The recommendation to lower saturated fat in the diet for the prevention of CHD has recently been challenged(Reference Siri-Tarino, Sun and Hu1). In their meta-analysis of prospective cohort studies, Siri-Tarino et al. (Reference Siri-Tarino, Sun and Hu2) found that dietary SFA were not associated with an increased risk of CHD. The same authors(Reference Siri-Tarino, Sun and Hu2) argued that substitution of saturated fat by carbohydrates, especially refined carbohydrates, may actually increase the risk of CHD. They attributed this to differential effects of dietary saturated fats and carbohydrates on concentrations of larger and smaller LDL particles and concluded that replacement of saturated fats by carbohydrates may increase CHD risk through atherogenic dyslipidemia(Reference Siri-Tarino, Sun and Hu2).
Besides the discussion on the controversial role of dietary saturated fat, the optimal amount of total fat and individual fatty acids for CHD prevention is also debated in the scientific community. Do we need specific quantitative recommendations for total fat and individual fatty acids, or do quantitative recommendations for some fatty acids and general statements for others suffice? The latter approach was taken in the Dietary Reference Values for fats of the European Food Safety Authority (EFSA)(3). Examples of the EFSA recommendations are ‘the lower the better’ for saturated fat, no recommendation for cis-MUFA and ‘at least 250 mg/d’ for the n-3 fatty acids EPA and DHA.
In this review, we challenge the interpretation of Siri-Tarino et al. about the absence of association between saturated fat and risk of CHD in prospective cohort studies, discuss the effect on CHD incidence of replacement of saturated fat by cis-unsaturated fat v. carbohydrates in controlled dietary experiments, and provide recommendations for fatty acids and dietary patterns for CHD prevention.
Lack of association between dietary saturated fat and CHD risk
The recently published meta-analysis of sixteen prospective cohort studies by Siri-Tarino et al. (Reference Siri-Tarino, Sun and Hu1) has not provided evidence that a high intake of saturated fat is associated with an increased risk of CHD. This meta-analysis included 214 182 subjects who were followed up for 5–23 years and developed 8644 cases of CHD. The median or mean of saturated fat intake in these studies varied between 12 and 20 % of energy. The pooled relative risk comparing extreme quantiles of saturated fat was 1·07 (95 % CI 0·96, 1·19; P = 0·22) for CHD. This finding is discordant with the classic diet–heart hypothesis that a high saturated fat intake increases the risk of CHD, mediated by raised serum cholesterol.
In his editorial accompanying the article by Siri-Tarino et al., Stamler(Reference Stamler4) argues that the diet–heart hypothesis is supported by a vast array of concordant evidence from multidisciplinary research. A major issue in this context is the accuracy of dietary data in epidemiological studies on saturated fat and CHD. Balogh et al. (Reference Balogh, Kahn and Medalie5) had already shown many years ago that twenty-two randomly collected 24-h dietary recalls are required to estimate the true individual mean intake within ± 20 %, while most studies have only one recall or a food frequency measure of saturated fat intake available. Therefore, the weak associations found may be explained by unreliability of this aspect of dietary information in observational studies. We discuss in more detail the methodological aspects related to dietary saturated fat, serum cholesterol and CHD.
Parallel to the lack of association between saturated fat intake and CHD risk, as shown in the meta-analysis, is the lack of association between dietary saturated fat and serum cholesterol in cross-sectional analyses of the Framingham, Tecumseh and Zutphen studies(Reference Kannel and Gordon6–Reference Kromhout8). Keys(Reference Keys9) explained the lack of an association between dietary saturated fat and serum cholesterol in cross-sectional studies by the large day-to-day variation within individuals in both saturated fat intake and serum cholesterol. He showed that the intra-individual variance of the SFA palmitic acid was more than twice as large as the inter-individual variance, based on two measurements. He also found that in healthy adults on an ostensibly constant diet, the average intra-individual standard deviation of serum cholesterol was approximately 200 mg/l (20 mg/dl), about half the total standard deviation(Reference Keys9).
The mathematical aspects of the zero or low-level correlation between dietary saturated fat and serum cholesterol were dealt with in a paper by Jacobs et al. (Reference Jacobs, Anderson and Blackburn10). Even with a fixed diet, serum cholesterol will vary due to differences in blood sampling, chemical analysis and variation in cholesterol levels unrelated to diet. Besides the variability in serum cholesterol there is also substantial variability in estimating dietary fatty acids. Jacobs et al. described the following sources of variation: (1) errors in identifying foods in food tables, (2) discrepancy between food table values and the true composition in foods eaten, (3) errors in estimating quantities of food eaten, (4) errors in remembering what was eaten and (5) differences in the food pattern of the observation period and that of the previous 2–4 weeks. These sources of error in both the dietary exposure and the effect measure will attenuate the correlation between dietary saturated fat and serum cholesterol. The attenuation of the true correlation is determined by the ratio of the variance between intra- and inter-individual variations. The larger the intra-individual variation, the larger the error term and the smaller the observed correlation.
It is therefore not a surprise that zero correlations were observed between dietary saturated fat and serum cholesterol in cross-sectional analyses. Jacobs et al. (Reference Jacobs, Anderson and Blackburn10) stated in their article ‘A corollary of the mathematical model here presented is that a correlation close to zero would likely be observed between diet e.g. dietary saturated fat and coronary heart disease incidence.’ The results of the meta-analysis by Siri-Tarino et al. (Reference Siri-Tarino, Sun and Hu1) are an illustration of their prophecy. Jacobs et al. (Reference Jacobs, Anderson and Blackburn10) concluded their article by saying that ‘An appropriate design for demonstrating or refuting diet and coronary heart disease incidence is a dietary experiment.’
Effect of different SFA on total and HDL-cholesterol
The classical controlled dietary experiments carried out before 1970 showed that replacing saturated fat, generally by starch or sucrose, decreased serum cholesterol, while replacing starch or sucrose by polyunsaturated fat also decreased serum cholesterol(Reference Keys, Anderson and Grande11, Reference Hegsted, McGandy and Myers12). The serum cholesterol-raising effect of saturated fat was twice as strong as the decreasing effect of polyunsaturated fat. Controlled dietary experiments carried out after 1970 showed a somewhat smaller serum cholesterol-raising effect when carbohydrate was replaced by saturated fat and only half of the serum cholesterol-decreasing effect after replacement of carbohydrate by polyunsaturated fat(Reference Mensink and Katan13). This was recently confirmed by the large multi-centre Reading University, Imperial College London, Surrey University, MRC Human Nutrition Research Cambridge and King's College London (RISCK) trial(Reference Jebb, Lovegrove and Griffin14).
A recent meta-analysis of controlled feeding experiments showed that SFA with twelve, fourteen and sixteen carbon atoms in contrast to the one with eighteen carbon atoms increased LDL-cholesterol when they isoenergetically replaced carbohydrate(Reference Micha and Mozaffarian15). All four SFA raised HDL-cholesterol, but the HDL-cholesterol-raising effect was greater as the chain length decreased. Overall, the total:HDL cholesterol ratio is not affected by the SFA with fourteen, sixteen and eighteen carbon atoms, but is significantly reduced when the SFA with twelve carbon atoms replaces carbohydrate. However, replacement of SFA with 12–18 carbon atoms by cis-MUFA- and -PUFA leads to lowering of total and LDL-cholesterol, while only slightly lowering HDL-cholesterol, and thus improving the total:HDL cholesterol ratio and CHD risk(Reference Mensink, Zock and Kester16).
Besides the effect of the chain length there may also be an effect of the source (e.g. animal v. plant origin and natural v. interesterified) of saturated fat on total and HDL-cholesterol. This was reviewed by Hayes & Pronczuk(Reference Hayes and Pronczuk17). They noted that typical diets provide 2–4 % of energy as stearic acid (18 : 0) from natural fats. When a hardened fat is needed (replacing food applications that until recently used trans-fats), an unmodified saturated fat, for example from a palm or coconut product, was seen as preferable to interesterified fat(Reference Hayes and Pronczuk17). At 2–4 % interesterified 18 : 0, effects on total and HDL-cholesterol and on glucose and insulin metabolism, immune function and liver enzymes are small. Detection of adverse effects starts at only approximately 7–8 % of energy or higher, although, similar to trans-fat, adverse effects of lower levels of interesterified fatty acids on other body systems could not be ruled out(Reference Hayes and Pronczuk17). We agree with Hayes & Pronczuk(Reference Hayes and Pronczuk17) that interesterified fatty acids should be used sparingly until more evidence about their health effects has become available.
Replacement of saturated fat by cis-unsaturated fat or carbohydrates and CHD risk in controlled dietary experiments
Different strategies are available for replacing the energy lost when lowering saturated fat intake. Among these strategies, saturated fat can be replaced by either cis-unsaturated fatty acids or carbohydrates. In most controlled dietary experiments of CHD risk, saturated fats were replaced by polyunsaturated vegetable oils(Reference Sacks and Katan18). The PUFA studied were mainly n-6 fatty acids (linoleic acid) and small amounts of the n-3 fatty acid α-linolenic acid (ALA) in some cases, e.g. when soyabean oil was used. Meta-analyses of short-term controlled dietary experiments lasting generally 4–6 weeks showed that these PUFA favourably influence the LDL:HDL cholesterol ratio(Reference Mensink and Katan13, Reference Mensink, Zock and Kester16). This ratio is a better predictor of CHD risk than total cholesterol or the individual lipoprotein fractions(Reference Mensink and Katan13).
Between 1968 and 1992 eight controlled dietary experiments of more than 1-year duration with hard coronary end points were reported(Reference Mozaffarian, Micha and Wallace19). In these trials, control diets were characterised by both a high total fat (35–45 % of energy) and a high saturated fat content (approximately 20 % of energy). The average polyunsaturated fat consumption was 15 % of energy in the intervention groups and 5 % of energy in the control groups. Replacement of saturated by polyunsaturated fat changed the polyunsaturated:saturated (P:S) ratio from approximately 0·2 to 2. The overall pooled risk reduction in CHD incidence was 19 % corresponding to 10 % reduced risk per 5 % energy of increased polyunsaturated fat intake. Study duration was an independent determinant of risk reduction, with studies of longer duration showing greater benefits(Reference Mozaffarian, Micha and Wallace19).
Ramsden et al. (Reference Ramsden, Hibbeln and Majchrzak20) concluded from the meta-analysis by Mozaffarian et al. (Reference Mozaffarian, Micha and Wallace19) that the effect of replacement of SFA by PUFA on CHD incidence could not be exclusively ascribed to an effect of n-6 PUFA. After an extensive literature search and dietary data extraction they concluded that in only three of the eight trials saturated fat was solely replaced by n-6 polyunsaturated fat and in the other five by a mixture of n-6 and n-3 fatty acids. In the latter trials, the CHD incidence was reduced by 22 % when saturated fat was replaced by a mixture of n-6 and n-3 PUFA. No effect was observed in the other three trials, but the number of studies was too small to draw conclusions. We agree with Ramsden et al. (Reference Ramsden, Hibbeln and Majchrzak20) that the effect of replacement of saturated fat by polyunsaturated fat on CHD incidence in the meta-analysis by Mozaffarian et al. (Reference Mozaffarian, Micha and Wallace19) should be ascribed to the combined effects of n-6 and n-3 PUFA. However, as Mozaffarian et al. (Reference Mozaffarian, Micha and Wallace19) showed, the effect on CHD incidence is in accord with the effect of the change in fatty acids on the total:HDL cholesterol ratio. Taking the results of these two meta-analyses together, we conclude that the effect of replacement of saturated fat by a mixture of n-6 and n-3 polyunsaturated fat can be ascribed to an effect of blood lipids and to that of n-3 PUFA independent of blood lipids, e.g. through prevention of ventricular arrhythmias(Reference McLennan, Owen and Slee21).
The alternative to replace energy from saturated fat with carbohydrates is much more complex, because ‘carbohydrates’ actually encompass a huge range of foods varying from high to low in micronutrients, phytochemicals and fibre (fruits, vegetables, whole grains v. sugar and refined grain). These two replacement strategies within the context of ‘carbohydrates’ are not likely to have the same effect on long-term risk. Only one controlled dietary experiment of long-term CHD risk has been carried out using this strategy(Reference Howard, Van Horn and Hsia22). Therefore, we also report here the recently published results of a Danish cohort study in which SFA in the statistical model were substituted for carbohydrates with low glycaemic index values (a marker of healthy carbohydrate-containing foods). This was associated with a lower risk of myocardial infarction(Reference Jakobsen, Dethlefsen and Joensen23). However, replacing SFA with carbohydrates with high glycaemic index values (a marker of unhealthy carbohydrate-containing foods) was associated with a higher risk(Reference Jakobsen, Dethlefsen and Joensen23). In a large long-term controlled dietary experiment carried out in 48 835 women aged 50–79 years, energy from fat decreased by 8 % and carbohydrates increased by the same percentage(Reference Howard, Van Horn and Hsia22). This trial population had a low habitual fibre intake (15 g/d) and the difference in fibre intake between the high- and low-carbohydrate groups was 2·4 g/d. Over 8 years of follow-up, there was no effect of diet on the total:HDL cholesterol ratio and CHD incidence(Reference Howard, Van Horn and Hsia22). More favourable results may be expected for low-fat diets with a P:S ratio of at least 1 and a high fibre content (>50 g/d), which would be rich in micronutrients and phytochemicals. However, controlled trials that tested the effect of this type of diet on CHD end points have not been carried out.
Given the paucity of controlled dietary experiments of CHD risk using the carbohydrate replacement strategy, the influence of low-fat, high-carbohydrate diets on lowering serum cholesterol is also of interest in this context. In a classic experiment reported in 1981, Lewis et al. (Reference Lewis, Hammett and Katan24) compared diets low in fat (27 % of energy with a P:S ratio of 1) and high in carbohydrates (59 % of energy) that were either low (20 g/d) or high (55 g/d) in fibre. The fibre-enriched diet contained more fruit and vegetables and substituted whole-wheat bread for white bread. The low-fat, high-fibre diet reduced LDL-cholesterol by 35 % and HDL-cholesterol by 11 % while the low-fat, low-fibre diet decreased LDL-cholesterol by 27 % and HDL-cholesterol by 12 %. These results suggest that the LDL:HDL cholesterol ratio is more favourably influenced by the low-fat, high-fibre diet than by the low-fat, low-fibre diet(Reference Lewis, Hammett and Katan24). These results were confirmed in a controlled dietary experiment using a dietary portfolio of cholesterol-lowering plant foods(Reference Jenkins, Kendall and Marchie25). The diet contained 29 % of energy from fat with a P:S ratio of 1·6 and 77 g fibre/d.
A food-based dietary experiment carried out in a sub-study of the Spanish PREvención con DIeta MEDiterránea (PREDIMED) Study is of further interest concerning fibre-rich diets, not necessarily in a low total fat context(Reference Estruch, Martinez-Gonzalez and Corella26). For this study, subjects with at least two CHD risk factors were randomised into three groups. The reference group was assigned to a low-fat diet and the other two groups were allocated to a recommended Mediterranean-style diet to which either 1 litre of extra virgin olive oil per week or 30 g nuts/d was added. Both the olive oil and the nut supplements were supplied to the participants by the investigators. Both olive oil and nuts are rich in MUFA and tree nuts also in n-6 and n-3 polyunsaturated fat. The diets with extra olive oil or nuts decreased LDL-cholesterol by − 39 and − 34 mg/l ( − 3·9 and − 3·4 mg/dl), and increased HDL-cholesterol by 29 and 16 mg/l (2·9 and 1·6 mg/dl)(Reference Estruch, Martinez-Gonzalez and Corella26). These results were comparable to those obtained in controlled dietary experiments(Reference Mensink and Katan13, Reference Mensink, Zock and Kester16).
In summary, replacement of saturated fat by polyunsaturated fat decreases the LDL:HDL cholesterol ratio and reduces the incidence of CHD. Replacement of saturated fat by cis-MUFA decreases the LDL:HDL cholesterol ratio. Low-fat diets high in carbohydrates but low in fibre do not change the LDL:HDL cholesterol ratio. In contrast, low-fat diets with a P:S ratio of at least 1 and high-fibre content do decrease the LDL:HDL cholesterol ratio. Diets high in cis-MUFA and low in fat with a P:S ratio of at least 1 and high fibre content may reduce CHD risk, although this has not been proven experimentally. This leads to the conclusion that different diets could be designed to prevent CHD. This potential diversity is crucial in engaging the diverse cultures and tastes worldwide in cardiovascular prevention(Reference Sacks and Katan18).
Optimal fatty acid composition and dietary patterns for CHD prevention
The ultimate question to be answered is what the optimal fatty acid composition of diets for CHD risk reduction could be? Based on eight carefully controlled studies, Sacks & Katan(Reference Sacks and Katan18) concluded that trans-fatty acids had the worst effect on blood lipids of all dietary fatty acids. cis-MUFA and n-6 PUFA reduce the total:HDL cholesterol ratio, whereas carbohydrates have a negligible effect on the ratio(Reference Mensink, Zock and Kester16). However, if low-fat, high-carbohydrate diets with a P:S ratio of at least 1 also had a high amount of fibre, a similar total:HDL cholesterol ratio was obtained when saturated fat was replaced by polyunsaturated fat(Reference Lewis, Hammett and Katan24, Reference Jenkins, Kendall and Marchie25). This suggests that for optimal CHD risk reduction not only the fatty acid composition but also the fibre content, probably indicating a composite of micronutrients and phytochemicals, is of importance.
Both saturated and trans-fatty acids not only have a detrimental effect on blood lipids but also increase CHD risk. This was observed in controlled dietary experiments in which saturated fat was replaced by vegetable oils rich in mostly n-6 PUFA. These diets reduced CHD incidence and the stronger the saturated fat reduction, the lower the CHD incidence(Reference Mozaffarian, Micha and Wallace19). When 5 % of energy from saturated fat was replaced by a similar amount of mostly n-6 polyunsaturated fat, CHD risk was reduced by 10 % (Table 1). Similar results were obtained in a meta-analysis of pooled data of eleven prospective cohort studies(Reference Jakobsen, O'Reilly and Heitmann27). Trans-fats increase the risk of CHD even more strongly than saturated fats. These fatty acids were introduced industrially in only a few products and tend to have relatively low within-person variance in observational data. For trans-fatty acids data from only prospective cohort studies are available. A meta-analysis of four studies showed that a reduction in trans-fatty acid intake of 2 % of energy is associated with a 24 % lower CHD risk (Table 1)(Reference Oomen, Ocke and Feskens28).
NA, not available.
* Based on both cohort studies and trials.
n-3 Fatty acids also contribute to an optimal fatty acid composition of the diet. The mother compound of these fatty acids is ALA, a PUFA with eighteen carbon atoms and three double bonds arising in plant oils, e.g. soyabean and linseed oil. A meta-analysis of prospective cohort studies showed that a high ALA intake compared to a low ALA intake was associated with a lower, though not statistically significant, risk of CHD mortality (relative risk 0·79; 95 % CI 0·60, 1·04)(Reference Brouwer, Katan and Zock29). Fish is an important source of the n-3 fatty acids EPA and DHA. Meta-analyses showed that persons who eat fish at least once a week have an approximately 15 % lower risk of fatal CHD(Reference He, Song and Daviglus30). Meta-analyses of randomised trials showed that an additional amount of EPA–DHA reduced the risk of both fatal CHD and sudden cardiac death(Reference León, Shibata and Sivakumaran31, Reference Marik and Varon32). In a meta-analysis of both prospective cohort studies and trials, Mozaffarian & Rimm(Reference Mozaffarian and Rimm33) showed that an increase from 0 to 250 mg EPA and DHA per d was associated with a 36 % lower CHD mortality risk (Table 1).
The results for the different fatty acids make clear the large potential of an optimal fatty acid composition for CHD prevention. Table 2 summarises broad recommendations for an optimal fatty acid composition in the context of nutritionally adequate diets(34, 35). For CHD prevention, a recommendation for total fat is not needed and may even be counterproductive. High-fat diets (30–45 % of energy) with a P:S ratio of more than 1 reduce the LDL:HDL cholesterol levels and CHD incidence compared with diets high in saturated fat with a P:S ratio of 0·2(Reference Mensink, Zock and Kester16, Reference Mozaffarian, Micha and Wallace19). Evidence is accumulating that this may also be the case for a low-fat diet with a P:S ratio of 1 and high-fibre content. However, low-fat, high-carbohydrate diets with low fibre content, have an unfavourable effect on blood lipoprotein fractions and therefore likely also on CHD risk(Reference Siri-Tarino, Sun and Hu2). Both SFA and, even more, trans-fatty acids increase CHD risk. Therefore, we see merit in the EFSA recommendation of an intake as low as possible for these fatty acids(3). However, this recommendation does not mean that people must avoid all foods high in SFA, such as chocolate, cheese, palm oil and coconut. We interpret the EFSA recommendation as an incentive to consume a nutritionally adequate diet with a low saturated fat content. To our opinion, there is no minimal amount of saturated fat that should be eaten. cis-MUFA compared with saturated and trans-fatty acids have a favourable effect on the LDL:HDL cholesterol ratio and possibly on CHD risk also. They are, however, not essential and, therefore, a recommendation is not needed.
* Adapted from the scientific opinion on dietary reference values for fats from the European Food Safety Authority Panel on Dietetic Products, Nutrition, and Allergies(3).
The most common n-6 fatty acid, linoleic acid, is essential. To prevent an essential fatty acid deficiency an intake of at least 2·5 % of energy is recommended(Reference Elmadfa and Kornsteiner36). On the basis of aggregate data from randomised trials, case–control and cohort studies, and long-term animal feeding experiments, Harris et al. (Reference Harris, Mozaffarian and Rimm37) concluded that an intake of at least 5 % of energy from linoleic acid is needed to reduce CHD risk. To prevent an essential fatty acid deficiency of the n-3 fatty acid ALA, an intake of at least 0·5 % of energy is recommended(Reference Elmadfa and Kornsteiner36). There is some but not yet convincing evidence from prospective cohort studies that a high ALA intake is associated with a lower CHD mortality risk(Reference Brouwer, Katan and Zock29). Upper limits for the intake of both linoleic acid and ALA are not needed for nutritionally adequate diets. Finally, there is convincing evidence from prospective cohort studies and trials that an intake of the n-3 fatty acids, EPA and DHA, of at least 250 mg/d is needed(Reference Mozaffarian and Rimm33). There is no evidence that a higher intake is needed for CHD prevention.
We propose to consume nutritionally adequate diets that are low in saturated fat and as low as possible in trans-fat. We raise concern about fabricated substitutes for trans-fat, such as interesterified fats. Nutritionally adequate diets should fulfil the requirements for the intake of n-6 and n-3 fatty acids. No recommendation is needed for the intake of cis-MUFA. Recommendations for fatty acid intake must be considered in the context of whole diets. Natural experiments showed that both traditional Mediterranean and Japanese diets were associated with a low risk of CHD(Reference Keys38, Reference Menotti, Kromhout and Blackburn39). The common feature of these diets was that they were both low in saturated and trans-fat, meat and dairy, and high in legumes, nuts and vegetables. The traditional Mediterranean diet was high in olive oil, whole grains and fruit, and moderate in fish while the traditional Japanese diet was high in fish and rice(Reference Kromhout, Keys and Aravanis40). This underscores that recommendations for fat intake must be made within a food-based approach to CHD prevention(Reference Jacobs and Steffen41, Reference Mozaffarian and Ludwig42).
D. K. and J. M. G. received an unrestricted grant from Unilever for the Alpha Omega Trial; J. M. G. also received an unrestricted grant from Alpro Foundation, Belgium, for epidemiological research on fatty acid intake and cardiovascular diseases; D. R. J. is a member of the Scientific Advisory Board of the California Walnut Commission (unpaid); A. M. has no conflicts of interest. D. K. conceived the review, carried out the literature research and drafted the manuscript; D. R. J. co-wrote part of the manuscript and revised it critically for important intellectual content; J. M. G. and A. M. contributed to the intellectual content. All the authors approved the final version of the review.