Intake of phytosterols from natural sources and risk of cardiovascular disease in the European Prospective Investigation into Cancer and Nutrition-the Netherlands (EPIC-NL) population

Abstract

Background

Phytosterols (PSs) are known to lower low-density lipoprotein cholesterol (LDL-C), an established risk factor for cardiovascular disease (CVD). Whether a high intake of PS reduces CVD risk is unknown. This observational study aimed to investigate the associations between intake of naturally occurring PSs, blood lipids and CVD risk.

Methods

The study included 35,597 Dutch men and women, participating in the European Prospective Investigation into Cancer and Nutrition-the Netherlands (EPIC-NL) study. At baseline, intakes of naturally occurring PSs were estimated with a validated food frequency questionnaire and non-fasting blood lipids were measured. Occurrence of CVD, coronary heart disease (CHD) and myocardial infarction (MI) was determined through linkage with registries.

Results

The average energy-adjusted PS intake at baseline was 296 mg/d (range: 83–966 mg/d). During 12.2 years of follow-up, 3047 CVD cases (8.6%) were documented. After adjustment for confounders, PS intake was not associated with risk of CVD, CHD or MI (p-value trend > 0.05); hazard ratios ranged from 0.90–0.99 for CVD, from 0.83–0.90 for CHD and from 0.80–0.95 for MI risk across quintiles of PS intake and were almost all non-significant. Higher PS intake was associated with lower total cholesterol (−0.06 mmol/l per 50 mg/d; p-value = 0.038) and lower LDL-C (−0.07 mmol/l; p-value = 0.007), particularly among men. In mediation analysis, LDL-C did not materially affect the association between PS intake and CVD risk.

Conclusions

In this population with a relatively narrow range of low naturally occurring PS intakes, intake of PS was not associated with reduced CVD risk despite lower LDL-C concentrations in men.

Phytosterols, blood lipids, cardiovascular disease

Introduction

Plant sterols and plant stanols (together they are referred to as phytosterols (PSs)) are bioactive compounds found in all foods of plant origin. PSs are well-known for their total cholesterol (TC)-lowering, and especially low-density lipoprotein cholesterol (LDL-C)-lowering properties; an average PS intake of 2 g/d lowers LDL-C by on average 8–10%.^1,2 Intakes of around 2 g/d of PSs cannot be achieved with habitual diets; PS intakes in the general population usually range between 200–400 mg/d.^3,4 With specific dietary habits such as vegetarians diets, higher PS intakes of 500–1000 mg/d can be reached.^5,6

Elevated LDL-C is an established risk factor for cardiovascular disease (CVD).⁷ As PSs lower LDL-C, one could assume that high intakes of PSs would reduce CVD risk. Direct evidence supporting such a reduced risk of CVD is however lacking. Given the difficulties in performing fully controlled CVD endpoint trials with PS intervention, observational studies could help to clarify whether intake of naturally occurring PSs is associated with blood lipid risk markers and incidence of CVD.

A few observational studies with dietary PS intakes have been performed and showed that people with higher intakes of naturally occurring PSs have lower concentrations of LDL-C^8–10 and tend to have a lower carotid intima-media thickness.⁹ A recent study showed that a high intake of naturally occurring PSs was related to a lower risk of a first myocardial infarction (MI).¹¹ However, this association was not apparent when PS intake was corrected for energy intake and no significant associations were observed in women.

We aimed to prospectively investigate the association between intake of PSs from natural sources and occurrence of cardiovascular events (total CVD, total coronary heart disease (CHD) and MI). As secondary objectives, we cross-sectionally investigated the association between naturally occurring PS intake and blood lipid concentrations at baseline and whether associations between PS intake and CVD were mediated through effects on LDL-C.

Subjects and methods

Study population

The European Prospective Investigation into Cancer and Nutrition-the Netherlands (EPIC-NL) cohort¹² consists of two contributions to the EPIC collaboration; the Monitoring Project on Risk Factors for Chronic Diseases (MORGEN) cohort and the Prospect cohort. The MORGEN cohort consists of 22,654 men and women, aged 20–64 years, recruited through random sampling from the general population between 1993–1997. Prospect is a cohort study among 17,357 women, aged 49–70 years, recruited during the same time period (1993–1997) through a breast cancer screening programme. The procedures in both cohorts were set up simultaneously and using similar methods with the exception of the blood pressure assessment. The data have been harmonised and merged in one database in 2006. The study complies with the Declaration of Helsinki and was approved by local medical ethical committees. All participants provided informed consent before study inclusion.

For the prospective analysis, the following exclusion criteria were applied: prevalent CVD based on self-report or identified through linkage with the National Medical Registry (1990–1997) (n = 1264), missing dietary intake data (n = 203), having extremely low or high reported energy intakes (i.e. ratio of energy intake over basal metabolic rate in the lowest or highest 0.5%) (n = 385), and missing follow-up data (n = 2562). Thus, in total 35,597 participants were included. For the cross-sectional analysis, we only included participants of a random 6.5% sample (n = 2604) for which we had data on complete blood lipid profile. Similar exclusion criteria were applied as mentioned above, except that participants were excluded when blood lipid data instead of follow-up data on CVD endpoints were missing. The numbers of participants included in the blood lipid analyses were 2417 for TC, 2383 for LDL-C, 2383 for high-density lipoprotein cholesterol (HDL-C) and 2410 for triglycerides (TGs).

Baseline assessments

At baseline, participants filled out a general questionnaire on demographics, disease history and lifestyle characteristics, a physical activity questionnaire and a validated food-frequency questionnaire (FFQ).¹³ A physical examination was performed as earlier described¹² and non-fasting venous blood samples were drawn at baseline. Physical activity was assessed by calculating the Cambridge Physical Activity Score. Smoking was classified as current, past or non-smoker and education level was categorised based on nine categories ranging from completion of primary education to university. Menopausal status was classified as pre-, peri-, or (surgical) postmenopausal; men were considered postmenopausal. Diagnosis of hyperlipidaemia was determined based on self-report (‘ever diagnosed?’ yes/no), whereas hypertension was determined based on measured blood pressure (BP) (>140 mm Hg systolic or >90 mm Hg diastolic BP), use of BP-lowering medication or self-report.

Assessment of nutrient and phytosterol intake

The self-administered FFQ contained questions on consumption frequency of 79 main food items during the past year.¹³ Additional questions were asked about sub-items, preparation methods or additions. Consumption of in total 178 foods when considering the sub-items could be calculated. Portion sizes were estimated using specified units or photographed portions. Energy and nutrient intakes were calculated based on the Dutch food composition table. Because this table does not contain information on PS content of foods, we estimated total PS intake by using a PS database that was developed by Ghent University, Belgium,⁴ based on the Finnish, UK and USA food composition tables,^14–16 scientific literature,¹⁷ Dutch recipes,¹⁸ ingredient lists on packaging, and known PS composition of equivalent foods. Intake data of individual PSs, such as sitosterol or campesterol, were not available. PS-enriched foods were not available on the market at the time of the dietary intake assessment and information on consumption of such products during later years was not available.

We used data from a previous validation study¹³ among 63 men and 58 women to estimate the relative validity of the PS intake as measured with the FFQ against 12 standardised 24-hour recalls. Reproducibility was tested against two other FFQs taken at six-month intervals. We observed a reasonable to good relative validity of the estimated PS intake with Pearson correlation coefficients of 0.72 for the crude PS intake and 0.59 for the energy-adjusted PS intake. Reproducibility was good with Pearson correlation coefficients ranging between 0.84–0.87 for the crude PS intake and 0.68–0.69 for the energy-adjusted PS intake.

Assessment of blood lipids

Data on baseline blood lipids were available for a random 6.5% sample of the total study population (n = 2604) representative of the full cohort,¹² and for all CVD cases that occurred until January 2006 (n = 2068). Non-fasting TC and TG were measured using enzymatic methods. Non-fasting HDL-C and LDL-C were measured using a homogeneous assay with enzymatic endpoint, on an autoanalyser (Beckman Coulter, Mijdrecht, the Netherlands).

Follow-up assessments

Participants were followed for occurrence of chronic diseases and death. Vital status was obtained through linkage with municipal population registries. Causes of death were obtained via ‘Statistics Netherlands’. Data on morbidity were obtained from the Dutch Hospital Association and Order of Medical Specialists. Registries were linked to the cohort based on a validated probabilistic method.¹⁹ Follow-up was complete until January 2008. Incidences of fatal and non-fatal events were combined, taking only the first-occuring events into account. The CVD events were coded according to ICD-9. CVD was based on codes 410–414 (ischaemic heart disease), 427.5 (cardiac arrest), 428 (heart failure), 415.1 (pulmonary disease), 443.9 (unspecified peripheral vascular disease), 430–438 (cerebrovascular disease), 440–442 (atherosclerosis and aneurysms), 444 (arterial embolism and thrombosis) and 798.1, 798.2 and 798.9 (sudden death), CHD based on codes 410–414, 427.5, 798.1, 798.2 and 798.9 and acute MI based on code 410.

Data analysis

Person-years were calculated from the date of return of the questionnaire until the date of CVD occurrence, date of death or 1 January 2008, whichever came first. Data on physical activity were missing in 14% of all participants. Missing values for physical activity were therefore imputed using the single imputation method (SPSS Missing Value Analysis). For all other variables, the few missing values (<0.5%) were imputed using the mean for continuous variables and a missing indicator for categorical variables. Nutrients were adjusted for energy intake using the regression residual method.²⁰ Blood lipid variables were log transformed in case of non-normally distributed data.

Cox proportional hazard models were used to prospectively analyse associations between intake of naturally occurring PSs and risk of total CVD, total CHD and MI. Associations were analysed categorically based on quintiles of energy-adjusted PS intake with the lowest quintile as the reference. All analyses were stratified for cohort (i.e. MORGEN or Prospect). Associations were adjusted for confounders. The first model adjusted for age and gender. The second model additionally adjusted for CVD risk factors, i.e. body mass index (BMI), education, smoking status, physical activity, menopause and total energy intake. The third, fully-adjusted, model additionally adjusted for dietary factors known to affect blood lipids and/or CVD risk, i.e. energy-adjusted intakes of saturated, polyunsaturated and monounsaturated fat, fibre, dietary cholesterol and alcohol. Two additional models were investigated to explore possible confounding by intake of sodium, retinol, β-carotene, vitamin D and vitamin E (model 4) or by hypertension (model 5). Effect modification by gender, waist circumference and hyperlipidaemia was investigated by including interaction terms with PS intake in the third model. Exploratory analyses were performed with stroke as outcome variable. The proportionality assumption was checked in the final models. Sensitivity analyses were performed to ensure robustness of the findings. We checked the impact of censoring at 2000 (i.e. the year that PS-enriched foods were introduced onto the market), exclusion of participants with cancer or diabetes at baseline, exclusion of energy under- and over-reporters as determined by the Goldberg criteria,²¹ exclusion of participants with a survival time <2 years (i.e. any undiagnosed illness preceding the early censoring may have changed a partipants’ diet) and additional adjustment for diabetic status/drug use.

Associations between energy-adjusted PS intake and blood lipids at baseline were analysed using linear regression analysis based on the same models as defined for the prospective analysis. Effect modification by gender, waist circumference and hyperlipidaemia was tested. To investigate whether associations between PS intake and CVD risk were mediated through effects on LDL-C, we applied a case-cohort design including all cases until January 2006 and the random 6.5% sample for which we had LDL-C data. Modified Cox proportional hazard models were used accounting for case-cohort design by Prentice-weighting;²² LDL-C was included in the third model to assess its mediation effect. Even if associations would be non-significant, mediation analysis could reveal relevant information as long as the hazard ratio (HR) is not 1.00.

A p-value below 0.05 was considered statistically significant. All statistical analyses were performed using the statistical package SAS (SAS version 9.2, SAS Institute).

Results

Overview of study population

Of the 35,597 participants, 25% were men and 75% were women (Table 1). The average age was 49.3 years. After a median of 12.2 years of follow-up, 3047 cases of CVD were documented, including 1807 cases of CHD and 606 cases of MI. The average baseline energy-adjusted PS intake in the whole population was 295.8 ± 49.2 mg/d (mean ± standard deviation (SD)). Average PS intakes ranged from 231.3 ± 22.0 to 366.0 ± 34.9 mg/d between the lowest and the highest quintiles. The most important dietary sources of PSs were fruits and vegetables (25.5%), bread and cereal products (25.1%), and fats, oils and sauces (19.3%). With higher naturally occurring PS intakes, participants were younger, more often female, had higher BMI, were more physically active, were lower educated and smoked less (Table 1). Furthermore, intakes of carbohydrates, mono- and polyunsaturated fat, and fibre were higher, whereas intakes of protein, saturated fat, cholesterol and alcohol were lower with higher PS intakes.

Table 1.

Overview of the study population when classified into categories of energy-adjusted phytosterol (PS) intake.

Characteristics	Quintiles based on energy-adjusted PS intakea					All participants
	Q1 (<257 mg/d)	Q2 (257–282 mg/d)	Q3 (283–305 mg/d)	Q4 (306–333 mg/d)	Q5 (>333 mg/d)
Demographics
Total n	7120	7118	7121	7118	7120	35597
CVD cases	713 (10.0)	588 (8.3)	553 (7.8)	567 (8.0)	626 (8.8)	3047 (8.6)
CHD cases	436 (6.1)	327 (4.6)	333 (4.7)	359 (5.0)	352 (4.9)	1807 (5.1)
MI cases	154 (2.2)	121 (1.7)	115 (1.6)	103 (1.5)	113 (1.6)	606 (1.7)
Stroke cases	124 (1.7)	118 (1.7)	113 (1.6)	93 (1.3)	132 (1.9)	580 (1.6)
Cohort (MORGEN)b	3830 (53.8)	3962 (55.7)	4046 (56.8)	4076 (57.3)	3839 (53.9)	19,753 (55.5)
Age (years)	50.4 ± 11.7	49.2 ± 12.0	48.9 ± 12.0	48.6 ± 11.9	49.3 ± 11.5	49.3 ± 11.9
Gender (male)	1943 (27.3)	1755 (24.7)	1847 (25.9)	1789 (25.1)	1592 (22.4)	8926 (25.1)
BMI (kg/m2)	25.6 ± 4.0	25.5 ± 13.8	25.6 ± 4.0	25.6 ± 3.8	26.0 ± 4.2	25.6 ± 4.0
Waist circumference (cm)	86.0 ± 11.8	84.9 ± 11.2	85.1 ± 11.2	84.7 ± 11.1	84.9 ± 11.5	85.1 ± 11.4
Smoking status (non-smoker)	2367 (33.2)	2699 (37.9)	2878 (40.4)	2868 (40.3)	2884 (40.5)	13696 (38.5)
Physically active	2710 (38.1)	2960 (41.6)	3089 (43.4)	3105 (43.6)	3100 (43.5)	14964 (42.0)
Education (higher level)	1502 (21.1)	1571 (22.1)	1501 (21.1)	1407 (19.8)	1279 (18.0)	7260 (20.4)
Pre-menopausal status	1336 (18.8)	1687 (23.7)	1735 (24.4)	1788 (25.1)	1781 (25.0)	8327 (23.4)
Hypertension	2641 (37.1)	2567 (36.1)	2613 (36.7)	2585 (36.3)	2719 (38.2)	13125 (36.9)
SBP (mm Hg)	127.2 ± 19.2	126.3 ± 19.0	126.3 ± 18.9	125.8 ± 18.3	126.2 ± 19.0	126.4 ± 18.9
DBP (mm Hg)	78.1 ± 10.7	77.7 ± 10.5	77.9 ± 10.7	77.8 ± 10.5	78.1 ± 10.6	77.9 ± 10.6
Hyperlipidaemia	522 (7.3)	502 (7.1)	532 (7.5)	611 (8.6)	607 (8.5)	2774 (7.8)
Dietc
Total energy intake (kcal/d)	2026 ± 595	2046 ± 584	2077 ± 593	2082 ± 609	2024 ± 640	2051 ± 605
Total carbohydrate intake (g/d)	214.2 ± 34.5	221.7 ± 30.3	223.5 ± 28.6	224.5 ± 28.6	224.9 ± 29.6	221.8 ± 30.7
Total protein intake (g/d)	78.0 ± 12.3	77.0 ± 10.9	76.1 ± 10.4	75.0 ± 10.0	73.1 ± 10.4	75.9 ± 11.0
Total fat intake (g/d)	76.1 ± 11.9	76.6 ± 10.9	77.4 ± 10.7	78.3 ± 11.0	80.1 ± 11.6	77.7 ± 11.3
SFA intake (g/d)	34.3 ± 6.6	32.9 ± 5.7	32.4 ± 5.4	32.0 ± 5.4	31.6 ± 5.8	32.6 ± 5.8
MUFA intake (g/d)	28.9 ± 5.2	29.1 ± 4.9	29.4 ± 4.9	29.6 ± 5.1	30.3 ± 5.3	29.5 ± 5.1
PUFA intake (g/d)	12.2 ± 2.9	14.0 ± 3.0	15.0 ± 3.2	16.0 ± 3.6	17.5 ± 4.2	14.9 ± 3.9
PS intake (mg/d)	231.3 ± 22.0	270.1 ± 7.3	293.4 ± 6.7	318.2 ± 8.1	366.0 ± 34.9	295.8 ± 49.2
Fibre intake (g/d)	19.9 ± 4.1	22.3 ± 3.9	23.7 ± 4.1	24.7 ± 4.2	26.4 ± 4.9	23.4 ± 4.8
Cholesterol intake (mg/d)	237.8 ± 63.8	221.1 ± 55.3	214.0 ± 54.3	210.4 ± 54.9	204.7 ± 59.6	217.6 ± 58.8
Alcohol intake (g/d)	17.0 ± 23.4	11.8 ± 16.7	9.9 ± 15.1	8.8 ± 13.2	7.7 ± 13.2	11.0 ± 17.1

BMI: body mass index; CHD: coronary heart disease; CVD: cardiovascular disease; DBP: diastolic blood pressure; MI: myocardial infarction; MORGEN: Monitoring Project on Risk Factors for Chronic Diseases; MUFA: monounsaturated fatty acid; PUFA: polyunsaturated fatty acid; SBP: systolic blood pressure; SFA: saturated fatty acid.

a

p-value for trend was <0.001, except for prevalence of stroke cases, DBP, prevalence of hypertension and total energy intake.

b

MORGEN is the name of one of the two cohorts that were part of this study.

c

All nutrients, except for total energy intake, were energy-adjusted.

Values are mean ± standard deviation (SD) or n (%).

Cardiovascular disease risk

In the fully-adjusted model (Table 2), no association was observed between energy-adjusted intake of naturally occurring PSs and total CVD risk (p_trend = 0.94) with non-significant HRs ranging between 0.90 and 0.99 across quintiles of PS intake. PS intake was also not associated with total CHD risk (p_trend = 0.17); however, PS intake was significantly associated with a lower risk of CHD in the second (HR = 0.83; 95% confidence interval (CI): 0.72–0.97) and in the third (0.84; 95% CI: 0.72–0.98) quintiles of PS intake vs the quintile with the lowest PS intake. In the fourth and fifth quintiles, HRs for CHD were 0.90 (95% CI: 0.76–1.06) and 0.84 (95% CI: 0.70–1.01). PS intake was not associated with MI risk (p_trend = 0.19) after adjustment for confounders; non-significant HRs ranged from 0.80–0.95 across quintiles of PS intake.

Table 2.

Energy-adjusted phytosterol (PS) intake and risk of cardiovascular disease, coronary heart disease and myocardial infarction in the European Prospective Investigation into Cancer and Nutrition-the Netherlands (EPIC-NL) cohort.

	Quintiles based on energy-adjusted PS intake					p-value for trend
	Q1	Q2	Q3	Q4	Q5
	(<257 mg/d)	(257–282 mg/d)	(283–305 mg/d)	(306–333 mg/d)	(>333 mg/d)
n	7120	7118	7121	7118	7120
HR (total CVD)
Model 1a	1.0	0.90	0.85	0.88	0.98	0.6033
		(0.80–1.00)	(0.76–0.95)	(0.79–0.98)	(0.88–1.09)
Model 2b	1.0	0.96	0.92	0.95	1.03	0.7334
		(0.86–1.07)	(0.82–1.03)	(0.85–1.06)	(0.92–1.15)
Model 3c	1.0	0.95	0.90	0.92	0.99	0.9401
		(0.84–1.06)	(0.79–1.02)	(0.81–1.05)	(0.86–1.14)
HR (total CHD)
Model 1	1.0	0.82	0.84	0.92	0.91	0.4232
		(0.71–0.95)	(0.73–0.97)	(0.80–1.06)	(0.79–1.04)
Model 2	1.0	0.88	0.90	0.98	0.94	0.7476
		(0.76–1.01)	(0.78–1.04)	(0.85–1.13)	(0.82–1.08)
Model 3	1.0	0.83	0.84	0.90	0.84	0.1722
		(0.72–0.97)	(0.72–0.98)	(0.76–1.06)	(0.70–1.01)
HR (MI)
Model 1	1.0	0.88	0.83	0.76	0.84	0.0763
		(0.69–1.11)	(0.65–1.06)	(0.59–0.97)	(0.66–1.07)
Model 2	1.0	0.99	0.95	0.86	0.93	0.3267
		(0.78–1.26)	(0.74–1.21)	(0.66–1.10)	(0.72–1.18)
Model 3	1.0	0.95	0.90	0.80	0.84	0.1878
		(0.74–1.22)	(0.68–1.17)	(0.59–1.06)	(0.62–1.15)

BMI: body mass index; CHD: coronary heart disease; CVD: cardiovascular disease; HR: hazard ratio; MI: myocardial infarction.

a

Model 1: corrected for age, gender and cohort (only for women).

b

Model 2: corrected for variables in model 1 + BMI, smoking status, education, physical activity level, menopausal status (only for women) and total energy intake.

c

Model 3: corrected for variables in model 2 + intake of saturated, polyunsaturated and monounsaturated fat, dietary cholesterol, fibre and alcohol.

Values are HR (95% confidence interval).

Models 4 and 5 showed essentially similar results indicating that a possible relation was not obscured by confounding of other dietary factors or hypertension. Interactions of PS intake with gender, waist circumference or hyperlipidaemia were not statistically significant. No associations were observed between PS intake and occurrence of stroke (see Supplementary Material, Appendix 1). In sensitivity analyses, censoring the analysis at year 2000, excluding participants with cancer or diabetes at baseline, excluding energy under- and over-reporters, excluding participants with a survival time <2 years and adjusting additionally for diabetic status/drug use did not change our results (data not shown).

Blood lipids

In the fully-adjusted model, energy-adjusted intake of naturally occurring PSs was significantly, inversely associated with TC, LDL-C and HDL-C (p < 0.05); each 50 mg/d incremental PS intake was significantly associated with a 0.06 mmol/l (95% CI: −0.11–0.00) lower TC, a 0.07 mmol/l (95% CI: −0.11; −0.02) lower LDL-C, and a 0.02 mmol/l (95% CI: −0.04–0.00) lower HDL-C. Furthermore, a significant association was observed between PS intake and TG concentrations (0.04; 95% CI: 0.01–0.06). Effect modification by gender was significant for LDL-C. When stratifying according to gender, PS intake was more strongly inversely associated with LDL-C in men (−0.18, 95% CI: −0.29– −0.08) than in women (−0.03, 95% CI: −0.08–0.03). Interactions of PS intake with waist circumference or hyperlipidaemia were not significant for the lipid parameters. An overview of the associations with blood lipids is provided in Supplementary Material, Appendix 2.

In mediation analysis, LDL-C hardly changed the association between PS intake and cardiovascular risk; the mediation effect was low for each quintile and at maximum 5% for total CVD, 3% for total CHD and 6% for MI risk. When analysing the mediation effect of LDL-C separately for men and women, we observed similar results.

Discussion

In this large cohort of 35,597 Dutch men and women, we observed no association between energy-adjusted intake of PSs from natural sources and CVD risk during 12 years of follow-up. However, higher naturally occurring PS intake was significantly associated with lower TC and LDL-C concentrations at baseline, particularly among men.

Intake of 2 g per day of PS has been shown to lower LDL-C by on average 10%.¹ Based on data from statin trials,²³ such a reduction in LDL-C could potentially reduce the absolute risk of CHD by ∼9%. Considering that the intakes of naturally occurring PSs are much lower than 2 g/d (i.e. on average 296 mg/d in the current study), only small risk reductions were expected: ∼2% lower CHD risk given the predicted LDL-C lowering effect for a difference in PS intake of 150 mg/d between the highest and lowest quintiles of PS intake or ∼4% lower CHD risk given the observed ∼5% lower LDL-C concentration between the highest and lowest quintiles of PS intake. In the current study, we observed surprisingly strong CHD hazard ratios ranging between 0.83–0.90 across quintiles of phytosterol intake, but these were not all statistically significant. Klingberg et al. recently showed in a nested case-referent study¹¹ that a high absolute PS intake was related to a reduced risk of a first MI in men with a 29% lower risk in the highest vs the lowest quartile. However, when corrected for total energy intake, this association of PS intake with MI was not significant anymore. In women, neither the absolute nor the energy-adjusted PS intakes were associated with risk of MI.¹¹ In our opinion, adjustment for total energy intake is required, since associations of PS intake with CVD risk may easily be confounded by energy intake. All in all, the findings of the current study are in line with previous investigation.¹¹

The significant associations between intake of naturally occurring PSs and lower TC and LDL-C concentrations were also found in previous observational studies with similar ranges of naturally occurring PS intakes.^8–10,24,25 It should be noted that in our study, the association with TC and LDL-C was only present in men whereas evidence from randomised controlled trials have shown that TC and LDL-C are lowered in both men and women.^1,2 It is not clear why this discrepancy exists. The association observed between PS intake and lower HDL-C concentrations was also found in other population studies, with some studies showing more pronounced effects in women¹⁰ (similar to our observation) whereas other studies showed more pronounced effects in men.^8,9 As randomised controlled trials clearly show that HDL-C concentrations are not changed upon PS intervention,²⁶ it might be that residual confounding has played a role in this association.

The mechanism by which PSs are expected to reduce CVD risk is their LDL-C-lowering effect. However, in our population with a relatively narrow range of low naturally occurring PS intakes, mediation analysis did not support the view that low dietary PS intakes are associated with reduced CVD risk through reductions in LDL-C. Whether higher intakes of PSs would eventually be significantly associated with reduced CVD risk through effects on LDL-C has yet to be investigated. This should preferably be done in populations with higher and broader ranges of PS intakes, for example by including people with diets containing predominantly rich sources of PSs (e.g. cereal products and vegetable oils) and users of foods enriched with PS. Users of PS-enriched foods consume much higher amounts of PS (∼1.0–1.3 g/d) and seem to have lower TC concentrations vs. non-users after five years of follow-up.^27,28

Strengths of this study include the large sample size and its continuous, prospective, and almost complete, follow-up for disease occurrence, but there are also some limitations. First, the intakes of PS from natural sources were low within a relatively narrow range, thereby limiting the capacity to detect an association between dietary PS intake and CVD risk. Second, dietary intake was assessed only at baseline. It cannot be ruled out that participants changed their dietary behaviours during follow-up thereby influencing the occurrence of disease and, thus, the findings of this study. However, excluding participants that most likely changed their dietary habits (those with chronic diseases at baseline and cases occurring during the first two years) did not alter our findings. Furthermore, assessment of the long-term reproducibility of the FFQ in the EPIC-Heidelberg cohort showed fairly high correlation between dietary assessments at baseline and at follow-up.²⁹ Related to this, we cannot exclude that our findings may be confounded by a small part of the study population that started using PS-enriched foods or cholesterol-lowering medication during follow-up. Although the proportion of people consuming PS-enriched foods was only ∼6% in a subset of our study population,²⁸ these foods contain high concentrations of PS and can therefore contribute considerably to the daily intake of PS. A sensitivity analysis with follow-up until 2000 (i.e. the year that PS-enriched foods were introduced onto the European market) however did not reveal different results. The influence of cholesterol-lowering medication use during follow-up could not be tested in sensitivity analysis and remains a limitation of our study. Third, because a national database with PS composition data did not exist for the Netherlands, a specific database was developed for the analysis.⁴ Although this database was developed with utmost care, some misclassification of the level of PS exposure may have occurred due to incomplete information on PS content in foods. Finally, food intake was estimated with FFQs that are vulnerable for misreporting. Exclusion of misreporters in sensitivity analysis did however not affect the results. Moreover, the main dietary PS sources (fruits and vegetables, cereal products and vegetable oils) and the average PS intakes in the current study were comparable to those observed in other populations.^3,4,30 Additionally, we could demonstrate good relative validity and reproducibility of the PS intake estimated with our FFQ.

In summary, intake of PSs from natural sources was not associated with a reduced CVD risk despite a lower LDL-C concentration particularly in men. Future studies should preferably investigate the association between PS intake and CVD risk in populations with higher and broader ranges of PS intake.

Acknowledgements

The authors would like to thank Statistics Netherlands and the PHARMO Institute for data on vital status and the incidence of non-fatal cardiovascular diseases. The results from the current study were presented at American Heart Association Cardiovascular Disease Epidemiology and Prevention – Nutrition, Physical Activity and Metabolism (AHA EPI-NPAM) meeting (18th of March 2014), San Francisco, USA; EuroPrevent meeting (10th of May 2014), Amsterdam, the Netherlands.

Funding

The EPIC-NL study was funded by the European Commission (the Public Health and Consumer Protection Directorate 1993-2004), Research D-G 2005, Dutch Ministry of Public Health, Welfare and Sports, Dutch Prevention Funds, Dutch Zorg Onderzoek Nederland and the World Cancer Research Fund (the Netherlands). No financial support from Unilever was provided for this research.

Conflicts of interest

RTR, EAT and PLZ are employees of Unilever Research and Development Vlaardingen. IS is financially supported by the Research Foundation-Flanders (Grant no. 1.2.683.11.N.00). YTS, GWD and JWJB have no conflicts of interest.