Nutrient v. food and diet-based recommendations for health
The hierarchy of evidence to support the relationship between diet and disease can be viewed as a pyramid of study designs, from which the strength of evidence increases from relatively weak cross-cultural and observational studies at its base, through prospective cohort trials, to the much stronger intervention studies at its peak (Fig. 1). While prospective cohort studies have provided a wealth of evidence to support associations between specific nutrients and disease endpoints( Reference Jakobsen, O'Reilly and Heitmann 1 – Reference Micha and Mozaffarian 3 ), the findings are prone to severe confounding by other categorical and continuous variables. Cohort studies( Reference Jakobsen, O'Reilly and Heitmann 1 , Reference Siri-Tarino, Sun and Hu 2 ) have also given little address to the impact of specific foods or whole diets, such that there is now a general lack information on how diets of different types influence cardiovascular health. In contrast, a relatively small number of dietary intervention trials have provided valuable information on mechanisms to support the cause and effect relationships with CVD; although the majority of interventions with individual nutrients have produced neutral or negative findings. Dietary interventions provide information relating to the feasibility of achieving dietary change within free-living populations, but are often restricted by their reliance on indirect biomarkers, rather than clinical endpoints of CVD.
Discordance between nutrient, food and dietary-based associations with CVD
Nutrients are ingested as complex mixtures in foods, which are consumed as mixed meals and diets. The biological effects of a nutrient in a food may be altered through interaction with other constituents within the food or other foods in the overall diet. It follows that because a food contains a nutrient, the food may not necessarily exert the effect typical of the single nutrient. This provides one obvious explanation for why associations between single nutrients and disease, which have been unequivocally confirmed in meta-analyses of observational cohort studies, do not necessarily translate into predictable outcomes in intervention studies with whole diets. The disparity between the impacts of single nutrients, foods and diets on health is highlighted by the potential health benefits of foods containing micronutrients, such as β-carotene and vitamin E( Reference Ness and Powles 4 , Reference Liu, Manson and Lee 5 ), and the neutral and negative outcomes from secondary prevention intervention trials with these single antioxidant micronutrients( Reference Miller, Pastor-Barriuso and Dalal 6 – Reference Eidelman, Hollar and Hebert 9 ). It is also evident from the inconsistency between the established cholesterol-raising effects of dietary SFA in experimental feeding studies, and effects produced by some foods that are rich in saturated fat, such as dairy products. There is emerging evidence that certain dairy foods produce little or no significant effect on serum LDL cholesterol, and as a food group, are associated with lower CVD mortality and associated risk factors, such as blood pressure( Reference Soedamah-Muthu, Ding and Al-Delaimy 10 , Reference Engberink, Hendriksen and Schouten 11 ). These findings suggest that recommendations based on single nutrients, particularly when additional nutrients are provided in the supplemental form, may be inappropriate for cardiovascular health. Guidelines based on foods, food groups and ultimately whole diets, may offer a more rational and comprehensive approach for the prevention of CVD. However, a stronger research base, grounded in a combination of observational and intervention data is required to substantiate the relationship between potentially beneficial diets and disease risk.
Existing evidence for the benefits of dietary patterns in reducing CVD risk
The Lyon Heart Study is one of the most successful trials to date to demonstrate the benefit of a dietary pattern in preventing CVD. It was a randomly controlled, secondary prevention study designed to examine the effects of a Mediterranean type of diet over 5 years in 605 patients who had suffered a previous myocardial infarction( Reference De Lorgeril, Salen and Martin 12 ). When primary and secondary endpoints were combined, the Mediterranean diet which was characterised by increased bread, root and green vegetables, fruit, fish, less red meat and fats high in olive oil and α-linolenic acid, produced an impressive 76% reduction in risk relative to the control diet, which was typical of that consumed in the UK and North America. More recently, when a typical Mediterranean diet was supplemented with either extra-virgin olive oil or nuts in a Spanish population, both experimental diets (extra-virgin olive oil n 2543 and nuts n 2454) produced highly significant reductions in the primary endpoint, which was a composite of myocardial infarction, stroke and death from cardiovascular causes over a follow-up of 4·8 years, in comparison with a control diet (n 2450)( Reference Estruch, Ros and Salas-Salvadó 13 ). However, in the Heart Institute of Spokane Diet Intervention and Evaluation Trial, in which an intervention with a Mediterranean-style diet was shown to be equivalent to a low-fat diet in increasing cardiovascular-event-free survival after myocardial infarction, close examination of the dietary data reveals that the Mediterranean-style diet targets were not achieved( Reference Tuttle, Shuler and Packard 14 ). In particular, the target of a two-fold increase in monounsaturated fat intake was not achieved, and at the end of the intervention with a Mediterranean-style diet, total fat intake and fatty acid profile were identical to that of the low-fat diet, apart from slightly higher intakes of ‘omega-3 fats’. These data and the difficulties inherent in ensuring compliance with specified dietary targets over a prolonged period, amply illustrate the challenges faced with these types of studies.
The Dietary Approaches to Stop Hypertension (DASH) trial has produced some of the most conclusive evidence for the benefits of a dietary pattern on blood pressure and other CVD risk factors. DASH examined the effects of two experimental diets against a control diet for 8 weeks, in 459 adults a proportion of which suffered moderate hypertension. The experimental diets consisted of either a daily intake of between eight and ten servings of fruits and vegetables, or a combination of this diet, plus 2·7 portions of low-fat dairy products, and reduced total and saturated fat. Sodium and body weight were maintained throughout the intervention( Reference Appel, Moore and Obarzanek 15 ). While both experimental diets produced dramatic reductions in systolic and diastolic blood pressure, the combination diet produced by far the greatest reductions in the total cohort (Fig. 2), and especially in a subgroup with hypertension. Other favourable effects included a significant increase in serum folate, and decreases in homocysteine( Reference Appel, Miller and Jee 16 ) and total serum and LDL cholesterol, but notably a potentially detrimental lowering of HDL cholesterol( Reference Obarzanek, Sacks and Vollmer 17 ). Predictably, the later addition of a low sodium intake to a DASH-type diet, showed even greater efficacy in lowering blood pressure, the population-wide application of which was estimated to translate into decreased risk of CHD and stroke of 15 and 27%, respectively( Reference Sacks, Svetkey and Vollmer 18 ). The impact of the DASH diet on the predicted 10-year risk of developing CHD has been calculated by the Framingham risk equations( Reference Chen, Maruthur and Appel 19 ). In comparison with the control and fruit and vegetable-rich diets, the DASH diet (rich in fruits, low-fat dairy, reduced total and saturated fat) was shown to reduce estimated 10-year CHD risk by 18 and 11% respectively, with evidence of a significantly greater reduction in CHD risk in Black African Americans. These findings provide further evidence to underpin national dietary guidelines in North America, which had already adopted the principles of the DASH diet some 5 years before this report.
Dietary pattern analysis: a new direction in nutritional epidemiology?
There are two main approaches for examining the relationship between dietary patterns and CVD. The first relies on the empirical application of cluster and principal components analyses to identify associations between dietary patterns and disease. This method has been used widely in cross-sectional studies to generate hypotheses on the potential effectiveness of dietary patterns on CVD reduction( Reference Bhupathiraju and Tucker 20 ). The second approach is based on the calculation of a dietary score or index that provides a summary of diet quality or adherence to specific dietary recommendations, the most established examples of which are scores for the DASH and Mediterranean diets. To investigate the long-term efficacy of the DASH diet on endpoints of CHD and stroke, a DASH diet score based on eight foods and nutrients( Reference Fung, Chiuve and McCullough 21 ) was calculated from FFQ on seven occasions, over 24 years in the Nurses’ Health Study. In this prospective, cohort study of 88 517 female nurses aged 34–59 years, the DASH diet was found to be associated with reductions in the relative risk for non-fatal and fatal CHD of 22 and 34%, respectively, and a 17% reduction in risk of stroke, across increasing quintiles of the DASH-diet score, after adjustment for age, smoking and other CVD risk factors. Other dietary scores include the Healthy Eating Index( Reference McCullough, Feskanich and Stampfer 22 ) and Recommended Food Score( Reference Kant, Schatzkin and Graubard 23 ). Diet Quality Index( Reference Patterson, Haines and Popkin 24 ) and the Mediterranean Diet Score( Reference Bach, Serra-Majem and Carrasco 25 ). A recent systematic review concluded that greater adherence to a Mediterranean-like dietary pattern was associated with significant reductions in overall mortality (9%) and mortality from CVD (9%; Fig. 3), cancer (6%), Parkinson's and Alzheimer's diseases (13%)( Reference Sofi, Cesari and Abbate 26 ).
Despite impressive evidence for the efficacy of dietary scores, one major drawback is their dependence on the current understanding of the relationship between diet and disease, making it impossible to identify and account for previously unknown factors. While it might seem intuitive that the overall effects of diet on CVD will be greater than that of single nutrients, there are coherent statistical and biological reasons to support this view. False-positive associations between a single nutrient and CHD can arise because the former acts as surrogate marker for other dietary components. A classic example of this would be dietary cholesterol in eggs. While consumption of eggs has been shown to be unrelated to CHD and stroke( Reference Scrafford, Tran and Barraj 27 , Reference Rong, Chen and Zhu 28 ), positive associations persist, in part, because eggs represent an easily quantified marker of a diet that is invariably energy dense and high in saturated fat. Likewise, multiple analyses of individual nutrients may also lead to false-positive findings simply by chance. Conversely, false-negative findings may arise when the effects of a single nutrient are confounded by other nutrients in the foods in which they are contained, through errors associated with its measurement and by misclassification of foods in FFQs. Although diet scores are less susceptible to confounding, they are subject to the same measurement errors, misclassifications and recording biases that characterise all diet and nutrient assessments, as well as an inherent inability to identify mechanisms that underlie the relationship between diet and disease. However, the classification of individuals according to dietary patterns which quantify the sum total effects of dietary bio-actives, is likely to result in the estimation of larger effect sizes due to the dietary exposure representing the cumulative, and possibly, synergistic effects of individual components. From a statistical perspective, there is also likely to be smaller measurement error, risk of confounding and greater chance of observing significant associations.
From a biological standpoint, atherosclerosis and thrombosis are complex, multi-factorial processes, which may be affected by nutrients, acting synergistically within whole diets, at numerous sites and at various stages of the disease. As well as their high potential for confounding and measurement type errors, investigation of single nutrient–disease relationships are aiming to delineate a single cause–effect pathway amongst an array of diet-related pathological processes. It is therefore unsurprising to find that many single nutrient intervention studies have failed to confirm the hypothetical relationship that has been demonstrated in observational studies.
What is the future for nutrition research?
Given the very strong evidence from cohort studies for the benefits of specific dietary patterns in prevention of CVD, the paucity of well-designed interventions with hard endpoints, and major implications for public health, there is arguably an urgent need for more research in this area. Dietary intervention trials, using approaches which can be feasibly achieved in free-living populations are required, not only to elucidate mechanisms of action of nutrients and food bio-actives in health and disease, but also to underpin future nutritional policy, which must recognise the limitations of single nutrient-based recommendations for cardiovascular health. High-quality randomised controlled dietary-intervention studies will always face the inevitable challenges associated with their prohibitive cost, and the feasibility of delivering a lifestyle modification in a free-living setting. To succeed, they must overcome behavioural barriers to dietary change, and the confounding influence of inter-individual variation in dietary response. Ideally they should include disease endpoints as the primary outcome, and above all, involve whole-diet interventions.
Financial support and conflicts of interest
There was no financial support for the writing of this article, and there are no conflicts of interest.
Author contribution
The manuscript is an overview of a presentation by Professor Christine Williams at the Winter Meeting of the Nutrition Society at the Royal Society of Medicine, London in December 2012. The manuscript was composed jointly by the three authors, C. M. W., J. A. L. and B. A. G., and was prepared and submitted by B. A. G.