- Robert Clarkea, research fellow,
- Chris Frostb, lecturer,
- Rory Collinsa, professor of medicine and epidemiology,
- Paul Applebya, research officer,
- Richard Petoa, professor of medical statistics and epidemiology
- a Clinical Trial Service Unit and Epidemiological Studies Unit, Nuffield Department of Clinical Medicine, Radcliffe Infirmary, Oxford OX2 6HE
- b Department of Statistics, London School of Hygiene and Tropical Medicine, London WC1E 7HT
- Correspondence to: Dr Clarke
- Accepted 28 October 1996
Objective: To determine the quantitative importance of dietary fatty acids and dietary cholesterol to blood concentrations of total, low density lipoprotein, and high density lipoprotein cholesterol.
Design: Meta-analysis of metabolic ward studies of solid food diets in healthy volunteers.
Subjects: 395 dietary experiments (median duration 1 month) among 129 groups of individuals.
Results: Isocaloric replacement of saturated fats by complex carbohydrates for 10% of dietary calories resulted in blood total cholesterol falling by 0.52 (SE 0.03) mmol/l and low density lipoprotein cholesterol falling by 0.36 (0.05) mmol/l. Isocaloric replacement of complex carbohydrates by polyunsaturated fats for 5% of dietary calories resulted in total cholesterol falling by a further 0.13 (0.02) mmol/l and low density lipoprotein cholesterol falling by 0.11 (0.02) mmol/l. Similar replacement of carbohydrates by monounsaturated fats produced no significant effect on total or low density lipoprotein cholesterol. Avoiding 200 mg/day dietary cholesterol further decreased blood total cholesterol by 0.13 (0.02) mmol/l and low density lipoprotein cholesterol by 0.10 (0.02) mmol/l.
Conclusions: In typical British diets replacing 60% of saturated fats by other fats and avoiding 60% of dietary cholesterol would reduce blood total cholesterol by about 0.8 mmol/l (that is, by 10-15%), with four fifths of this reduction being in low density lipoprotein cholesterol.
The quantitative importance of diet to blood cholesterol remains uncertain because non-experimental dietary studies in community subjects are unreliable and experimental (“metabolic ward”) studies have been too small to be separately reliable
We conducted a meta-analysis of 395 published, metabolic ward experiments of the effects of various dietary lipids on blood cholesterol
Isocaloric increases in saturated fat intake were associated with increases in total and low density lipoprotein cholesterol and smaller increases in high density lipoprotein cholesterol; increased polyunsaturated fat intake decreased total and low density lipoprotein cholesterol and increased high density lipoprotein cholesterol; monounsaturated fat had no significant effect on total and low density lipoprotein cholesterol but increased high density lipoprotein cholesterol
In the average British diet replacement of 60% of the saturated fat by other dietary fats and avoidance of 60% of dietary cholesterol would reduce blood cholesterol by about 0.8 mmol/l (that is, by 10-15%), with four fifths of this reduction being in low density lipoprotein cholesterol
The effect on vascular disease of a prolonged difference of 0.8 mmol/l in blood cholesterol concentration depends on the relative importance at different ages of the benefits of reducing low density lipoprotein cholesterol and the hazards of reducing high density lipoprotein cholesterol, which require further study
The quantitative importance of diet to blood total cholesterol and, more importantly, to its fractions (low density lipoprotein and high density lipoprotein cholesterol), remains uncertain.12345 This is partly because metabolic mechanisms are not clear, partly because non-experimental dietary studies in community subjects are unreliable,67891011 and partly because previous experimental (that is, “metabolic ward”) studies have been too small to be reliable. Hence, selective emphasis on particular studies can lead to conflicting conclusions about important issues12–for example, whether isocalorically replacing complex carbohydrates by monounsaturated fats affects blood total cholesterol or its fractions.13141516 The aim of this meta-analysis of metabolic ward studies is to provide reliable quantitative estimates of the relevance of dietary intake of fatty acids and dietary cholesterol to blood concentrations of total cholesterol and cholesterol fractions.
Published reports of dietary intervention studies conducted under controlled conditions that ensured compliance (metabolic ward studies) with diets persisting at least two weeks were systematically sought by Medline searches, scanning relevant reference lists, and handsearching nutrition journals. We excluded studies if they were of subjects selected for some disorder (such as diabetes or dyslipidaemia), if the dietary changes were deliberately confounded by other interventions (such as weight reduction or exercise), or if there were no data available about dietary fatty acids or dietary cholesterol. The search strategy (details of which are available on request) did not, however, require the availability of data on changes in body weight. Because the dietary changes were to be isocaloric and most experimental periods lasted only a few weeks, substantial weight changes were not expected. Hence, many relevant publications did not include data on body weight after the experimental periods (and, of those that did, many found no material differences17 18).
Among the 81 eligible reports identified (see appendix), we excluded one long term multicentre study for poor compliance. Solid food diets were assessed in 72 of these reports among 129 groups of subjects in 395 experiments with various designs (109 randomised crossover, 57 randomised or matched parallel, 77 non-randomised Latin square, and 152 non-randomised sequential). Details of the major individual fatty acids in the diets were available for 134 experiments, and blood concentrations of high density lipoprotein and low density lipoprotein fractions of cholesterol were available for 227. Liquid formula diets, which were assessed in 32 experiments in eight reports, were examined separately because the effects of such diets may differ from those of solid food diets.19
From each publication, we sought information about mean age and weight, the experimental diets (caloric intake; intake of dietary cholesterol; and percentage of calories as total, saturated, polyunsaturated, and monounsaturated fat), and blood cholesterol concentrations (total, low density lipoprotein, and high density lipoprotein) in plasma or serum at the end of the experiments. For some experiments, the mean weight, age, or dietary cholesterol had to be estimated from median or mid-range values. Fatty acids were classified by carbon chain lengths (with C18, for example, indicating 18 carbons) and by the number of carbon-carbon double bonds (such as C18:1). Saturated fatty acids have no such double bonds, monounsaturates have one, and polyunsaturates have more than one. Double bonds are cis if the two hydrogen atoms at each end are on the same side of the double bond and trans if otherwise.
We used “Multilevel” regression analyses (MLN-software, London University Education Institute) that included age, weight, and dietary intake of nutrients as well as one term per study to ensure that people within any one study were compared directly only with each other. Such analyses assessed different sources of variability: (a) within group, between experiments; (b) within study, between matched groups; (c) within study, between unmatched groups; and (d) between studies.
Univariate and multivariate analyses for total cholesterol
Figure 1 plots the mean dietary saturated fat in each experiment against the mean blood total cholesterol concentration at its end, with separate plots for different types of experimental design with solid food diets and for liquid formula diets, but without adjustment for factors other than intake of saturated fat. Crude correlations might misrepresent the real relations as some experimental periods were part of the same study, but the multilevel regression analyses that are plotted take appropriate account of such differences. The four regression slopes for the different types of solid food experiments were similar, and so the results of these 395 experiments were combined. The overall effects of saturated fat were less in the liquid formula experiments, and so these are considered separately. Univariate regressions for the solid food experiments indicated that dietary intakes of saturated fat, cholesterol, and total fat were each associated with highly significant increases in blood total cholesterol, while intake of polyunsaturated fat was associated with a highly significant decrease and intake of monounsaturated fat produced no significant effect on blood total cholesterol (table 1).
Such univariate analyses might be misleading, however, because isocaloric increases in one type of fat in many of the experiments were accompanied by decreases in other dietary fats or in dietary cholesterol. Multivariate analyses were therefore also performed, which assessed isocaloric replacement of complex carbohydrates by particular lipids after simultaneous adjustment for other dietary factors (and, less importantly, for age and initial body weight). Thus, for example, in the multivariate analyses “effects of saturated fat” would actually mean “effects of replacing carbohydrate isocalorically by saturated fat.” These produced smaller regression coefficients for blood total cholesterol than did the univariate analyses (table 1). Different types of experimental design with solid foods produced similar multivariate coefficients for the effects of specific fats on blood total cholesterol (fig 2), with smaller effects for the liquid formula diets. (Similar multivariate associations of blood total cholesterol with intake of various types of fat were observed in men and women; over or under age 35; over or under 70 kg body weight; over or under 2800 daily dietary calories; over or under 300 mg daily dietary cholesterol; and in studies of 2-4, 4-6, and over 6 weeks' duration (data not shown).) If, to ensure that attention is restricted only to randomised evidence, all results from studies of Latin square or sequential designs are ignored, then the overall findings for solid food diets would not be materially altered (fig 2).
Multivariate analyses for low density lipoprotein and high density lipoprotein cholesterol
A subset of 227 of the solid food experiments also reported blood low density lipoprotein and high density lipoprotein cholesterol. Multivariate analyses indicated that isocaloric increases in saturated fat intake were associated with highly significant increases in low density lipoprotein cholesterol and smaller increases in high density lipoprotein cholesterol, as were increases in dietary cholesterol (table 1). Conversely, increases in polyunsaturated fat intake significantly decreased low density lipoprotein cholesterol and increased high density lipoprotein cholesterol, while monounsaturated fat had no significant effect on low density lipoprotein cholesterol but did increase high density lipoprotein cholesterol.
Effects of realistic dietary changes
The multivariate analyses indicated that isocaloric replacement of saturated fats equivalent to 10% of dietary calories by complex carbohydrates would typically be associated with a reduction in blood total cholesterol of 0.52 (SE 0.03) mmol/l (table 2). Isocaloric replacement by polyunsaturated fat of carbohydrates equivalent to 5% of dietary calories would be expected to reduce blood cholesterol by a further 0.13 (0.02) mmol/l, whereas changes in intake of monounsaturated fat seemed to have little effect. Isocaloric replacement of saturated fats by unsaturated fats produced about three times the reduction in blood cholesterol produced by replacement of total fat by complex carbohydrates (table 2). A reduction of 200 mg/day in dietary cholesterol would be associated with a further reduction in blood cholesterol of 0.13 (0.02) mmol/l. Together, the estimated effect of these isocaloric dietary changes would be to reduce blood total cholesterol by 0.76 mmol/l (SE 0.03, 99% confidence interval 0.67 to 0.85) (fig 3), consisting of a reduction in low density lipoprotein cholesterol of 0.62 (0.04) mmol/l and in high density lipoprotein cholesterol of 0.10 (0.02) mmol/l (table 2).
Individual saturated fats
Altogether, 134 of the solid food experiments provided information on dietary intake of the major individual saturated fatty acids: laurate (C12:0), myristate (C14:0), palmitate (C16:0), and stearate (C18:0). Intake of the first three was positively related with blood cholesterol concentration, such that halving intake of each isocalorically would be expected to reduce blood total cholesterol by 0.32 (0.04) mmol/l (table 3). Stearic acid, however, which accounts for about a quarter of dietary saturated fats, did not seem to be significantly related to blood cholesterol concentration.
Forty of the solid food experiments provided information on the dietary intake of trans monounsaturated fats (mainly trans C18:1; elaidate). Multivariate regression coefficients for blood total cholesterol concentration (adjusted for other dietary fats, cholesterol intake, age, and weight) were 0.038 (SE 0.10) for trans monounsaturated fat, which is similar to 0.047 (0.008) for saturated fat in the same analyses. But, trans fatty acids account for only 2% of calories in the British diet,20 so replacing half isocalorically by carbohydrates would be expected to reduce blood total cholesterol by only 0.05 (0.01) mmol/l.
Because of the opposing effects on vascular disease of reductions in low density lipoprotein and high density lipoprotein, it is not sufficient to consider the effects of diet only on total blood cholesterol. Overall, because their effect on blood cholesterol is strong and the Western dietary intake is substantial, the key dietary factor in these studies was the intake of saturated fats. The present results indicate that isocaloric replacement of 60% of saturated fat by complex carbohydrates in the British diet would reduce blood total cholesterol by 0.5 mmol/l and low density lipoprotein cholesterol by 0.4 mmol/l, irrespective of sex, age, and body weight. Intake of polyunsaturated fat is also important, with effects that are about half as strong–in the opposite direction–as those of saturated fats. Intake of monounsaturated fat had no significant effect on total or low density lipoprotein cholesterol despite raising high density lipoprotein cholesterol by about as much as polyunsaturates. The combined effect of changing the type, but not the amount, of dietary fat by replacement of 10% of dietary calories from saturates by monounsaturates (5%) and by polyunsaturates (5%), together with consuming 200 mg less dietary cholesterol, would be a reduction in blood cholesterol of about 0.8 mmol/l, with the reduction chiefly in low density lipoprotein cholesterol (fig 3).
Isocaloric dietary changes such as these, which affect only the type but not the amount of dietary fat, are relevant to the control of plasma low density lipoprotein cholesterol but not to the control of obesity. Caloric restriction should eventually lead to weight loss, but this review has studied only isocaloric change. Table 2 indicates that if total dietary fat is reduced isocalorically by replacement with complex carbohydrates rather than with unsaturated fats, then the unwanted reduction in high density lipoprotein cholesterol could be so great that the ratio of low density lipoprotein to high density lipoprotein cholesterol would not be much affected.
Saturated fats with different chain lengths may have different effects, but, although intake of C18:0 fatty acids seemed less relevant than C12:0-C16:0, the evidence was limited and further direct comparisons are needed. The evidence on cis and trans unsaturated fats is also limited: trans unsaturated fats may be similar in effect to saturates, but intake of them is generally low in most people, and so their relevance to blood lipid concentrations is probably small. This review has not, however, tested the hypothesis21 that trans unsaturates from hydrogenated vegetable oil differ in their effects from natural trans unsaturates.
The reduction in blood cholesterol concentration shown by this review with isocaloric replacement of saturated by unsaturated fats appears within just a few weeks and is greater than is sometimes appreciated.5 Previous reviews of the effects of dietary fatty acids have yielded slightly different results from ours. An analysis of 27 studies involving 65 experiments also concluded that replacement of saturates by unsaturates produced substantial changes in the blood lipoprotein profile,3 but the size of changes suggested by our overview are greater. Another review of 248 metabolic ward experiments yielded similar conclusions for the effects of fatty acids on blood total cholesterol but was unable to reach any conclusions for lipoprotein fractions.4 Discrepant results from earlier reviews or individual studies3 4 15 16 22 reinforce the need for periodically updated meta-analyses12 of all available evidence from metabolic ward studies. (We restricted our attention to such studies because the weaker results from non-experimental dietary studies in community subjects may chiefly reflect poor compliance567891011.)
Although dietary change is difficult, these findings are relevant to countries such as Britain, where saturated fats provide 15-17% of dietary calories (chiefly in dairy produce, meat, oils, spreads, eggs, and confectionery) and where the daily cholesterol intake is 280-390 mg (chiefly in meat and eggs).20 23 Table 2 and figure 3 show that changes in dietary fats and cholesterol that many should find practicable would typically reduce blood cholesterol by about 0.8 mmol/l (equivalent to a 10-15% reduction), with large beneficial decreases low density lipoprotein cholesterol concentration and only small adverse decreases in high density lipoprotein cholesterol. The effect on vascular disease of a prolonged difference of 0.8 mmol/l in blood cholesterol concentration depends on the relative importance at different ages of the benefits of reducing low density lipoprotein and the hazards of reducing high density lipoprotein cholesterol, which require further study.
Alison Mills and Joyce Hughes, MAFF Nutrition Branch of the Ministry of Agriculture, Fisheries and Foods, provided British dietary data. Vivien Reid, Anthony Keech, and Jane Armitage provided helpful comments, as did the anonymous referees.
Funding: British Heart Foundation and Medical Research Council.
Conflict of interest: None.