Reduced or modified dietary fat for preventing cardiovascular disease (Review)

BACKGROUND
Reduction or modification of dietary fat can improve total cholesterol levels, but may also have a variety of effects, both positive and negative, on other cardiovascular risk factors.


OBJECTIVES
The aim of this systematic review was to assess the effect of reduction or modification of dietary fats on total and cardiovascular mortality and cardiovascular morbidity over at least 6 months, using all available randomized clinical trials.


SEARCH STRATEGY
The Cochrane Library, MEDLINE, EMBASE, CAB Abstracts, CVRCT registry and related Cochrane Groups' trial registers were searched through spring 1998, SIGLE to January 1999. Trials known to experts in the field and biographies were included through May 1999.


SELECTION CRITERIA
Trials fulfilled the following criteria: 1) randomized with appropriate control group, 2) intention to reduce or modify fat or cholesterol intake (excluding exclusively omega-3 fat interventions), 3) not multi factorial, 4) healthy adult humans, 5) intervention at least six months, 6) mortality or cardiovascular morbidity data available. Inclusion decisions were duplicated, disagreement resolved by discussion or a third party.


DATA COLLECTION AND ANALYSIS
Rate data were extracted by two independent reviewers and meta-analysis performed using random effects methodology. Meta-regression and funnel plots were used.


MAIN RESULTS
Twenty seven studies were included (40 intervention arms, 30,901 person-years). There was no significant effect on total mortality (rate ratio 0.98, 95% CI 0.86 to 1.12), a trend towards protection form cardiovascular mortality (rate ratio 0.91, 95% CI 0.77 to 1.07), and significant protection from cardiovascular events (rate ratio 0.84, 95% CI 0.72 to 0.99). The latter became non-significant on sensitivity analysis. Trials where participants were involved for more than 2 years showed significant reductions in the rate of cardiovascular events and a suggestion of protection from total mortality. The degree of protection from cardiovascular events appeared similar in high and low risk groups, but was statistically significant only in the former.


REVIEWER'S CONCLUSIONS
The findings are suggestive of a small but potentially important reduction in cardiovascular risk in trials longer than two years. Lifestyle advice to all those at high risk of cardiovascular disease (especially where statins are unavailable or rationed), and to lower risk population groups, should continue to include permanent reduction of dietary saturated fat and partial replacement by unsaturates.


B A C K G R O U N D
There has been a great deal of research carried out in the area of diet and cardiovascular disease, the diet-heart hypothesis. Much of this has been invested in long term prospective observational studies looking at dietary patterns and subsequent cardiovascular events. This work is powerful at providing associations between dietary factors and cardiovascular risk. However, intervention studies are needed to clarify cause and effect, and it is essential that intervention trials form the basis of evidence based practice in this area.
Most intervention studies which have been carried out have studied the effect of dietary interventions on risk factors for heart disease, and separate work ties the effect of altering these risk factors to changes in disease incidence and mortality. Systematic review in this area follows the same pattern, so that there are reviews of the effect of dietary advice on change on lipid levels (Brunner 1997;Clarke 1997;Denke 1995;Mensink 1992) and reviews on the effect of lipid level alterations on cardiovascular morbidity and mortality (Law 1994;Walsh 1995;Rubins 1995). Other risk factors dealt with in a similar way are blood pressure measurements (Bucher 1996;Law 1991), body weight (SIGN 1996), angiographic measurements (Marchioli 1994), antioxidant intake (Ness 1997) and alcohol (Rimm 1996).
A problem with this two-level approach is that any single dietary alteration may have effects over a wide range of risk factors for cardiovascular disease. An example of this is the choice of substitution of saturated fats by carbohydrate, polyunsaturated fats or monounsaturated fats in the diet. This choice will strongly affect lipid profile, and may also affect oxidative state, rate of cholesterol efflux from fibroblasts, blood pressure, weight, insulin resistance, post-prandial triacylglycerol response, blood clotting factors and platelet aggregation. There may also be other effects which we are not yet aware of. Evidence of beneficial effect on one risk factor does not rule out an opposite effect on another unstudied risk factor, and therefore an overall null (or harmful) effect of intervention. The best way of combining the effects on all of these risk factors is to not study risk factors, but to study the effects of dietary change on important outcomes, on cardiovascular morbidity and mortality, and on total mortality. diet which aims to modify serum lipid levels. This has crystallized as the American Heart Association Step 1 and 2 diets. These still form the basis of more extensive dietary recommendations by the American Heart Association (Stone 1996;Krauss 1996). How effective are these alterations in dietary fat at reducing cardiovascular morbidity and mortality?

O B J E C T I V E S
The aim of this systematic review was to assess the effect of change in dietary fats, which would be expected to result in lipid lowering, on mortality and cardiovascular morbidity, using all available randomized clinical trials.

Types of studies
Randomized controlled trials only. Randomization of individuals was accepted, or of larger groups where there were at least 6 of these groups randomized. Randomization was excluded where it was clear that allocation concealment did not occur (e.g. divisions based on days of the week or first letter of the family name were excluded).

Types of participants
Studies of adults (18 years or older, no upper age limit) at any risk of cardiovascular disease (with or without existing cardiovascular disease) were accepted. Participants could be of either gender, but those who were acutely ill, pregnant or lactating were excluded.

Types of interventions
All randomized controlled trials of interventions stating an intention to reduce or modify dietary fat or cholesterol, such as would be expected to result in improvement of serum lipid profile, were considered. The intervention had to be dietary advice, supplementation (of fats, oils or modified or low fat foods) or a provided diet, and the control group usual diet, placebo or a control diet. Interventions excluded (unless they were present in addition to those above) were addition of alpha-linolenic acid, omega-3 fats or fish oils (as the mechanism of action of these fats is probably mainly anti-thrombotic or anti-arrhythmic), high fibre diets and garlic (as pulses, fruits and vegetables may have various effects other than lipid lowering), low calorie diets or exploration of varying forms of carbohydrate (unless also specifically low in fat or fat modified).
Also excluded were all multiple risk factor interventions other than diet or supplementation (unless the effects of diet or supplementation could be separated). Trials were only included where primary outcome data (mortality or cardiovascular morbidity) could be collected (by communication with authors if necessary).

Types of outcome measures
Primary outcomes: The main outcomes were total and cardiovascular mortality. The other important outcome was combined cardiovascular events, which included any of the following data available from a trial: cardiovascular deaths, cardiovascular morbidity (non-fatal myocardial infarction, angina, stroke, heart failure, peripheral vascular events) and unplanned cardiovascular interventions (coronary artery bypass surgery or angioplasty). Secondary outcomes: Secondary outcomes included risk factor changes (weight, blood pressure, total, LDL or HDL cholesterol and triglyceride levels) and quality of life measures (feelings of health, time off work).
Cochrane Review Groups in areas related to this review include the Diabetes Group (now the proposed Endocrine and Metabolic Disorders Group), Stroke Group, Renal Group, Hypertension Group and Peripheral Vascular Disease Group. The groups were contacted and asked to search their trial registers for relevant trials. Bibliographies of all identified systematic reviews, major non-systematic reviews and included trials were searched for further trials. Experts in the field were contacted (May 1999) for references to studies not yet identified by the search process. The 60 experts were defined as persons who served as author (not necessarily the primary author) on a trial meeting inclusion criteria for the review, or the contact author for any relevant systematic review or extensive non-systematic review. All contacted authors of trials were also asked whether they knew of trials which may have been missed. Attempts were made to obtain translations of relevant non-English articles, or contact with the author was established to enable assessment of eligibility.

Data collection and analysis
DATA COLLECTION Articles were only rejected on initial screen if the reviewer could determine from the title and abstract that the article was not a report of a randomized controlled trial; or the trial did not address a low or modified fat diet; or the trial was exclusively in children less than 18 years old, pregnant women or the critically ill; or the trial was of less than 6 months duration; or the intervention was multi-factorial. When a title/abstract could not be rejected with certainty, the full text of the article was obtained for further evaluation. The inclusion of studies was assessed independently by two assessors (LH and RLT) and differences between reviewers' results resolved by discussion and, when necessary, in consultation with a third reviewer (RAR). Trials were categorised as "possible" (where all inclusion criteria appeared to be met or where the ascertainment, or otherwise, of outcome events was uncertain, to be resolved by writing to the author) or "excluded". Attempts were made to contact all authors of "possible" trials in order to confirm or ascertain whether inclusion criteria were met. A data extraction form was designed for this review. Data concerning participants, interventions and outcomes, trial quality characteristics (Chalmers 1990), data on potential effect modifiers including participants baseline risk of cardiovascular disease, trial duration, intensity of intervention (dietary advice, diet provided, dietary advice plus supplementation, supplementation alone), medications used (particularly lipid lowering medication) and smoking status, numbers of events and total patient years in trial were extracted. Where provided, data on risk factors for cardiovascular disease including blood pressure, lipids and weight were collected Baseline risk of cardiovascular disease was defined as follows: high risk are participants with existing vascular disease including a history of myocardial infarction, stroke, peripheral vascular disease, angina, heart failure or previous coronary artery bypass grafting or angioplasty; moderate risk are participants with a familial risk, dyslipidaemia, diabetes mellitus, hypertension, chronic renal failure; low risk are other participants or mixed population groups. Original reports of trial results were extracted by two reviewers (LH and RLT). Differences were resolved by discussion. DATA SYNTHESIS Primary measures of interest were the effect of intervention on 1. total and cardiovascular mortality 2. combined cardiovascular events (including cardiovascular deaths, non-fatal myocardial infarction, stroke, angina, heart failure, peripheral vascular disease, angioplasty and coronary artery bypass grafting) 3. quality of life measures. Pre-specified analyses included: Meta-analysis of data on the following outcomes: · total mortality · cardiovascular mortality · combined cardiovascular events Each of these was ranked according to the percentage energy from fat in the control group, starting high. Meta-analysis, sub grouping by · trials with mean follow-up time over 2 years · initial level of risk (low, medium, high) · mode of intervention (advice, supplementation or provision of diet) for total mortality and total cardiovascular events as outcomes. Meta-regression on total mortality outcome and total cardiovascular events by change in · difference in total fat as a percentage of energy between the intervention and control groups · difference in total serum cholesterol between the intervention and control groups The data were in the form of rates. Treatment effect was measured as a rate ratio and meta-analysis performed as a weighted average of (ln) rate ratios (as described by Hasselblad 1995). For trials with a zero in one arm of the data a small number (0.5) was added to the number of events in both groups. Trials where it was known that there were no events in either intervention group were included in the review for completeness, but could not be included in the meta-analysis. Where trials ran one control group and more than one included intervention group, data from each intervention group were used and the events and patient-years in the control group were divided into equal shares. This resulted in fractional numbers of events in some cases. It was planned that if trials randomized by cluster were identified the patient numbers would be reduced to an "effective sample size" (as described by Hauck 1991), however none were identified that were both included and had cardiovascular events or deaths. Meta-analysis was performed (by JPTH) using random effects methodology (DerSimonian 1986) within S-PLUS (Higgins 1999). Random effects meta-regression (Berkley 1995) was per-formed using the STATA command metareg (Sharp 1998). Funnel plots were drawn to examine the possibility of publication bias (Egger 1997).

Description of studies
See: Characteristics of included studies; Characteristics of ongoing studies. Twenty seven studies are included in the review and are described in the table 'characteristics of included studies'. Four more trials are ongoing, described in the table 'characteristics of ongoing studies'. 219 trials have been excluded, and the reasons for these exclusions are described in the list of references. After the trial author and year the number in brackets refers to the reason for exclusion. These are as follows: 1: the trial was not randomized, or was not adequately randomized, or there were less than six groups for cluster randomization, 2: there was no control group, or no usual or control diet or placebo group for the dietary intervention arm of the trial, 3: the stated aim of the intervention was not reduction or modification of dietary fat or cholesterol intake (increasing omega-3 fats was excluded), 4: the intervention was multi factorial and the effects of the dietary intervention could not be separated from those of other types of intervention, 5: the intervention group were not adult humans, or were acutely ill or were pregnant, 6: the intervention (diet provided or supplementation) did not continue for at least 6 months (or 26 weeks or 180 days) or (dietary advice) the participants were not followed up for at least 6 months, 7: neither mortality nor cardiovascular morbidity data were available (this was only decided definitely after contact with at least one author), trials where it was known that no events occurred were included. A few studies remain where contact with the authors has not yet been established, or contact with the authors has not yet ascertained whether it is known that events occurred. These studies have been included on all criteria above except for number 7. They are at the end of the list of Studies awaiting assessment, labelled 'Z pending', other studies in this list have not yet been assessed for inclusion in duplicate. The 27 included trials comprise 40 distinct intervention arms. Papers describing these trials range in publication date from 1964 to 1998, and were conducted in North America (11), Europe (15) and Australia (1). Seven of the trials include only people at high risk of cardiovascular disease, six at moderate risk, 14 at low risk. All of the high risk trials were men only, women were included in five low risk trials and eleven low or medium risk trials. Thus most of the included events occurred in men. Of the 40 intervention arms only 17 provide useable event data, and only 8 provide data on more than 10 events in total. Dietary interventions varied from trials which provided the majority of food for their participants over several years of study (2), trials which advised diets with dietary fat restriction or modification (17), and those which provided a combination of dietary advice and supplementation (8). The goals of the dietary alteration varied enormously, aiming for fat levels between 15 and 45 per cent of dietary energy, either reducing total fat or replacing saturated by unsaturated fats, sometimes aiming to reduce dietary cholesterol. Specific advice was sometimes given as to type of carbohydrates to be used, calorie restrictions, amounts of fibre, amounts of fruit and vegetables, poultry and fish. Supplements included oil (to be drunk daily or used in cooking), low or modified fat foods supplied by a trial shop, margarines, milk, oily fish, vitamin supplements and fibrous biscuits. Of the 40 intervention arms included 15 aimed only to lower total fat intake (of which two had cardiovascular events), 14 aimed to modify the type of fat eaten (of which ten had cardiovascular events), nine arms aimed to both lower total fat and modify the type of fat eaten (of which five had cardiovascular events), one aimed only to lower dietary cholesterol intake and the final arm did not state what its dietary aims were (neither of these two had events). Of these trials only nine stated that an intended outcome was to assess mortality or cardiovascular morbidity of some sort. A further 13 intended to monitor lipid or cardiovascular risk factor outcomes, and the remainder aimed to assess the following outcomes: bile acid kinetics, feasibility of dietary intervention, occurrence of retinopathy or skin cancer and recurrence of neoplastic polyps.

Risk of bias in included studies
All trials included were randomized controlled trials. Those with detected pseudo random allocation (for example where participants are randomized according to birth date or alphabetically from their name) were excluded. It is often difficult to assess whether the allocation group was concealed from the person deciding on eligibility for the trial, but the actual phrase describing the process of randomization (from published or unpublished material) is included in the table on characteristics of included studies. Allocation concealment was not duplicate data extracted, but for most studies it would probably be 'unclear'. Physician blinding (for the purpose of diagnosing outcomes) makes little difference where total mortality is the outcome, but is important for all other outcomes. Blinding was adequate for 11 trials, inadequate for three and unclear for 13. Participant blinding is difficult in dietary trials, but possible where all or some food is provided by the trial. Participant blinding was adequate for three trials, inadequate for 22 and unclear for two.
A systematic difference in care between the control and intervention groups (such that any differences in the results of the trial might result from these differences and not the dietary intervention) was not present in 17 trials, "minor" in eight, present in one and unclear in one. There was never any indication that there was a difference in the use of medications between the control and intervention groups (which could potentially have swamped out any differential effects of diet).

Effects of interventions
Overall 18,196 people were included in the 27 included trials (8647 in the control groups, 9549 in the intervention groups), over 30,901 person-years of observation (15096 control, 15806 intervention). Details of the extracted rate data are seen in Table  1, 'Outcome data from included trials'. 1430 total deaths were documented (520 in high risk groups), 812 cardiovascular deaths (393 in high risk groups) and 1216 combined cardiovascular events (721 in high risk groups). Of the 27 trials, 13 were documented as having had no mortality and/or no cardiovascular events. Three trials had known events, but it has not been possible to ascertain the randomization group for these people (Oxford Retinopathy 34 deaths, BDIT Pilot Studies three deaths, Low Fat in Breast CA at least two deaths). All of the seven trials which included high risk participants (DART, London Corn/Olive, London Low Fat, MRC Soya, Oslo Diet-Heart, STARS, Sydney Diet-Heart) did have documented events, as did four trials in low risk groups (Minnesota Coronary, National Diet-Heart, Veterans Admin and Veterans Skin CA). Data on quality of life outcomes were only found for one trial, and so were not extracted. Funnel Plot A funnel plot was drawn to indicate whether publication bias was likely (using total mortality data). Only trials with events can be plotted by this method. The funnel plot appears fairly symmetrical suggesting an absence of serious bias. The funnel plot can be viewed on the web site of the Cochrane Heart Group (http: //www.epi.bris.ac.uk/cochrane/heart.htm). Inter-rater agreement The kappa statistic for inter-rater agreement on including or excluding potential trials was 0.61.

Meta-analyses
The numerical results of all meta-analyses performed are shown in Table 2, 'Results of random effects meta-analyses and sub grouping'. (As meta-analysis was performed using rate data it is not possible to display the pictorial results of these calculations within the Cochrane Library. They can be viewed on the web site of the Cochrane Heart Group (http://www.epi.bris.ac.uk/cochrane/ heart.htm). Meta-analysis suggests that, over 30,901 person-years of observation, for people of varying risk of cardiovascular disease, there is no significant effect of alteration in quantity and/or quality of di-etary fat on total mortality. Our best estimate of the rate ratio is 0.98 (95% CI 0.86 to 1.12). A rate ratio of 1.0 would indicate no effect, less than 1.0 suggests benefit from the intervention (in this case dietary fat modification), and greater than 1.0 suggests harm from the intervention. The effect on cardiovascular mortality suggests a trend towards protection by modification of dietary fat, but this is not statistically significant, rate ratio 0.91 (95% CI 0.77 to 1.07). The trend towards protection is strengthened when the effect on combined cardiovascular events is considered, this is significant, with a rate ratio of 0.84 (95% CI 0.719 to 0.986). Meta-analysis was repeated excluding the results of the Oslo Diet-Heart trial which provided oily fish to participants in the low dietary fat arm. As oily fish appears to reduce mortality and cardiovascular events in high risk people (Hooper 1999), it may be the oily fish rather than the low fat diet providing the observed effect. Removing the Oslo Diet-Heart trial attenuated the rate ratios for all three main outcomes (total mortality 1.02, 95% CI 0.91, 1.14; cardiovascular mortality 0.94, 95% CI 0.79, 1.11; combined cardiovascular events 0.86, 95% CI 0.72, 1.03). The rate ratio for combined cardiovascular events was no longer significant. Each of the intervention arms in each of the trials was ranked according to the percentage energy from fat in the control group, starting high, for each of the three preceding meta-analyses. There was no obvious trend (to the eye) as a result of this ordering. Sub grouping Within the above meta-analyses there was no significant statistical heterogeneity, however the trials performed varying interventions on groups at very different cardiovascular risk so that some clinical heterogeneity was certainly present. For this reason random effects meta-analysis was performed (Mosteller 1992), and the effects of this clinical heterogeneity was explored by sub-group analysis and meta-regression (Thompson 1991). Sub grouping was by mean follow-up time in trial, by initial level of cardiovascular risk and by the style of dietary intervention. These were each explored for two outcomes, total mortality and combined cardiovascular events. Exploring heterogeneity through sub grouping for length of time in trial, for level of cardiovascular risk and for style of intervention still offered no significant effects on mortality, although there was a suggestion of increased protection during trials of more than 2 years. Trials where participants were involved for more than 2 years on average did show significant reductions in the rate of combined cardiovascular events (the pooled estimate of the rate ratio was 0.76, 95% CI 0.65 to 0.90, compared to a rate ratio of 0.84 for all trials combined). The reduction in events remained statistically significant when the results of the Oslo Diet-Heart trial were omitted from the analysis. Trials of those at high initial cardiovascular risk (pooled estimate of the rate ratio was 0.84, 95% CI 0.70 to 0.99) suggested very similar levels of protection from combined cardiovascular events as trials of those at low cardiovascular risk (the pooled estimate of the rate ratio was 0.82, 95% CI 0.56 to 1.20, compared to a rate ratio of 0.84 for all trials combined) despite the estimate for those at low cardiovascular risk not being statistically significant. The style of dietary modification (dietary advice, supplementation or diet provided) did not influence rate ratios. Meta-regression Meta-regression was used to explore the effects of changing the percentage of energy from fat and of altering serum cholesterol levels on two outcomes, total mortality and combined cardiovascular events. For the two calculations involving percentage of energy from total fat, the information extracted from each trial was the percentage of energy from fat achieved in the intervention group, minus the percentage of energy from fat achieved in the control group (so that where the fat intake is lower in the intervention group, the number used is negative). Similarly, for the two calculations involving serum total cholesterol, the information extracted was the serum total cholesterol (in mmol/litre) achieved in the intervention group, minus the serum total cholesterol achieved in the control group (so that where the serum cholesterol is lower in the intervention group, the number used is negative). The numerical results of all the meta-regressions performed are shown in Table 3, 'Results of random effects meta-regression', the visual representations can be viewed on the web site of the Cochrane Heart Group (http://www.epi.bris.ac.uk/cochrane/ heart.htm). Rate ratios for total mortality and total cardiovascular events dropped as the percentage energy from fat fell, and as total serum cholesterol levels fell. However, none of the trends were statistically significant. This may be in part due to the large differences in interventions between the trials. It is also the case that the larger trials did not reduce dietary fat extensively, so that these trials are clustered together leaving only smaller trials to suggest the actual slope of the relationship, making a statistically significant correlation less likely.

D I S C U S S I O N
This review suggests that dietary fat reduction or modification may be protective of cardiovascular events, but this is still not clear.
The National Diet-Heart study (published in 1968) was carried out as a pilot study for a large scale test of the efficacy of dietary fat modification in the general (male) population on cardiovascular morbidity and mortality. The definitive trial was never begun due to cost considerations. It is unlikely now that this failure to conduct the definitive trial will ever be rectified.

Length of follow-up
If dietary fat modification has some immediate effect on mortality or cardiovascular morbidity (for example, by altering clotting) then number of years observation on each individual may not be very important, and effect could be seen in a trial with many participants followed over a short time, as well as in a trial with fewer participants followed over a long time. If their main effect takes some time to manifest itself (for example by slowly altering degree or type of an atherosclerotic plaque) then the effect may be seen after a "lag" period. In this case the trial with many participants followed over a short time may show no effect at all, but the trial with fewer participants followed over a longer time period will be more likely to show an effect, even if the total number of personyears of observation is the same.
In the 4S trial (4S 1994) 4444 participants were followed for roughly 19,339 person-years of observation, a mean of 4.35 years each. The Kaplan-Meier curve for all-cause mortality for the 4S trial only shows a clear separation between the two randomisation groups at roughly 2 years. For this reason trials within the systematic review were grouped into those with a mean follow-up of two years or less, and those with mean follow-up of more than two years.
Pooled results of dietary fat trials indicate that reduction or modification of dietary fat intake does significantly reduce the incidence of combined cardiovascular events. The effect is consistent with a benefit as large as a 28 per cent reduction in events, with a best estimate of 16 per cent reduction in events. This effect is seen almost exclusively in those who continue to modify their diet over at least two years. The trials with follow-up times from 6 months to 2 years may be diluting the effect of the trials with more than two years follow-up in the overall meta-analysis, but data on time to event were not available (the rate ratio for combined cardiovascular events is 0.84 overall, 0.96 in trials with mean follow-up of two years or less, 0.76 in trials with a mean follow-up of more than two years).
Total mortality was examined as there is no likelihood of ascertainment or diagnostic bias which may occur with cause-specific event outcomes. The data follow a similar trend, with no effect in the shorter trials and a suggestion of benefit in the trials of more than two years, but here the trend is not significant (the rate ratio for total mortality is 0.98 overall, 1.04 in trials with mean followup of two years or less, 0.93 in trials with a mean follow-up of more than two years).
This suggests that the effects of dietary fat modification will take time to manifest themselves, and there is little evidence of immediate effects on factors such as thrombosis. The main effects of dietary fat reduction and modification are likely to be on the scale and type of atherosclerotic plaque, but other mechanisms may be operating.

Degree of lipid lowering
Following the 4S trial it is well established that lipid lowering through use of statins does have a protective effect on people at high risk of cardiovascular disease. This and more recent statin trials have shown a highly significant 25 per cent fall in coronary heart disease mortality (Ebrahim 1998).
If the protective effect of statins relates to their lipid lowering effect then the extent of lipid lowering within the dietary trials might be important. A summary of the lipid lowering effects of the major intervention trials of statins (Ebrahim 1998) suggests an average reduction of total serum cholesterol of over 20 per cent. Within the set of dietary trials used for this review the mean individual initial total serum cholesterol level was 5.8 mmol/litre, and the average change over the trial was a fall of 0.64 mmol/litre (11.1%). This is a much smaller effect on serum cholesterol than that of the statins, and is similar to the fall provided by bibrates which do not appear to reduce clinical events (Ebrahim 1998).
Rather surprisingly much of the total cholesterol reduction in the dietary trials comes from a low risk trial, the Minnesota Coronary trial, as modified institutional food was provided to a vast number of low risk people over only one year on average, resulting in a large reduction in total cholesterol, but with little change in cardiovascular events and a slight increase in mortality. If the Minnesota Coronary trial is excluded the initial total serum cholesterol level within the dietary trials is 6.46 mmol/litre and the mean change in total cholesterol between the control and intervention groups is a fall of 0.47 mmol/litre (7.3%) in the intervention groups, only a third of the total serum cholesterol fall expected with statin therapy.
This relatively small degree of lipid lowering may be a reason that no significant effect of dietary fat intervention was seen on total or cardiovascular mortality in the short term. The larger number of total cardiovascular events than of deaths provides greater statistical power. There was a suggestion from the meta-regression that a greater degree of reduction of total serum cholesterol resulted in a greater reduction in events.

Participants level of risk
As the rate of events will be higher in high risk groups, it should be possible to see the effect of an intervention more rapidly in a high risk group of participants (Davey Smith 1993). There have been suggestions that randomized controlled trials are unsuitable for assessing the effectiveness of interventions with very modest levels of effect in low risk populations, because of the huge numbers of person-years of observation needed to gain sufficient statistical power to avoid Type II errors (Ebrahim 1997). It may be very difficult to disprove effectiveness even when such interventions are clinically useless. In this review a similar level of risk reduction of combined cardiovascular events is seen in both high and low risk groups, but this effect only reaches statistical significance in the high risk partici-pants. This is likely to be due to a relative lack of endpoints in the lower risk population.
When endpoints such as mortality are used the situation becomes more difficult as in low risk groups the proportion of deaths which are unrelated to cardiovascular disease (and unlikely to be influenced by dietary fat changes) rises, again diluting any differences in the numbers of deaths between intervention and control groups. It is more likely that significant changes in cardiovascular deaths will be seen than in total mortality. The trend is certainly in this direction (pooled rate ratio for total mortality 0.98, for cardiovascular mortality 0.91). Our best estimate is that dietary fat reduction and modification result in a reduction of 9 per cent in deaths due to cardiovascular disease, and a reduction of 2 per cent in total deaths, but the confidence intervals are wide.
The high risk participants in the dietary fat trials all show evidence of cardiovascular disease. Under current guidelines most high risk participants with raised lipid levels should be on statin therapy (Wood 1998). This raises the question of whether there is any additional advantage of adherence to a low or modified fat diet in addition to statin therapy. Little evidence exists at present to answer this question. However, in all parts of the world where drug budgets are restricted and use of statins remains rationed even for those at high risk the use of low or modified fat diets would appear to be a cost-effective option leading to considerable reductions in cardiovascular events (and so in health budgets) in only a few years.
The low risk participants are unlikely to be on statin therapy under current guidelines. The suggestion of protection of this group from cardiovascular events, with a reduction of roughly 18 per cent of events, by dietary fat modification (even though this does not reach statistical significance, but taking into account the lack of power) would appear to merit continued public health action.
Low fat or modified fat diets An individuals dietary intake is a complex mixture of foods, each of which is a complex mixture of nutrients. Altering one dietary component leads to unintentional alterations in many others, each of which may have positive or negative effects on several risk factors and, eventually, health.
The fat interventions included in this review are low fat diets (where total fat is reduced, and energy is usually replaced by increasing carbohydrate intake), modified fat diets (where a proportion of saturated fat is replaced by unsaturated fats, and total fat intakes do not alter) and combinations of the two (with some fat reduction and some replacement with unsaturates). Whilst these diets have similar effects on total serum cholesterol levels it may be that their effects on cardiovascular disease incidence and mortality are different. For example, low fat, high carbohydrate diets are likely to result in higher triglyceride and lower HDL cholesterol levels than a diet where saturated fats are wholly replaced by unsaturated fats (Mensink 1992). As only two low fat trial arms with events are included in the meta-analyses, it would not be possible to separate out the effects of the various types of dietary fat changes on mortality and morbidity within this review. But if we aim to achieve best cardiovascular protection (rather than the best cholesterol reduction) we must be clear about exactly what dietary advice is advocated. Further large scale, long term trials with disease end points would be needed to clarify this, but are unlikely to be mounted given the feasibility and cost considerations. However, results of large scale, long term ongoing trials like the Polyp Prevention trial (due to report soon), WINS and Canadian DBCP may help to clarify the effect of low fat diets on total mortality, and also on cardiovascular events in those at low risk of cardiovascular disease.

Improved interventions
Interventions on dietary fat need to result in useful levels of cholesterol reduction and these must be sustained for at least two years to have an impact on levels of cardiovascular events. Systematic reviews of the effect of diet on serum cholesterol levels have suggested that levels of serum cholesterol reduction are much lower in free-living low risk groups than in high risk groups (Ebrahim 1998). We might expect reductions in serum cholesterol of only 3 to 5 per cent even with quite intensive interventions in the general population (Brunner 1997;Tang 1998). Interventions in high risk populations appear to reduce serum cholesterol levels by about ten per cent. This difference is likely to be because of lower levels of motivation and long term dietary compliance in those who have not experienced cardiovascular disease themselves (Tang 1998).
Effective interventions tend to focus specifically on dietary (rather than multiple lifestyle) changes, to incorporate behavioural theories and goals, use active involvement and specific behaviour change strategies, personalise the intervention, provide feedback and multiple contacts and build support through contact with family, colleagues or local leaders. Changing the environment by increasing availability of healthy choices, using simple signs to identify them and/or manipulating food composition without publicising the fact may also be productive (Roe 1997).
There is also confusion about whether low fat or modified fat changes are most effective. It is important that individuals and populations are receiving clear, evidence-based advice about the types of dietary fat changes which are most effective in reducing cardiovascular risk, as well as ways to achieve those changes. Further research comparing low fat and modified fat changes on cardiovascular disease risk factors would be feasible and helpful.
Most of the events analysed in this review come from male participants. It may be that the effect of dietary fats on women's risk of cardiovascular events is distinct from those of men.

Other systematic review results
This review aimed to find all relevant dietary intervention trials which reduced or modified dietary fat intake, followed its participants for at least six months and collected mortality or morbidity data, even when the individual trials were not powered to come to any conclusions about mortality or morbidity.
Results of this systematic review are similar to those of a less rigorous systematic review by Ebrahim and Davey Smith (Ebrahim 1996). They examined ten unifactorial dietary trial arms which resulted in serum cholesterol lowering. They found nonsignificant reductions in both total mortality (odds ratio 0.96, 95% CI 0.85 to 1.08) and coronary heart disease mortality (odds ratio 0.98, 95% CI 0.83 to 1.15).
Results of these systematic reviews conflict with a previous systematic review by Truswell on dietary interventions and mortality or morbidity outcomes (Truswell 1994) which found that dietary interventions significantly reduced total mortality (pooled odds ratio 0.94). However his inclusion criteria were very different, including interventions which did not aim to alter dietary fat or serum cholesterol, multifactorial interventions and a non-randomized trial (the Finnish Mental Hospital trial, Miettinen 1972 (1)), and excluding at least one relevant intervention arm (the olive oil arm of the London Corn/Olive trial).
This review is suggestive that dietary fat alteration is protective against combined cardiovascular events. No significant effect on total mortality is seen, probably because the analysis is under powered (with only half of the high risk observation years of the 4S study, less than half of its total cholesterol lowering effect and few participants involved for long enough to see any effect), but the suggestion is a reduction in total mortality in those following a reduced and/or modified fat diet for at least two years. However, it may be that there is no effect of dietary fat reduction and/or modification on total mortality.

Implications for practice
Dietary change to reduce or modify dietary fat intake appears to reduce the incidence of combined cardiovascular events. This trend is statistically significant for all trials, but when a trial which also increased omega-3 fat intake in the intervention group is excluded the results are no longer statistically significant. The protective effect is seen almost exclusively in those who continue to modify their diet over at least two years. The extent of this protection appears similar in both high and low risk populations, although the relationship does not achieve statistical significance in low risk participants. Dietary advice to those at high risk of cardiovascular disease (particularly where statins may not be available), and probably also to lower risk population groups, should continue to include dietary fat modification and it should be stressed that this is a permanent pattern of eating.
There is a suggestion that dietary fat modification has protective effects on total mortality and on cardiovascular mortality when the dietary modification is followed for at least two years, however this trend is not statistically significant. It may be that not enough people were involved in long term trials to show the protective effect of a change in dietary fat, or it may be that there is no such effect.

Implications for research
The financial implications (costs and savings) of appropriate advice and legislation to modify fat intake in those at various levels of cardiovascular risk should be assessed and reflected in health policy.
It is not clear whether there is additional benefit of modifying dietary fat in those at high risk of cardiovascular disease who are on statins to reduce their cholesterol levels. Most of the trials of statins required participants in both control and intervention groups to receive dietary fat advice. Further research to examine the need for maintenance of dietary fat modification whilst on statins would only be feasible using serum cholesterol changes, but the issue is not of major importance.
Whilst interventions to alter dietary fat intake in individuals at high cardiovascular risk have been fairly successful, such health promotion initiatives in the general population have been less successful. Further work is needed to help high and low risk individuals to make effective changes to dietary fat and to maintain these changes over their lifetimes. Research into the effects of improved labelling, pricing initiatives and improved availability of healthier foods, linking food production and processing into the health agenda may yield huge advances in this area.
It is not clear whether a low fat diet, a modified fat diet, or a combination of both is most protective of cardiovascular events. Results from ongoing trials which are assessing the effects of low fat diets on certain cancers may help to clarify the different effects of low and modified fat diets on mortality.

A C K N O W L E D G E M E N T S
The help of the following investigators in providing information about their own and others trials is gratefully acknowledged: SAA Beresford ( Love and thanks to Richard, Rowan and Robin for their support and understanding.

D A T A A N D A N A L Y S E S
This review has no analyses.

C O N T R I B U T I O N S O F A U T H O R S
All co-reviewers were active in the design of the review and in providing critical revisions of the manuscript. Julian Higgins also performed the statistical analyses, Rachel Thompson duplicated the inclusion / exclusion and data extraction of all studies and Rudolph Riemersma arbitrated on study inclusion where necessary. Shah Ebrahim and Carolyn Summerbell were primary advisors. Lee Hooper originated and was primarily responsible for planning and carrying out the review and was the principal author.

D E C L A R A T I O N S O F I N T E R E S T
LH was employed as a dietitian working in the area of cardiac rehabilitation for much of the duration of this review. RLT and CDS are also dietitians.

Internal sources
• University of Manchester, UK.

External sources
• Studentship, Systematic Reviews Training Unit, Institute of Child Health, University of London, UK.