- Lee Hooper, research associate in evidence based care and systematic review ()a,
- Carolyn D Summerbell, reader in human nutritionb,
- Julian P T Higgins, statisticianc,
- Rachel L Thompson, senior research fellow in public health nutritiond,
- Nigel E Capps, consultant chemical pathologist for Shropshiree,
- George Davey Smith, professor of clinical epidemiologyf,
- Rudolph A Riemersma, senior lecturer in cardiac biochemistryg,
- Shah Ebrahim, professor in epidemiology of ageingf
- a Manchester Dental and Education Centre (MANDEC), University Dental Hospital of Manchester, Manchester M15 6FH
- b School of Health, University of Teesside, Middlesbrough, Cleveland TS1 3BA
- c Systematic Reviews Training Unit, Institute of Child Health, London WC1N 1EH
- d Public Health Nutrition, Institute of Human Nutrition, Level B, South Academic Block, Southampton General Hospital, Southampton SO16 6YD
- e Department of Clinical Biochemistry, Princess Royal Hospital NHS Trust, Apley Castle, Telford TF6 6TF
- f Department of Social Medicine, University of Bristol, Canynge Hall, Bristol BS8 2PR
- g Cardiovascular Research, University of Edinburgh, Edinburgh EH8 9XF
- Correspondence to: L Hooper
- Accepted 16 January 2001
Objective: To assess the effect of reduction or modification of dietary fat intake on total and cardiovascular mortality and cardiovascular morbidity.
Design: Systematic review.
Data sources: Cochrane Library, Medline, Embase, CAB abstracts, SIGLE, CVRCT registry, and biographies were searched; trials known to experts were included.
Included studies: Randomised controlled trials stating intention to reduce or modify fat or cholesterol intake in healthy adult participants over at least six months. Inclusion decisions, validity, and data extraction were duplicated. Meta-analysis (random effects methodology), meta-regression, and funnel plots were performed.
Results: 27 studies (30 902 person years of observation) were included. Alteration of dietary fat intake had small effects on total mortality (rate ratio 0.98; 95% confidence interval 0.86 to 1.12). Cardiovascular mortality was reduced by 9% (0.91; 0.77 to 1.07) and cardiovascular events by 16% (0.84; 0.72 to 0.99), which was attenuated (0.86; 0.72 to 1.03) in a sensitivity analysis that excluded a trial using oily fish. Trials with at least two years' follow up provided stronger evidence of protection from cardiovascular events (0.76; 0.65 to 0.90).
Conclusions: There is a small but potentially important reduction in cardiovascular risk with reduction or modification of dietary fat intake, seen particularly in trials of longer duration.
What is already known on this topic
What is already known on this topic The epidemiological relation between dietary fat intake and cardiovascular disease is central in strategies aimed at risk reduction in populations and individuals
Systematic review of randomised controlled trials supports manipulation of dietary fat to control serum lipid concentrations, though evidence of effect on one risk factor does not rule out an opposite or reinforced effect on another unstudied risk factor
Randomised controlled trials of dietary fat reduction or modification have shown varying results on cardiovascular morbidity and mortality
What this study adds
What this study adds Systematic review of trials of modified fat intake shows that reduction or modification of dietary fat intake results in reductions in cardiovascular events, but only in trials of at least two years' duration
There is little effect on total mortality
Despite decades of effort and many thousands of people randomised, there is still only limited and inconclusive evidence of the effects of modification of total, saturated, monounsaturated, or polyunsaturated fats on cardiovascular morbidity and mortality
For half a century the relation between dietary fat and cardiovascular disease, the “diet-heart” hypothesis, has been a central tenet of strategies for risk reduction in individuals and populations.1 Observational studies2 and systematic reviews of clinical trials with risk factors as end points3–7 support this relation. However, evidence of a beneficial effect in observational studies does not provide convincing evidence. For example, the protective effect of β carotene in coronary heart disease was strongly supported by observational evidence, but large randomised controlled trials showed no protective effects on morbidity or mortality.8
Previous investigators have extrapolated the reduction in coronary heart disease that might be expected from changes in blood cholesterol concentration, 4 7 even though there is direct evidence from randomised controlled trials of the effect of modification or reduction of intake of dietary fats. We therefore performed a systematic review to assess the effect of change in dietary fat intake, which would be expected to result in a lowering of cholesterol concentration, on mortality and cardiovascular morbidity, using all available randomised clinical trials. The interventions included any of the following: reduction in intake of total fat; reduction in intake of saturated fat; reduction in intake of dietary cholesterol; or a shift from saturated to unsaturated fat.
Much of the methodology has been reported previously.9 Briefly, we developed a search strategy to search for nutrition based randomised controlled trials on the Cochrane Library, Medline, Embase, CAB abstracts, CVRCT registry (inception of database to mid-1998), and SIGLE (January 1999). We searched bibliographies and contacted related Cochrane Review Groups and 60 experts (May 1999) for further trials. There were no language restrictions.
The inclusion criteria were adequate randomisation; usual or control diet or placebo group; stated aim of intervention was reduction or modification of intake of dietary fat or cholesterol, unless the intervention was exclusively omega 3 fatty acids; intervention was not multifactorial; the intervention group were not children, acutely ill, or pregnant; the intervention (diet provided or supplementation) continued for at least six months or follow up (after dietary advice) was at least six months; and data on mortality or cardiovascular morbidity were available.
Our primary outcomes of interest were effects of intervention on total mortality, cardiovascular mortality, combined cardiovascular events (including all available data on cardiovascular deaths, non-fatal myocardial infarction, stroke, angina, heart failure, peripheral vascular disease, angioplasty, and coronary artery bypass grafting), and quality of life. Event data were included only when they occurred during provision of diet or supplement (when these were provided) or while randomisation and blinding were maintained (in dietary advice trials).
Inclusion was assessed independently by two assessors (LH, RLT) and differences were resolved by discussion, with, if necessary, a third reviewer (RAR). Data extracted independently in duplicate (by LH and RLT) included type of participants, interventions, and outcomes; characteristics of trial quality; numbers of events; total patient years in trial; and data on potential effect modifiers. Effect modifiers included participants' baseline risk of cardiovascular disease, trial duration, mode of intervention, change in intake of dietary fats, and serum cholesterol concentration achieved.
Criteria for assessment of trial quality included method of randomisation, physician blinding, participant blinding, and any systematic difference in care between the intervention groups.
We used meta-analysis to explore the primary hypothesis, which was that reduction or modification of dietary fat intake affects mortality, cardiovascular mortality, and cardiovascular events. We examined the effects of duration of intervention, initial level of cardiovascular risk and dietary fat intake, type of intervention, and changes in intake of total, saturated, monounsaturated, and polyunsaturated fats and blood cholesterol concentration using subgrouping of trials and meta-regression. We excluded trials that aimed to alter omega 3 intake as the method of action (if any) is probably different from any action caused by the reduction in intake of total or saturated fat—that is, it does not primarily lower low density lipoprotein cholesterol.
Treatment effect was measured as a rate ratio and meta-analysis performed as a weighted average of log rate ratios.10 The meta-analysis used random effects methodology11 within S-PLUS.12 Meta-analysis pools results of individual trials, with weighting so that results with higher precision (related to sample size) contribute more to the combined outcome. We used the STATA command metareg13 for random effects meta-regression.14 Meta-regression investigates the effect of one or more study characteristics on the size of treatment effect, taking precision into account. A genuine relation may be inferred (for example, if the extent of reduction in total fat achieved by the intervention is associated with the degree of reduction in mortality) when a slope is significantly different from zero. We used funnel plot asymmetry to detect any bias in the trials retrieved.15
We identified 27 studies, comprising 40 distinct intervention arms over 30 902 person years of observation.16–42 A table with details of all 27 studies can be found on the BMJ's website. Figure 1 gives a flow chart of studies assessed and excluded at various stages of the review. Table 1 gives details of extracted data. We found almost no data on quality of life or levels of trans fats. The κ statistic for inter-rater agreement on inclusion or exclusion of potential trials was 0.61.43
Reduction or modification of dietary fat intake
The pooled rate ratio for total mortality was 0.98 (95% confidence interval 0.86 to 1.12), which indicates little, if any, effect (fig 2). The data on cardiovascular mortality indicate a slight (9%) protection from modification of intake of dietary fat (0.91; 0.77 to 1.07) (fig 3) and a 16% reduction in cardiovascular events (0.84; 0.72 to 0.99) (fig 4). A funnel plot of effect size versus sample size did not show any evidence of bias (data not shown).
The Oslo diet-heart trial provided some oily fish (a source of omega 3 fatty acids) to participants in the intervention group.33 Exclusion of the results of this trial attenuated the rate ratios for all three main outcomes (total mortality 1.02 (0.91 to 1.14); cardiovascular mortality 0.94 (0.79 to 1.11); combined cardiovascular events 0.86 (0.72 to 1.03)).
Duration of follow up, initial level of cardiovascular risk, type of intervention, and dietary fat intake at baseline
We obtained results for specific length of time on diet, initial level of cardiovascular risk, or mode of intervention through subgrouping of trials. Trials in which the mean length of follow up exceeded two years showed somewhat larger reductions in combined cardiovascular events (0.76 (0.65 to 0.90) v 0.96 (0.75 to 1.23) than trials of less than two years' duration). When we excluded data from the Oslo diet-heart study, the rate ratio for combined cardiovascular events for trials with mean follow up longer than two years was not altered (0.77; 0.62 to 0.96). Total mortality was not substantially influenced by mean follow up.
The relation between mean follow up and cardiovascular events was further explored through meta-regression. The negative slope (ln (rate ratio) v mean follow up time, slope −0.096; −0.190 to −0.002) suggests fewer cardiovascular events at longer mean follow up times in the groups with reduced or modified fat intake.
Trials with participants at high initial cardiovascular risk suggested very similar levels of protection from combined cardiovascular events (rate ratio 0.84; 0.70 to 0.99) as did trials with participants at low cardiovascular risk (0.82; 0.56 to 1.20). Neither the method of dietary modification (by diet advice, advice plus a supplement, or diet provided) nor the initial level of intake of dietary fat influenced rate ratios.
Changes in fat intake and blood cholesterol concentration
We used meta-regression to explore the effects on total mortality and combined cardiovascular events of changing the proportion of energy from total fat and of altering serum cholesterol concentrations. Meta-regressions also explored the relation between change in proportion of saturated fat, polyunsaturated fat, and monounsaturated fat on cardiovascular events. These suggested that total mortality and cardiovascular events were reduced as energy from fat, and as serum cholesterol concentrations, fell. Similarly, cardiovascular events were reduced as each of the fat subfractions decreased. However, none of these relations were convincing (none of the slopes were different from zero at a P<0.05, see table 2), except for that of monounsaturated fat. Here the slope depended heavily on one medium sized trial, St Thomas's atherosclerosis regression study (STARS),37 and should therefore be treated with caution.
Pooled results of dietary fat trials indicate that reduction or modification of intake of dietary fat reduces the incidence of combined cardiovascular events by 16% (rate ratio 0.84; 95% confidence interval 0.72 to 0.99) and cardiovascular deaths by 9% (0.91; 0.77 to 1.07). No effect was seen on total mortality.
Exclusion of data from the Oslo diet study33 attenuated the effects on cardiovascular events. In this trial participants in the intervention group were supplied with oily fish, in addition to dietary advice. Oily fish, which is rich in omega 3 fatty acids, seems to have independent beneficial effects,44 so the benefits seen in this trial may have been due to the fish oils and not to cholesterol lowering or alterations in other dietary fats.
Duration of trial effect
Subgrouping by length of mean follow up suggests that virtually all protection from cardiovascular events occurs in trials of at least two years' duration. In trials with mean follow up of two years or less the reduction in events was only 4%, whereas in trials with longer follow up reductions of up to 24% were seen. This effect was confirmed in a meta-regression.
The division at two years was chosen after we studied initial data from the Scandinavian simvastatin survival study (4S), which assessed the effect of simvastatin on mortality and morbidity after myocardial infarction.45 In that study little difference in event rates between treated and control groups emerged until after two years. The similar time lag between the effects of dietary intervention and statins suggests a common mechanism, perhaps through effects on the scale, type, and stability of atherosclerotic plaques (rather than rapid alteration in blood coagulability), but other mechanisms may also be operating.
It is not clear whether it is the duration of the intervention or the duration of the follow up that determines whether the intervention is effective. Sustained change in dietary behaviour, promoted by long interventions, is probably necessary to achieve reduction in cardiovascular events, and lengthy follow up is probably required before pathophysiological changes that reduce clinical events are observed.
Initial level of cardiovascular risk
Subgrouping suggested no effect of initial level of risk on outcomes. Both high and low risk groups had a similar level of risk reduction for combined cardiovascular events. This is in contrast to results of previous studies,46 though initial levels of risk in studies in this systematic review were generally high (people in control group at “low risk” of cardiovascular events suffered 2.57 events per 100 people per year and those at “high risk” 7.62 events per 100 people per year).
Type of intervention
Subgrouping by mode of intervention showed no benefit of providing the entire diet compared with dietary advice or advice and a supplement. This was surprising as we expected that provision of food would have a more powerful effect on events than dietary advice alone.
Changes in blood cholesterol concentration
Meta-regression provided weak evidence that a greater reduction in serum cholesterol concentrations in the intervention group compared with the control group resulted in a slightly greater reduction in cardiovascular events and mortality.
A reduction of over 20% in total serum cholesterol concentration can result in a 25% fall in mortality from coronary heart disease.47 Within dietary trials in this review the mean initial serum cholesterol concentration was 5.8 mmol/l, with a fall of 0.64 mmol/l (11%). This degree of reduction is similar to that achieved by early fibrates, which show reductions in rates of coronary heart disease only among those at high risk.46 However, early fibrates seemed to increase mortality from other causes, and hence any benefit from this reduction in cholesterol concentration was outweighed by adverse effects of the drugs themselves. In dietary modification this degree of cholesterol lowering is probably of modest overall benefit.
Methodology of the review
In this review we aimed to find trials that modified or reduced dietary fat intake for at least six months, even when mortality and morbidity were not reported. We tried to contact trial authors to ascertain whether it was known if any deaths or cardiovascular events had occurred. For this reason, many of the included trials are small but are included in the hope of augmenting the data from larger trials and reducing bias. Although small trials cannot individually hope to achieve useful data on rare events, in meta-analysis we are increasing the power of the group of studies by pooling.
We included factorial trials (for example, the diet and reinfarction trial (DART)17) when it was possible to separate out the data on the effect of change in dietary fat from the other interventions (which are randomly distributed across the fat control and intervention groups). We excluded multiple risk factor intervention trials as it is not possible to disaggregate the effects of change in dietary fat intake from other, potentially effective interventions such as increasing exercise, stopping smoking, and change in other dietary constituents. There is already a systematic review on this topic.48 We excluded the Lyon heart study49 as inclusion in our review was decided by the stated intention of the trial.49 This trial nowhere states the intention to reduce or modify dietary fat intake and the major effect on diet was to increase the omega 3 fat intake. We excluded the Finnish mental hospital study50 as it was not a truly randomised design (a cluster randomised trial with only two clusters).
Although we amassed 30 902 person years of observation, 1430 deaths, and 1216 cardiovascular events, considerable uncertainty exists over the size of effects and the means by which they were achieved given the scarcity of trials longer than two years' duration. It is unlikely that further resources will now be forthcoming to perform a large trial of dietary modification. Consequently the evidence pooled in this review contributes all that is available to guide clinical practice and health policy on dietary manipulation. However, publication of data on deaths or cardiovascular events in the large ongoing trials of dietary fat lowering with cancer end points would augment currently available data.
In this review we have tried to separate out whether changes in individual fatty acid fractions are responsible for any benefits to health (using the technique of meta-regression). The answers are not definitive, the data being too sparse to be convincing. We are left with a suggestion that less total fat or less of any individual fatty acid fraction in the diet is beneficial.
The findings on cardiovascular events are broadly in keeping with benefits that might be expected from modest lowering of cholesterol concentration and certainly provide support, at an individual level, for the central role of dietary fat intake in the causation of cardiovascular disease.
This study was conducted as a Cochrane systematic review under the auspices of the Cochrane Heart Group, whose assistance is gratefully acknowledged. We thank Gill Clements for her comments throughout the review, all those at the Systematic Reviews Training Unit for their help and support, and all the trialists and experts who kindly provided information and advice. The original version of this review is available in the Cochrane Library (Hooper L, Summerbell CD, Higgins JPT, Thompson RL, Clements G, Capps N, et al. Reduced or modified dietary fat for prevention of cardiovascular disease. Cochrane Database Syst Rev 2000;(2):CD002137) but does not include some post-hoc analysis performed as a result of BMJ reviewer's comments.
Contributors: All authors were actively involved in the design of the review and provision of critical revisions to the manuscript. JPTH performed the statistical analyses, RLT duplicated trial inclusion/exclusion and data extraction, and RAR arbitrated on study inclusion when necessary. SE and CDS were primary advisers. GDS provided epidemiological expertise and NEC provided a clinical perspective. LH originated and was primarily responsible for planning and carrying out the review, was the principal author, and is guarantor for the paper.
Funding LH's work on this review was partly funded by a studentship from the Systematic Reviews Training Unit, Institute of Child Health, University of London, and Shropshire Health Authority, Shrewsbury, contributed to travel expenses. RAR is supported by the British Heart Foundation.
Competing interests None declared.
A table giving details of included studies is available on the BMJ's website