFA-3
0.0025
0.03
6.0
0.003
0.0013
0.27
6.4
0.001
C22:5 C22:6
0.0014 0.0024
0.14 0.05
4.2 4.2
0.02 0.01
0.0008 0.0010
0.45 0.36
5.8 4.8
0.003 0.004
For full names of fatty acid variables and adjustments, see Table 2. Dc, delta c statistic (area under the receiver operating characteristic curve); R, relative integrated discrimination index (%). The c-statistic for Framingham variables is 0.7425 and for ‘all’ variables is 0.7594. a Linear terms + quadratic terms where quadratic effect P-value is ,0.05 in Table 2. b For a list of adjustments see Table 2.
ATFAs, such as weight fluctuations, the use of supplements (e.g. fish oil capsules) and the procedures used. The tissue-sampling site (arm, buttock, or abdomen is known to cause variations in sampling results,19,32 but it is unknown what effects this would have on CVD-risk estimation. The arm was chosen in this study because it is the most practical site for extraction in large clinical studies. Degradation during long-term storage could also affect sample values. To reduce such issues, we used standardized procedures for adipose tissue biopsies and assays were all done within 3 years of sampling. This study has two strengths that are essential for robust inference: relatively low sampling error (due to the large sample size, leading to narrow CIs) and relatively low chance of measurement error. The latter is a combination of the objective measures of ATFA measurement, the exclusion of those with pre-existing CVD at baseline and the prospective and impersonal nature of event ascertainment. As well as being the only cohort study of ATFAs, as far as we are aware this is also the first study that uses modern statistical methods to investigate whether objective measures of diet add to discrimination and reclassification over and above routine risk scores. To have shown that they do add to all the variables included in the two most appropriate risk scores moves the debate over whether diet is an independent risk factor for CVD one important, positive, step further. The study has several limitations. First, the ATFAs were only measured once, yet diets are sure to have changed during follow-up. Thus we can estimate how well the accumulated
effects of diet over the years preceding examination predict future CVD, after adding to risk scores, but not how changes in diet predict CVD. However, it is only the first of these goals that these analyses seek to address. Risk scores are conventionally based upon single enumerations of risk factors, because they give the risk of CVD in the future according to results collected today. Thus, although the Framingham study has multiple measurements of blood pressure, cholesterol, smoking, etc., the Framingham risk score only uses risk factor data collected at a single time. Hence, for the current purpose, a one-off measurement is sufficient. We cannot, however, say how changes in diet may have affected the risk of CVD. We also cannot estimate how the relative effects of components of the diet may change with the introduction of new treatments over time, most obviously following the widespread introduction of statins from the mid-1990s in Scotland. It is conceivable that statins may affect ATFAs—although the Oxford Cholesterol Study found no significant effects of simvastatin on individual free fatty acids in serum33—but we have no reason to suppose that statin use would cause important variations in the associations between ATFAs and CVD. Another limitation is that some important ATFAs are missing from the panel that was measured .20 years ago, when technology was less welldeveloped. This includes the n-3 variables: 18:3, n-3 (alpha linolenic acid) and 20:5, n-3 (eicosapentaenoic acid, EPA) and restricts the scope of our conclusions regarding the effects of ATFAs. Furthermore, modern practice would be to store specimens at 2808C, rather than 2408C, when kept unthawed for .6 months. This
The prognostic value of ATFAs for incident CVD
may have introduced some measurement error into our analyses, which will have caused associations with CVD to tend to the null. Another issue may be that our ATFA measures are proportional, not absolute. Also, we have no external data set upon which to test our findings—instead we use the most efficient internal method of bias correction that we are aware of.28 Although the bootstrap distributions sometimes showed moderate skewness, in sensitivity analyses, percentile-based CIs based on repeated bootstrapping with 500 resamples, and on 1000 resamples, were always similar to the normal-based CIs, suggesting that the estimates and approximate P-values quoted in Table 3 provide robust inferences. Finally, despite their attractiveness as objective estimates of long-term dietary habits, ATFAs have the practical disadvantage that they are invasive and not suitable for widespread use in clinical check-ups. This is emphasized by the large number of missing values (.60%) in our ATFA data, of which about a third were due to subjects refusing the test.
Conclusions Using biomarkers, such as adipose tissue, allows investigation of the long-term exposure to fatty acids. We have shown that both n-6 and n-3 PUFA are protective for incident CVD and increase the discrimination and classification capacities of standard cardiovascular risk equations. Even if the routine use of tissue biopsy is not warranted, these results help to better understand the mechanisms involved in the protective effects of n-6 and n-3 PUFA on CVD.
Supplementary material Supplementary material is available at European Heart Journal online.
Funding This study was supported by the Chief Scientist Office of the Scottish Health Department and subsequently by the British Heart Foundation. M.W.’s research is supported by the Australian National Health and Medical Research Council’s program grant 571281. G.D.B. is a Wellcome Trust Career Development Fellow (WBS U.1300.00.006.00012.01). F.H.’s research is supported by NIH grant HL60712. Conflict of interest: none declared.
References 1. Thompson FE, Byers T. Dietary assessment resource manual. J Nutr 1994;124: 2245S– 2317S. 2. Mozaffarian D, Micha R, Wallace S. Effects on coronary heart disease of increasing polyunsaturated fat in place of saturated fat: a systematic review and meta-analysis of randomized controlled trials. PLoS Med 2010;7:e1000252. 3. Harris WS, Poston WC, Haddock CK. Tissue n-3 and n-6 fatty acids and risk for coronary heart disease events. Atherosclerosis 2007;193:1 –10. 4. Arab L. J Nutr 2003;133:925S –932S. 5. Øverby NC, Serra-Majem L, Andersen LF. Dietary assessment methods on n-3 fatty acid intake: a systematic review. Br J Nutr 2009;102:S56 – S63. 6. Campos H, Baylin A, Willett WC. Alpha-linolenic acid and risk of nonfatal acute myocardial infarction. Circulation 2008;118:339 –345. 7. Lopes C, Aro A, Azevedo A, Ramos E, Barros H. Intake and adipose tissue composition of fatty acids and risk of myocardial infarction in a male Portuguese community sample. J Am Diet Assoc 2007;107:276 –286. 8. Ghahremanpour F, Firoozrai M, Darabi M, Zavarei A, Mohebbi A. Adipose tissue trans fatty acids and risk of coronary artery disease: a case-control study. Ann Nutr Metab 2008;52:24–28.
1423 9. Pedersen JI, Ringstad J, Almendingen K, Haugen TS, Stensvold I, Thelle DS. Adipose tissue fatty acids and risk of myocardial infarction—a case –control study. Eur J Clin Nutr 2000;54:618 – 625. 10. Baylin A, Kabagambe EK, Ascherio A, Spiegelman D, Campos H. Adipose tissue alphalinolenic acid and nonfatal acute myocardial infarction in Costa Rica. Circulation 2003;107:1586 – 1591. 11. Baylin A, Campos H. Arachidonic acid in adipose tissue is associated with nonfatal acute myocardial infarction in the central valley of Costa Rica. J Nutr 2004;134: 3095 –3099. 12. Clifton PM, Keogh JB, Noakes M. Trans fatty acids in adipose tissue and the food supply are associated with myocardial infarction. J Nutr 2004;134:874–879. 13. Guallar E, Aro A, Jime´nez FJ, Martı´n-Moreno JM, Salminen I, van’t Veer P, Kardinaal AF, Go´mez-Aracena J, Martin BC, Kohlmeier L, Kark JD, Mazaev VP, Ringstad J, Guille´n J, Riemersma RA, Huttunen JK, Thamm M, Kok FJ. Omega-3 fatty acids in adipose tissue and risk of myocardial infarction. The EURAMIC Study. Arterioscl Thromb Vasc Biol 1999;19:1111 –1118. 14. Kark JD, Kaufmann NA, Binka F, Goldberger N, Berry EM. Adipose tissue n-6 fatty acids and acute myocardial infarction in a population consuming a diet high in polyunsaturated fatty acids. Am J Clin Nutr 2003;77:796 –802. 15. Roberts TL, Wood DA, Riemersma RA, Gallagher PJ, Lampe FC. Linoleic acid and risk of sudden cardiac death. Br Heart J 1993;70:524 –529. 16. Wood DA, Riemersma RA, Butler S, Thomson M, Macintyre C, Elton RA, Oliver MF. Linoleic and eicosapentaenoic acids in adipose tissue and platelets and risk of coronary heart disease. Lancet 1987;1:177–183. 17. Smit LA, Baylin A, Campos H. Conjugated linoleic acid in adipose tissue and risk of myocardial infarction. Am J Clin Nutr 2010;92:34 –40. 18. Woodward M. Epidemiology: Study Design and Data Analysis. 2nd ed. Boca Raton: Chapman and Hall/CRC Press; 2005. 19. Tavendale R, Lee AJ, Smith WCS, Tunstall-Pedoe H. Adipose tissue fatty acids in Scottish men and women: results from the Scottish Heart Health Study. Atherosclerosis 1992;94:161 –169. 20. Bolton-Smith C, Woodward M, Tavendale R. Evidence for age-related differences in the fatty acid composition of human adipose tissue, independent of diet. Eur J Clin Nutrition 1997;51:619–624. 21. D’Agostino RB Sr, Vasan RS, Pencina MJ, Wolf PA, Cobain M, Massaro JM, Kannel WB. General cardiovascular risk profile for use in primary care: the Framingham Heart Study. Circulation 2008;117:743 –753. 22. Woodward M, Brindle P, Tunstall-Pedoe H. Adding social deprivation and family history to cardiovascular risk assessment: the ASSIGN score from the Scottish Heart Health Extended Cohort (SHHEC). Heart 2007;93:172 –176. 23. Tunstall-Pedoe H, Woodward M, Tavendale R, A’Brook R, McCluskey MK. Comparison of the prediction by 27 different factors of coronary heart disease and death in men and women of the Scottish Heart Health Study. Brit Med J 1997; 315:722–729. 24. Bolton-Smith C, Smith WCS, Woodward M, Tunstall-Pedoe H. Nutrient intakes of different social class groups: Results from the Scottish Heart Health Study. Brit J Nutrition 1991;65:321 –335. 25. Smith WCS, Tavendale R, Tunstall-Pedoe H. Simplified subcutaneous fat biopsy for nutritional surveys. Hum Nutr Clin Nutr 1986;40:323 – 325. 26. Prineas RJ, Crow RS, Blackburn H. The Minnesota Code Manual of Electrocardiographic Findings: Standards and Procedures for Measurement and Classification. Bristol: John Wright; 1982. 27. Pencina MJ, D’Agostino RBD Sr, D’Agostino RBD Jr, Vasan RS. Evaluating the added predictive ability of a new marker: from area under the ROC curve to reclassification and beyond. Stat Med 2008;27:157 –212. 28. Harrell FE Jr, Lee KL, Mark DB. Multivariable prognostic models: issues in developing models, evaluating assumptions and adequacy, and measuring and reducing errors. Stat Med 1996;15:361 –387. 29. Woodward M, Welsh P, Rumley A, Tunstall-Pedoe H, Lowe GD. Do inflammatory biomarkers add to the discrimination of cardiovascular disease after allowing for social deprivation? Results from a 10-year cohort study in Glasgow, Scotland. Eur Heart J 2010;31:2669 –2675. 30. Woodward M, Tunstall-Pedoe H, Rumley A, Lowe G. Does fibrinogen add to prediction of cardiovascular disease? Results from the Scottish Heart Health Extended Cohort Study. Brit J Haematol 2009;146:442–446. 31. Shearer GC, Pottala JV, Spertus JA, Harris WS. Red blood cell fatty acid patterns and acute coronary syndrome. PLoS One 2009;4:e5444. 32. Schafer L, Overvad K. Subcutaneous adipose tissue fatty acids and vitamin E in humans: relation to diet and sampling site. Am J Clin Nutr 1990;52:486 –490. 33. Mitropoulos KA, Armitage JM, Collins R, Meade TW, Reeves BEA, Wallendszus KR, Wilson SS, Lawson A, Peto R. Randomized placebo-controlled study of the effects of simvastatin on haemostatic variables, lipoproteins and free fatty acids. Eur Heart J 1997;18:235 –241.
