Reduction in saturated fat intake for cardiovascular disease

Summary of findings 1. Effect of reducing saturated fat compared to usual saturated fat on CVD risk in adults (note: for the full set of GRADE tables see additional tables 24 to 28)

Low saturated fat compared with usual saturated fat for CVD risk
Patient or population: people at any baseline risk of CVD Intervention: lower saturated fat intake Comparison: higher saturated fat intake Settings: Any, including community‐dwelling and institutions. Included RCTs were conducted in North America, Europe and Australia/New Zealand, no studies were carried out in industrialising or developing countries.
Outcomes	Relative effect (95% CI)	Anticipated absolute effects (95% CI)		No of Participants (studies)	Quality of the evidence (GRADE)	Comments
Outcomes	Relative effect (95% CI)	Risk with higher SFA intake	Risk with lower SFA intake	No of Participants (studies)	Quality of the evidence (GRADE)	Comments
All‐cause mortality follow‐up mean duration 56 months¹	RR 0.96 (0.90 to 1.03)	62 per 1000	60 per 1000 (56 to 64)	55,858 (12)	⊕⊕⊕⊝ Moderate^2,3,4,5,6	Critical importance. Reducing saturated fat intake probably makes little or no difference to all‐cause mortality.
Cardiovascular mortality follow‐up mean duration 53 months¹	RR 0.94 (0.78 to 1.13)	19 per 1000	18 per 1000 (15 to 22)	53,421 (11)	⊕⊕⊕⊝ Moderate^2,3,4,6,7	Critical importance. Reducing saturated fat intake probably makes little or no difference to cardiovascular mortality.
Combined cardiovascular events follow‐up mean duration 52 months¹	RR 0.83 (0.70 to 0.98)	85 per 1000	70 per 1000 (59 to 83)	53,758 (13)	⊕⊕⊕⊝ Moderate^4,8,9,10,11	Critical importance. Reducing saturated fat intake probably reduces cardiovascular events (to a greater extent with greater cholesterol reduction).
Myocardial infarctions follow‐up mean duration 55 months	RR 0.90 (0.80 to 1.01)	32 per 1000	29 per 1000 (25 to 32)	53,167 (11)	⊕⊝⊝⊝ VeryLow ^3,4,5,11,12	Critical importance. The effect of reducing saturated fat intake on risk of myocardial infarction is unclear as the evidence is of very low quality.
Non‐fatal MI follow‐up mean duration 55 months¹	RR 0.97 (0.87 to 1.07)	26 per 1000	25 per 1000 (23 to 28)	52,834 (8)	⊕⊕⊝⊝ Low^3,4,5,6,13	Critical importance. Reducing saturated fat may have little or no effect on risk of non‐fatal myocardial infarction.
Stroke follow‐up mean duration 59 months¹	RR 0.92 (0.68 to 1.25)	22 per 1000	20 per 1000 (15 to 27)	50,952 (7)	⊕⊝⊝⊝ VeryLow ^3,4,6,13,14	Critical importance. The effect of reducing saturated fat on the risk of stroke is unclear as the evidence was of very low quality.
CHD mortality follow‐up mean duration 65 months¹	RR 0.97 (0.82 to 1.16)	16 per 1000	16 per 1000 (13 to 19)	53,159 (9)	⊕⊕⊝⊝ Low^2,3,4,6,14	Critical importance. Reducing saturated fat intake may have little or no effect on CHD mortality.
CHD events follow‐up mean duration 59 months¹	RR 0.83 (0.68 to 1.01)	42 per 1000	35 per 1000 (29 to 43)	53,199 (11)	⊕⊝⊝⊝ Verylow ^4,5,6,12,15	Critical importance. The effect of reducing saturated fat on risk of CHD events is unclear as the evidence is of very low quality.
*The risk in the intervention group (and its 95% confidence interval) is based on the assumed risk in the comparison group and the relative effect of the intervention (and its 95% CI). CI: Confidence interval; RR: Risk Ratio; CHD: coronary heart disease.
GRADE Working Group grades of evidence High quality: Further research is very unlikely to change our confidence in the estimate of effect. Moderate quality: Further research is likely to have an important impact on our confidence in the estimate of effect and may change the estimate. Low quality: Further research is very likely to have an important impact on our confidence in the estimate of effect and is likely to change the estimate. Very low quality: We are very uncertain about the estimate.
¹Minimum study duration was 24 months. ²Risk of bias. Limiting trials to those at low summary risk of bias also suggested little or no effect. Not downgraded. ³Inconsistency. We found no important heterogeneity; I² ≤ 30%. Not downgraded. ⁴Indirectness. These RCTs directly assessed the effect of lower vs higher saturated fat intake on health outcomes of interest. Participants included men and women with and without CVD at baseline (also some participants with CVD risk factors like diabetes, or at risk of cancers). However, no trials included participants from developing countries. Not downgraded. ⁵Imprecision. The 95% CI includes both no effect and a benefit. Downgraded once. ⁶Publication bias. The funnel plot, and comparison of fixed‐ and random‐effects meta‐analyses did not suggest major small‐study (publication) bias. Not downgraded. ⁷Imprecision. The 95% CI includes both harm and benefit. Downgraded once. ⁸Risk of bias. Limiting trials to those at low summary risk of bias suggested a smaller and non‐statistically significant effect (RR 0.96, 95% CI 0.76 to 1.20) suggesting little or no effect on risk of CVD events. Downgraded once (along with publication bias). ⁹Inconsistency. Although heterogeneity was high, I² = 65%, this was mostly explained by the degree of cholesterol‐lowering (a dose effect). Not downgraded. ¹⁰Imprecision. The 95% CI includes only benefit or minimal effect. Not downgraded. ¹¹Publication bias. The funnel plot did not suggest publication bias, but comparison of fixed‐ and random‐effects meta‐analyses suggested possible small‐study (publication) bias. Downgraded once (along with risk of bias, downgraded once in total). ¹²Risk of bias. Limiting trials to those at low summary risk of bias moved the RR slightly towards 1.0, suggesting little or no effect on total MI. Downgraded once. ¹³Risk of bias. Limiting trials to those at low summary risk of bias moved the RR slightly away from 1.0, suggesting that reducing SFA reduces the risk of non‐fatal MI. This was also seen in several other sensitivity analyses. Downgraded once. ¹⁴Imprecision. The 95% CI includes both important benefits and important harms. Downgraded twice. ¹⁵Inconsistency. Heterogeneity was high, I² = 65%. Downgraded once.

Background

In 1949, Ryle and Russell in Oxford documented a dramatic increase in coronary heart disease (CHD), and the Registrar General’s Statistical Tables of 1920 to 1955 showed that there had been a 70‐fold increase in coronary deaths during this 35‐year period (Oliver 2000; Ryle 1949). This sudden surge in coronary heart disease sparked research into its causes. A case‐control study published in 1953 of 200 post‐myocardial infarction patients and age‐matched controls established that those with disease had higher plasma cholesterol levels (Oliver 1953).

Meanwhile in 1949 in the USA, Gofman had separated lipids into lipoprotein classes through ultra centrifugation, describing the LDL as ‘atherosclerogenic’ (Gofman 1949). The following year Keys 1950 proposed that the concentration of plasma cholesterol was proportional to dietary saturated fatty acids (SFA) intake. This relationship was confirmed in work by Hegsted (Hegsted 1965; Hegsted 2000), who published an equation explaining the relationship in 1965 and subsequently in 2000. The equation suggests that dietary saturated fat increases serum cholesterol and so increases cardiovascular (CV) risk, while polyunsaturated fats (PUFA) reduce both. This has since been further refined:

Δ serum cholesterol (in mg/dL) = 2.16 * Δ dietary saturated fat intake (as percentage of energy) – 1.65 * Δ dietary PUFA intake (as percentage of energy, %E) + 6.77 * Δ dietary cholesterol intake (in units of 100 mg/day) – 0.53

The Seven Countries Study compared CHD mortality in 12,000 men aged 40 to 59 in seven countries, and found positive correlations between CHD mortality and total fat intake in 1970, then in 1986 between CHD mortality and saturated fat intake (Keys 1986; Thorogood 1996). A migrant study of Japanese men living in different cultures confirmed in 1974 that men in California had the diet richest in saturated fat and cholesterol, and the highest CHD rates, those in Hawaii had intermediate saturated fat and CHD rates, and those in Japan had a diet lowest in saturated fat and cholesterol, and the least CHD (Kagan 1974; Robertson 1977). However, systematic reviews of the observational data have not confirmed these early studies. Skeaff 2009 included 28 USA and European cohorts (including 6600 CHD deaths among 280,000 participants) investigating the effects of total, saturated, monounsaturated, trans and omega‐3 fats on CHD deaths and events. They found no clear relationship between total, saturated or monounsaturated fat (MUFA) intake and coronary heart disease events or deaths. There was evidence that trans fats increased both coronary heart disease events and deaths, and that total PUFAs and omega‐3 fats decreased them. Intervention studies are needed to clarify cause and effect, to ensure that confounding is not hiding true relationships, or suggesting relationships where they do not exist. Trials also directly address the issue of whether altering dietary saturated fat in adults is helpful in reducing the risk of CVD in the general population and in those at high risk. Intervention trials are crucial in forming the basis of evidence‐based practice in this area.

Most intervention studies have assessed effects 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 reviews in this area follow the same pattern. There are systematic reviews of the effect of dietary fat advice on serum lipid levels (Brunner 1997; Clarke 1997; Denke 1995; Kodama 2009; Malhotra 2014; Mensink 1992; Mensink 2003; Rees 2013; Weggemans 2001; Yu‐Poth 1999), suggesting that dietary changes cause changes in serum lipids. There are also systematic reviews on the effect of lipid level alterations on CV morbidity and mortality (Briel 2009; De Caterina 2010; Law 1994; Robinson 2009; Rubins 1995; Walsh 1995), suggesting that changes in lipids do affect CVD risk. Other risk factors dealt with in a similar way are blood pressure (Bucher 1996; Law 1991; Shah 2007), body weight or fatness (Astrup 2000; Hession 2009; SIGN 1996), angiographic measurements (Marchioli 1994), antioxidant intake (Ness 1997), metabolic profile (Kodama 2009) and alcohol intake (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 CVD. An example of this is the choice of substitution of saturated fats by carbohydrate, PUFAs, MUFAs or protein in the diet. This choice may alter lipid profile, and may also affect blood pressure, body weight, oxidative state, rate of cholesterol efflux from fibroblasts, insulin resistance, post‐prandial triacylglycerol response, blood clotting factors, and platelet aggregation. There may also be further risk factors of which we are not yet aware. 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. While understanding the effects of dietary advice on intermediate risk factors helps to ensure diets are truly altered by advice, and illuminates mechanisms, 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 CV morbidity and mortality, and on total mortality.

Substantial randomised controlled trial data on the effects of dietary fat on mortality and morbidity do exist and have been previously reviewed (Abdelhamid 2020; Abdelhamid 2018; Abdelhamid 2019; Brainard 2020; Brown 2019; Deane 2019; Hanson 2020; Hooper 2018; Hooper 2019; Hooper 2012). A recent very large trial, the Women's Health Initiative, that included over 2000 women with, and over 48,000 women without, CVD at baseline for over eight years (WHI 2006)) has raised many questions about both the effects of fat on health and on how we best conduct research to understand the relationship (Astrup 2011; Michels 2009; Prentice 2007; Stein 2006; Yngve 2006). We incorporated these findings into an update of a Cochrane review on dietary fat and CVD risk with a search in 2010 (Hooper 2012), finding reductions in cardiovascular events in studies that modified dietary fat, and in studies of at least two years' duration, but not in studies of fat reduction or studies with less than two years' follow‐up.

Why it is important to do this review

Public health dietary advice on prevention of cardiovascular disease (CVD) has changed over time, with a focus on fat modification during the 1960s and fat reduction during the 1990s following the introduction of USA and UK dietary guidance on fat reduction, limiting saturated fat intake to 10% of energy (Harcombe 2015). In 2006, recommendations by the American Heart Association suggested that, among other dietary measures, Americans should "limit intake of saturated fat to 7% of energy, trans fat to 1% of energy, and cholesterol to 300 mg/day by choosing lean meats and vegetable alternatives, fat‐free (skim) or low‐fat (1% fat) dairy products and minimise intake of partially hydrogenated fats" (Lichtenstein 2006). Current American Heart Association guidelines suggest that Americans should "Aim for a dietary pattern that achieves 5% to 6% of calories from saturated fat" and "Reduce percent of calories from saturated fat" (both graded as strong evidence on the basis of effects on serum lipids ‐ trials with cardiovascular outcomes are not referenced or discussed, Eckel 2013). European guidance on the treatment of dyslipidaemia is similarly based on dietary effects on lipids, recommending reduction in saturated fats (ESC/EAS 2011) and referencing Mensink 2003, while the Joint British Societies' guidance on preventing CVD recommends a healthy diet including low saturated fat intake (Mach 2019), referencing a variety of evidence including several recent systematic reviews. This is reflected in UK Scientific Advisory Committee on Nutrition recommendations that "dietary reference value for saturated fats remains unchanged: the [population] average contribution of saturated fatty acids to [total] dietary energy be reduced to no more than about 10%", and that "saturated fats are substituted with unsaturated fats. More evidence is available supporting substitution with PUFA than substitution with MUFA" (SACN 2019).

Recent UK National Institute for Health and Care Excellence(NICE) guidance suggests that for people at high risk of or with CVD that they "eat a diet in which total fat intake is 30% or less of total energy intake, saturated fats are 7% or less of total energy intake, intake of dietary cholesterol is less than 300 mg/day and where possible saturated fats are replaced by monounsaturated and polyunsaturated fats". This statement was based on long‐term randomised controlled trials reporting hard outcomes, and NICE separately assessed effects of high polyunsaturated diets, including four of the trials included in this review (NICE 2014).

We were interested in assessing the direct evidence from trials of the effects of reducing saturated fats, and considering what macronutrients the saturated fats were replaced by, updating Hooper 2015a. This update also supports a request from the World Health Organization Nutrition Guidance Expert Advisory Group (WHO NUGAG) to more accurately assess effects of reducing saturated fats on all‐cause mortality, CV morbidity and other health outcomes, and to consider the differential effects on health outcomes of replacement of the energy from saturated fat by other fats, carbohydrates or protein.

Objectives

To assess the effect of reducing saturated fat intake and replacing it with carbohydrate (CHO), polyunsaturated (PUFA) or monounsaturated fat (MUFA) and/or protein on mortality and cardiovascular morbidity, using all available randomised clinical trials.

Additional World Health Organization Nutrition Guidance Expert Advisory Group (WHO NUGAG) specific questions included:

In adults, what is the effect in the population of reduced percentage of energy (%E) intake from saturated fatty acids (SFA) relative to higher intake for reduction in risk of noncommunicable diseases (NCDs)?
What is the effect on coronary heart disease mortality and coronary heart disease events?
What is the effect in the population of replacing SFA with polyunsaturated fats (PUFAs), monounsaturated fats (MUFAs), carbohydrates (CHO) (refined versus unrefined), protein or trans fatty acids (TFAs) relative to no replacement for reduction in risk of NCDs?
What is the effect in the population of consuming < 10%E as SFA relative to > 10%E as SFA for reduction in risk of NCDs?
What is the effect in the population of a reduction in %E from SFA from 10% in gradual increments relative to higher intake for reduction in risk of NCDs?

Methods

Criteria for considering studies for this review

Types of studies

Randomised controlled trials only. We accepted randomisation of individuals, or of larger groups (clusters) where there were at least six of these groups randomised. We excluded studies where allocation was not truly randomised (e.g. divisions based on days of the week or first letter of the family name), or where allocation was not stated as randomised, and no further information was available from the authors.

Types of participants

We included studies of adults (18 years or older, no upper age limit) at any risk of cardiovascular disease, with or without existing cardiovascular disease, using or not using lipid‐lowering medication. Participants could be of either gender, but we excluded those who were acutely ill, pregnant or lactating.

Types of interventions

We included randomised controlled trials stating an intention to reduce saturated fat (SFA) intake (by suggesting appropriate nutrient‐based or food‐based aims) OR which provided a general dietary aim, such as improving heart health or reducing total fat, that also achieved a statistically significant saturated fat reduction (P < 0.05) during the trial in the intervention arm compared with the control arm. The intervention had to be dietary advice, supplementation of fats, oils or modified or low‐fat foods, or a provided diet, compared to higher saturated fat intake which could be usual diet, higher saturated fat, placebo or a control diet. Intended duration of the dietary intervention was at least two years (24 months or 104 weeks).

We excluded multiple risk factor interventions other than diet or supplementation (unless effects of diet or supplementation could be separated, as in a factorial design, so that the additional intervention was consistent or randomised between the intervention or control groups) and studies that aimed for weight loss in one arm but not the other. Atkins‐type diets aiming to increase protein and fat intake were excluded, as were studies where fat was reduced by means of a fat‐substitute (like Olestra). Enteral and parenteral feeds were excluded, as were formula weight reducing diets.

Examples: studies that reduced saturated fats and encouraged physical activity in one arm and compared with encouraging physical activity in the control were included; studies that reducedsaturated fats and encouraged physical activity in one arm and compared with no intervention in the control were excluded; studies that reduced saturated fats and encouraged fruit and vegetables in one arm and compared with no intervention in the control were included.

Types of outcome measures

Primary outcomes

All‐cause mortality (deaths from any cause)
Cardiovascular (CVD) mortality (deaths from myocardial infarction, stroke, and/or sudden death)
Combined CVD events. These included data available on number of people experiencing any of the following: cardiovascular death, cardiovascular morbidity (non‐fatal myocardial infarction, angina, stroke, heart failure, peripheral vascular events, atrial fibrillation) and unplanned cardiovascular interventions (coronary artery bypass surgery or angioplasty).

To meet our inclusion criteria, trials had to report either deaths or CVD events. These could be reported as serious adverse events (SAEs) or via communication with authors.

Secondary outcomes

Additional health events; the outcomes CHD mortality and CHD events were added at the request of the WHO NUGAG group, and were not present in the original overarching systematic review. For each of these, we assessed number of participants experiencing any of these:
- Myocardial infarction, total (fatal and non‐fatal)
- Myocaridal infarction, non‐fatal
- Stroke
- CHD mortality, which includes death from myocardial infarction or sudden CVD death
- CHD events, which include any of the following: fatal or non‐fatal myocardial infarction, angina or sudden CVD death
- type II diabetes incidence
Blood measures including serum blood lipids
- total cholesterol (TC, mmol/L)
- low‐density lipoprotein (LDL) cholesterol, mmol/L
- high‐density lipoprotein (HDL) cholesterol, mmol/L
- triglyceride (TG), mmol/L
- TG/HDL ratio
- LDL/HDL ratio
- total/HDL ratio
- lipoprotein (a) (Lp(a)), mmol/L
- insulin sensitivity including glucose tolerance (homeostatic model assessment (HOMA), intravenous glucose tolerance test (IV‐GTT), clamp, glycosylated haemoglobin (HbA1C))
Other outcomes including adverse effects reported by study authors
- cancer diagnoses
- cancer deaths
- body weight, kg
- body mass index (BMI, kg/m²)
- systolic blood pressure (sBP, mmHg)
- diastolic blood pressure (dBP, mmHg)
- quality of life (any measure)

As all trials collect data on deaths and cardiovascular events (as serious adverse events if not as planned outcome measures), we only included trials where we knew that at least one primary outcome occurred, by communication with authors if necessary. Where we knew that at least one primary outcome occurred, we included the study even where we were unable to use that data in meta‐analysis. We excluded studies where we knew that no primary outcome events occurred (for a study to be excluded in this way the paper needed to be very explicit about the lack of all outcomes or we received confirmation from the authors) and this was noted as the reason for exclusion. Lack of a single primary outcome only occurs in very small studies or in young cohorts, so omitting these studies will make no difference to effect sizes and very little difference to absolute effect sizes (NNTs etc). All other trials were considered unclear and where we could not gain clarification on events from authors, they were classified as “awaiting assessment”.

For composite outcomes (like CVD events), we worked to collect data on the number of participants in each arm who experienced any type of CVD event, and did not double‐count people (so that a person experiencing a stroke and two heart attacks during a trial was counted as one person experiencing CVD events, not as three CVD events).

We extracted event and continuous outcome data for the latest time point available within the trial, and always at least 24 months from inception. We collected change data (with a measure of variance) for continuous outcomes where these were available, and end data where change data were not provided in usable format.

Search methods for identification of studies

Electronic searches

The updated searches were run on 15 October 2019 on the following databases:

CENTRAL (Issue 10 of 12, 2019, Cochrane Library)
MEDLINE (Epub Ahead of Print, In‐Process & Other Non‐Indexed Citations, MEDLINE Daily and MEDLINE, Ovid, 1946 to October 14, 2019)
Embase (Ovid, 1980 to 2019 week 41).

For this update, we introduced searches of two trials registers on 17 October 2019; Clinicaltrials.gov (www.clinicaltrials.gov) and WHO International Clinical Trials Registry Platform (ICTRP) (apps.who.int/trialsearch/). The searches are described in Appendix 1. The RCT filter for MEDLINE was the Cochrane sensitivity and precision‐maximising RCT filter (Lefebvre 2011), and for Embase, terms as recommended in the Cochrane Handbook were applied (Lefebvre 2011).

As we were updating another Cochrane review relating to dietary fat (Hooper 2015b) at the same time, results of the searches for both reviews were combined and de‐duplicated before assessment of titles and abstracts.

The search to 2014 is described in Hooper 2015a, and previous searches in Hooper 2012. .

Searching other resources

We searched for recent publications of the included studies, to ensure the best possible data set for each study.

Data collection and analysis

Selection of studies

Search results were loaded into Covidence software. All authors independently assessed titles and abstracts from the search, differences were resolved by discussion and, when the findings were not clear cut, the full text was collected for assessment. We only rejected articles on initial screen if the author could determine from the title and abstract that the article was not a report of a randomised controlled trial; the trial did not address a low or modified fat diet; the trial was exclusively in children less than 18 years old, pregnant women or the critically ill; the trial was of less than 24 months duration; or the intervention was multifactorial. When we could not reject a title/abstract with certainty, we obtained the full text of the article for further evaluation.

Data extraction and management

We used a data extraction form designed for earlier versions of this review. We extracted 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 participant years in trial. Where provided, we collected data on risk factors for cardiovascular disease including blood pressure, lipids and weight.

We defined baseline risk of cardiovascular disease 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. Those at low or moderate risk combined are primary prevention trials.

Data were extracted independently in duplicate by AA, FOJ and/or LH, alongside assessment of risk of bias.

Assessment of risk of bias in included studies

We carried out 'Risk of bias' assessment independently in duplicate as part of data extraction. We assessed trial risk of bias using the Cochrane tool for assessment of risk of bias (Higgins 2011). For included RCTs, we also assessed whether each study:

was free of systematic differences in care,
aimed to reduce SFA intake,
achieved SFA reduction, or
achieved total serum cholesterol reduction.

We used the category 'other bias' to note any further issues of methodological concern. Funding was not formally a part of our assessment of bias in RCTs as it is not a core part of the Cochrane 'Risk of bias' tool, but was reported in the Characteristics of included studies.

Two authors (LH, NM) independently extracted validity data from studies identified by the previous search, and resolved differences by discussion.

Poorly concealed allocation is associated with a 40% greater effect size (Schulz 1995), so randomisation and allocation concealment are core issues for all trials. Lack of blinding is associated with bias, though smaller levels of bias than lack of allocation concealment (Savovic 2012), especially in studies with objectively measured outcomes (Wood 2008).

For this review, we introduced the concept of summary risk of bias for whole trials. We considered dietary advice or all‐food‐provided type trials to be at low summary risk of bias where we judged randomisation, allocation concealment, and blinding of outcome assessors to be adequate. Summary risk of bias was considered moderate to high in all other included trials.

Measures of treatment effect

The effect measures of choice were risk ratios (RR) for dichotomous data and mean difference (MD) for continuous data.

Unit of analysis issues

We did not include any cluster‐randomised trials in this review, as no relevant studies included at least six clusters.

Where there was more than one relevant intervention arm but only one control arm, we either pooled the relevant intervention arms to create a single pairwise comparison (where the intervention arms were equivalently appropriate for this review) as described in the Cochrane Handbook (Higgins 2011), or we excluded intervention arms that were not appropriate for this review, or less appropriate than another arm. When two arms were appropriate for different subgroups (Rose corn oil 1965; Rose olive 1965), then we used the control group once with each intervention arm, and divided the number of events in the control group, and the number of participants in the control group, evenly between the two study comparisons.

In the previous version of this review, data for WHI 2006 were presented separately for those without baseline CVD, and with baseline CVD, for most outcomes. This has been altered in this version of the review, so that both sets of data are presented as a single trial except when subgrouping by CVD risk. This has the effect of representing this study in the same way others are represented (which is appropriate), and slightly reducing the weight of the WHI 2006 study in random‐effects meta‐analysis, altering the numbers in the analysis.

When assessing event data, we aimed to assess number of participants experiencing an event (rather than numbers of events), to avoid counting more than one outcome event for any one individual within any one comparison. Where we were unclear (for example, where a paper reported numbers of myocardial infarcts but not by arm), we asked authors for further information.

Dealing with missing data

Where trials satisfied the inclusion criteria of our review but did not report mortality and morbidity, or not by study arm, we tried to contact study authors. This allowed inclusion of studies that would otherwise have had to be excluded. We excluded studies which were otherwise relevant but where we could not establish the presence or absence of primary outcomes, despite multiple attempts at author contact.

It was often unclear whether data on primary or secondary outcome events may still have been missing, and so we did not impute data for this review.

Where included studies used methods to infer missing data (such as carrying the latest measurement forward), then we used these data in analyses. Where this was not done, we used the data as presented.

Assessment of heterogeneity

We examined heterogeneity using the I² test, and considered it important where greater than 50% (Higgins 2003; Higgins 2011). Where we identified important clinical or unexplained statistical heterogeneity, we did not pool but instead summarised the studies in a narrative format. We used the assessment of heterogeneity in our GRADE assessments, so that the quality of evidence was downgraded where heterogeneity was important, and not explained by subgrouping or meta‐regression.

Assessment of reporting biases

We used funnel plots to examine the possibility of small study bias, including publication bias (Egger 1997), for the primary outcomes of total mortality and combined cardiovascular events. For this update, we also compared findings of fixed‐ and random‐effects meta‐analysis since the two methods weight small trials differently, and different effect sizes suggest potential small study bias (Page 2019).

Data synthesis

We carried out data synthesis in the absence of clinical heterogeneity. We used numbers of events in each study arm, and total number of participants randomised, where extracted, and Mantel‐Haenszel random‐effects meta‐analysis carried out in Review Manager 5 software, to assess risk ratios. We extracted event and continuous outcome data for the latest time point available within the trial, and always at least 24 months from inception.

We excluded trials where we knew that there were no events in either group. Where trials ran one control group and more than one included intervention group, we used data from the intervention group providing the comparison that best assessed the effect of altering dietary fat. Where the intervention groups appeared equal in this respect, we merged the intervention groups (simply added for dichotomous data, and using the techniques described in Higgins 2011 for continuous data). We had planned that if we identified trials randomised by cluster we would reduce the participant numbers to an "effective sample size" (as described by Hauck 1991); however, we found none that were both included and had cardiovascular events or deaths.

To assess the WHO NUGAG question on the effect of consuming < 10%E as SFA relative to > 10%E as SFA on the risk of noncommunicable diseases (NCDs) in the population, we combined studies with a control group saturated fat intake of > 10%E and an intervention group saturated fat intake of < 10%E. To assess the effect of a reduction in %E from SFA from 10% in gradual increments relative to higher intake, we repeated this with saturated fat cut‐offs between 7%E and 13%E.

Subgroup analysis and investigation of heterogeneity

Prespecified analyses included:

Effects of SFA reduction compared with usual or standard diet on all (primary and secondary) outcomes and potential adverse effects. This main analysis addressed the main objective of the review and the first WHO specific question.

Prespecified subgroups for all outcomes included:

energy substitution ‐ we intended to subgroup studies according to the main energy replacement for SFA ‐ PUFA, MUFA, CHO (refined or unrefined), protein, trans fats, a mixture of these, or unclear. However, when we presented these data to the WHO NUGAG group, they suggested that this subgrouping be altered. They suggested that we use all studies where SFA was reduced and any of PUFA, MUFA, CHO or protein were statistically significantly increased (P < 0.05) in the intervention compared to the control group to assess the effects of replacement by each, regardless of whether or not it constituted the main replacement for SFA. This meant that some studies appeared in more than one subgroup. As there were almost no data in the studies on trans fats, or on refined and unrefined carbohydrates, we did not include a trans group or distinguish by carbohydrate type. This subgrouping addresses the main objective of the review, and the third WHO specific question.

Further subgroups, run for primary and CVD health‐related secondary outcomes only, included:

Prespecified:

Baseline SFA intake, represented by control group SFA intake (up to 12%E from SFA, > 12 to 15%E, > 15 to 18%E, > 18%E from SFA, or unclear)
Sex (men, women and mixed populations)
Baseline CVD risk (low‐risk or general populations, moderate‐risk populations which were defined by risk factors for CVD such as hypertension or diabetes, high‐risk populations with existing CVD at baseline)
Duration in study (mean duration in trial up to 24 months, > 24 to 48 months, > 48 months, and unclear). Duration was a prespecified subgroup that we used in earlier versions of this review to separate studies with duration of less than two years from those of at least two years. As we have excluded shorter studies from this review, and have access to longer studies, we have explored duration over longer time spans. As some long studies had a high proportion of participants whose time in trial was censored, and we wanted to express mean experience of the trial, we used mean duration of participants in the study, rather than the formal study duration for this subgrouping, so that some two‐year intervention trials, because they had some deaths or dropouts, had a mean duration in trial of 21 or 22 months.

WHO NUGAG added subgroups:

Degree of SFA reduction, represented by the difference between SFA intake in the intervention and control groups during the study (up to 4%E from SFA reduction achieved, > 4 to 8% reduction achieved, > 8% reduction achieved, unclear). We prespecified that we intended to explore the degree of SFA reduction in meta‐regression, but its addition as a subgroup was post hoc, and requested by the WHO NUGAG group.
Serum total cholesterol reduction achieved (reduced by a mean of at least 0.2 mmol/L, reduced by less than 0.2 mmol/L or unclear). We prespecified that we intended to explore the degree of serum total cholesterol reduction in meta‐regression.
Ethnic group. Insufficient information was presented to make this feasible. Hence, we report ethnicity information in the Characteristics of included studies.

We explored the effects of different levels of SFA, PUFAs, MUFAs and total dietary fats, and CHO achieved in trials (all as difference between the intervention and control groups, as %E, and for SFA as a percentage of SFA in the intervention compared with control), baseline SFA intake (as %E), change in total cholesterol (difference between intervention and control groups, in mmol/L), sex, study duration in months, and baseline CVD risk using meta‐regression on total cardiovascular events. We performed random‐effects meta‐regression (Berkley 1995) using the STATA command metareg (Sharp 1998; Sterne 2001; Sterne 2009).

To explore the WHO NUGAG specific question about the effect of the population consuming < 10%E as SFA relative to > 10%E SFA, we assessed effects of all studies where the mean assessed intervention SFA intake was < 10%E and the mean control SFA intake was > 10%E. We explored the effect of reduction of %E from SFA in gradual increments by using cut‐offs of 7%E (where all studies with a mean intervention SFA intake < 7%E and mean control SFA intake > 7%E were pooled), 8%, 9%, 10%, 11%, 12% and 13%. We omitted studies where SFA intakes were not reported from these analyses. For each primary outcome, we plotted the pooled risk ratio of that outcome against the cut‐off, %E from SFA.

Referee‐added subgroups:

In response to the suggestion of a referee of this systematic review, and to better understand the effect of use of statins since the 1990s, we subgrouped studies by decade of publication.

Sensitivity analysis

We carried out sensitivity analyses for primary outcomes assessing the effect of:

Excluding studies which did not state an aim to reduce SFA
Excluding studies which did not report SFA intake during the trial, or did not find a statistically significant reduction in SFA in the intervention compared to the control
Excluding studies where total cholesterol (TC) was not reduced (statistically significant reduction of TC, or of LDL where TC was not reported (considered reduced where P < 0.05), or where reduction was not at least 0.2 mmol/L in intervention compared to control where variance was not reported)
Excluding the largest study (WHI 2006)
Analysis run with Mantel‐Haenszel fixed‐effect model
Analysis run with Peto fixed‐effect model

For this update we also introduced sensitivity analysis excluding trials not at low summary risk of bias. We used results of these analysis to inform GRADE assessment of risk of bias.

GRADE

All primary outcomes, and secondary additional health events, were represented in the 'Summary of findings' table, and underwent GRADE assessment. The GRADE Working Group has developed a common, sensible and transparent approach to grading quality of evidence and strength of recommendations (www.gradeworkinggroup.org/; GRADE 2004). The evidence within this systematic review was first assessed using the GRADE system by the review authors and then discussed and modified by the WHO NUGAG group.

Outcome data were interpreted as follows:

Is there an effect? (options were ‘increased risk’, ‘decreased risk’, or ‘little or no effect’). Our main outcome measure was RR so we decided on existence of an effect using RR. RR > 8% (RR < 0.92 or > 1.08) for the highest quality evidence suggested increased or decreased risk (otherwise little or no effect). The presence or not of an effect was decided on the RR for the main analysis and sensitivity analyses, the highest quality evidence (the main analysis, the sensitivity analyses of trials at low summary risk of bias and at low risk of compliance problems).
For continuous outcomes, reducing SFA was considered to have little or no effect unless effect sizes represented at least 5% change from baseline (or 2% in the case of cumulative outcomes such as adiposity).
Quality of evidence was assessed using GRADE assessment (GRADE 2004) for key outcomes. We used the five GRADE considerations (risk of bias, consistency of effect, imprecision, indirectness and publication bias) to assess the quality of the body of evidence as it related to the studies that contributed data to the meta‐analyses for the prespecified outcomes. We used methods and recommendations described in Section 8.5 and Chapter 12 of the Cochrane Handbook for Systematic Reviews of Interventions (Higgins 2011), plus GRADEpro GDT software (GRADEpro 2015). We justified all decisions to downgrade the quality of studies using footnotes and made comments to aid reader's understanding of the review.
Where there was a suggested effect, the size of effect was assessed using the number needed to treat for an additional beneficial outcome (NNTB), number needed to treat for an additional harmful outcome (NNTH) or absolute risk reduction (ARR).

Results

Description of studies

Results of the search

Figure 1 displays the flow diagram for inclusion of studies. We assessed the 7991 titles and abstracts from the updated electronic search, as well as assessing the 8930 titles and abstracts from the search for our sister review (Hooper 2015b), which de‐duplicated to 15,314 titles and abstracts. Of these, 530 were considered potentially relevant to one or both reviews, so were collected as full text. Ten publications were considered relevant for this systematic review, and these were grouped into:

Figure 1

Study flow diagram for this systematic review (update searches run October 2019).

three new publications for two already included trials (WHI 2006; WINS 2006),
two publications for one study awaiting assessment (not enough details to confirm inclusion, ICFAMED), and
five publications for four ongoing trials (ENAbLE due unclear; NCT02481466 due 2020; NCT02938832 due 2023; NEW Soul Study due 2022).

There were no new included trials, but there were new data for WHI 2006 and WINS 2006, as well as the ongoing studies and the study awaiting assessment. This resulted in an updated review including 15 RCTs (16 comparisons as the Rose trial has two comparisons, Rose corn oil 1965 and Rose olive 1965).

Included studies

We included 15 randomised controlled trials (RCTs) randomising 56,675 participants in the review (Included studies), and describe them in Characteristics of included studies. The interventions are compared in Table 1.

Table 1. Comparison of study interventions for included RCTs

Reference	Population	CVD risk category	Is intervention delivered to Individual or group?	intervention given by?	Face‐to‐face or other?	Number of visits	Is intervention advice only or other intervention?
Black 1994	People with non‐melanoma skin cancer	Low	Unclear	Dietitian	Face‐to‐face	8 x weekly classes then monthly follow‐up sessions	Advice (behaviour techniques learning)
DART 1989	Men recovering from a MI	High	Individual	Dietitian	Face‐to‐face	9	Advice (diet advice, recipes and encouragement)
Houtsmuller 1979	Adults with newly‐diagnosed diabetes	Moderate	Unclear	Dietitian	Unclear	Unclear	Advice?
Ley 2004	People with impaired glucose intolerance or high normal blood glucose	Moderate	Small group	Unclear	Face‐to‐face	Monthly meetings	Advice (education, personal goal‐setting, self‐monitoring)
Moy 2001	Middle‐aged siblings of people with early CHD, with at least 1 CVD risk factor	Moderate	Individual	Trained nurse	Face‐to‐face	6 ‐ 8 weekly for 2 years	Advice (individualised counselling sessions)
MRC 1968	Free‐living men who have survived a 1st MI	High	Individual	Dietitian	Face‐to‐face	Unclear	Advice and supplement (soy oil)
Oslo Diet‐Heart 1966	Men with previous MI	High	Individual	Dietitian	Face‐to‐face and other	Unclear	Advice and supplement (food)
Oxford Retinopathy 1978	Newly‐diagnosed non‐insulin‐dependent diabetics	Moderate	Individual	Diabetes dietitian	Face‐to‐face	After 1 month then at 3‐month intervals	Advice
Rose corn oil 1965	Men (?) with angina or following MI	High	Unclear	Unclear	Unclear	Follow‐up clinic monthly, then every 2 months	Advice and supplement (oil)
Rose olive 1965	Men (?) with angina or following MI	High	Unclear	Unclear	Unclear	Follow‐up clinic monthly, then every 2 months	Advice and supplement (oil)
Simon 1997	Women with a high risk of breast cancer	Low	Individual followed by individual or group	Dietitian	Face‐to‐face	Bi‐weekly over 3 months followed by monthly	Advice (individualised eating plan and counselling sessions)
STARS 1992	Men with angina referred for angiography	High	Individual	Dietitian	Face‐to‐face	Clinic visits at 3‐month intervals	Advice
Sydney Diet‐Heart 1978	Men with angina referred for angiography	High	Individual	Unclear	Face‐to‐face	3 times in 1st year and twice annually thereafter	Advice
Veterans Admin 1969	Men living at the Veterans Administration Center	Low	Individual	Unclear (whole diet provided)	N/A	N/A	Diet provided
WHI 2006	Postmenopausal women aged 50 ‐ 79 with or without CVD at baseline	Low and High	Group	Nutritionists	Face‐to‐face	18 sessions/1st yr and quarterly maintenance sessions after	Advice
WINS 2006	Women with localised resected breast cancer	Low	Individual followed by group	Dietitian	Face‐to‐face	8 bi‐weekly sessions, then 3‐monthly contact and optional monthly sessions	Advice

MI: myocardial infarction
N/A: not applicable

The main study papers ranged in publication date from 1965 to 2006, but with supplementary publications included up to 2019. The RCTs were conducted in North America (six), Europe (seven), and Australia/New Zealand (two); no studies were carried out in industrialising or developing countries. Six RCTs included only people at high risk of cardiovascular disease, four at moderate risk, and four at low risk (three with raised cancer risk or cancer diagnosis, one with no specific health risks), while one trial included participants at low and high CVD risk (WHI 2006, Table 1; this trial made assessments in each of these groups). Seven studies included only men, three only women, and five both men and women. However, as the largest trial (WHI 2006) was in women only, women are the largest group represented. Trial duration ranged from two to more than eight years, with a mean duration of 4.7 years.

The form of interventions varied (Table 1). Interventions were of advice to alter intake in 15 of the 16 intervention arms, and additional supplements such as oil or other foods were provided in three trials (four arms: MRC 1968; Oslo Diet‐Heart 1966; Rose corn oil 1965; Rose olive 1965), while all food was provided in a residential facility in one RCT (Veterans Admin 1969). Of the 15 arms with an advice element, most interventions were delivered face‐to‐face, but this was unclear in three arms (Houtsmuller 1979; Rose corn oil 1965; Rose olive 1965). Advice was provided individually in nine intervention arms (followed by later group sessions in two arms), in groups only in two trials (Ley 2004; WHI 2006), and was unclear in three RCTs (Black 1994; Houtsmuller 1979; Rose corn oil 1965; Rose olive 1965). Advice was provided by a dietitian in nine arms, a nutritionist in one, a trained nurse in one and was unclear in four. Frequency of study visits for advice and follow‐up varied between three times in the first year and twice annually thereafter up to 18 sessions in the first year and quarterly maintenance visits thereafter.

Of the 15 included studies (16 intervention arms), 11 RCTs (12 comparisons) provided data on all‐cause mortality (including 55,858 participants and 3518 deaths), 10 RCTs (11 comparisons) on CV mortality (53,421 participants and 1096 cardiovascular deaths), and 11 RCTs (12 comparisons) on combined cardiovascular CVD events (53,300 participants, of whom 4476 participants experienced at least one CVD event) (Table 2). In two included studies, it was clear that events had occurred, but it was not clear in which arm(s) the events had occurred (Oxford Retinopathy 1978; Simon 1997), so that we could not include the data in the meta‐analyses. Secondary health events and other secondary outcomes were reported in varying number of studies (between 1 and 15 studies reported on any single outcome, see Table 2 and Table 3).

Table 2. Number of participants and number of outcomes for dichotomous variables (by intervention arm)

	Participants	All‐cause mortality	CV mortality	CVD events	MI	Non‐fatal MI	Stroke	CHD mortality	CHD events	Diabetes Diagnoses
Black 1994	133	133	133	133	0	0	0	0	0	0
DART 1989	2033	2033	2033	2033	2033	2033	0	2033	2033	0
Houtsmuller 1979	102	0	0	102	102	0	0	102	102	0
Ley 2004	176	176	176	176	176	0	176	0	176	0
Moy 2001	267	0	0	235	235	235	235	0	267	0
MRC 1968	393	393	393	393	393	393	393	393	393	0
Oslo Diet‐Heart 1966	412	412	412	412	412	412	412	412	412	0
Oxford Retinopathy 1978	249 (data not provided by arm)	0	0	0	0	0	0	0	0	0
Rose corn oil 1965	41	41	41	41	41	41	0	41	41	0
Rose olive 1965	39	39	39	39	39	39	0	39	39	0
Simon 1997	194 (data not provided by arm)	0	0	0	0	0	0	0	0	0
STARS 1992	60	55	55	55	55	0	55	0	55	0
Sydney Diet‐Heart 1978	458	458	458	458	0	0	0	458	0	0
Veterans Admin 1969	846	846	846	846	846	846	846	846	846	0
WHI 2006	48,835	48,835	48,835	48,835	48,835	48,835	48,835	48,835	48,835	48,835
WINS 2006	2437	2437	0	0	0	0	0	0	0	0
Total Participants	56,675	55,858	53,421	53,758	53,167	52,834	50,952	53,159	53,199	48,835
Percent of participants for this outcome	100%	99%	94%	95%	94%	93%	90%	94%	94%	86%

These numbers are the numbers of participants in each study who were available for assessment of outcomes within meta‐analysis (not necessarily the number of participants randomised within the trial).

CHD: coronary heart disease
CV: cardiovascular
CVD: cardiovascular disease

See: Summary of findings 1 Effect of reducing saturated fat compared to usual saturated fat on CVD risk in adults (note: for the full set of GRADE tables see additional tables 24 to 28)

Table 3. Number of participants and number of participants with data for continuous outcomes (by intervention arm)

	Participants	Total cholesterol	LDL cholesterol	HDL cholesterol	Triglycerides	TG/HDL ratio	Total cholesterol/HDL ratio	LDL/HDL ratio	LP (a)	Insulin sensitivity
Black 1994	133	0	0	0	0	0	0	0	0	0
DART 1989	2033	1855	0	1855	0	0	0	0	0	0
Houtsmuller 1979	102	96	0	0	96	0	0	0	0	96
Ley 2004	176	103	103	103	103	0	103	0	0	103
Moy 2001	267	0	235	235	235	0	0	0	0	0
MRC 1968	393	177	0	0	0	0	0	0	0	0
Oslo Diet‐Heart 1966	412	329	0	0	0	0	0	0	0	0
Oxford Retinopathy 1978	249	58	0	0	0	0	0	0	0	0
Rose corn oil 1965	41	22	0	0	0	0	0	0	0	0
Rose olive 1965	39	24	0	0	0	0	0	0	0	0
Simon 1997	194	72	71	72	71	0	0	0	0	0
STARS 1992	60	50	50	50	50	0	50	50	50	50
Sydney Diet‐Heart 1978	458	458	0	0	458	0	0	0	0	0
Veterans Admin 1969	846	843	0	0	0	0	0	0	0	0
WHI 2006	48,835	2832	2832	2832	2832	0	2832	0	2832	2832
WINS 2006	2437	196	0	0	0	0	0	0	0	0
Total Participants	56,675	7115	3291	5147	3845	0	2985	50	2882	3081
Percent of participants for this outcome	100%	13%	6%	9%	7%	0%	5%	0.1%	5%	5%

HDL: high density lipoprotein
LDL: low density lipoprotein
Lp(a): lipoprotein (a)
TG: triglyceride

Excluded studies

We excluded 520 full‐text publications at this update, having assessed the full texts in duplicate. We describe the reasons for some of these exclusions in Characteristics of excluded studies tables. We excluded 29 studies where data on events were not reported in publications and contact with authors confirmed that there had been no deaths or cardiovascular events, where contact with authors confirmed that data were not available, or where we could not establish contact with authors.

Risk of bias in included studies

We display 'Risk of bias' assessments in the individual included study arms in Figure 2.

Figure 2

Methodological quality summary: review authors' judgements about each methodological quality item for each included study. Please note that while Rose 1965 (Rose corn oil 1965; Rose olive 1965) appears twice in this summary, it is a single trial. Rose 1965 was a 3‐arm trial and we have used the two intervention arms separately in the review.

Allocation

All the included trials were randomised controlled trials, and some detail of the randomisation process was provided for all studies, so all were considered at low risk of bias. We excluded those with detected pseudo‐random allocation (for example where participants are randomised according to birth date or alphabetically from their name). We judged allocation concealment to be well done in eight RCTs (eight comparisons, Ley 2004; Oslo Diet‐Heart 1966; Oxford Retinopathy 1978; STARS 1992; Sydney Diet‐Heart 1978; Veterans Admin 1969; WHI 2006; WINS 2006), and unclear in the remainder.

Blinding

Blinding of participants is not easy in dietary studies, as the participants usually have to follow instructions to attain the specific dietary goals. However, it is feasible in some circumstances, including when food is provided via an institutional setting, or meals provided at a central setting and remaining meals packed to take away. It can also be achieved through use of a trial shop, where very specific food‐based dietary advice is provided for all participants, or where the same dietary advice is provided to both groups but a different supplement (e.g. dietary advice to reduce fats, then provision of different oils or fats) is provided. Where participants are not blinded, it is difficult to ensure that study staff, healthcare providers and outcome assessors are blinded. The single RCT that appears to have had adequate participant and study personnel blinding was Veterans Admin 1969, and we judged blinding of participants to be inadequate in the remaining studies.

Blinding of outcome assessment was assessed separately for mortality and CVD outcomes. Blinding is not relevant in assessing all‐cause mortality, so all trials were considered at low risk of bias for detection bias for this outcome. For CVD outcomes, nine trials were at low risk of detection bias, one was at high risk and the remainder were unclear.

Incomplete outcome data

Assessing whether incomplete outcome data had been addressed was difficult, as the primary outcomes for this review (mortality and cardiovascular events) were often reported as dropouts and exclusions from the original studies, rather than as the primary outcomes of these trials. When mortality or cardiovascular events or both were noted in any one study, it is still feasible that some participants left that study feeling unwell or because the diet was inconvenient, so were simply lost to follow‐up from the perspective of the study, and later died or experienced a cardiovascular event. However, six of the studies checked medical records or death registers to ensure that such events were all collected (Black 1994, DART 1989; Oslo Diet‐Heart 1966; Sydney Diet‐Heart 1978; Veterans Admin 1969; WINS 2006). Within one study, there was extensive tracking of medical records, with assessment of health status by blinded trained adjudicators (WHI 2006), so few major events were likely to have been missed. In the other eight studies, it is not possible to know whether additional deaths or cardiovascular events occurred, that were not counted or ascertained within this review.

Selective reporting

Assessment of selective reporting is difficult when the outcome of interest was simply considered a cause of dropouts in most included studies. We tried to contact all of the trialists to ask about deaths and outcome events, but it is possible that some trialists did not reply as they felt that their data did not reflect the expected or hoped‐for pattern of events. All of the included studies have either reported that the participants did not experience any of our primary outcomes, have published their outcome data, or have provided the data they did possess. For this reason, we have graded all the included studies as at low risk of selective reporting.

Other potential sources of bias

Systematic differences in care. We assessed the studies for risk of bias in relation to systematic differences in care. The three RCTs (four comparisons) that appeared at low risk of systematic differences in care between the study arms included Rose corn oil 1965; Rose olive 1965; Oxford Retinopathy 1978; Veterans Admin 1969, while 11 RCTs clearly did have differences in care, such as differential time provided for those on the intervention to learn a new diet, and/or differential medical follow‐up, and one was unclear (Houtsmuller 1979).

Aim to reduce saturated fat. As several studies did not provide clear aims for their interventions (other than to alter specific dietary components, for example), we assessed whether the study stated an aim to reduce saturated fat. Ten RCTs (11 comparisons) clearly aimed to reduce saturated fat in their intervention arms, either directly or indirectly, for example, by stating food goals (DART 1989; Houtsmuller 1979; Moy 2001; MRC 1968; Oslo Diet‐Heart 1966; Rose corn oil 1965; Rose olive 1965; STARS 1992; Sydney Diet‐Heart 1978; Veterans Admin 1969; WHI 2006), while the remaining five did not (although they did achieve SFA reduction).

Successful saturated fat reduction. Eleven RCTs (11 comparisons) assessed SFA intake during the study period and showed that SFA intake in the intervention arm was statistically significantly lower than that in the control arm (Black 1994; DART 1989; Ley 2004; Moy 2001; Oxford Retinopathy 1978; Simon 1997; STARS 1992; Sydney Diet‐Heart 1978; Veterans Admin 1969; WHI 2006; WINS 2006). The remaining studies did not report SFA intake, so we rated them as unclear.

Successful cholesterol reduction. We would expect saturated fat reduction to be reflected in total or LDL cholesterol reductions, which may be more accurate assessments than self‐reported saturated fat intake. Nine RCTs (10 comparisons) provided information on serum total or LDL cholesterol levels in the intervention and control arms during the study, and found a reduction in the intervention arm compared to the control (P < 0.05, or where variances were not provided showed a reduction of at least 0.2 mmol/L in the mean intervention measure compared with control). The studies that successfully reduced serum total cholesterol in lower saturated fat arms compared with higher saturated fat arms were DART 1989; Houtsmuller 1979; Simon 1997; STARS 1992; Sydney Diet‐Heart 1978; WHI 2006, while Moy 2001 did not report total cholesterol (TC) but showed statistically significant reductions in LDL, and two studies (MRC 1968; Oslo Diet‐Heart 1966) did not report variances but did reduce mean TC in the intervention arm compared with control by at least 0.2 mm0l/L. One study (Black 1994) did not report lipid levels during the study, while five others did report lipid levels but did not suggest clear differences between lower and higher saturated fat arms (Ley 2004; Oxford Retinopathy 1978; Rose corn oil 1965; Rose olive 1965; Veterans Admin 1969; WINS 2006).

Dietary changes other than saturated fat. Some trials were partially confounded by aiming to make dietary changes other than those directly related to dietary fat intakes; for example, some studies encouraged intervention participants to make changes to their fat intake as well as changes to fruit and vegetable or fibre or salt intakes. In these studies, any effect on outcomes could be a result of other dietary changes, not of changes in saturated fat intake. The 11 studies (12 comparisons) that appeared free of such differences included Black 1994; DART 1989; Houtsmuller 1979; Ley 2004; MRC 1968; Oxford Retinopathy 1978; Rose corn oil 1965; Rose olive 1965; Simon 1997; Sydney Diet‐Heart 1978; Veterans Admin 1969; WINS 2006. This factor was not considered alongside others in the formal risk of bias assessment (Figure 2) so is described here. We did not identify any further methodological issues.

Summary risk of bias. We considered dietary advice or all‐food‐provided type trials to be at low summary risk of bias where we judged randomisation, allocation concealment, and blinding of outcome assessors to be adequate. For CVD outcomes, five trials were assessed as at low summary risk of bias: Ley 2004; Sydney Diet‐Heart 1978; Veterans Admin 1969; WHI 2006; WINS 2006. For all‐cause mortality (and lipid outcomes) where blinding of outcome assessors is not important, a further three trials were also at low summary risk of bias, eight in total: Ley 2004; Oslo Diet‐Heart 1966; Oxford Retinopathy 1978; STARS 1992; Sydney Diet‐Heart 1978; Veterans Admin 1969; WHI 2006; WINS 2006.

Effects of interventions

Primary outcomes

All‐cause mortality

GRADE assessment suggests that reducing saturated fat intake probably makes little or no difference to all‐cause mortality (moderate‐quality evidence, downgraded once for imprecision).

There was little or no effect of lower saturated fat compared to higher saturated fat intake on mortality (risk ratio (RR) 0.96, 95% confidence interval (CI) 0.90 to 1.03, I² = 2%, 55,858 participants, 3518 deaths, 11 RCTs, P_effect = 0.42, Analysis 1.1). This lack of effect was confirmed in sensitivity analyses including only trials at low summary risk of bias (Analysis 1.2), that aimed to reduce saturated fat (Analysis 1.3), that significantly reduced saturated fat intake (Analysis 1.4), that achieved a reduction in total or LDL cholesterol (Analysis 1.5), excluding the largest trial (WHI 2006, Analysis 1.6), or analysing using Mantel‐Haenszel or Peto fixed‐effect analysis (Analysis 1.7; Analysis 1.8).

Small study bias was assessed using a funnel plot and comparing the results of fixed‐ and random‐effects meta‐analysis. The funnel plot did not suggest any small study bias (Figure 3), and the results of fixed‐ and random‐effects meta‐analyses were very similar, suggesting that small study bias was not an issue.

Figure 3

Funnel plot of comparison: fat modification or reduction vs usual diet ‐ total mortality.

There was little or no effect, regardless of what nutrients were used to replace the saturated fat removed, including replacement with PUFA, MUFA, CHO and/or protein (Analysis 1.9). Effects did not differ by main substitution (Analysis 1.10), study duration (Analysis 1.11), baseline saturated fat intake (Analysis 1.12), degree of difference in saturated fat between arms (Analysis 1.13), participant sex (Analysis 1.14), by baseline CVD risk (Analysis 1.15), by degree of cholesterol reduction (Analysis 1.16) or by decade of publication (Analysis 1.17, Chi² test for differences between subgroups all P > 0.05).

Cardiovascular mortality

GRADE assessment suggests that reducing saturated fat intake probably makes little or no difference to cardiovascular mortality (moderate‐quality evidence, downgraded once for imprecision).

There was little or no effect of SFA reduction on cardiovascular mortality (RR 0.95, 95% CI 0.80 to 1.12, I² = 30%, 10 RCTs, 53,421 participants, 1096 cardiovascular deaths, Analysis 1.18). This lack of effect was confirmed in sensitivity analyses limiting to trials at low summary risk of bias (Analysis 1.19), explicitly aiming to reduce saturated fat (Analysis 1.20), achieving statistically significant saturated fat reduction (Analysis 1.21), achieving cholesterol reduction (Analysis 1.22), or running fixed‐effect analysis (Analysis 1.24; Analysis 1.25). However, excluding the largest single trial (WHI 2006) suggested that reducing saturated fat intake reduced the risk of CVD mortality (Analysis 1.23).

The funnel plot did not suggest small study bias (Figure 4), and the similarity in effect sizes between fixed‐ and random‐effects analysis suggests that small study bias is not important here.

Figure 4

Funnel plot of comparison: fat modification or reduction vs usual diet ‐ cardiovascular mortality

Subgrouping did not suggest important effects of reduced SFA on cardiovascular mortality, regardless of what was substituted for SFA (Analysis 1.26). When subgrouping by main substitution (Analysis 1.27), duration (Analysis 1.28), baseline SFA intake (Analysis 1.29), by difference in SFA (Analysis 1.30), participant sex (Analysis 1.31), baseline CVD risk (Analysis 1.32), or degree of cholesterol reduction (Analysis 1.33), there were no statistically significant differences between subgroups. There was a marginally significant difference between subgroups when ordered by decade of publication, but no clear pattern of effect, so we assumed the effect was probably spurious (Analysis 1.34). Additionally, effects did not appear to relate to statin use, as there was a reduction in risk of CVD mortality in studies published in the 1960s and a marginal increase in risk in the one trial published during the 1970s (although the 95% confidence interval did include 1.0), both well before statins were in common use (the 4S trial which first showed that use of statins reduced mortality was published in 1994, 4S 1994).

Cardiovascular events

GRADE assessment suggests that reducing SFA intake probably reduces cardiovascular events, to a greater extent with greater cholesterol reduction (moderate‐quality evidence, downgraded once for risk of bias and publication bias combined).

There was a 17% reduction in cardiovascular events in people who had reduced SFA compared with those on higher SFA (RR 0.83, 95% CI 0.70 to 0.98, I² = 67%, 12 RCTs, 53,758 participants, 4538 people with cardiovascular events, P_effect = 0.03, Analysis 1.35). This protective effect was confirmed in sensitivity analyses including only trials that aimed to reduce saturated fat (Analysis 1.37), that significantly reduced saturated fat intake (Analysis 1.38), that achieved a reduction in total or LDL cholesterol (Analysis 1.39), or excluding the largest trial (WHI 2006, Analysis 1.40). Analysing including only trials at low summary risk of bias, or using Mantel‐Haenszel or Peto fixed‐effect analysis suggested more marginal protection(Analysis 1.36; Analysis 1.41; Analysis 1.42).

A funnel plot did not suggest severe small‐study bias (Figure 5), but fixed‐effect analyses suggested slightly smaller effects (Analysis 1.41; Analysis 1.42), suggesting that smaller studies with more cardiovascular events in the intervention groups may be missing. Adding any such studies back would tend to moderate the protective effect of reducing SFA.

Figure 5

Funnel plot of comparison: fat modification or reduction vs usual diet ‐ combined cardiovascular events.

Sensitivity analysis omitting trials which included dietary interventions in addition to changes to dietary fat (for example, changes to fruit and vegetable or fibre intake) we excluded three trials (Oslo Diet‐Heart 1966; STARS 1992; WHI 2006). This analysis also suggested that reducing saturated fat (rather than other dietary changes) reduced risk of cardiovascular events: RR 0.86 (95% CI 0.67 to 1.09, Analysis 1.43).

When we subgrouped according to replacement for SFA, the PUFA replacement group suggested a 21% reduction in cardiovascular events, a 16% reduction in studies replacing SFA with carbohydrate, and little or no effect of other replacements, but without statistically significant effects between subgroups (Analysis 1.44). Similarly, there were no statistically significant differences between subgroups by main replacement (Analysis 1.45), by study duration (Analysis 1.46), in men or women (Analysis 1.49) or by baseline CVD risk (Analysis 1.50). When subgrouping, there was a suggestion of greater effects when baseline SFA was higher (Analysis 1.47), with greater reduction of SFA (Analysis 1.48), and with greater cholesterol reduction (Analysis 1.51). There were different effects by decade of publication, but no suggestion of a trend or a change following wider introduction of statins in the mid‐1990s (Analysis 1.52).

We further explored the effects of dietary fats on cardiovascular events, by using meta‐regression of the difference between the control and intervention of total fat intake, SFA intake, MUFA intake, PUFA intake, CHO intake (all by percentage of energy (%E)), serum total cholesterol (in mmol/L) achieved in trials, as well as baseline SFA intake, sex, study duration in months, and CVD risk of participants at baseline (Table 4). As we included only 13 studies for this outcome, we ran meta‐regressions exploring single explanatory factors at once, and as data were limited, with many studies not reporting dietary intake data, these analyses were limited in power to assess outcomes. The data suggested that greater reductions in total serum cholesterol levels reduced CVD events more. Greater baseline SFA intake and greater reduction in SFA were also associated with greater improvement in CVD events with SFA reduction, and increases in PUFA and MUFA intakes were slightly protective of CVD events, but none of these relationships were statistically significant. Overall, the relationship with serum total cholesterol was clearest (P = 0.04, accounting for 99% of between‐study variation). Sex, study duration and baseline cardiovascular risk did not appear to influence effect size. Apparent heterogeneity was accounted for by a dose‐effect; where SFA reduction resulted in greater serum cholesterol reduction, the reduction in CVD events was greater.

Table 4. Meta‐regression of effects of SFA reduction on cardiovascular events

Regression factor	No. of studies	Constant	Coefficient (95% CI)	P value	Proportion of between study variation explained
Change in SFA as %E	8	0.01	0.05 (‐0.03 to 0.13)	0.16	89%
Change in SFA as % of control	8	0.26	0.01 (‐0.01 to 0.03)	0.14	89%
Baseline SFA as %E	8	0.68	‐0.06 (‐0.15 to 0.04)	0.19	81%
Change in TC, mmol/L	12	0.03	0.69 (0.05 to 1.33)	0.04	99%
Change in PUFA as %E	5	‐0.01	‐0.02 (‐0.08 to 0.03)	0.25	100%
Change in MUFA as %E	5	‐0.26	‐0.03 (‐0.14 to 0.09)	0.50	‐87%
Change in CHO as %E	7	‐0.11	‐0.00 (‐0.05 to 0.05)	0.92	‐273%
Change in total fat intake as %E	9	‐0.17	‐0.01 (‐0.03 to 0.01)	0.28	100%
Gender*	13	‐0.17	‐0.14 (‐0.63 to 0.35)	0.55	‐13%
Study duration	13	‐0.47	0.00 (‐0.01 to 0.02)	0.76	‐24.8%
CVD risk at baseline**	13	‐0.44	0.03 (‐0.48 to 0.55)	0.89	‐39%

^{*Gender was coded as follows: 0 = women, 1 = mixed, 2 = men
**CVD risk at baseline was coded as follows: 1 = Low CVD risk, 2 = Moderate CVD risk, 3 = existing CVD}
^{CHO: carbohydrate
CI: confidence interval
CVD: cardiovascular disease
E: energy
MUFA: monounsaturated fatty acid
PUFA: polyunsaturated fatty fat
SFA: saturated fatty acid
TC: total cholesterol}

This 17% reduction in risk of CVD events translated into a number needed to treat for an additional beneficial outcome (NNTB) of 56 in primary prevention trials, so that 56 people need to reduce their saturated fat intake over around four years for one person to avoid experiencing a CVD event. In secondary prevention trials, the NNTB was 53.

Secondary outcomes ‐ health events

Myocardial Infarction (fatal and non‐fatal)

GRADE assessment suggested that the effect of reducing saturated fat intake on risk of myocardial infarction is unclear as the evidence was of very low‐quality (downgraded once each for risk of bias, imprecision and publication bias).

There was a small protective effect of SFA reduction on myocardial infarction (fatal and non‐fatal, RR 0.90, 95% CI 0.80 to 1.01, I² = 10%, 10 RCTs (11 comparisons) including 53,167 participants, 1714 people experiencing MI, Analysis 2.1). This protective effect was slightly modified in sensitivity analyses, and confirmed in analyses limited to trials that aimed to reduce saturated fat (Analysis 2.3), that achieved a reduction in total or LDL cholesterol (Analysis 2.5), and excluding the largest trial (WHI 2006, Analysis 2.6). Sensitivity analyses including only trials at low summary risk of bias (RR 0.93, 95% CI 0.81 to 1.08, Analysis 2.2), that significantly reduced saturated fat intake (Analysis 2.4), analysed using Mantel‐Haenszel or Peto fixed‐effect analysis (Analysis 2.7; Analysis 2.8) suggested little or no effect, though risk ratios were still all < 1.0.

The funnel plot was difficult to interpret, but did not raise major concerns about small‐study bias (not shown). While effects of random‐ and fixed‐effect meta‐analysis were only slightly different, they fell each side of the line suggesting an effect (Analysis 2.7; Analysis 2.8). There may be a small amount of small study bias.

The protective effect of replacing SFA with PUFA appeared to explain the reduction in MI (Analysis 2.9), but there were no statistically significant differences between subgroups by replacement (Analysis 2.10), duration (Analysis 2.11), baseline SFA intake (Analysis 2.12), change in SFA intake (Analysis 2.13), participant sex (Analysis 2.14), baseline CVD risk (Analysis 2.15), cholesterol reduction (Analysis 2.16) or decade of publication (Analysis 2.17).

Myocardial Infarction (non‐fatal only)

GRADE assessment suggests that reducing saturated fat may have little or no effect on risk of non‐fatal myocardial infarction (low‐quality evidence, downgraded once each for risk of bias and imprecision).

There was no clear effect of SFA reduction compared to usual diet on non‐fatal myocardial infarction (RR 0.97, 95% CI 0.87 to 1.07, I² = 0%, 7 RCTs, 52,834 participants, 1385 people with at least one non‐fatal MI, Analysis 2.18). This lack of effect was not altered in sensitivity analyses retaining only those that aimed to reduce SFA (Analysis 2.20), those showing a reduction in serum cholesterol (Analysis 2.22), or fixed‐effect analysis (Analysis 2.24; Analysis 2.25). However, sensitivity analyses retaining only trials at low summary risk of bias (RR 0.89, 95% CI 0.58 to 1.35, Analysis 2.19), those showing a significant reduction in SFA (Analysis 2.21), and omitting the largest trial (WHI 2006, Analysis 2.23) all suggested a reduction in non‐fatal MI with reduced SFA.

The funnel plot did not raise major concerns about small‐study bias (not shown), and effects of fixed‐ and random‐effects analyses were very similar, reinforcing the lack of small study bias.

Subgrouping by any replacement for SFA suggested reductions in non‐fatal MI when replaced by PUFA, but not other replacements (Analysis 2.26). Subgrouping by main substitution (Analysis 2.27), duration (Analysis 2.28), baseline SFA intake (Analysis 2.29), degree of SFA reduction (Analysis 2.30), sex (Analysis 2.31), baseline CVD risk (Analysis 2.32), degree of cholesterol reduction (Analysis 2.33) and decade of publication (Analysis 2.34) did not suggest significant differences between subgroups.

Stroke (any type, fatal or non‐fatal)

GRADE assessment suggests that the effect of reducing SFA intake on stroke is unclear as the evidence is of very low‐quality (downgraded twice for imprecision and once for risk of bias).

As data on stroke were sparse, it was not possible to tease out differential effects on ischaemic or haemorrhagic strokes, or whether a stroke was fatal. For this analysis, we combined all stroke data from any study. There was little or no effect of SFA reduction compared to usual diet on stroke of any type with any outcome (RR 0.92, 95% CI 0.68 to 1.25, I² = 9%, 7 RCTs, 50,952 participants, 1118 people with stroke, Analysis 2.35). This lack of effect was not altered in sensitivity analyses retaining only those that aimed to reduce SFA (Analysis 2.37), those showing a reduction in serum cholesterol (Analysis 2.39), or fixed‐effect analysis (Analysis 2.41; Analysis 2.42). However, for sensitivity analyses retaining only trials at low summary risk of bias (RR 0.76, 95% CI 0.42 to 1.38, Analysis 2.36), those showing a significant reduction in SFA (Analysis 2.38), and omitting the largest trial (WHI 2006, Analysis 2.40) the best estimate of effect always suggested a reduction in stroke with reduced SFA, though they were not statistically significant.

We did not create a funnel plot as the analysis only included data from seven RCTs, however RRs generated using fixed‐effect analyses were much closer to 1.0 than the random‐effects meta‐analysis (suggesting a small amount of publication bias), though both suggested little or no effect.

Subgrouping by any substitution for SFA suggested reduction in risk of stroke whether SFA was replaced by PUFA, CHO or protein (Analysis 2.43). Subgrouping did not suggest significant differences between subgroups by main substitution (Analysis 2.44), duration (Analysis 2.62), baseline SFA (Analysis 2.46), SFA change (Analysis 2.47), sex (Analysis 2.48), CVD risk (Analysis 2.49), cholesterol reduction (Analysis 2.50) or decade of publication (Analysis 2.51).

Coronary heart disease (CHD) mortality

GRADE assessment suggests that reducing saturated fat intake may have little or no effect on CHD mortality (low‐quality evidence, downgraded twice for imprecision).

Eight RCTs (9 comparisons) suggest little or no effect of reducing saturated fat on risk of CHD mortality (RR 0.97, 95% CI 0.82 to 1.16, I² = 28%, 53,159 participants, 927 people died of coronary heart disease, Analysis 2.52), and this was not altered in any sensitivity analyses (Analysis 2.53; Analysis 2.54; Analysis 2.55; Analysis 2.56; Analysis 2.57; Analysis 2.58; Analysis 2.59).

We did not create a funnel plot as the analysis only included data from seven RCTs, but the results of fixed‐ and random‐effects analyses were nearly identical, suggesting that small study bias is not an issue here.

There was no suggestion of an effect of reducing SFA on CHD mortality regardless of what replaced the SFA (Analysis 2.60). There were no statistically significant differences between subgrouping in any analysis (Analysis 2.61; Analysis 2.62; Analysis 2.63; Analysis 2.64; Analysis 2.65; Analysis 2.66; Analysis 2.67; Analysis 2.68).

Coronary heart disease events

GRADE assessment suggested that the effect of reducing saturated fat on CHD events is unclear as the evidence is of very low‐quality (downgraded once each for imprecision, risk of bias and inconsistency).

There was the suggestion of a 17% reduction in CHD events as a result of saturated fat reduction in the main analysis (RR 0.83, 95% CI 0.68 to 1.01, I² = 62%, 53,199 participants, 2261 people had at least one coronary heart disease event in 10 RCTs, Analysis 2.69). This did not differ in sensitivity analyses (Analysis 2.74; Analysis 2.72; Analysis 2.73; Analysis 2.71; Analysis 2.75; Analysis 2.76) except when limiting to trials at low summary risk of bias (RR 0.92, 95% CI 0.77 to 1.10, Analysis 2.70).

The funnel plot did not appear unbalanced, and the results of fixed‐ and random‐effects analyses were different, though both suggested that reducing SFA resulted in lower risk of CHD.

Subgrouping by any replacement for SFA suggested that replacement by PUFA may lead to reduced risk of CHD events (Analysis 2.77). There were no statistically significant differences between any other subgroups (Analysis 2.78; Analysis 2.79; Analysis 2.80; Analysis 2.81; Analysis 2.82; Analysis 2.83; Analysis 2.84) except by decade of publication, though this did not suggest any sequence or step change (Analysis 2.85).

Type 2 diabetes, new diagnoses

Only one RCT reported on diagnosis of diabetes (WHI 2006). There was little or no effect of reducing SFA intakes on diagnosis of diabetes in this study (RR 0.96, 95% CI 0.90 to 1.02, 48,835 participants, 3342 developed diabetes, Analysis 2.86). WHI 2006 was assessed at low summary risk of bias, aimed to reduce SFA, and demonstrated significant SFA and cholesterol reduction. With only one trial, we were not able to assess publication bias or carry out subgrouping.

Secondary outcomes ‐ blood levels

Serum blood lipids

Total cholesterol (TC): There was a reduction in TC in participants with reduced SFA compared to higher SFA (mean difference (MD) ‐0.24 mmol/L, 95% CI ‐0.36 to ‐0.13, I² = 60%, 13 RCTs, 7115 participants, Analysis 3.1). We did not conduct sensitivity analyses or most subgroupings on secondary outcomes, but there was no clear differential effect on TC depending on the replacement for SFA (PUFA, MUFA, CHO or a mixture, Analysis 3.2; Analysis 3.3). The funnel plot did not raise concerns about small‐study bias (not shown).

Low‐density lipoprotein (LDL): There was a reduction in LDL in participants with reduced SFA compared to higher SFA (MD ‐0.19 mmol/L, 95% CI ‐0.33 to ‐0.05, I² = 37%, 5 RCTs, 3291 participants, Analysis 3.4). There was no clear differential effect on LDL depending on the replacement for SFA (PUFA, MUFA, CHO or a mixture, Analysis 3.5; Analysis 3.6). We could not interpret the funnel plot due to sparsity of studies (not shown).

High‐density lipoprotein (HDL): There was little or no effect of reducing SFA intakes on HDL (MD ‐0.01 mmol/L, 95% CI ‐0.02 to 0.01, I² = 0%, 7 RCTs, 5147 participants, Analysis 3.7). There was no clear differential effect on HDL depending on the replacement for SFA (PUFA, MUFA, CHO or a mixture, Analysis 3.8; Analysis 3.9). We could not interpret the funnel plot due to sparsity of studies (not shown).

Triglycerides (TG): There was little or no effect of reducing SFA intakes on TG (MD ‐0.08 mmol/L, 95% CI ‐0.21 to 0.04, I² = 51%, 7 RCTs, 3845 participants, Analysis 3.10). There was no clear differential effect on TG depending on the replacement for SFA (PUFA, MUFA, CHO or a mixture, Analysis 3.11; Analysis 3.12). We could not interpret the funnel plot due to sparsity of studies (not shown).

TG/HDL ratio: We did not find any studies that reported TG/HDL ratio.

TC/HDL ratio: Only three RCTs reported on TC/HDL ratio. There was little or no effect of reducing SFA intakes on TC/HDL (MD ‐0.10, 95% CI ‐0.33 to 0.13, I² = 24%, 2985 participants, Analysis 3.13). There were no clear differential effects of replacement on TC/HDL (Analysis 3.14; Analysis 3.15). We could not interpret the funnel plot due to sparsity of studies (not shown).

LDL/HDL ratio: Only one RCT reported on LDL/HDL ratio. There was no clear effect of reducing SFA intakes on LDL/HDL in this study (MD ‐0.36, 95% CI ‐0.92 to 0.20, 50 participants, Analysis 3.16). This study replaced SFA with CHO (mainly) and PUFA.

Lipoprotein (a) (Lp(a)): Only two RCTs reported on lipoprotein (a), but these included 28,820 participants. There was little or no effect of reducing SFA intakes on Lp(a) (MD 0.00, 95% CI ‐0.00 to 0.00, I² = 0%, Analysis 3.17). There was no suggestion of differential effects of replacement on Lp(a) (Analysis 3.18; Analysis 3.19). We could not interpret the funnel plot due to sparsity of studies (not shown).

Homeostatic model assessment (HOMA): Only one RCT reported on the effects of reducing SFA on insulin resistance using HOMA. There was little or no effect of reducing SFA intakes compared to usual diet on HOMA in this study (MD ‐0.00, 95% CI ‐0.04 to 0.04, 2832 participants, Analysis 3.20).

Glucose at two hours post‐glucose tolerance test (GTT): Only three RCTs reported on glucose two hours post‐GTT. There was a reduction in glucose after reducing SFA intakes compared to usual diet (MD ‐1.69 mmol/L, 95% CI ‐2.55 to ‐0.82, I² = 45%, 249 participants, Analysis 3.20). We could not interpret the funnel plot due to sparsity of studies (not shown).

HbA1c (glycosylated haemoglobin): HbA1c was not measured in any included RCT.

Secondary outcomes ‐ other outcomes and potential harms

There was little or no effect of reducing SFA intakes on cancer diagnoses of any type (RR 0.94, 95% CI 0.83 to 1.07, I² = 33%, 4 RCTs, 52,294 participants, 5476 cancer diagnoses, Analysis 4.1); cancer deaths (RR 1.00, 95% CI 0.61 to 1.64, I² = 49%, 5 RCTs, 52,283 participants, 2472 cancer deaths, Analysis 4.2); systolic blood pressure (MD ‐0.19 mmHg, 95% CI ‐1.36 to 0.97, I² = 0%, 5 RCTs, 3812 participants, Analysis 4.5); diastolic blood pressure (MD ‐0.36 mmHg, 95% CI ‐1.03 to 0.32, I² = 0%, 5 RCTs, 3812 participants, Analysis 4.6).

There was evidence that reducing SFA intake resulted in small reductions in body weight (MD ‐1.97 kg, 95% CI ‐3.67 to ‐0.27, I² = 72%, 6 RCTs, 4541 participants, Analysis 4.3), and body mass index (MD ‐0.50, 95% CI ‐0.82 to ‐0.19, I² = 55%, 6 RCTs, 5553 participants, Analysis 4.4).

Only one RCT reported assessing quality of life. The Women's Health Initiative (WHI 2006) assessed quality of life at baseline using the SF‐36 tool). They found that being in the lower SFA arm resulted in a small improvement in Global Quality of Life at trial close‐out (on a scale of 0 worst to 10 best, MD 0.04, 95% CI 0.01 to 0.07, Analysis 4.7). This very small effect (less than 1% change) was statistically significant but unlikely to be relevant to individuals. However, it suggests no reduction in quality of life in those reducing their saturated fat.

Other results

To assess the effect in the population of consuming < 10%E as SFA relative to > 10%E as SFA for reduction in risk of noncommunicable diseases (NCDs), we combined studies with a control group saturated fat intake of > 10%E and an intervention group saturated fat intake of < 10%E for all‐cause mortality, cardiovascular and coronary heart disease mortality and events, myocardial infarctions, non‐fatal myocardial infarctions, and stroke. To assess the effect in the population of a reduction in %E from SFA from 10% in gradual increments relative to higher intake we repeated this with saturated fat cut‐offs between 7%E and 13%E. The data for these cut‐offs are shown in Table 5, and were plotted for a visual overview (Figure 6). The figure suggests reductions in cardiovascular outcomes in studies where saturated fat intake was greater than 10%E in control groups, and less than 10%E in intervention groups.

Table 5. SFA cut‐off data

Cut‐ off	RR of all‐cause mortality	RR of CVD mortality	RR of CVD events	RR of MI	RR of non‐fatal MI	RR of stroke	RR of CHD mortality	RR of CHD events
7%E	0.89 (0.66 to 1.20)	0.20 (0.01 to 4.15)	0.20 (0.01 to 4.15)	N/A	N/A	N/A	N/A	N/A
8%E	0.89 (0.66 to 1.20)	0.20 (0.01 to 4.15)	0.20 (0.01 to 4.15)	N/A	N/A	N/A	N/A	N/A
9%E	0.96 (0.83 to 1.10)	0.69 (0.51 to 0.94)	0.79 (0.62 to 0.99)	0.76 (0.55 to 1.05)	0.62 (0.31 to 1.21)	0.59 (0.30 to 1.15)	0.82 (0.55 to 1.21)	0.77 (0.56 to 1.04)
10%E	0.99 (0.90 to 1.09)	0.95 (0.67 to 1.35)	0.88 (0.66 to 1.18)	0.93 (0.80 to 1.08)	0.89 (0.58 to 1.35)	0.87 (0.58 to 1.33)	1.05 (0.77 to 1.43)	0.82 (0.60 to 1.13)
11%E	0.99 (0.88 to 1.12)	0.92 (0.65 to 1.31)	0.86 (0.66 to 1.13)	0.94 (0.84 to 1.06)	0.89 (0.58 to 1.35)	0.76 (0.45 to 1.30)	1.02 (0.84 to 1.24)	0.85 (0.63 to 1.15)
12%E	0.98 (0.91 to 1.07)	0.95 (0.75 to 1.21)	0.90 (0.74 to 1.08)	0.94 (0.85 to 1.04)	0.90 (0.72 to 1.14)	0.93 (0.55 to 1.25)	1.02 (0.84 to 1.24)	0.90 (0.77 to 1.06)
13%E	1.02 (0.83 to 1.25)	0.93 (0.63 to 1.38)	0.87 (0.65 to 1.17)	0.87 (0.73 to 1.04)	0.72 (0.50 to 1.03)	0.54 (0.29 to 1.00)	1.06 (0.76 to 1.48)	0.84 (0.63 to 1.12)

CHD: coronary heart disease
CVD: cardiovascular disease
E: energy
MI: myocardial infarction
N/A: not applicable (no relevant studies)
RR: risk ratio
SFA: saturated fat, as percentage of energy

Figure 6

Exploration of saturated fat cut‐offs

Additional WHO NUGAG specific questions:

In adults what is the effect in the population of reduced percentage of energy (%E) intake from saturated fatty acids (SFA) relative to higher intake for reduction in risk of noncommunicable diseases (NCDs)?

We found that reducing saturated fat for at least two years suggested no clear effects on all‐cause or cardiovascular mortality, but a 17% reduction in combined cardiovascular events. Heterogeneity in this result was partially explained by greater reductions in cardiovascular events in studies with greater serum total cholesterol reductions (implying greater reductions in SFA intake). Effects of reducing SFA on other cardiovascular and cancer outcomes were either very small or unclear (as the evidence was of very low quality), but it should be noted that risk ratios were all 1.0 or lower ‐ no harm was indicated. Effects on NCD risk factors were small but positive (serum total cholesterol, LDL cholesterol, systolic and diastolic blood pressure, weight and BMI) or neutral (HDL cholesterol and TGs).

What is the effect of reducing SFA on coronary heart disease mortality and coronary heart disease events?

We found little or no effect of reducing SFA on non‐fatal MI and CHD events, but the evidence on MI, stroke and CHD events was of very low quality. However, all risk ratios were less than 1.0.

What is the effect in the population of replacing SFA with PUFAs, MUFAs, CHO (refined versus unrefined), protein or trans fatty acids (TFAs) relative to no replacement for reduction in risk of NCDs?

We found greater reductions in cardiovascular events in studies that replaced saturated fats by PUFAs or CHO than in studies with replacement with MUFAs or protein, where there was little evidence of any effect.

What is the effect of replacing some saturated fat with PUFA on risk of CVD in adults?

There is moderate‐quality evidence that replacing saturated fat with PUFA probably reduces the risk of CVD events. Replacing SFA with PUFA also appears to reduce the risk of total MI, non‐fatal MI, stroke and CHD events, but has little or no effect on all‐cause mortality, CVD mortality and CHD mortality.

What is the effect of replacing some saturated fat with MUFA on risk of CVD in adults?

The evidence for effects of replacing SFA with MUFA was very limited, so assessment of health effects was not possible.

What is the effect of replacing some saturated fat with CHO on risk of CVD in adults?

While studies that replaced SFA with CHO reduced CVD events and stroke, effects on all‐cause mortality and other CVD outcomes suggested little or no effect.

What is the effect of replacing some saturated fat with protein on risk of CVD in adults?

There was no evidence suggesting that replacing SFA with protein reduced all‐cause mortality or any CVD outcomes, but the evidence was limited.

What is the effect in the population of consuming < 10%E as SFA relative to > 10%E as SFA for reduction in risk of NCDs?

Cut‐off data were difficult to interpret, and confidence intervals were wide, but they suggested greater reductions in cardiovascular events in studies where saturated fat intake was greater than 10%E in control groups, and less than 10%E in intervention groups (see Figure 6).

What is the effect in the population of a reduction in %E from SFA from 10% in gradual increments relative to higher intake for reduction in risk of NCDs?

The data from RCTs are too limited to be able to address this question.

Discussion

Summary of main results

This systematic review of long‐term randomised controlled trials of SFA reduction suggests that reducing saturated fat for at least two years probably has little or no effect on all‐cause or cardiovascular mortality, but probably caused a 17% (95% CI 2 to 30%, I² = 67%, moderate‐quality evidence) reduction in people experiencing cardiovascular events. The heterogeneity in this relationship was explained by greater reduction in CVD events in trials with greater serum cholesterol lowering. This effect on cardiovascular events was retained in most sensitivity analyses, but not when limiting to studies at low summary risk of bias. Subgrouping suggested that there was a 21% (95% CI 0 to 38%) reduction in cardiovascular events in studies that replaced saturated fats by PUFAs, and a 16% (95% CI ‐6 to 33%) reduction in studies replacing with CHO, with little information on the effect of replacing with MUFAs or protein. The difference between subgroups was not statistically significant. We could not explore data on trans fats due to lack of data. Meta‐regression and subgrouping suggested that greater reductions in SFA intake, greater reductions in total serum cholesterol levels, higher baseline SFA intake and greater increases in PUFA and MUFA intakes reduced CVD events more, but the strongest factor was the degree of cholesterol lowering. This suggests that the cardiovascular effects of reducing saturated fat rely on changes in atherosclerosis via serum cholesterol. The degree of cholesterol lowering reflects greater reduction in SFA and greater increase in PUFA (Hegsted 2000).

There may be little or no effect of SFA reduction on non‐fatal MI or CHD mortality (both low‐quality evidence), and effects on fatal and non‐fatal MI, stroke and CHD events were unclear (as the evidence was of very low‐quality). However, risk ratios were less than 1.00 for all of these. While we found small reductions in body weight and body mass index with advice to reduce saturated fats, there was little or no effect on diabetes diagnoses, cancer diagnoses or cancer deaths, or on systolic or diastolic blood pressure.

Reducing saturated fat caused reductions in serum total and LDL cholesterol, which did not differ according to type of replacement. There was little or no effect of saturated fat reduction on serum HDL cholesterol or triglyceride. Data on lipid ratios, Lp(a) and HOMA were very limited and effects unclear, but SFA reduction appears to reduce glucose two hours after a glucose load.

Overall completeness and applicability of evidence

The review included adult participants at varying levels of risk of cardiovascular disease, men and women, with mean ages from 46 to 66 years at baseline, in free‐living and institutional settings, and across the past 50 years. All the studies were conducted in industrialised countries, and no data were available from developing or transitional countries. The effectiveness of SFA reduction has been well assessed, with trials of at least 24 months including more than 50,000 participants for all primary and secondary CVD outcomes. Three thousand five hundred and eighteen participants in the included trials died, 1096 died of a cardiovascular cause, and 4538 experienced at least one cardiovascular event.

The external validity of the review in industrialised countries, men and women, people with low, moderate and high risk of cardiovascular disease was high, but it is not clear how this evidence relates to diets in developing and transitional countries.

Quality of the evidence

All 15 trials (16 comparisons) included were randomised controlled trials, allocation concealment was judged well done in eight RCTs and blinding of outcome assessors adequate in nine trials assessing CVD outcomes (and all trials assessing all‐cause mortality). Blinding of participants is difficult and expensive in dietary fat trials, but was adequate in one trial. We judged incomplete outcome data not to be a problem in seven RCTs, and selective reporting was not a problem in any trial. Three trials were free of differences in care between the intervention and control arms, 10 RCTs stated an aim to reduce saturated fat, 11 showed evidence they had reduced SFA intake (all studies did one or the other), and nine studies showed clear reductions in total cholesterol. Five trials were at low summary risk of bias.

The lack of blinding of participants in most dietary trials is unlikely to alter outcome assessment when outcomes include death and cardiovascular events (although it could potentially affect assessment of worsening of angina, or increased dose of antianginals), but lack of blinding in participants may have led those in the control groups to alter their lifestyle and dietary practices (for example, feeling that they have not been helped to reduce their cardiovascular risk, they may act to reduce their own risk by altering other lifestyle behaviours such as smoking or exercise, leading to a potential lessening of the apparent effect of the dietary intervention). Systematic differences in care between arms may have led to intervention groups receiving additional support in areas like self efficacy and gaining support from new social circles, potentially beneficial to health regardless of dietary fat intake, or gaining additional healthcare professional time, possibly leading to earlier diagnosis and treatment of other risk factors such as raised blood pressure. Additional dietary messages such as those around fruit and vegetable intake, fibre, alcohol and sugars, present in many studies, may have been protective, or may have diluted the effect or attainability or both of the saturated fat goals.

The quality of evidence balances the uncertainty over allocation concealment, lack of blinding and presence of systematic differences in care and additional dietary differences between arms (Figure 2) with the scale and consistency of the evidence across studies and across decades, despite very different designs and design flaws. For this reason, there is moderate‐quality evidence that interventions that reduced dietary saturated fat intake reduced the risk of cardiovascular events.

Complex interventions

With complex interventions, such as dietary interventions, there are additional questions that need to be asked about included studies. Important issues to consider include defining the intervention, searching for and identifying all relevant studies, selecting studies for inclusion and data synthesis (Lenz 2007; Sheppherd 2009), as well as questions around whether the intended intervention was realised in study participants during the study.

For this review, we have worked to define the interventions clearly (see Characteristics of included studies), providing information on the type of intervention, stating the study aims and methods for each arm and the assessed total and saturated fat intakes attained within the study. However, while we have characterised the interventions, no two studies that reduced SFA had exactly the same dietary goals for the intervention groups. Methods of attaining the dietary goals varied from providing a whole diet over several years (in studies based in institutions) to providing advice on diet alongside supplementary foods such as margarines or oils, to providing dietary advice with or without supplementary support in the way of group sessions, cooking classes, shopping tours, feedback, self‐efficacy sessions and/or individual counselling. We aimed to use this variety to support generalisability for the effects of the interventions.

We aimed to identify all the relevant studies through use of a broad search strategy, which was time‐consuming. However, we believe that we have included most relevant trials. We also carefully defined acceptable interventions for each arm, to simplify decisions on inclusion, and the two independent assessors often agreed. We augmented data synthesis by subgrouping and meta‐regression, to help us understand the effects of individual elements of dietary fat changes.

A study that sets out to assess the effect of a 30% reduction in saturated fat intake may attain this level of reduction in some participants, exceed it in some and not achieve it at all in others. The actual mean change attained in the intervention group may be less dramatic than that aimed for, and the participants in the control group may also reduce their saturated fat intake by a small amount, narrowing the difference in saturated fat between the groups further and so reducing the scale of any outcome. This can be dealt with in the systematic review if we meta‐regress the difference in saturated fat intake between the intervention and control group with the scale of the outcome (assuming a linear dose response), still allowing us to understand the effect of altering saturated fat intake. However, it is difficult to measure actual saturated fat intake achieved. Some trials did not report it, either because they did not assess it, or did assess it but didn't report this relatively uninteresting outcome. Other trials did report the results of asking people what they were eating, using a food frequency questionnaire or several 24‐hour food recalls. However, there is good evidence to believe that asking people how they are eating may produce somewhat biased information (Kristal 2005; Schatzkin 2003), and this may be a greater problem where the participant has been recently urged to eat in a particular way, as in a dietary trial. Assessment of change in total cholesterol is a way to get over self‐reporting of dietary intake as reducing saturated fat reduces total and LDL cholesterol. This review suggests that the relationship between saturated fat reduction and CVD events is moderated by the degree of cholesterol lowering, which is exactly what would be expected of a true effect.

The interventions used in the studies included in this review were varied, with some participants given all their food over a long period of time in an institutional setting, while most participants were given advice on how to achieve dietary changes, with or without the support of supplements such as oils and foods (Table 1). Advice was provided by a variety of health professionals, and with different levels of intensity. The effect of this was that different degrees of saturated fat reduction were achieved in different studies. The level of compliance with interventions involving long‐term behaviour change, such as those used in these studies, can vary widely. This is likely to attenuate the pooled effect and bias it towards the null. Insofar as we were able to understand this issue, subgrouping and meta‐regression suggested that greater reductions in saturated fats were associated with greater reductions in the risk of cardiovascular disease events. This suggestion of a dose response strengthens our belief that there is a true effect of reducing saturated fat on CVD events.

Potential biases in the review process

In compiling the included studies, we worked hard to locate randomised studies that altered dietary SFA intake for at least 24 months, even when cardiovascular events were not reported in study publications, or where such events were reported incidentally as reasons for participant dropouts. We attempted to contact all authors of potentially includable studies to verify the presence or absence of our outcomes. In many studies, no outcomes relevant to this review occurred or were recorded, and the numbers of events occurring within single studies varied from none to over 2500 deaths, over 500 cardiovascular deaths, and over 3000 participants experiencing at least one cardiovascular event (all within WHI 2006, the largest single study with almost 50,000 female participants for many years).

The number of cardiovascular deaths across the review was relatively small (1096), so while we can be quite confident in reporting a reduction in participants experiencing cardiovascular events (4476 events) with SFA reduction, and a lack of effect on total mortality (3518 deaths) within the studies' time scales, the effect on cardiovascular mortality is less clear. The risk ratio of 0.94 (95% CI 0.78 to 1.13, Analysis 1.18) may translate into a small protective effect, but this is unclear.

The lack of effect on individual cardiovascular events is harder to explain; there were 1714 people experiencing MIs, 1118 strokes and 1385 non‐fatal MIs, 2472 cancer deaths, 3342 diabetes diagnoses and 5476 cancer diagnoses. Lack of clear effects on any of these outcomes is surprising, given the effects on total cardiovascular events, but may be due to the relatively short timescale of the included studies, compared to a usual lifespan during which risks of chronic illnesses develop over decades, and to relatively small reductions in saturated fat (and serum cholesterol) in some trials. Some of the events included within combined cardiovascular events, such as new or worsening angina or increased anti‐anginal treatment, could potentially be influenced by allocated study arm, and so might increase bias within unblinded trials (although they also add power to see potential effects). There are difficulties in finding data on the number of people experiencing composite end points such as cardiovascular events. This end point represents the number of people experiencing any of the following: cardiovascular death, cardiovascular morbidity (non‐fatal myocardial infarction, angina, stroke, heart failure, peripheral vascular events, atrial fibrillation) and unplanned cardiovascular interventions (coronary artery bypass surgery or angioplasty). Adding up the number of events is easy, but a single participant may have experienced a stroke, an MI and atrial fibrillation during a trial ‐ and we need to take care not to count this individual three times. So finding such composite end point data involves using the best published composite end point data and supplementing this with author contact where possible. We have underestimated such composite end points rather than overestimated them where exact data are not available. Added to this complex picture, it needs to be remembered that definitions and diagnoses of some end points have altered over time.

Where the funnel plots and comparison of fixed‐ and random‐effects meta‐analyses suggest small‐study bias, we have downgraded the quality of the evidence in GRADE, but effects of any such small study bias appear small.

Some trials were partially confounded by aiming to make dietary changes other than those directly related to dietary fat intakes; for example, some studies encouraged intervention participants to make changes to their fat intake as well as changes to fruit and vegetable, fibre or salt intakes. In these studies, any effect on outcomes could be a result of other dietary changes, not of changes in saturated fat intake. The 11 studies (12 comparisons) that appeared free of such differences included Black 1994; DART 1989; Houtsmuller 1979; Ley 2004; MRC 1968; Oxford Retinopathy 1978; Rose corn oil 1965; Rose olive 1965; Simon 1997; Sydney Diet‐Heart 1978; Veterans Admin 1969; WINS 2006. On the basis of reviewer comments, we assessed effects of reducing saturated fat intake on combined CVD events including only these trials free of additional interventions. Omitting trials with additional interventions (Oslo Diet‐Heart 1966; STARS 1992; WHI 2006) leaves nine studies (ten arms) randomising 4456 participants of whom 812 experienced a CVD event, suggesting a similar reduction in CVD events (RR 0.86, 95% CI 0.67 to 1.09, I² = 59%, Analysis 1.43) to the main analysis (RR 0.83, 95% CI 0.70 to 0.98, I² = 67%, > 53,000 participants randomised, Analysis 1.35). This suggests that effects on combined CVD events are not driven by interventions other than reductions in saturated fats and any energy replacements.

One surprising element of this review is the lack of new trials identified in the 2019 update, and small numbers of potential ongoing trials. This is likely to be because well‐powered trials on cardiovascular end points will need to be large and carried out over several years, so expensive. As the effects of saturated fats are felt to be established and understood, trialists and funders may feel that the money would be better invested in answering other questions. For most of the ongoing trials, information is limited and these trials may or may not be included when fully published. Perhaps the current evidence set is as definitive as we will achieve during the 'statin era'.

Agreements and disagreements with other studies or reviews

In this review, saturated fat reduction had little or no effect on all‐cause or cardiovascular mortality but did appear to reduce the risk of cardiovascular events by 17%, although effects on MI and stroke individually were less clear. This result was rather different from those of Siri‐Tarino 2010, who systematically reviewed cohort studies that assessed relationships between saturated fat and cardiovascular events. They included 21 studies and did not find associations between saturated fat intake and cardiovascular disease (RR 1.0, 95% CI 0.89 to 1.11). This meta‐analysis has been criticised (Katan 2010; Scarborough 2010; Stamler 2010), as results of half of the studies included in their meta‐analysis were adjusted for serum cholesterol concentrations, while there is an established relation between saturated fat intake and cholesterol level. The issue of what factors should be adjusted for, and what not adjusted for, in observational studies when dietary factors are very tightly correlated is a thorny one, and one of the reasons why trial data may be helpful. The studies included in the meta‐analysis also varied widely in the method used to assess intake, as half of the studies collected one‐day intake data. However, as with our review, they found little or no relationship between saturated fat intake and coronary heart disease (RR 1.07, 95% CI 0.96 to 1.19) though their data did suggest a (non‐statistically significant) reduction in stroke risk with higher saturated fat intake (RR 0.81, 95% CI 0.62 to 1.05, Siri‐Tarino 2010).

In this review, we found that replacing saturated fat with PUFAs (a modified‐fat diet) appeared more protective of cardiovascular events than replacement with carbohydrates (a low‐fat diet, Analysis 1.44; Analysis 1.45). This was similar to results within our closely allied systematic review assessing health effects of total fat reduction, where modified‐fat diets were protective and low‐fat diets were not (Hooper 2012). Meta‐regression did not suggest any relationship between either PUFAs or MUFAs and cardiovascular events in this review, although the analysis was underpowered. Alonso 2006 suggested a protective role for MUFA from olive oil, but not from meat sources (the main source of MUFA in the USA and Northern Europe). Our systematic review was not able to explore this issue as we included only one small study (underpowered to assess health outcomes on its own) that replaced SFA with MUFA, using an olive oil supplement (Rose olive 1965). A review by Mozaffarian 2010, which again included very similar studies to the last version of this review, with the Finnish Mental Hospital study and Women's Health Initiative data added, stated that their findings provided evidence that consuming PUFAs in place of saturated fat would reduce coronary heart disease. However, their evidence for this was limited, as they found that modifying fat reduced the risk of myocardial infarction or coronary heart disease death (combined) by 19% (similar to our result). As the mean increase in PUFAs in these studies was 9.9% of energy, they infer an effect of increasing PUFAs by 5% of energy of 10% reduction in risk of myocardial infarction or coronary heart disease death. They provided no suggestion or evidence of a relationship between degree of PUFAs increase and level of risk reduction. Another review carried out during updating of the Nordic Nutritional Recommendations (Schwab 2014) included observational as well as intervention studies, and concluded that there was convincing evidence that partial replacement of SFA with PUFA decreases risk of CVD while replacement with CHO is associated with increased CVD risk. The review included studies performed solely in white participants or with a clear white majority.

Within the meta‐regression, we hoped to combine studies that effectively altered saturated fat by different degrees, so that studies that reduced saturated fat very little and studies that reduced it a great deal would all offer data points for the meta‐regression against mortality and morbidity end points, and similarly for total fat, polyunsaturated, monounsaturated and trans fats. Unfortunately many of the included studies did not report data on assessed dietary intake during the trial, reducing the quantity of data available to understand the relationships. Another limitation in understanding effects of individual classes of fatty acids on mortality and morbidity (both in trials and in observational studies) was our ability to correctly assess participants' intake. We could overcome this by using biomarkers such as serum LDL cholesterol (differences between the LDL concentration in the intervention and control arms could be seen as a reasonable and independent approximation of saturated fat intake); however, as many studies were carried out in the 1960s to 1990s, few measured and reported LDL cholesterol. We used meta‐regression with serum total cholesterol (although this is a composite marker and so less related to saturated fat intake), but although this was available for more studies than LDL it was not available for all studies. Despite the limited data, there was a clear suggestion from meta‐regression that there was greater reduction of risk of cardiovascular events in studies with greater total serum cholesterol reduction, supporting the central role of serum lipids in the link between dietary saturated fats and cardiovascular events.

Participants' level of risk

As the rate of events is higher in high‐risk groups (by definition), it should require smaller sample sizes and shorter follow‐up to observe an effect of an intervention in a high‐risk group of participants (Davey Smith 1993). There have been suggestions that randomised 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). However, with the publication of the Women's Health Initiative trial (WHI 2006) we now have data on more people experiencing cardiovascular events who were originally at low risk of cardiovascular disease than in people with moderate or high risk. The same is true for cardiovascular deaths and total mortality.

When end points such as total mortality are used, the situation becomes more difficult, as in low‐risk groups the proportion of deaths which are unrelated to cardiovascular disease (and perhaps 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 changes in cardiovascular deaths will be seen than in total mortality. The trend is certainly in this direction, with the pooled risk ratio for total mortality 0.96 (95% CI 0.90 to 1.03, Analysis 1.1), and for cardiovascular mortality RR 0.94 (95% CI 0.78 to 1.13, Analysis 1.18). Our best estimate is that SFA reduction results in a reduction of 6% in deaths due to cardiovascular disease, and a reduction of 4% in total deaths, but these are small effects with wide confidence intervals.

The high‐risk participants all showed evidence of cardiovascular disease at baseline. Under current guidelines, most high‐risk participants with raised lipid levels should be on lipid‐lowering medication (Grundy 2019; NICE 2014; O'Gara 2014). This raises the question of whether there is any additional advantage of adherence to a reduced SFA 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 lipid‐lowering medication remains rationed even for those at high risk, the use of reduced SFA diets would appear to be a cost‐effective option leading to considerable reductions in cardiovascular events for populations (and so in health budgets) in only a few years.

Low‐risk participants are unlikely to be on lipid‐lowering medication under current guidelines. The suggestion of protection of low‐risk individuals from cardiovascular events with a reduction of roughly 17% of events in just a few years of intervention, as there is no evidence that effects in the low‐CVD‐risk group are different from effects in the higher‐risk groups, would appear to merit continued public health action. Recent guidelines recommend saturated fat reduction in general populations (SACN 2019).

A factor that may affect participant risk of cardiovascular disease, and also the effectiveness of reducing saturated fat intake, that has altered over time is the level of use of statins to control serum lipids in people at moderate and high risk of CVD. The 4S 1994 trial, which was the first trial to show that use of statins could reduce mortality in people with coronary heart disease, was published in 1994 and led to an explosion of the use of statins. For most health outcomes, we saw no clear effect of a decade of publication on risk, but for combined CVD events and CHD events, there were differences between subgroups. For combined CVD events, there were reductions in risk with reduced saturated fat intakes in the 1960s, 1970s and 1990s (both trials published early in the decade), but no clear effect of reducing saturated fat in the 1980s (one trial with 283 events) or 2000s (three large trials). It is possible (but not clear) that participants in the trials published in the 2000s were protected by higher levels of statin use (statins were allowed in participants in the largest trial, WHI 2006).

Figure 1

Study flow diagram for this systematic review (update searches run October 2019).

Figure 2

Figure 3

Funnel plot of comparison: fat modification or reduction vs usual diet ‐ total mortality.

Figure 4

Funnel plot of comparison: fat modification or reduction vs usual diet ‐ cardiovascular mortality

Figure 5

Funnel plot of comparison: fat modification or reduction vs usual diet ‐ combined cardiovascular events.

Figure 6

Exploration of saturated fat cut‐offs

Analysis 1.1

Comparison 1: SFA reduction vs usual diet ‐ primary outcomes, Outcome 1: ALL‐CAUSE MORTALITY

Analysis 1.2

Comparison 1: SFA reduction vs usual diet ‐ primary outcomes, Outcome 2: All‐cause mortality, SA low summary risk of bias

Analysis 1.3

Comparison 1: SFA reduction vs usual diet ‐ primary outcomes, Outcome 3: All‐cause mortality, SA aim to reduce SFA

Analysis 1.4

Comparison 1: SFA reduction vs usual diet ‐ primary outcomes, Outcome 4: All‐cause mortality, SA statistically significant SFA reduction

Analysis 1.5

Comparison 1: SFA reduction vs usual diet ‐ primary outcomes, Outcome 5: All‐cause mortality, SA TC reduction

Analysis 1.6

Comparison 1: SFA reduction vs usual diet ‐ primary outcomes, Outcome 6: All‐cause mortality, SA excluding WHI

Analysis 1.7

Comparison 1: SFA reduction vs usual diet ‐ primary outcomes, Outcome 7: All‐cause mortality, SA Mantel‐Haenszel fixed‐effect

Analysis 1.8

Comparison 1: SFA reduction vs usual diet ‐ primary outcomes, Outcome 8: All‐cause mortality, SA Peto fixed‐effect

Analysis 1.9

Comparison 1: SFA reduction vs usual diet ‐ primary outcomes, Outcome 9: All‐cause mortality, subgroup by any substitution

Analysis 1.10

Comparison 1: SFA reduction vs usual diet ‐ primary outcomes, Outcome 10: All‐cause mortality, subgroup by main substitution

Analysis 1.11

Comparison 1: SFA reduction vs usual diet ‐ primary outcomes, Outcome 11: All‐cause mortality, subgroup by duration

Analysis 1.12

Comparison 1: SFA reduction vs usual diet ‐ primary outcomes, Outcome 12: All‐cause mortality, subgroup by baseline SFA

Analysis 1.13

Comparison 1: SFA reduction vs usual diet ‐ primary outcomes, Outcome 13: All‐cause mortality, subgroup by SFA change

Analysis 1.14

Comparison 1: SFA reduction vs usual diet ‐ primary outcomes, Outcome 14: All‐cause mortality, subgroup by sex

Analysis 1.15

Comparison 1: SFA reduction vs usual diet ‐ primary outcomes, Outcome 15: All‐cause mortality, subgroup by CVD risk

Analysis 1.16

Comparison 1: SFA reduction vs usual diet ‐ primary outcomes, Outcome 16: All‐cause mortality, subgroup by TC reduction

Analysis 1.17

Comparison 1: SFA reduction vs usual diet ‐ primary outcomes, Outcome 17: All‐cause mortality, subgroup decade of publication

Analysis 1.18

Comparison 1: SFA reduction vs usual diet ‐ primary outcomes, Outcome 18: CARDIOVASCULAR MORTALITY

Analysis 1.19

Comparison 1: SFA reduction vs usual diet ‐ primary outcomes, Outcome 19: CVD mortality, SA low summary risk of bias

Analysis 1.20

Comparison 1: SFA reduction vs usual diet ‐ primary outcomes, Outcome 20: CVD mortality, SA aim to reduce SFA

Analysis 1.21

Comparison 1: SFA reduction vs usual diet ‐ primary outcomes, Outcome 21: CVD mortality, SA statistically significant SFA reduction

Analysis 1.22

Comparison 1: SFA reduction vs usual diet ‐ primary outcomes, Outcome 22: CVD mortality, SA TC reduction

Analysis 1.23

Comparison 1: SFA reduction vs usual diet ‐ primary outcomes, Outcome 23: CVD mortality, SA excluding WHI

Analysis 1.24

Comparison 1: SFA reduction vs usual diet ‐ primary outcomes, Outcome 24: CVD mortality, SA Mantel‐Haenszel fixed‐effect

Analysis 1.25

Comparison 1: SFA reduction vs usual diet ‐ primary outcomes, Outcome 25: CVD mortality, SA Peto fixed‐effect

Analysis 1.26

Comparison 1: SFA reduction vs usual diet ‐ primary outcomes, Outcome 26: CVD mortality, subgroup by any substitution

Analysis 1.27

Comparison 1: SFA reduction vs usual diet ‐ primary outcomes, Outcome 27: CVD mortality, subgroup by main substitution

Analysis 1.28

Comparison 1: SFA reduction vs usual diet ‐ primary outcomes, Outcome 28: CVD mortality, subgroup by duration

Analysis 1.29

Comparison 1: SFA reduction vs usual diet ‐ primary outcomes, Outcome 29: CVD mortality, subgroup by baseline SFA

Analysis 1.30

Comparison 1: SFA reduction vs usual diet ‐ primary outcomes, Outcome 30: CVD mortality, subgroup by SFA change

Analysis 1.31

Comparison 1: SFA reduction vs usual diet ‐ primary outcomes, Outcome 31: CVD mortality, subgroup by sex

Analysis 1.32

Comparison 1: SFA reduction vs usual diet ‐ primary outcomes, Outcome 32: CVD mortality, subgroup by CVD risk

Analysis 1.33

Comparison 1: SFA reduction vs usual diet ‐ primary outcomes, Outcome 33: CVD mortality, subgroup by TC reduction

Analysis 1.34

Comparison 1: SFA reduction vs usual diet ‐ primary outcomes, Outcome 34: CVD mortality, subgroup decade of publication

Analysis 1.35

Comparison 1: SFA reduction vs usual diet ‐ primary outcomes, Outcome 35: COMBINED CARDIOVASCULAR EVENTS

Analysis 1.36

Comparison 1: SFA reduction vs usual diet ‐ primary outcomes, Outcome 36: CVD events, SA low summary risk of bias

Analysis 1.37

Comparison 1: SFA reduction vs usual diet ‐ primary outcomes, Outcome 37: CVD events, SA aim to reduce SFA

Analysis 1.38

Comparison 1: SFA reduction vs usual diet ‐ primary outcomes, Outcome 38: CVD events, SA statistically significant SFA reduction

Analysis 1.39

Comparison 1: SFA reduction vs usual diet ‐ primary outcomes, Outcome 39: CVD events, SA TC reduction

Analysis 1.40

Comparison 1: SFA reduction vs usual diet ‐ primary outcomes, Outcome 40: CVD events, SA excluding WHI

Analysis 1.41

Comparison 1: SFA reduction vs usual diet ‐ primary outcomes, Outcome 41: CVD events, SA Mantel‐Haenszel fixed‐effect

Analysis 1.42

Comparison 1: SFA reduction vs usual diet ‐ primary outcomes, Outcome 42: CVD events, SA Peto fixed‐effect

Analysis 1.43

Comparison 1: SFA reduction vs usual diet ‐ primary outcomes, Outcome 43: CVD events, SA excluding trials with additional interventions

Analysis 1.44

Comparison 1: SFA reduction vs usual diet ‐ primary outcomes, Outcome 44: CVD events, subgroup by any substitution

Analysis 1.45

Comparison 1: SFA reduction vs usual diet ‐ primary outcomes, Outcome 45: CVD events, subgroup by main substitution

Analysis 1.46

Comparison 1: SFA reduction vs usual diet ‐ primary outcomes, Outcome 46: CVD events, subgroup by duration

Analysis 1.47

Comparison 1: SFA reduction vs usual diet ‐ primary outcomes, Outcome 47: CVD events, subgroup by baseline SFA

Analysis 1.48

Comparison 1: SFA reduction vs usual diet ‐ primary outcomes, Outcome 48: CVD events, subgroup by SFA change

Analysis 1.49

Comparison 1: SFA reduction vs usual diet ‐ primary outcomes, Outcome 49: CVD events, subgroup by sex

Analysis 1.50

Comparison 1: SFA reduction vs usual diet ‐ primary outcomes, Outcome 50: CVD events, subgroup by CVD risk

Analysis 1.51

Comparison 1: SFA reduction vs usual diet ‐ primary outcomes, Outcome 51: CVD events, subgroup by TC reduction

Analysis 1.52

Comparison 1: SFA reduction vs usual diet ‐ primary outcomes, Outcome 52: CVD events, subgroup decade of publication

Analysis 2.1

Comparison 2: SFA reduction vs usual diet ‐ secondary health events, Outcome 1: MYOCARDIAL INFARCTION

Analysis 2.2

Comparison 2: SFA reduction vs usual diet ‐ secondary health events, Outcome 2: MI, SA by low summary risk of bias

Analysis 2.3

Comparison 2: SFA reduction vs usual diet ‐ secondary health events, Outcome 3: MI, SA aim to reduce SFA

Analysis 2.4

Comparison 2: SFA reduction vs usual diet ‐ secondary health events, Outcome 4: MI, SA statistically significant SFA reduction

Analysis 2.5

Comparison 2: SFA reduction vs usual diet ‐ secondary health events, Outcome 5: MI, SA by TC reduction

Analysis 2.6

Comparison 2: SFA reduction vs usual diet ‐ secondary health events, Outcome 6: MI, SA excluding WHI

Analysis 2.7

Comparison 2: SFA reduction vs usual diet ‐ secondary health events, Outcome 7: MI, SA Mantel‐Haenszel fixed‐effect

Analysis 2.8

Comparison 2: SFA reduction vs usual diet ‐ secondary health events, Outcome 8: MI, SA Peto fixed‐effect

Analysis 2.9

Comparison 2: SFA reduction vs usual diet ‐ secondary health events, Outcome 9: MI, subgroup by any substitution

Analysis 2.10

Comparison 2: SFA reduction vs usual diet ‐ secondary health events, Outcome 10: MI, subgroup by main substitution

Analysis 2.11

Comparison 2: SFA reduction vs usual diet ‐ secondary health events, Outcome 11: MI, subgroup by duration

Analysis 2.12

Comparison 2: SFA reduction vs usual diet ‐ secondary health events, Outcome 12: MI, subgroup by baseline SFA

Analysis 2.13

Comparison 2: SFA reduction vs usual diet ‐ secondary health events, Outcome 13: MI, subgroup by SFA change

Analysis 2.14

Comparison 2: SFA reduction vs usual diet ‐ secondary health events, Outcome 14: MI, subgroup by sex

Analysis 2.15

Comparison 2: SFA reduction vs usual diet ‐ secondary health events, Outcome 15: MI, subgroup by CVD risk

Analysis 2.16

Comparison 2: SFA reduction vs usual diet ‐ secondary health events, Outcome 16: MI, subgroup by TC reduction

Analysis 2.17

Comparison 2: SFA reduction vs usual diet ‐ secondary health events, Outcome 17: MI, subgroup decade of publication

Analysis 2.18

Comparison 2: SFA reduction vs usual diet ‐ secondary health events, Outcome 18: NON‐FATAL MYOCARDIAL INFARCTION

Analysis 2.19

Comparison 2: SFA reduction vs usual diet ‐ secondary health events, Outcome 19: Non‐fatal MI, SA by low summary risk of bias

Analysis 2.20

Comparison 2: SFA reduction vs usual diet ‐ secondary health events, Outcome 20: Non‐fatal MI, SA aim to reduce SFA

Analysis 2.21

Comparison 2: SFA reduction vs usual diet ‐ secondary health events, Outcome 21: Non‐fatal MI, SA statistically significant SFA reduction

Analysis 2.22

Comparison 2: SFA reduction vs usual diet ‐ secondary health events, Outcome 22: Non‐fatal MI, SA by TC reduction

Analysis 2.23

Comparison 2: SFA reduction vs usual diet ‐ secondary health events, Outcome 23: Non‐fatal MI, SA excluding WHI

Analysis 2.24

Comparison 2: SFA reduction vs usual diet ‐ secondary health events, Outcome 24: Non‐fatal MI, SA Mantel‐Haenszel fixed‐effect

Analysis 2.25

Comparison 2: SFA reduction vs usual diet ‐ secondary health events, Outcome 25: Non‐fatal MI, SA Peto fixed‐effect

Analysis 2.26

Comparison 2: SFA reduction vs usual diet ‐ secondary health events, Outcome 26: Non‐fatal MI, subgroup by any substitution

Analysis 2.27

Comparison 2: SFA reduction vs usual diet ‐ secondary health events, Outcome 27: Non‐fatal MI, subgroup by main substitution

Analysis 2.28

Comparison 2: SFA reduction vs usual diet ‐ secondary health events, Outcome 28: Non‐fatal MI, subgroup by duration

Analysis 2.29

Comparison 2: SFA reduction vs usual diet ‐ secondary health events, Outcome 29: Non‐fatal MI, subgroup by baseline SFA

Analysis 2.30

Comparison 2: SFA reduction vs usual diet ‐ secondary health events, Outcome 30: Non‐fatal MI, subgroup by SFA change

Analysis 2.31

Comparison 2: SFA reduction vs usual diet ‐ secondary health events, Outcome 31: Non‐fatal MI, subgroup by sex

Analysis 2.32

Comparison 2: SFA reduction vs usual diet ‐ secondary health events, Outcome 32: Non‐fatal MI, subgroup by CVD risk

Analysis 2.33

Comparison 2: SFA reduction vs usual diet ‐ secondary health events, Outcome 33: Non‐fatal MI, subgroup by TC reduction

Analysis 2.34

Comparison 2: SFA reduction vs usual diet ‐ secondary health events, Outcome 34: Non‐fatal MI, subgroup decade of publication

Analysis 2.35

Comparison 2: SFA reduction vs usual diet ‐ secondary health events, Outcome 35: STROKE

Analysis 2.36

Comparison 2: SFA reduction vs usual diet ‐ secondary health events, Outcome 36: Stroke, SA by low summary risk of bias

Analysis 2.37

Comparison 2: SFA reduction vs usual diet ‐ secondary health events, Outcome 37: Stroke, SA aim to reduce SFA

Analysis 2.38

Comparison 2: SFA reduction vs usual diet ‐ secondary health events, Outcome 38: Stroke, SA statistically significant SFA reduction

Analysis 2.39

Comparison 2: SFA reduction vs usual diet ‐ secondary health events, Outcome 39: Stroke, SA by TC reduction

Analysis 2.40

Comparison 2: SFA reduction vs usual diet ‐ secondary health events, Outcome 40: Stroke, SA excluding WHI

Analysis 2.41

Comparison 2: SFA reduction vs usual diet ‐ secondary health events, Outcome 41: Stroke, SA Mantel‐Haenszel fixed‐effect

Analysis 2.42

Comparison 2: SFA reduction vs usual diet ‐ secondary health events, Outcome 42: Stroke, SA Peto fixed‐effect

Analysis 2.43

Comparison 2: SFA reduction vs usual diet ‐ secondary health events, Outcome 43: Stroke, subgroup by any substitution

Analysis 2.44

Comparison 2: SFA reduction vs usual diet ‐ secondary health events, Outcome 44: Stroke, subgroup by main substitution

Analysis 2.45

Comparison 2: SFA reduction vs usual diet ‐ secondary health events, Outcome 45: Stroke, subgroup by duration

Analysis 2.46

Comparison 2: SFA reduction vs usual diet ‐ secondary health events, Outcome 46: Stroke, subgroup by baseline SFA

Analysis 2.47

Comparison 2: SFA reduction vs usual diet ‐ secondary health events, Outcome 47: Stroke, subgroup by SFA change

Analysis 2.48

Comparison 2: SFA reduction vs usual diet ‐ secondary health events, Outcome 48: Stroke, subgroup by sex

Analysis 2.49

Comparison 2: SFA reduction vs usual diet ‐ secondary health events, Outcome 49: Stroke, subgroup by CVD risk

Analysis 2.50

Comparison 2: SFA reduction vs usual diet ‐ secondary health events, Outcome 50: Stroke, subgroup by TC reduction

Analysis 2.51

Comparison 2: SFA reduction vs usual diet ‐ secondary health events, Outcome 51: Stroke, subgroup decade of publication

Analysis 2.52

Comparison 2: SFA reduction vs usual diet ‐ secondary health events, Outcome 52: CORONARY HEART DISEASE MORTALITY

Analysis 2.53

Comparison 2: SFA reduction vs usual diet ‐ secondary health events, Outcome 53: CHD mortality, SA by low summary risk of bias

Analysis 2.54

Comparison 2: SFA reduction vs usual diet ‐ secondary health events, Outcome 54: CHD mortality, SA aim to reduce SFA

Analysis 2.55

Comparison 2: SFA reduction vs usual diet ‐ secondary health events, Outcome 55: CHD mortality, SA statistically significant SFA reduction

Analysis 2.56

Comparison 2: SFA reduction vs usual diet ‐ secondary health events, Outcome 56: CHD mortality, SA by TC reduction

Analysis 2.57

Comparison 2: SFA reduction vs usual diet ‐ secondary health events, Outcome 57: CHD mortality, SA excluding WHI

Analysis 2.58

Comparison 2: SFA reduction vs usual diet ‐ secondary health events, Outcome 58: CHD mortality, SA Mantel‐Haenszel fixed‐effect

Analysis 2.59

Comparison 2: SFA reduction vs usual diet ‐ secondary health events, Outcome 59: CHD mortality, SA Peto fixed‐effect

Analysis 2.60

Comparison 2: SFA reduction vs usual diet ‐ secondary health events, Outcome 60: CHD mortality, subgroup by any substitution

Analysis 2.61

Comparison 2: SFA reduction vs usual diet ‐ secondary health events, Outcome 61: CHD mortality, subgroup by main substitution

Analysis 2.62

Comparison 2: SFA reduction vs usual diet ‐ secondary health events, Outcome 62: CHD mortality, subgroup by duration

Analysis 2.63

Comparison 2: SFA reduction vs usual diet ‐ secondary health events, Outcome 63: CHD mortality, subgroup by baseline SFA

Analysis 2.64

Comparison 2: SFA reduction vs usual diet ‐ secondary health events, Outcome 64: CHD mortality, subgroup by SFA change

Analysis 2.65

Comparison 2: SFA reduction vs usual diet ‐ secondary health events, Outcome 65: CHD mortality, subgroup by sex

Analysis 2.66

Comparison 2: SFA reduction vs usual diet ‐ secondary health events, Outcome 66: CHD mortality, subgroup by CVD risk

Analysis 2.67

Comparison 2: SFA reduction vs usual diet ‐ secondary health events, Outcome 67: CHD mortality, subgroup by TC reduction

Analysis 2.68

Comparison 2: SFA reduction vs usual diet ‐ secondary health events, Outcome 68: CHD mortality, subgroup decade of publication

Analysis 2.69

Comparison 2: SFA reduction vs usual diet ‐ secondary health events, Outcome 69: CORONARY HEART DISEASE EVENTS

Analysis 2.70

Comparison 2: SFA reduction vs usual diet ‐ secondary health events, Outcome 70: CHD events, SA by low summary risk of bias

Analysis 2.71

Comparison 2: SFA reduction vs usual diet ‐ secondary health events, Outcome 71: CHD events, SA excluding WHI

Analysis 2.72

Comparison 2: SFA reduction vs usual diet ‐ secondary health events, Outcome 72: CHD events, SA statistically significant SFA reduction

Analysis 2.73

Comparison 2: SFA reduction vs usual diet ‐ secondary health events, Outcome 73: CHD events, SA by TC reduction

Analysis 2.74

Comparison 2: SFA reduction vs usual diet ‐ secondary health events, Outcome 74: CHD events, SA aim to reduce SFA

Analysis 2.75

Comparison 2: SFA reduction vs usual diet ‐ secondary health events, Outcome 75: CHD events, SA Mantel‐Haenszel fixed‐effect

Analysis 2.76

Comparison 2: SFA reduction vs usual diet ‐ secondary health events, Outcome 76: CHD events, SA Peto fixed‐effect

Analysis 2.77

Comparison 2: SFA reduction vs usual diet ‐ secondary health events, Outcome 77: CHD events, subgroup by any substitution

Analysis 2.78

Comparison 2: SFA reduction vs usual diet ‐ secondary health events, Outcome 78: CHD events, subgroup by main substitution

Analysis 2.79

Comparison 2: SFA reduction vs usual diet ‐ secondary health events, Outcome 79: CHD events, subgroup by duration

Analysis 2.80

Comparison 2: SFA reduction vs usual diet ‐ secondary health events, Outcome 80: CHD events, subgroup by baseline SFA

Analysis 2.81

Comparison 2: SFA reduction vs usual diet ‐ secondary health events, Outcome 81: CHD events, subgroup by SFA change

Analysis 2.82

Comparison 2: SFA reduction vs usual diet ‐ secondary health events, Outcome 82: CHD events, subgroup by sex

Analysis 2.83

Comparison 2: SFA reduction vs usual diet ‐ secondary health events, Outcome 83: CHD events, subgroup by CVD risk

Analysis 2.84

Comparison 2: SFA reduction vs usual diet ‐ secondary health events, Outcome 84: CHD events, subgroup by TC reduction

Analysis 2.85

Comparison 2: SFA reduction vs usual diet ‐ secondary health events, Outcome 85: CHD events, subgroup decade of publication

Analysis 2.86

Comparison 2: SFA reduction vs usual diet ‐ secondary health events, Outcome 86: DIABETES DIAGNOSES

Analysis 3.1

Comparison 3: SFA reduction vs usual diet ‐ secondary blood outcomes, Outcome 1: Total cholesterol, mmol/L

Analysis 3.2

Comparison 3: SFA reduction vs usual diet ‐ secondary blood outcomes, Outcome 2: TC, mmol/L, subgroup by any replacement

Analysis 3.3

Comparison 3: SFA reduction vs usual diet ‐ secondary blood outcomes, Outcome 3: TC, mmol/L, subgroup by main replacement

Analysis 3.4

Comparison 3: SFA reduction vs usual diet ‐ secondary blood outcomes, Outcome 4: LDL cholesterol, mmol/L

Analysis 3.5

Comparison 3: SFA reduction vs usual diet ‐ secondary blood outcomes, Outcome 5: LDL, mmol/L, subgroup by any replacement

Analysis 3.6

Comparison 3: SFA reduction vs usual diet ‐ secondary blood outcomes, Outcome 6: LDL, mmol/L, subgroup by main replacement

Analysis 3.7

Comparison 3: SFA reduction vs usual diet ‐ secondary blood outcomes, Outcome 7: HDL cholesterol, mmol/L

Analysis 3.8

Comparison 3: SFA reduction vs usual diet ‐ secondary blood outcomes, Outcome 8: HDL, mmol/L, subgroup by any replacement

Analysis 3.9

Comparison 3: SFA reduction vs usual diet ‐ secondary blood outcomes, Outcome 9: HDL, mmol/L, subgroup by main replacement

Analysis 3.10

Comparison 3: SFA reduction vs usual diet ‐ secondary blood outcomes, Outcome 10: Triglycerides, mmol/L

Analysis 3.11

Comparison 3: SFA reduction vs usual diet ‐ secondary blood outcomes, Outcome 11: TG, mmol/L, subgroup by any replacement

Analysis 3.12

Comparison 3: SFA reduction vs usual diet ‐ secondary blood outcomes, Outcome 12: TG, mmol/L, subgroup by main replacement

Analysis 3.13

Comparison 3: SFA reduction vs usual diet ‐ secondary blood outcomes, Outcome 13: total cholesterol /HDL ratio

Analysis 3.14

Comparison 3: SFA reduction vs usual diet ‐ secondary blood outcomes, Outcome 14: TC /HDL ratio, subgroup by any replacement

Analysis 3.15

Comparison 3: SFA reduction vs usual diet ‐ secondary blood outcomes, Outcome 15: TC /HDL ratio, subgroup by main replacement

Analysis 3.16

Comparison 3: SFA reduction vs usual diet ‐ secondary blood outcomes, Outcome 16: LDL /HDL ratio

Analysis 3.17

Comparison 3: SFA reduction vs usual diet ‐ secondary blood outcomes, Outcome 17: Lp(a), mmol/L

Analysis 3.18

Comparison 3: SFA reduction vs usual diet ‐ secondary blood outcomes, Outcome 18: Lp(a), mmol/L, subgroup by any replacement

Analysis 3.19

Comparison 3: SFA reduction vs usual diet ‐ secondary blood outcomes, Outcome 19: Lp(a), mmol/L, subgroup by main replacement

Analysis 3.20

Comparison 3: SFA reduction vs usual diet ‐ secondary blood outcomes, Outcome 20: Insulin sensitivity

Analysis 4.1

Comparison 4: SFA reduction vs usual diet ‐ secondary outcomes including potential adverse effects, Outcome 1: Cancer diagnoses

Analysis 4.2

Comparison 4: SFA reduction vs usual diet ‐ secondary outcomes including potential adverse effects, Outcome 2: Cancer deaths

Analysis 4.3

Comparison 4: SFA reduction vs usual diet ‐ secondary outcomes including potential adverse effects, Outcome 3: Weight, kg

Analysis 4.4

Comparison 4: SFA reduction vs usual diet ‐ secondary outcomes including potential adverse effects, Outcome 4: BMI, kg/m2

Analysis 4.5

Comparison 4: SFA reduction vs usual diet ‐ secondary outcomes including potential adverse effects, Outcome 5: Systolic Blood Pressure, mmHg

Analysis 4.6

Comparison 4: SFA reduction vs usual diet ‐ secondary outcomes including potential adverse effects, Outcome 6: Diastolic Blood Pressure, mmHg

Analysis 4.7

Comparison 4: SFA reduction vs usual diet ‐ secondary outcomes including potential adverse effects, Outcome 7: Quality of Life

Summary of findings 1. Effect of reducing saturated fat compared to usual saturated fat on CVD risk in adults (note: for the full set of GRADE tables see additional tables 24 to 28)

Low saturated fat compared with usual saturated fat for CVD risk
Patient or population: people at any baseline risk of CVD Intervention: lower saturated fat intake Comparison: higher saturated fat intake Settings: Any, including community‐dwelling and institutions. Included RCTs were conducted in North America, Europe and Australia/New Zealand, no studies were carried out in industrialising or developing countries.
Outcomes	Relative effect (95% CI)	Anticipated absolute effects (95% CI)		No of Participants (studies)	Quality of the evidence (GRADE)	Comments
Outcomes	Relative effect (95% CI)	Risk with higher SFA intake	Risk with lower SFA intake	No of Participants (studies)	Quality of the evidence (GRADE)	Comments
All‐cause mortality follow‐up mean duration 56 months¹	RR 0.96 (0.90 to 1.03)	62 per 1000	60 per 1000 (56 to 64)	55,858 (12)	⊕⊕⊕⊝ Moderate^2,3,4,5,6	Critical importance. Reducing saturated fat intake probably makes little or no difference to all‐cause mortality.
Cardiovascular mortality follow‐up mean duration 53 months¹	RR 0.94 (0.78 to 1.13)	19 per 1000	18 per 1000 (15 to 22)	53,421 (11)	⊕⊕⊕⊝ Moderate^2,3,4,6,7	Critical importance. Reducing saturated fat intake probably makes little or no difference to cardiovascular mortality.
Combined cardiovascular events follow‐up mean duration 52 months¹	RR 0.83 (0.70 to 0.98)	85 per 1000	70 per 1000 (59 to 83)	53,758 (13)	⊕⊕⊕⊝ Moderate^4,8,9,10,11	Critical importance. Reducing saturated fat intake probably reduces cardiovascular events (to a greater extent with greater cholesterol reduction).
Myocardial infarctions follow‐up mean duration 55 months	RR 0.90 (0.80 to 1.01)	32 per 1000	29 per 1000 (25 to 32)	53,167 (11)	⊕⊝⊝⊝ VeryLow ^3,4,5,11,12	Critical importance. The effect of reducing saturated fat intake on risk of myocardial infarction is unclear as the evidence is of very low quality.
Non‐fatal MI follow‐up mean duration 55 months¹	RR 0.97 (0.87 to 1.07)	26 per 1000	25 per 1000 (23 to 28)	52,834 (8)	⊕⊕⊝⊝ Low^3,4,5,6,13	Critical importance. Reducing saturated fat may have little or no effect on risk of non‐fatal myocardial infarction.
Stroke follow‐up mean duration 59 months¹	RR 0.92 (0.68 to 1.25)	22 per 1000	20 per 1000 (15 to 27)	50,952 (7)	⊕⊝⊝⊝ VeryLow ^3,4,6,13,14	Critical importance. The effect of reducing saturated fat on the risk of stroke is unclear as the evidence was of very low quality.
CHD mortality follow‐up mean duration 65 months¹	RR 0.97 (0.82 to 1.16)	16 per 1000	16 per 1000 (13 to 19)	53,159 (9)	⊕⊕⊝⊝ Low^2,3,4,6,14	Critical importance. Reducing saturated fat intake may have little or no effect on CHD mortality.
CHD events follow‐up mean duration 59 months¹	RR 0.83 (0.68 to 1.01)	42 per 1000	35 per 1000 (29 to 43)	53,199 (11)	⊕⊝⊝⊝ Verylow ^4,5,6,12,15	Critical importance. The effect of reducing saturated fat on risk of CHD events is unclear as the evidence is of very low quality.
*The risk in the intervention group (and its 95% confidence interval) is based on the assumed risk in the comparison group and the relative effect of the intervention (and its 95% CI). CI: Confidence interval; RR: Risk Ratio; CHD: coronary heart disease.
GRADE Working Group grades of evidence High quality: Further research is very unlikely to change our confidence in the estimate of effect. Moderate quality: Further research is likely to have an important impact on our confidence in the estimate of effect and may change the estimate. Low quality: Further research is very likely to have an important impact on our confidence in the estimate of effect and is likely to change the estimate. Very low quality: We are very uncertain about the estimate.
¹Minimum study duration was 24 months. ²Risk of bias. Limiting trials to those at low summary risk of bias also suggested little or no effect. Not downgraded. ³Inconsistency. We found no important heterogeneity; I² ≤ 30%. Not downgraded. ⁴Indirectness. These RCTs directly assessed the effect of lower vs higher saturated fat intake on health outcomes of interest. Participants included men and women with and without CVD at baseline (also some participants with CVD risk factors like diabetes, or at risk of cancers). However, no trials included participants from developing countries. Not downgraded. ⁵Imprecision. The 95% CI includes both no effect and a benefit. Downgraded once. ⁶Publication bias. The funnel plot, and comparison of fixed‐ and random‐effects meta‐analyses did not suggest major small‐study (publication) bias. Not downgraded. ⁷Imprecision. The 95% CI includes both harm and benefit. Downgraded once. ⁸Risk of bias. Limiting trials to those at low summary risk of bias suggested a smaller and non‐statistically significant effect (RR 0.96, 95% CI 0.76 to 1.20) suggesting little or no effect on risk of CVD events. Downgraded once (along with publication bias). ⁹Inconsistency. Although heterogeneity was high, I² = 65%, this was mostly explained by the degree of cholesterol‐lowering (a dose effect). Not downgraded. ¹⁰Imprecision. The 95% CI includes only benefit or minimal effect. Not downgraded. ¹¹Publication bias. The funnel plot did not suggest publication bias, but comparison of fixed‐ and random‐effects meta‐analyses suggested possible small‐study (publication) bias. Downgraded once (along with risk of bias, downgraded once in total). ¹²Risk of bias. Limiting trials to those at low summary risk of bias moved the RR slightly towards 1.0, suggesting little or no effect on total MI. Downgraded once. ¹³Risk of bias. Limiting trials to those at low summary risk of bias moved the RR slightly away from 1.0, suggesting that reducing SFA reduces the risk of non‐fatal MI. This was also seen in several other sensitivity analyses. Downgraded once. ¹⁴Imprecision. The 95% CI includes both important benefits and important harms. Downgraded twice. ¹⁵Inconsistency. Heterogeneity was high, I² = 65%. Downgraded once.

Summary of findings 1. Effect of reducing saturated fat compared to usual saturated fat on CVD risk in adults (note: for the full set of GRADE tables see additional tables 24 to 28)

Table 1. Comparison of study interventions for included RCTs

Reference	Population	CVD risk category	Is intervention delivered to Individual or group?	intervention given by?	Face‐to‐face or other?	Number of visits	Is intervention advice only or other intervention?
Black 1994	People with non‐melanoma skin cancer	Low	Unclear	Dietitian	Face‐to‐face	8 x weekly classes then monthly follow‐up sessions	Advice (behaviour techniques learning)
DART 1989	Men recovering from a MI	High	Individual	Dietitian	Face‐to‐face	9	Advice (diet advice, recipes and encouragement)
Houtsmuller 1979	Adults with newly‐diagnosed diabetes	Moderate	Unclear	Dietitian	Unclear	Unclear	Advice?
Ley 2004	People with impaired glucose intolerance or high normal blood glucose	Moderate	Small group	Unclear	Face‐to‐face	Monthly meetings	Advice (education, personal goal‐setting, self‐monitoring)
Moy 2001	Middle‐aged siblings of people with early CHD, with at least 1 CVD risk factor	Moderate	Individual	Trained nurse	Face‐to‐face	6 ‐ 8 weekly for 2 years	Advice (individualised counselling sessions)
MRC 1968	Free‐living men who have survived a 1st MI	High	Individual	Dietitian	Face‐to‐face	Unclear	Advice and supplement (soy oil)
Oslo Diet‐Heart 1966	Men with previous MI	High	Individual	Dietitian	Face‐to‐face and other	Unclear	Advice and supplement (food)
Oxford Retinopathy 1978	Newly‐diagnosed non‐insulin‐dependent diabetics	Moderate	Individual	Diabetes dietitian	Face‐to‐face	After 1 month then at 3‐month intervals	Advice
Rose corn oil 1965	Men (?) with angina or following MI	High	Unclear	Unclear	Unclear	Follow‐up clinic monthly, then every 2 months	Advice and supplement (oil)
Rose olive 1965	Men (?) with angina or following MI	High	Unclear	Unclear	Unclear	Follow‐up clinic monthly, then every 2 months	Advice and supplement (oil)
Simon 1997	Women with a high risk of breast cancer	Low	Individual followed by individual or group	Dietitian	Face‐to‐face	Bi‐weekly over 3 months followed by monthly	Advice (individualised eating plan and counselling sessions)
STARS 1992	Men with angina referred for angiography	High	Individual	Dietitian	Face‐to‐face	Clinic visits at 3‐month intervals	Advice
Sydney Diet‐Heart 1978	Men with angina referred for angiography	High	Individual	Unclear	Face‐to‐face	3 times in 1st year and twice annually thereafter	Advice
Veterans Admin 1969	Men living at the Veterans Administration Center	Low	Individual	Unclear (whole diet provided)	N/A	N/A	Diet provided
WHI 2006	Postmenopausal women aged 50 ‐ 79 with or without CVD at baseline	Low and High	Group	Nutritionists	Face‐to‐face	18 sessions/1st yr and quarterly maintenance sessions after	Advice
WINS 2006	Women with localised resected breast cancer	Low	Individual followed by group	Dietitian	Face‐to‐face	8 bi‐weekly sessions, then 3‐monthly contact and optional monthly sessions	Advice
MI: myocardial infarction N/A: not applicable

Table 1. Comparison of study interventions for included RCTs

Table 2. Number of participants and number of outcomes for dichotomous variables (by intervention arm)

	Participants	All‐cause mortality	CV mortality	CVD events	MI	Non‐fatal MI	Stroke	CHD mortality	CHD events	Diabetes Diagnoses
Black 1994	133	133	133	133	0	0	0	0	0	0
DART 1989	2033	2033	2033	2033	2033	2033	0	2033	2033	0
Houtsmuller 1979	102	0	0	102	102	0	0	102	102	0
Ley 2004	176	176	176	176	176	0	176	0	176	0
Moy 2001	267	0	0	235	235	235	235	0	267	0
MRC 1968	393	393	393	393	393	393	393	393	393	0
Oslo Diet‐Heart 1966	412	412	412	412	412	412	412	412	412	0
Oxford Retinopathy 1978	249 (data not provided by arm)	0	0	0	0	0	0	0	0	0
Rose corn oil 1965	41	41	41	41	41	41	0	41	41	0
Rose olive 1965	39	39	39	39	39	39	0	39	39	0
Simon 1997	194 (data not provided by arm)	0	0	0	0	0	0	0	0	0
STARS 1992	60	55	55	55	55	0	55	0	55	0
Sydney Diet‐Heart 1978	458	458	458	458	0	0	0	458	0	0
Veterans Admin 1969	846	846	846	846	846	846	846	846	846	0
WHI 2006	48,835	48,835	48,835	48,835	48,835	48,835	48,835	48,835	48,835	48,835
WINS 2006	2437	2437	0	0	0	0	0	0	0	0
Total Participants	56,675	55,858	53,421	53,758	53,167	52,834	50,952	53,159	53,199	48,835
Percent of participants for this outcome	100%	99%	94%	95%	94%	93%	90%	94%	94%	86%
These numbers are the numbers of participants in each study who were available for assessment of outcomes within meta‐analysis (not necessarily the number of participants randomised within the trial). CHD: coronary heart disease CV: cardiovascular CVD: cardiovascular disease

Table 2. Number of participants and number of outcomes for dichotomous variables (by intervention arm)

Table 3. Number of participants and number of participants with data for continuous outcomes (by intervention arm)

	Participants	Total cholesterol	LDL cholesterol	HDL cholesterol	Triglycerides	TG/HDL ratio	Total cholesterol/HDL ratio	LDL/HDL ratio	LP (a)	Insulin sensitivity
Black 1994	133	0	0	0	0	0	0	0	0	0
DART 1989	2033	1855	0	1855	0	0	0	0	0	0
Houtsmuller 1979	102	96	0	0	96	0	0	0	0	96
Ley 2004	176	103	103	103	103	0	103	0	0	103
Moy 2001	267	0	235	235	235	0	0	0	0	0
MRC 1968	393	177	0	0	0	0	0	0	0	0
Oslo Diet‐Heart 1966	412	329	0	0	0	0	0	0	0	0
Oxford Retinopathy 1978	249	58	0	0	0	0	0	0	0	0
Rose corn oil 1965	41	22	0	0	0	0	0	0	0	0
Rose olive 1965	39	24	0	0	0	0	0	0	0	0
Simon 1997	194	72	71	72	71	0	0	0	0	0
STARS 1992	60	50	50	50	50	0	50	50	50	50
Sydney Diet‐Heart 1978	458	458	0	0	458	0	0	0	0	0
Veterans Admin 1969	846	843	0	0	0	0	0	0	0	0
WHI 2006	48,835	2832	2832	2832	2832	0	2832	0	2832	2832
WINS 2006	2437	196	0	0	0	0	0	0	0	0
Total Participants	56,675	7115	3291	5147	3845	0	2985	50	2882	3081
Percent of participants for this outcome	100%	13%	6%	9%	7%	0%	5%	0.1%	5%	5%
These numbers are the numbers of participants in each study who were available for assessment of outcomes within meta‐analysis (not necessarily the number of participants randomised within the trial). HDL: high density lipoprotein LDL: low density lipoprotein Lp(a): lipoprotein (a) TG: triglyceride

Table 3. Number of participants and number of participants with data for continuous outcomes (by intervention arm)

Table 4. Meta‐regression of effects of SFA reduction on cardiovascular events

Regression factor	No. of studies	Constant	Coefficient (95% CI)	P value	Proportion of between study variation explained
Change in SFA as %E	8	0.01	0.05 (‐0.03 to 0.13)	0.16	89%
Change in SFA as % of control	8	0.26	0.01 (‐0.01 to 0.03)	0.14	89%
Baseline SFA as %E	8	0.68	‐0.06 (‐0.15 to 0.04)	0.19	81%
Change in TC, mmol/L	12	0.03	0.69 (0.05 to 1.33)	0.04	99%
Change in PUFA as %E	5	‐0.01	‐0.02 (‐0.08 to 0.03)	0.25	100%
Change in MUFA as %E	5	‐0.26	‐0.03 (‐0.14 to 0.09)	0.50	‐87%
Change in CHO as %E	7	‐0.11	‐0.00 (‐0.05 to 0.05)	0.92	‐273%
Change in total fat intake as %E	9	‐0.17	‐0.01 (‐0.03 to 0.01)	0.28	100%
Gender*	13	‐0.17	‐0.14 (‐0.63 to 0.35)	0.55	‐13%
Study duration	13	‐0.47	0.00 (‐0.01 to 0.02)	0.76	‐24.8%
CVD risk at baseline**	13	‐0.44	0.03 (‐0.48 to 0.55)	0.89	‐39%
^{Gender was coded as follows: 0 = women, 1 = mixed, 2 = men *CVD risk at baseline was coded as follows: 1 = Low CVD risk, 2 = Moderate CVD risk, 3 = existing CVD} ^{CHO: carbohydrate CI: confidence interval CVD: cardiovascular disease E: energy MUFA: monounsaturated fatty acid PUFA: polyunsaturated fatty fat SFA: saturated fatty acid TC: total cholesterol}

Table 4. Meta‐regression of effects of SFA reduction on cardiovascular events

Table 5. SFA cut‐off data

Cut‐ off	RR of all‐cause mortality	RR of CVD mortality	RR of CVD events	RR of MI	RR of non‐fatal MI	RR of stroke	RR of CHD mortality	RR of CHD events
7%E	0.89 (0.66 to 1.20)	0.20 (0.01 to 4.15)	0.20 (0.01 to 4.15)	N/A	N/A	N/A	N/A	N/A
8%E	0.89 (0.66 to 1.20)	0.20 (0.01 to 4.15)	0.20 (0.01 to 4.15)	N/A	N/A	N/A	N/A	N/A
9%E	0.96 (0.83 to 1.10)	0.69 (0.51 to 0.94)	0.79 (0.62 to 0.99)	0.76 (0.55 to 1.05)	0.62 (0.31 to 1.21)	0.59 (0.30 to 1.15)	0.82 (0.55 to 1.21)	0.77 (0.56 to 1.04)
10%E	0.99 (0.90 to 1.09)	0.95 (0.67 to 1.35)	0.88 (0.66 to 1.18)	0.93 (0.80 to 1.08)	0.89 (0.58 to 1.35)	0.87 (0.58 to 1.33)	1.05 (0.77 to 1.43)	0.82 (0.60 to 1.13)
11%E	0.99 (0.88 to 1.12)	0.92 (0.65 to 1.31)	0.86 (0.66 to 1.13)	0.94 (0.84 to 1.06)	0.89 (0.58 to 1.35)	0.76 (0.45 to 1.30)	1.02 (0.84 to 1.24)	0.85 (0.63 to 1.15)
12%E	0.98 (0.91 to 1.07)	0.95 (0.75 to 1.21)	0.90 (0.74 to 1.08)	0.94 (0.85 to 1.04)	0.90 (0.72 to 1.14)	0.93 (0.55 to 1.25)	1.02 (0.84 to 1.24)	0.90 (0.77 to 1.06)
13%E	1.02 (0.83 to 1.25)	0.93 (0.63 to 1.38)	0.87 (0.65 to 1.17)	0.87 (0.73 to 1.04)	0.72 (0.50 to 1.03)	0.54 (0.29 to 1.00)	1.06 (0.76 to 1.48)	0.84 (0.63 to 1.12)
CHD: coronary heart disease CVD: cardiovascular disease E: energy MI: myocardial infarction N/A: not applicable (no relevant studies) RR: risk ratio SFA: saturated fat, as percentage of energy

Table 5. SFA cut‐off data

Comparison 1. SFA reduction vs usual diet ‐ primary outcomes

Outcome or subgroup title	No. of studies	No. of participants	Statistical method	Effect size
1.1 ALL‐CAUSE MORTALITY Show forest plot	12	55858	Risk Ratio (M‐H, Random, 95% CI)	0.96 [0.90, 1.03]

1.2 All‐cause mortality, SA low summary risk of bias Show forest plot	7	53219	Risk Ratio (M‐H, Random, 95% CI)	0.95 [0.84, 1.08]

1.3 All‐cause mortality, SA aim to reduce SFA Show forest plot	9	53112	Risk Ratio (M‐H, Random, 95% CI)	0.97 [0.89, 1.06]

1.4 All‐cause mortality, SA statistically significant SFA reduction Show forest plot	8	54973	Risk Ratio (M‐H, Random, 95% CI)	0.98 [0.92, 1.04]

1.5 All‐cause mortality, SA TC reduction Show forest plot	8	53073	Risk Ratio (M‐H, Random, 95% CI)	0.97 [0.88, 1.07]

1.6 All‐cause mortality, SA excluding WHI Show forest plot	11	7023	Risk Ratio (M‐H, Random, 95% CI)	0.95 [0.83, 1.07]

1.7 All‐cause mortality, SA Mantel‐Haenszel fixed‐effect Show forest plot	12	55858	Risk Ratio (M‐H, Fixed, 95% CI)	0.97 [0.91, 1.03]

1.8 All‐cause mortality, SA Peto fixed‐effect Show forest plot	12	55858	Peto Odds Ratio (Peto, Fixed, 95% CI)	0.96 [0.90, 1.04]

1.9 All‐cause mortality, subgroup by any substitution Show forest plot	12		Risk Ratio (M‐H, Random, 95% CI)	Subtotals only

1.9.1 replaced by PUFA	7	4238	Risk Ratio (M‐H, Random, 95% CI)	0.96 [0.82, 1.13]
1.9.2 replaced by MUFA	1	52	Risk Ratio (M‐H, Random, 95% CI)	3.00 [0.33, 26.99]
1.9.3 replaced by CHO	6	53669	Risk Ratio (M‐H, Random, 95% CI)	0.97 [0.90, 1.04]
1.9.4 replaced by protein	5	53614	Risk Ratio (M‐H, Random, 95% CI)	0.97 [0.90, 1.04]
1.9.5 replacement unclear	0	0	Risk Ratio (M‐H, Random, 95% CI)	Not estimable
1.10 All‐cause mortality, subgroup by main substitution Show forest plot	12		Risk Ratio (M‐H, Random, 95% CI)	Subtotals only

1.10.1 replaced by PUFA	6	4183	Risk Ratio (M‐H, Random, 95% CI)	0.97 [0.82, 1.14]
1.10.2 replaced by MUFA	1	52	Risk Ratio (M‐H, Random, 95% CI)	3.00 [0.33, 26.99]
1.10.3 replaced by CHO	5	51636	Risk Ratio (M‐H, Random, 95% CI)	0.97 [0.90, 1.04]
1.10.4 replaced by protein	0	0	Risk Ratio (M‐H, Random, 95% CI)	Not estimable
1.10.5 replacement unclear	0	0	Risk Ratio (M‐H, Random, 95% CI)	Not estimable
1.11 All‐cause mortality, subgroup by duration Show forest plot	12	55858	Risk Ratio (M‐H, Random, 95% CI)	0.96 [0.90, 1.03]

1.11.1 up to 24mo	4	2246	Risk Ratio (M‐H, Random, 95% CI)	0.99 [0.78, 1.26]
1.11.2 >24 to 48mo	3	1294	Risk Ratio (M‐H, Random, 95% CI)	0.96 [0.83, 1.12]
1.11.3 >48mo	4	52142	Risk Ratio (M‐H, Random, 95% CI)	0.96 [0.79, 1.16]
1.11.4 unclear duration	1	176	Risk Ratio (M‐H, Random, 95% CI)	0.33 [0.07, 1.61]
1.12 All‐cause mortality, subgroup by baseline SFA Show forest plot	12	55858	Risk Ratio (M‐H, Random, 95% CI)	0.96 [0.90, 1.03]

1.12.1 up to 12%E SFA baseline	1	2437	Risk Ratio (M‐H, Random, 95% CI)	0.90 [0.67, 1.21]
1.12.2 >12 to 15%E SFA baseline	5	51635	Risk Ratio (M‐H, Random, 95% CI)	1.01 [0.86, 1.19]
1.12.3 >15 to 18%E SFA baseline	1	55	Risk Ratio (M‐H, Random, 95% CI)	0.35 [0.04, 3.12]
1.12.4 >18%E SFA baseline	1	846	Risk Ratio (M‐H, Random, 95% CI)	0.98 [0.83, 1.15]
1.12.5 unclear	4	885	Risk Ratio (M‐H, Random, 95% CI)	0.80 [0.62, 1.04]
1.13 All‐cause mortality, subgroup by SFA change Show forest plot	12	55858	Risk Ratio (M‐H, Random, 95% CI)	0.96 [0.90, 1.03]

1.13.1 up to 4%E difference	5	53939	Risk Ratio (M‐H, Random, 95% CI)	0.99 [0.86, 1.13]
1.13.2 >4 to 8%E difference	2	188	Risk Ratio (M‐H, Random, 95% CI)	0.41 [0.08, 2.07]
1.13.3 >8%E difference	1	846	Risk Ratio (M‐H, Random, 95% CI)	0.98 [0.83, 1.15]
1.13.4 unclear	4	885	Risk Ratio (M‐H, Random, 95% CI)	0.80 [0.62, 1.04]
1.14 All‐cause mortality, subgroup by sex Show forest plot	12	55858	Risk Ratio (M‐H, Random, 95% CI)	0.96 [0.90, 1.03]

1.14.1 Men	9	4410	Risk Ratio (M‐H, Random, 95% CI)	0.96 [0.83, 1.11]
1.14.2 Women	2	51272	Risk Ratio (M‐H, Random, 95% CI)	0.97 [0.90, 1.05]
1.14.3 Mixed, men and women	1	176	Risk Ratio (M‐H, Random, 95% CI)	0.33 [0.07, 1.61]
1.15 All‐cause mortality, subgroup by CVD risk Show forest plot	12	55858	Risk Ratio (M‐H, Random, 95% CI)	0.96 [0.90, 1.03]

1.15.1 Low CVD risk	4	52251	Risk Ratio (M‐H, Random, 95% CI)	0.97 [0.91, 1.04]
1.15.2 Moderate CVD risk	1	176	Risk Ratio (M‐H, Random, 95% CI)	0.33 [0.07, 1.61]
1.15.3 Existing CVD disease	7	3431	Risk Ratio (M‐H, Random, 95% CI)	0.97 [0.76, 1.24]
1.16 All‐cause mortality, subgroup by TC reduction Show forest plot	12		Risk Ratio (M‐H, Random, 95% CI)	Subtotals only

1.16.1 serum chol reduced by at least 0.2mmol/L	7	4238	Risk Ratio (M‐H, Random, 95% CI)	0.96 [0.81, 1.14]
1.16.2 serum chol reduced by <0.2mmol/L	4	51487	Risk Ratio (M‐H, Random, 95% CI)	0.97 [0.90, 1.04]
1.16.3 serum chol reduction unclear	1	133	Risk Ratio (M‐H, Random, 95% CI)	0.51 [0.05, 5.46]
1.17 All‐cause mortality, subgroup decade of publication Show forest plot	12	55858	Risk Ratio (M‐H, Random, 95% CI)	0.96 [0.90, 1.03]

1.17.1 1960s	5	1731	Risk Ratio (M‐H, Random, 95% CI)	0.92 [0.80, 1.07]
1.17.2 1970s	1	458	Risk Ratio (M‐H, Random, 95% CI)	1.49 [0.95, 2.34]
1.17.3 1980s	1	2033	Risk Ratio (M‐H, Random, 95% CI)	0.98 [0.76, 1.25]
1.17.4 1990s	2	188	Risk Ratio (M‐H, Random, 95% CI)	0.41 [0.08, 2.07]
1.17.5 2000s	3	51448	Risk Ratio (M‐H, Random, 95% CI)	0.97 [0.88, 1.05]
1.18 CARDIOVASCULAR MORTALITY Show forest plot	11	53421	Risk Ratio (M‐H, Random, 95% CI)	0.94 [0.78, 1.13]

1.19 CVD mortality, SA low summary risk of bias Show forest plot	4	50315	Risk Ratio (M‐H, Random, 95% CI)	0.96 [0.67, 1.38]

1.20 CVD mortality, SA aim to reduce SFA Show forest plot	9	53112	Risk Ratio (M‐H, Random, 95% CI)	0.95 [0.79, 1.14]

1.21 CVD mortality, SA statistically significant SFA reduction Show forest plot	7	52536	Risk Ratio (M‐H, Random, 95% CI)	0.95 [0.75, 1.21]

1.22 CVD mortality, SA TC reduction Show forest plot	8	53073	Risk Ratio (M‐H, Random, 95% CI)	0.95 [0.78, 1.15]

1.23 CVD mortality, SA excluding WHI Show forest plot	10	4586	Risk Ratio (M‐H, Random, 95% CI)	0.92 [0.72, 1.18]

1.24 CVD mortality, SA Mantel‐Haenszel fixed‐effect Show forest plot	11	53421	Risk Ratio (M‐H, Fixed, 95% CI)	0.95 [0.85, 1.07]

1.25 CVD mortality, SA Peto fixed‐effect Show forest plot	11	53421	Peto Odds Ratio (Peto, Fixed, 95% CI)	0.95 [0.84, 1.08]

1.26 CVD mortality, subgroup by any substitution Show forest plot	11		Risk Ratio (M‐H, Random, 95% CI)	Subtotals only

1.26.1 replaced by PUFA	7	4251	Risk Ratio (M‐H, Random, 95% CI)	0.95 [0.73, 1.25]
1.26.2 replaced by MUFA	1	52	Risk Ratio (M‐H, Random, 95% CI)	3.00 [0.33, 26.99]
1.26.3 replace by CHO	5	51232	Risk Ratio (M‐H, Random, 95% CI)	0.99 [0.85, 1.14]
1.26.4 replaced by protein	4	51177	Risk Ratio (M‐H, Random, 95% CI)	0.99 [0.86, 1.14]
1.26.5 replacement unclear	0	0	Risk Ratio (M‐H, Random, 95% CI)	Not estimable
1.27 CVD mortality, subgroup by main substitution Show forest plot	11		Risk Ratio (M‐H, Random, 95% CI)	Subtotals only

1.27.1 replaced by PUFA	6	4196	Risk Ratio (M‐H, Random, 95% CI)	0.97 [0.73, 1.28]
1.27.2 replaced by MUFA	1	52	Risk Ratio (M‐H, Random, 95% CI)	3.00 [0.33, 26.99]
1.27.3 replace by CHO	4	49199	Risk Ratio (M‐H, Random, 95% CI)	0.78 [0.42, 1.46]
1.27.4 replaced by protein	0	0	Risk Ratio (M‐H, Random, 95% CI)	Not estimable
1.27.5 replacement unclear	0	0	Risk Ratio (M‐H, Random, 95% CI)	Not estimable
1.28 CVD mortality, subgroup by duration Show forest plot	11	53447	Risk Ratio (M‐H, Random, 95% CI)	0.95 [0.78, 1.16]

1.28.1 up to 24mo	4	2272	Risk Ratio (M‐H, Random, 95% CI)	1.26 [0.54, 2.94]
1.28.2 >24 to 48mo	3	1294	Risk Ratio (M‐H, Random, 95% CI)	0.79 [0.57, 1.08]
1.28.3 >48 mo	3	49705	Risk Ratio (M‐H, Random, 95% CI)	1.02 [0.73, 1.43]
1.28.4 unclear duration	1	176	Risk Ratio (M‐H, Random, 95% CI)	0.25 [0.03, 2.19]
1.29 CVD mortality, subgroup by baseline SFA Show forest plot	11	53447	Risk Ratio (M‐H, Random, 95% CI)	0.95 [0.78, 1.16]

1.29.1 up to 12%E SFA baseline	0	0	Risk Ratio (M‐H, Random, 95% CI)	Not estimable
1.29.2 >12 to 15%E SFA baseline	5	51635	Risk Ratio (M‐H, Random, 95% CI)	1.06 [0.84, 1.32]
1.29.3 >15 to 18%E SFA baseline	1	55	Risk Ratio (M‐H, Random, 95% CI)	0.35 [0.04, 3.12]
1.29.4 >18%E SFA baseline	1	846	Risk Ratio (M‐H, Random, 95% CI)	0.70 [0.51, 0.96]
1.29.5 unclear	4	911	Risk Ratio (M‐H, Random, 95% CI)	1.00 [0.61, 1.66]
1.30 CVD mortality, subgroup by SFA change Show forest plot	11	53447	Risk Ratio (M‐H, Random, 95% CI)	0.95 [0.78, 1.16]

1.30.1 up to 4%E difference	4	51502	Risk Ratio (M‐H, Random, 95% CI)	1.07 [0.85, 1.33]
1.30.2 >4 to 8%E difference	2	188	Risk Ratio (M‐H, Random, 95% CI)	0.29 [0.05, 1.70]
1.30.3 >8%E difference	1	846	Risk Ratio (M‐H, Random, 95% CI)	0.70 [0.51, 0.96]
1.30.4 unclear	4	911	Risk Ratio (M‐H, Random, 95% CI)	1.00 [0.61, 1.66]
1.31 CVD mortality, subgroup by sex Show forest plot	11	53447	Risk Ratio (M‐H, Random, 95% CI)	0.95 [0.78, 1.16]

1.31.1 Men	9	4436	Risk Ratio (M‐H, Random, 95% CI)	0.96 [0.73, 1.25]
1.31.2 Women	1	48835	Risk Ratio (M‐H, Random, 95% CI)	1.00 [0.84, 1.19]
1.31.3 Mixed, men and women	1	176	Risk Ratio (M‐H, Random, 95% CI)	0.25 [0.03, 2.19]
1.32 CVD mortality, subgroup by CVD risk Show forest plot	11	53447	Risk Ratio (M‐H, Random, 95% CI)	0.96 [0.80, 1.14]

1.32.1 Low CVD risk	3	47537	Risk Ratio (M‐H, Random, 95% CI)	0.84 [0.60, 1.16]
1.32.2 Moderate CVD risk	1	176	Risk Ratio (M‐H, Random, 95% CI)	0.25 [0.03, 2.19]
1.32.3 Existing CVD disease	8	5734	Risk Ratio (M‐H, Random, 95% CI)	1.04 [0.83, 1.31]
1.33 CVD mortality, subgroup by TC reduction Show forest plot	11		Risk Ratio (M‐H, Random, 95% CI)	Subtotals only

1.33.1 serum chol reduced by at least 0.2mmol/L	7	4251	Risk Ratio (M‐H, Random, 95% CI)	0.95 [0.73, 1.25]
1.33.2 serum chol reduced by <0.2mmol/L	3	49063	Risk Ratio (M‐H, Random, 95% CI)	0.97 [0.47, 2.01]
1.33.3 serum chol reduction unclear	1	133	Risk Ratio (M‐H, Random, 95% CI)	0.20 [0.01, 4.15]
1.34 CVD mortality, subgroup decade of publication Show forest plot	11	53421	Risk Ratio (M‐H, Random, 95% CI)	0.94 [0.78, 1.13]

1.34.1 1960s	5	1731	Risk Ratio (M‐H, Random, 95% CI)	0.78 [0.63, 0.97]
1.34.2 1970s	1	458	Risk Ratio (M‐H, Random, 95% CI)	1.59 [0.99, 2.55]
1.34.3 1980s	1	2033	Risk Ratio (M‐H, Random, 95% CI)	1.01 [0.77, 1.31]
1.34.4 1990s	2	188	Risk Ratio (M‐H, Random, 95% CI)	0.29 [0.05, 1.70]
1.34.5 2000s	2	49011	Risk Ratio (M‐H, Random, 95% CI)	0.78 [0.27, 2.21]
1.35 COMBINED CARDIOVASCULAR EVENTS Show forest plot	13	53758	Risk Ratio (M‐H, Random, 95% CI)	0.83 [0.70, 0.98]

1.36 CVD events, SA low summary risk of bias Show forest plot	4	50315	Risk Ratio (M‐H, Random, 95% CI)	0.96 [0.76, 1.20]

1.37 CVD events, SA aim to reduce SFA Show forest plot	11	53449	Risk Ratio (M‐H, Random, 95% CI)	0.84 [0.70, 1.00]

1.38 CVD events, SA statistically significant SFA reduction Show forest plot	8	52771	Risk Ratio (M‐H, Random, 95% CI)	0.90 [0.74, 1.08]

1.39 CVD events, SA TC reduction Show forest plot	10	53410	Risk Ratio (M‐H, Random, 95% CI)	0.83 [0.69, 1.00]

1.40 CVD events, SA excluding WHI Show forest plot	12	4923	Risk Ratio (M‐H, Random, 95% CI)	0.79 [0.64, 0.98]

1.41 CVD events, SA Mantel‐Haenszel fixed‐effect Show forest plot	13	53758	Risk Ratio (M‐H, Fixed, 95% CI)	0.94 [0.89, 0.99]

1.42 CVD events, SA Peto fixed‐effect Show forest plot	13	53758	Peto Odds Ratio (Peto, Fixed, 95% CI)	0.93 [0.88, 0.99]

1.43 CVD events, SA excluding trials with additional interventions Show forest plot	10	4456	Risk Ratio (M‐H, Random, 95% CI)	0.86 [0.67, 1.09]

1.44 CVD events, subgroup by any substitution Show forest plot	13		Risk Ratio (M‐H, Random, 95% CI)	Subtotals only

1.44.1 replaced by PUFA	8	4353	Risk Ratio (M‐H, Random, 95% CI)	0.79 [0.62, 1.00]
1.44.2 replaced by MUFA	1	52	Risk Ratio (M‐H, Random, 95% CI)	1.00 [0.53, 1.89]
1.44.3 replace by CHO	5	51232	Risk Ratio (M‐H, Random, 95% CI)	0.84 [0.67, 1.06]
1.44.4 replaced by protein	4	51177	Risk Ratio (M‐H, Random, 95% CI)	0.97 [0.91, 1.03]
1.44.5 replacement unclear	1	235	Risk Ratio (M‐H, Random, 95% CI)	1.68 [0.41, 6.87]
1.45 CVD events, subgroup by main substitution Show forest plot	13		Risk Ratio (M‐H, Random, 95% CI)	Subtotals only

1.45.1 replaced by PUFA	7	4298	Risk Ratio (M‐H, Random, 95% CI)	0.84 [0.66, 1.06]
1.45.2 replaced by MUFA	1	52	Risk Ratio (M‐H, Random, 95% CI)	1.00 [0.53, 1.89]
1.45.3 replace by CHO	4	49199	Risk Ratio (M‐H, Random, 95% CI)	0.67 [0.39, 1.16]
1.45.4 replaced by protein	0	0	Risk Ratio (M‐H, Random, 95% CI)	Not estimable
1.45.5 replacement unclear	1	235	Risk Ratio (M‐H, Random, 95% CI)	1.68 [0.41, 6.87]
1.46 CVD events, subgroup by duration Show forest plot	13	53758	Risk Ratio (M‐H, Random, 95% CI)	0.83 [0.70, 0.98]

1.46.1 up to 24mo	5	2481	Risk Ratio (M‐H, Random, 95% CI)	0.96 [0.78, 1.16]
1.46.2 >24 to 48mo	3	1294	Risk Ratio (M‐H, Random, 95% CI)	0.73 [0.56, 0.95]
1.46.3 >48mo	3	49705	Risk Ratio (M‐H, Random, 95% CI)	0.97 [0.72, 1.33]
1.46.4 unclear duration	2	278	Risk Ratio (M‐H, Random, 95% CI)	0.43 [0.17, 1.08]
1.47 CVD events, subgroup by baseline SFA Show forest plot	13	53758	Risk Ratio (M‐H, Random, 95% CI)	0.83 [0.70, 0.98]

1.47.1 up to 12%E SFA baseline	0	0	Risk Ratio (M‐H, Random, 95% CI)	Not estimable
1.47.2 >12 to 15%E SFA baseline	6	51870	Risk Ratio (M‐H, Random, 95% CI)	0.99 [0.85, 1.15]
1.47.3 >15 to 18%E SFA baseline	1	55	Risk Ratio (M‐H, Random, 95% CI)	0.41 [0.22, 0.78]
1.47.4 >18%E SFA baseline	1	846	Risk Ratio (M‐H, Random, 95% CI)	0.79 [0.63, 1.00]
1.47.5 unclear	5	987	Risk Ratio (M‐H, Random, 95% CI)	0.72 [0.51, 1.03]
1.48 CVD events, subgroup by SFA change Show forest plot	13	53758	Risk Ratio (M‐H, Random, 95% CI)	0.83 [0.70, 0.98]

1.48.1 up to 4%E difference	5	51737	Risk Ratio (M‐H, Random, 95% CI)	1.00 [0.86, 1.16]
1.48.2 >4 to 8%E difference	2	188	Risk Ratio (M‐H, Random, 95% CI)	0.40 [0.22, 0.74]
1.48.3 >8%E difference	1	846	Risk Ratio (M‐H, Random, 95% CI)	0.79 [0.63, 1.00]
1.48.4 unclear	5	987	Risk Ratio (M‐H, Random, 95% CI)	0.72 [0.51, 1.03]
1.49 CVD events, subgroup by sex Show forest plot	13	53758	Risk Ratio (M‐H, Random, 95% CI)	0.83 [0.70, 0.98]

1.49.1 Men	9	4410	Risk Ratio (M‐H, Random, 95% CI)	0.85 [0.71, 1.03]
1.49.2 Women	1	48835	Risk Ratio (M‐H, Random, 95% CI)	0.98 [0.92, 1.04]
1.49.3 Mixed, men and women	3	513	Risk Ratio (M‐H, Random, 95% CI)	0.59 [0.23, 1.49]
1.50 CVD events, subgroup by CVD risk Show forest plot	13	53758	Risk Ratio (M‐H, Random, 95% CI)	0.86 [0.74, 1.00]

1.50.1 Low CVD risk	3	47537	Risk Ratio (M‐H, Random, 95% CI)	0.89 [0.75, 1.06]
1.50.2 Moderate CVD risk	3	513	Risk Ratio (M‐H, Random, 95% CI)	0.59 [0.23, 1.49]
1.50.3 Existing CVD disease	8	5708	Risk Ratio (M‐H, Random, 95% CI)	0.91 [0.75, 1.12]
1.51 CVD events, subgroup by TC reduction Show forest plot	13		Risk Ratio (M‐H, Random, 95% CI)	Subtotals only

1.51.1 serum chol reduced by at least 0.2mmol/L	9	4575	Risk Ratio (M‐H, Random, 95% CI)	0.79 [0.63, 1.00]
1.51.2 serum chol reduced by <0.2mmol/L	3	49050	Risk Ratio (M‐H, Random, 95% CI)	0.98 [0.91, 1.04]
1.51.3 serum chol reduction unclear	1	133	Risk Ratio (M‐H, Random, 95% CI)	0.20 [0.01, 4.15]
1.52 CVD events, subgroup decade of publication Show forest plot	13	53758	Risk Ratio (M‐H, Random, 95% CI)	0.83 [0.70, 0.98]

1.52.1 1960s	5	1731	Risk Ratio (M‐H, Random, 95% CI)	0.79 [0.69, 0.91]
1.52.2 1970s	2	560	Risk Ratio (M‐H, Random, 95% CI)	0.66 [0.12, 3.80]
1.52.3 1980s	1	2033	Risk Ratio (M‐H, Random, 95% CI)	0.92 [0.74, 1.15]
1.52.4 1990s	2	188	Risk Ratio (M‐H, Random, 95% CI)	0.40 [0.22, 0.74]
1.52.5 2000s	3	49246	Risk Ratio (M‐H, Random, 95% CI)	0.98 [0.91, 1.04]

Comparison 1. SFA reduction vs usual diet ‐ primary outcomes

Comparison 2. SFA reduction vs usual diet ‐ secondary health events

Outcome or subgroup title	No. of studies	No. of participants	Statistical method	Effect size
2.1 MYOCARDIAL INFARCTION Show forest plot	11	53167	Risk Ratio (M‐H, Random, 95% CI)	0.90 [0.80, 1.01]

2.2 MI, SA by low summary risk of bias Show forest plot	3	49857	Risk Ratio (M‐H, Random, 95% CI)	0.93 [0.81, 1.08]

2.3 MI, SA aim to reduce SFA Show forest plot	10	52991	Risk Ratio (M‐H, Random, 95% CI)	0.89 [0.78, 1.02]

2.4 MI, SA statistically significant SFA reduction Show forest plot	6	52180	Risk Ratio (M‐H, Random, 95% CI)	0.94 [0.85, 1.04]

2.5 MI, SA by TC reduction Show forest plot	9	52952	Risk Ratio (M‐H, Random, 95% CI)	0.88 [0.77, 1.01]

2.6 MI, SA excluding WHI Show forest plot	10	4332	Risk Ratio (M‐H, Random, 95% CI)	0.85 [0.73, 0.98]

2.7 MI, SA Mantel‐Haenszel fixed‐effect Show forest plot	11	53167	Risk Ratio (M‐H, Fixed, 95% CI)	0.92 [0.84, 1.01]

2.8 MI, SA Peto fixed‐effect Show forest plot	11	53167	Peto Odds Ratio (Peto, Fixed, 95% CI)	0.92 [0.83, 1.01]

2.9 MI, subgroup by any substitution Show forest plot	11		Risk Ratio (M‐H, Random, 95% CI)	Subtotals only

2.9.1 replaced by PUFA	7	3895	Risk Ratio (M‐H, Random, 95% CI)	0.83 [0.67, 1.02]
2.9.2 replaced by MUFA	1	52	Risk Ratio (M‐H, Random, 95% CI)	1.40 [0.51, 3.85]
2.9.3 replace by CHO	4	51099	Risk Ratio (M‐H, Random, 95% CI)	0.96 [0.86, 1.06]
2.9.4 replaced by protein	3	51044	Risk Ratio (M‐H, Random, 95% CI)	0.96 [0.86, 1.07]
2.9.5 replacement unclear	1	235	Risk Ratio (M‐H, Random, 95% CI)	2.02 [0.19, 21.94]
2.10 MI, subgroup by main substitution Show forest plot	11		Risk Ratio (M‐H, Random, 95% CI)	Subtotals only

2.10.1 replaced by PUFA	6	3840	Risk Ratio (M‐H, Random, 95% CI)	0.83 [0.67, 1.04]
2.10.2 replaced by MUFA	1	52	Risk Ratio (M‐H, Random, 95% CI)	1.40 [0.51, 3.85]
2.10.3 replace by CHO	3	49066	Risk Ratio (M‐H, Random, 95% CI)	0.97 [0.86, 1.09]
2.10.4 replaced by protein	0	0	Risk Ratio (M‐H, Random, 95% CI)	Not estimable
2.10.5 replacement unclear	1	235	Risk Ratio (M‐H, Random, 95% CI)	2.02 [0.19, 21.94]
2.11 MI, subgroup by duration Show forest plot	11	53167	Risk Ratio (M‐H, Random, 95% CI)	0.90 [0.80, 1.01]

2.11.1 up to 24mo	4	2348	Risk Ratio (M‐H, Random, 95% CI)	0.95 [0.77, 1.17]
2.11.2 >24 to 48mo	3	1294	Risk Ratio (M‐H, Random, 95% CI)	0.83 [0.64, 1.06]
2.11.3 >48mo	2	49247	Risk Ratio (M‐H, Random, 95% CI)	0.81 [0.54, 1.24]
2.11.4 unclear	2	278	Risk Ratio (M‐H, Random, 95% CI)	0.41 [0.02, 7.73]
2.12 MI, subgroup by baseline SFA Show forest plot	11	53167	Risk Ratio (M‐H, Random, 95% CI)	0.90 [0.80, 1.01]

2.12.1 up to 12%E SFA baseline	0	0	Risk Ratio (M‐H, Random, 95% CI)	Not estimable
2.12.2 >12 to 15%E SFA baseline	4	51279	Risk Ratio (M‐H, Random, 95% CI)	0.96 [0.87, 1.07]
2.12.3 >15 to 18%E SFA baseline	1	55	Risk Ratio (M‐H, Random, 95% CI)	0.52 [0.05, 5.39]
2.12.4 >18%E SFA baseline	1	846	Risk Ratio (M‐H, Random, 95% CI)	0.76 [0.55, 1.05]
2.12.5 unclear	5	987	Risk Ratio (M‐H, Random, 95% CI)	0.84 [0.54, 1.30]
2.13 MI, subgroup by SFA change Show forest plot	11	53167	Risk Ratio (M‐H, Random, 95% CI)	0.90 [0.80, 1.01]

2.13.1 up to 4%E difference	4	51279	Risk Ratio (M‐H, Random, 95% CI)	0.96 [0.87, 1.07]
2.13.2 >4 to 8%E difference	1	55	Risk Ratio (M‐H, Random, 95% CI)	0.52 [0.05, 5.39]
2.13.3 >8%E difference	1	846	Risk Ratio (M‐H, Random, 95% CI)	0.76 [0.55, 1.05]
2.13.4 unclear	5	987	Risk Ratio (M‐H, Random, 95% CI)	0.84 [0.54, 1.30]
2.14 MI, subgroup by sex Show forest plot	11	53167	Risk Ratio (M‐H, Random, 95% CI)	0.90 [0.80, 1.01]

2.14.1 Men	7	3819	Risk Ratio (M‐H, Random, 95% CI)	0.85 [0.73, 0.98]
2.14.2 Women	1	48835	Risk Ratio (M‐H, Random, 95% CI)	0.97 [0.86, 1.09]
2.14.3 Mixed, men and women	3	513	Risk Ratio (M‐H, Random, 95% CI)	0.75 [0.13, 4.47]
2.15 MI, subgroup by CVD risk Show forest plot	11	53167	Risk Ratio (M‐H, Random, 95% CI)	0.90 [0.80, 1.01]

2.15.1 Low CVD risk	2	49681	Risk Ratio (M‐H, Random, 95% CI)	0.90 [0.72, 1.13]
2.15.2 Moderate CVD risk	3	513	Risk Ratio (M‐H, Random, 95% CI)	0.75 [0.13, 4.47]
2.15.3 Existing CVD disease	6	2973	Risk Ratio (M‐H, Random, 95% CI)	0.87 [0.74, 1.03]
2.16 MI, subgroup by TC reduction Show forest plot	11		Risk Ratio (M‐H, Random, 95% CI)	Subtotals only

2.16.1 serum chol reduced by at least 0.2mmol/L	8	4117	Risk Ratio (M‐H, Random, 95% CI)	0.83 [0.70, 0.98]
2.16.2 serum chol reduced by <0.2mmol/L	3	49050	Risk Ratio (M‐H, Random, 95% CI)	0.98 [0.87, 1.10]
2.16.3 serum chol reduction unclear	0	0	Risk Ratio (M‐H, Random, 95% CI)	Not estimable
2.17 MI, subgroup decade of publication Show forest plot	11	53167	Risk Ratio (M‐H, Random, 95% CI)	0.90 [0.80, 1.01]

2.17.1 1960s	5	1731	Risk Ratio (M‐H, Random, 95% CI)	0.80 [0.64, 1.00]
2.17.2 1970s	1	102	Risk Ratio (M‐H, Random, 95% CI)	0.08 [0.00, 1.33]
2.17.3 1980s	1	2033	Risk Ratio (M‐H, Random, 95% CI)	0.91 [0.73, 1.14]
2.17.4 1990s	1	55	Risk Ratio (M‐H, Random, 95% CI)	0.52 [0.05, 5.39]
2.17.5 2000s	3	49246	Risk Ratio (M‐H, Random, 95% CI)	0.98 [0.87, 1.10]
2.18 NON‐FATAL MYOCARDIAL INFARCTION Show forest plot	8	52834	Risk Ratio (M‐H, Random, 95% CI)	0.97 [0.87, 1.07]

2.19 Non‐fatal MI, SA by low summary risk of bias Show forest plot	2	49681	Risk Ratio (M‐H, Random, 95% CI)	0.89 [0.58, 1.35]

2.20 Non‐fatal MI, SA aim to reduce SFA Show forest plot	8	52834	Risk Ratio (M‐H, Random, 95% CI)	0.97 [0.87, 1.07]

2.21 Non‐fatal MI, SA statistically significant SFA reduction Show forest plot	4	51949	Risk Ratio (M‐H, Random, 95% CI)	0.90 [0.72, 1.14]

2.22 Non‐fatal MI, SA by TC reduction Show forest plot	7	52795	Risk Ratio (M‐H, Random, 95% CI)	0.97 [0.87, 1.07]

2.23 Non‐fatal MI, SA excluding WHI Show forest plot	7	3999	Risk Ratio (M‐H, Random, 95% CI)	0.81 [0.64, 1.04]

2.24 Non‐fatal MI, SA Mantel‐Haenszel fixed‐effect Show forest plot	8	52834	Risk Ratio (M‐H, Fixed, 95% CI)	0.97 [0.87, 1.08]

2.25 Non‐fatal MI, SA Peto fixed‐effect Show forest plot	8	52834	Peto Odds Ratio (Peto, Fixed, 95% CI)	0.97 [0.87, 1.08]

2.26 Non‐fatal MI, subgroup by any substitution Show forest plot	8		Risk Ratio (M‐H, Random, 95% CI)	Subtotals only

2.26.1 replaced by PUFA	5	3738	Risk Ratio (M‐H, Random, 95% CI)	0.80 [0.63, 1.03]
2.26.2 replaced by MUFA	1	52	Risk Ratio (M‐H, Random, 95% CI)	1.20 [0.42, 3.45]
2.26.3 replace by CHO	2	50868	Risk Ratio (M‐H, Random, 95% CI)	0.93 [0.72, 1.21]
2.26.4 replaced by protein	2	50868	Risk Ratio (M‐H, Random, 95% CI)	0.93 [0.72, 1.21]
2.26.5 replacement unclear	1	235	Risk Ratio (M‐H, Random, 95% CI)	2.02 [0.19, 21.94]
2.27 Non‐fatal MI, subgroup by main substitution Show forest plot	8		Risk Ratio (M‐H, Random, 95% CI)	Subtotals only

2.27.1 replaced by PUFA	5	3738	Risk Ratio (M‐H, Random, 95% CI)	0.80 [0.63, 1.03]
2.27.2 replaced by MUFA	1	52	Risk Ratio (M‐H, Random, 95% CI)	1.20 [0.42, 3.45]
2.27.3 replace by CHO	1	48835	Risk Ratio (M‐H, Random, 95% CI)	1.01 [0.90, 1.13]
2.27.4 replaced by protein	0	0	Risk Ratio (M‐H, Random, 95% CI)	Not estimable
2.27.5 replacement unclear	1	235	Risk Ratio (M‐H, Random, 95% CI)	2.02 [0.19, 21.94]
2.28 Non‐fatal MI, subgroup by duration Show forest plot	8	52834	Risk Ratio (M‐H, Random, 95% CI)	0.97 [0.87, 1.07]

2.28.1 up to 24mo	4	2348	Risk Ratio (M‐H, Random, 95% CI)	0.83 [0.57, 1.22]
2.28.2 >24 to 48mo	2	1239	Risk Ratio (M‐H, Random, 95% CI)	0.82 [0.53, 1.27]
2.28.3 >48mo	2	49247	Risk Ratio (M‐H, Random, 95% CI)	0.99 [0.88, 1.12]
2.28.4 unclear	0	0	Risk Ratio (M‐H, Random, 95% CI)	Not estimable
2.29 Non‐fatal MI, subgroup by baseline SFA Show forest plot	8	52834	Risk Ratio (M‐H, Random, 95% CI)	0.97 [0.87, 1.07]

2.29.1 up to 12%E SFA baseline	0	0	Risk Ratio (M‐H, Random, 95% CI)	Not estimable
2.29.2 >12 to 15%E SFA baseline	3	51103	Risk Ratio (M‐H, Random, 95% CI)	0.97 [0.83, 1.13]
2.29.3 >15 to 18%E SFA baseline	0	0	Risk Ratio (M‐H, Random, 95% CI)	Not estimable
2.29.4 >18%E SFA baseline	1	846	Risk Ratio (M‐H, Random, 95% CI)	0.62 [0.31, 1.21]
2.29.5 unclear	4	885	Risk Ratio (M‐H, Random, 95% CI)	0.91 [0.65, 1.27]
2.30 Non‐fatal MI, subgroup by SFA change Show forest plot	8	52834	Risk Ratio (M‐H, Random, 95% CI)	0.97 [0.87, 1.07]

2.30.1 up to 4%E difference	3	51103	Risk Ratio (M‐H, Random, 95% CI)	0.97 [0.83, 1.13]
2.30.2 >4 to 8%E difference	0	0	Risk Ratio (M‐H, Random, 95% CI)	Not estimable
2.30.3 >8%E difference	1	846	Risk Ratio (M‐H, Random, 95% CI)	0.62 [0.31, 1.21]
2.30.4 unclear	4	885	Risk Ratio (M‐H, Random, 95% CI)	0.91 [0.65, 1.27]
2.31 Non‐fatal MI, subgroup by sex Show forest plot	8	52834	Risk Ratio (M‐H, Random, 95% CI)	0.97 [0.87, 1.07]

2.31.1 Men	6	3764	Risk Ratio (M‐H, Random, 95% CI)	0.81 [0.63, 1.03]
2.31.2 Women	1	48835	Risk Ratio (M‐H, Random, 95% CI)	1.01 [0.90, 1.13]
2.31.3 Mixed, men and women	1	235	Risk Ratio (M‐H, Random, 95% CI)	2.02 [0.19, 21.94]
2.32 Non‐fatal MI, subgroup by CVD risk Show forest plot	8	52834	Risk Ratio (M‐H, Random, 95% CI)	0.95 [0.80, 1.13]

2.32.1 Low CVD risk	2	47404	Risk Ratio (M‐H, Random, 95% CI)	0.87 [0.68, 1.12]
2.32.2 Moderate CVD risk	1	235	Risk Ratio (M‐H, Random, 95% CI)	2.02 [0.19, 21.94]
2.32.3 Existing CVD disease	6	5195	Risk Ratio (M‐H, Random, 95% CI)	1.00 [0.76, 1.31]
2.33 Non‐fatal MI, subgroup by TC reduction Show forest plot	8		Risk Ratio (M‐H, Random, 95% CI)	Subtotals only

2.33.1 serum chol reduced by at least 0.2mmol/L	6	3960	Risk Ratio (M‐H, Random, 95% CI)	0.80 [0.62, 1.03]
2.33.2 serum chol reduced by <0.2mmol/L	2	48874	Risk Ratio (M‐H, Random, 95% CI)	1.01 [0.90, 1.13]
2.33.3 serum chol reduction unclear	0	0	Risk Ratio (M‐H, Random, 95% CI)	Not estimable
2.34 Non‐fatal MI, subgroup decade of publication Show forest plot	8	52834	Risk Ratio (M‐H, Random, 95% CI)	0.97 [0.87, 1.07]

2.34.1 1960s	5	1731	Risk Ratio (M‐H, Random, 95% CI)	0.84 [0.62, 1.13]
2.34.2 1970s	0	0	Risk Ratio (M‐H, Random, 95% CI)	Not estimable
2.34.3 1980s	1	2033	Risk Ratio (M‐H, Random, 95% CI)	0.74 [0.48, 1.14]
2.34.4 1990s	0	0	Risk Ratio (M‐H, Random, 95% CI)	Not estimable
2.34.5 2000s	2	49070	Risk Ratio (M‐H, Random, 95% CI)	1.01 [0.90, 1.13]
2.35 STROKE Show forest plot	7	50952	Risk Ratio (M‐H, Random, 95% CI)	0.92 [0.68, 1.25]

2.36 Stroke, SA by low summary risk of bias Show forest plot	3	49857	Risk Ratio (M‐H, Random, 95% CI)	0.76 [0.42, 1.38]

2.37 Stroke, SA aim to reduce SFA Show forest plot	6	50776	Risk Ratio (M‐H, Random, 95% CI)	1.01 [0.90, 1.14]

2.38 Stroke, SA statistically significant SFA reduction Show forest plot	5	50147	Risk Ratio (M‐H, Random, 95% CI)	0.83 [0.55, 1.25]

2.39 Stroke, SA by TC reduction Show forest plot	6	50776	Risk Ratio (M‐H, Random, 95% CI)	1.01 [0.90, 1.14]

2.40 Stroke, SA excluding WHI Show forest plot	6	2117	Risk Ratio (M‐H, Random, 95% CI)	0.63 [0.35, 1.14]

2.41 Stroke, SA Mantel‐Haenszel fixed‐effect Show forest plot	7	50952	Risk Ratio (M‐H, Fixed, 95% CI)	1.01 [0.89, 1.13]

2.42 Stroke, SA Peto fixed‐effect Show forest plot	7	50952	Peto Odds Ratio (Peto, Fixed, 95% CI)	1.01 [0.89, 1.14]

2.43 Stroke, subgroup by any substitution Show forest plot	7		Risk Ratio (M‐H, Random, 95% CI)	Subtotals only

2.43.1 replaced by PUFA	4	1706	Risk Ratio (M‐H, Random, 95% CI)	0.68 [0.37, 1.27]
2.43.2 replaced by MUFA	0	0	Risk Ratio (M‐H, Random, 95% CI)	Not estimable
2.43.3 replace by CHO	3	49066	Risk Ratio (M‐H, Random, 95% CI)	0.73 [0.29, 1.87]
2.43.4 replaced by protein	2	49011	Risk Ratio (M‐H, Random, 95% CI)	0.65 [0.15, 2.75]
2.43.5 replacement unclear	1	235	Risk Ratio (M‐H, Random, 95% CI)	1.01 [0.06, 15.93]
2.44 Stroke, subgroup by main substitution Show forest plot	7		Risk Ratio (M‐H, Random, 95% CI)	Subtotals only

2.44.1 replaced by PUFA	3	1651	Risk Ratio (M‐H, Random, 95% CI)	0.92 [0.31, 2.69]
2.44.2 replaced by MUFA	0	0	Risk Ratio (M‐H, Random, 95% CI)	Not estimable
2.44.3 replace by CHO	3	49066	Risk Ratio (M‐H, Random, 95% CI)	0.73 [0.29, 1.87]
2.44.4 replaced by protein	0	0	Risk Ratio (M‐H, Random, 95% CI)	Not estimable
2.44.5 replacement unclear	1	235	Risk Ratio (M‐H, Random, 95% CI)	1.01 [0.06, 15.93]
2.45 Stroke, subgroup by duration Show forest plot	6	50559	Risk Ratio (M‐H, Random, 95% CI)	0.91 [0.67, 1.23]

2.45.1 up to 24mo	1	235	Risk Ratio (M‐H, Random, 95% CI)	1.01 [0.06, 15.93]
2.45.2 >24 to 48mo	2	901	Risk Ratio (M‐H, Random, 95% CI)	0.57 [0.30, 1.11]
2.45.3 >48mo	2	49247	Risk Ratio (M‐H, Random, 95% CI)	1.03 [0.91, 1.16]
2.45.4 unclear duration	1	176	Risk Ratio (M‐H, Random, 95% CI)	0.20 [0.02, 1.68]
2.46 Stroke, subgroup by baseline SFA Show forest plot	6	50559	Risk Ratio (M‐H, Random, 95% CI)	0.91 [0.67, 1.23]

2.46.1 up to 12%E SFA baseline	0	0	Risk Ratio (M‐H, Random, 95% CI)	Not estimable
2.46.2 >12 to 15%E SFA baseline	3	49246	Risk Ratio (M‐H, Random, 95% CI)	0.91 [0.50, 1.66]
2.46.3 >15 to 18%E SFA baseline	1	55	Risk Ratio (M‐H, Random, 95% CI)	0.35 [0.01, 8.12]
2.46.4 >18%E SFA baseline	1	846	Risk Ratio (M‐H, Random, 95% CI)	0.59 [0.30, 1.15]
2.46.5 unclear	1	412	Risk Ratio (M‐H, Random, 95% CI)	2.00 [0.18, 21.89]
2.47 Stroke, subgroup by SFA change Show forest plot	6	50559	Risk Ratio (M‐H, Random, 95% CI)	0.91 [0.67, 1.23]

2.47.1 up to 4%E difference	3	49246	Risk Ratio (M‐H, Random, 95% CI)	0.91 [0.50, 1.66]
2.47.2 >4 to 8%E difference	1	55	Risk Ratio (M‐H, Random, 95% CI)	0.35 [0.01, 8.12]
2.47.3 >8%E difference	1	846	Risk Ratio (M‐H, Random, 95% CI)	0.59 [0.30, 1.15]
2.47.4 unclear	1	412	Risk Ratio (M‐H, Random, 95% CI)	2.00 [0.18, 21.89]
2.48 Stroke, subgroup by sex Show forest plot	6	50559	Risk Ratio (M‐H, Random, 95% CI)	0.91 [0.67, 1.23]

2.48.1 Men	3	1313	Risk Ratio (M‐H, Random, 95% CI)	0.63 [0.33, 1.18]
2.48.2 Women	1	48835	Risk Ratio (M‐H, Random, 95% CI)	1.03 [0.91, 1.16]
2.48.3 Mixed, men and women	2	411	Risk Ratio (M‐H, Random, 95% CI)	0.37 [0.07, 1.97]
2.49 Stroke, subgroup by CVD risk Show forest plot	6	50559	Risk Ratio (M‐H, Random, 95% CI)	1.00 [0.89, 1.11]

2.49.1 Low CVD risk	2	47404	Risk Ratio (M‐H, Random, 95% CI)	0.86 [0.52, 1.42]
2.49.2 Moderate CVD risk	2	411	Risk Ratio (M‐H, Random, 95% CI)	0.37 [0.07, 1.97]
2.49.3 Existing CVD disease	3	2744	Risk Ratio (M‐H, Random, 95% CI)	1.01 [0.86, 1.18]
2.50 Stroke, subgroup by TC reduction Show forest plot	7		Risk Ratio (M‐H, Random, 95% CI)	Subtotals only

2.50.1 serum chol reduced by at least 0.2mmol/L	5	1941	Risk Ratio (M‐H, Random, 95% CI)	0.70 [0.38, 1.28]
2.50.2 serum chol reduced by <0.2mmol/L	2	49011	Risk Ratio (M‐H, Random, 95% CI)	0.65 [0.15, 2.75]
2.50.3 serum chol reduction unclear	0	0	Risk Ratio (M‐H, Random, 95% CI)	Not estimable
2.51 Stroke, subgroup decade of publication Show forest plot	7	50952	Risk Ratio (M‐H, Random, 95% CI)	0.92 [0.68, 1.25]

2.51.1 1960s	3	1651	Risk Ratio (M‐H, Random, 95% CI)	0.92 [0.31, 2.69]
2.51.2 1970s	0	0	Risk Ratio (M‐H, Random, 95% CI)	Not estimable
2.51.3 1980s	0	0	Risk Ratio (M‐H, Random, 95% CI)	Not estimable
2.51.4 1990s	1	55	Risk Ratio (M‐H, Random, 95% CI)	0.35 [0.01, 8.12]
2.51.5 2000s	3	49246	Risk Ratio (M‐H, Random, 95% CI)	0.91 [0.50, 1.66]
2.52 CORONARY HEART DISEASE MORTALITY Show forest plot	9	53159	Risk Ratio (M‐H, Random, 95% CI)	0.97 [0.82, 1.16]

2.53 CHD mortality, SA by low summary risk of bias Show forest plot	3	50139	Risk Ratio (M‐H, Random, 95% CI)	1.05 [0.77, 1.43]

2.54 CHD mortality, SA aim to reduce SFA Show forest plot	9	53159	Risk Ratio (M‐H, Random, 95% CI)	0.97 [0.82, 1.16]

2.55 CHD mortality, SA statistically significant SFA reduction Show forest plot	4	52172	Risk Ratio (M‐H, Random, 95% CI)	1.02 [0.84, 1.24]

2.56 CHD mortality, SA by TC reduction Show forest plot	8	53120	Risk Ratio (M‐H, Random, 95% CI)	0.97 [0.81, 1.16]

2.57 CHD mortality, SA excluding WHI Show forest plot	8	4324	Risk Ratio (M‐H, Random, 95% CI)	0.97 [0.76, 1.24]

2.58 CHD mortality, SA Mantel‐Haenszel fixed‐effect Show forest plot	9	53159	Risk Ratio (M‐H, Fixed, 95% CI)	0.97 [0.86, 1.10]

2.59 CHD mortality, SA Peto fixed‐effect Show forest plot	9	53159	Peto Odds Ratio (Peto, Fixed, 95% CI)	0.97 [0.85, 1.11]

2.60 CHD mortality, subgroup by any substitution Show forest plot	9		Risk Ratio (M‐H, Random, 95% CI)	Subtotals only

2.60.1 replaced by PUFA	7	4298	Risk Ratio (M‐H, Random, 95% CI)	0.98 [0.74, 1.28]
2.60.2 replaced by MUFA	1	52	Risk Ratio (M‐H, Random, 95% CI)	3.00 [0.33, 26.99]
2.60.3 replaced by CHO	2	50868	Risk Ratio (M‐H, Random, 95% CI)	0.99 [0.85, 1.16]
2.60.4 replaced by protein	2	50868	Risk Ratio (M‐H, Random, 95% CI)	0.99 [0.85, 1.16]
2.60.5 replacement unclear	0	0	Risk Ratio (M‐H, Random, 95% CI)	Not estimable
2.61 CHD mortality, subgroup by main substitution Show forest plot	9		Risk Ratio (M‐H, Random, 95% CI)	Subtotals only

2.61.1 replaced by PUFA	7	4298	Risk Ratio (M‐H, Random, 95% CI)	0.98 [0.74, 1.28]
2.61.2 replaced by MUFA	1	52	Risk Ratio (M‐H, Random, 95% CI)	3.00 [0.33, 26.99]
2.61.3 replaced by CHO	1	48835	Risk Ratio (M‐H, Random, 95% CI)	0.99 [0.82, 1.20]
2.61.4 replaced by protein	0	0	Risk Ratio (M‐H, Random, 95% CI)	Not estimable
2.61.5 replacement unclear	0	0	Risk Ratio (M‐H, Random, 95% CI)	Not estimable
2.62 CHD mortality, subgroup by duration Show forest plot	9		Risk Ratio (M‐H, Random, 95% CI)	Subtotals only

2.62.1 up to 24mo	3	2113	Risk Ratio (M‐H, Random, 95% CI)	1.02 [0.78, 1.33]
2.62.2 >24 to 48months	2	1239	Risk Ratio (M‐H, Random, 95% CI)	0.87 [0.64, 1.19]
2.62.3 >48 months	3	49705	Risk Ratio (M‐H, Random, 95% CI)	1.02 [0.72, 1.45]
2.62.4 unclear duration	1	102	Risk Ratio (M‐H, Random, 95% CI)	0.09 [0.01, 1.60]
2.63 CHD mortality, subgroup by baseline SFA Show forest plot	9		Risk Ratio (M‐H, Random, 95% CI)	Subtotals only

2.63.1 up to 12%E SFA baseline	0	0	Risk Ratio (M‐H, Random, 95% CI)	Not estimable
2.63.2 >12% to 15%E SFA baseline	3	51326	Risk Ratio (M‐H, Random, 95% CI)	1.07 [0.86, 1.34]
2.63.3 >15 to 18%E SFA baseline	0	0	Risk Ratio (M‐H, Random, 95% CI)	Not estimable
2.63.4 >18%E SFA baseline	1	846	Risk Ratio (M‐H, Random, 95% CI)	0.82 [0.55, 1.21]
2.63.5 unclear	5	987	Risk Ratio (M‐H, Random, 95% CI)	0.85 [0.56, 1.29]
2.64 CHD mortality, subgroup by SFA change Show forest plot	9		Risk Ratio (M‐H, Random, 95% CI)	Subtotals only

2.64.1 up to 4%E difference	3	51326	Risk Ratio (M‐H, Random, 95% CI)	1.07 [0.86, 1.34]
2.64.2 >4 to 8%E difference	0	0	Risk Ratio (M‐H, Random, 95% CI)	Not estimable
2.64.3 >8%E difference	1	846	Risk Ratio (M‐H, Random, 95% CI)	0.82 [0.55, 1.21]
2.64.4 unclear	5	987	Risk Ratio (M‐H, Random, 95% CI)	0.85 [0.56, 1.29]
2.65 CHD mortality, subgroup by sex Show forest plot	9		Risk Ratio (M‐H, Random, 95% CI)	Subtotals only

2.65.1 Men	7	4222	Risk Ratio (M‐H, Random, 95% CI)	0.98 [0.79, 1.23]
2.65.2 Women	1	48835	Risk Ratio (M‐H, Random, 95% CI)	0.99 [0.82, 1.20]
2.65.3 Mixed, men and women	1	102	Risk Ratio (M‐H, Random, 95% CI)	0.09 [0.01, 1.60]
2.66 CHD mortality, subgroup by CVD risk Show forest plot	9		Risk Ratio (M‐H, Random, 95% CI)	Subtotals only

2.66.1 Low CVD risk	2	47404	Risk Ratio (M‐H, Random, 95% CI)	0.95 [0.78, 1.16]
2.66.2 Moderate CVD risk	1	102	Risk Ratio (M‐H, Random, 95% CI)	0.09 [0.01, 1.60]
2.66.3 Existing CVD disease	7	5653	Risk Ratio (M‐H, Random, 95% CI)	1.03 [0.83, 1.27]
2.67 CHD mortality, subgroup by TC reduction Show forest plot	9		Risk Ratio (M‐H, Random, 95% CI)	Subtotals only

2.67.1 serum chol reduced by at least 0.2mmol/L	7	4285	Risk Ratio (M‐H, Random, 95% CI)	0.96 [0.75, 1.24]
2.67.2 serum chol reduced by <0.2mmol/L	2	48874	Risk Ratio (M‐H, Random, 95% CI)	0.99 [0.82, 1.20]
2.67.3 serum chol reduction unclear	0	0	Risk Ratio (M‐H, Random, 95% CI)	Not estimable
2.68 CHD mortality, subgroup decade of publication Show forest plot	9	53159	Risk Ratio (M‐H, Random, 95% CI)	0.97 [0.82, 1.16]

2.68.1 1960s	5	1731	Risk Ratio (M‐H, Random, 95% CI)	0.84 [0.66, 1.06]
2.68.2 1970s	2	560	Risk Ratio (M‐H, Random, 95% CI)	0.54 [0.03, 9.26]
2.68.3 1980s	1	2033	Risk Ratio (M‐H, Random, 95% CI)	1.00 [0.76, 1.30]
2.68.4 1990s	0	0	Risk Ratio (M‐H, Random, 95% CI)	Not estimable
2.68.5 2000s	1	48835	Risk Ratio (M‐H, Random, 95% CI)	0.99 [0.82, 1.20]
2.69 CORONARY HEART DISEASE EVENTS Show forest plot	11	53199	Risk Ratio (M‐H, Random, 95% CI)	0.83 [0.68, 1.01]

2.70 CHD events, SA by low summary risk of bias Show forest plot	3	49857	Risk Ratio (M‐H, Random, 95% CI)	0.92 [0.77, 1.10]

2.71 CHD events, SA excluding WHI Show forest plot	10	4364	Risk Ratio (M‐H, Random, 95% CI)	0.80 [0.62, 1.03]

2.72 CHD events, SA statistically significant SFA reduction Show forest plot	6	52212	Risk Ratio (M‐H, Random, 95% CI)	0.91 [0.77, 1.06]

2.73 CHD events, SA by TC reduction Show forest plot	9	52984	Risk Ratio (M‐H, Random, 95% CI)	0.80 [0.65, 0.99]

2.74 CHD events, SA aim to reduce SFA Show forest plot	10	53023	Risk Ratio (M‐H, Random, 95% CI)	0.82 [0.67, 1.00]

2.75 CHD events, SA Mantel‐Haenszel fixed‐effect Show forest plot	11	53199	Risk Ratio (M‐H, Fixed, 95% CI)	0.91 [0.84, 0.99]

2.76 CHD events, SA Peto fixed‐effect Show forest plot	11	53199	Peto Odds Ratio (Peto, Fixed, 95% CI)	0.90 [0.83, 0.99]

2.77 CHD events, subgroup by any substitution Show forest plot	11		Risk Ratio (M‐H, Random, 95% CI)	Subtotals only

2.77.1 replaced by PUFA	7	3895	Risk Ratio (M‐H, Random, 95% CI)	0.76 [0.57, 1.00]
2.77.2 replaced by MUFA	1	52	Risk Ratio (M‐H, Random, 95% CI)	1.50 [0.62, 3.61]
2.77.3 replaced by CHO	4	51104	Risk Ratio (M‐H, Random, 95% CI)	0.93 [0.78, 1.11]
2.77.4 replaced by protein	3	51044	Risk Ratio (M‐H, Random, 95% CI)	0.96 [0.88, 1.05]
2.77.5 replacement unclear	1	267	Risk Ratio (M‐H, Random, 95% CI)	2.93 [0.31, 27.84]
2.78 CHD events, subgroup by main substitution Show forest plot	11		Risk Ratio (M‐H, Random, 95% CI)	Subtotals only

2.78.1 replaced by PUFA	6	3840	Risk Ratio (M‐H, Random, 95% CI)	0.79 [0.60, 1.04]
2.78.2 replaced by MUFA	1	52	Risk Ratio (M‐H, Random, 95% CI)	1.50 [0.62, 3.61]
2.78.3 replaced by CHO	3	49071	Risk Ratio (M‐H, Random, 95% CI)	0.82 [0.39, 1.72]
2.78.4 replaced by protein	0	0	Risk Ratio (M‐H, Random, 95% CI)	Not estimable
2.78.5 replacement unclear	1	267	Risk Ratio (M‐H, Random, 95% CI)	2.93 [0.31, 27.84]
2.79 CHD events, subgroup by duration Show forest plot	11		Risk Ratio (M‐H, Random, 95% CI)	Subtotals only

2.79.1 up to 24 months	4	2380	Risk Ratio (M‐H, Random, 95% CI)	1.01 [0.76, 1.35]
2.79.2 >24 to 48 months	3	1294	Risk Ratio (M‐H, Random, 95% CI)	0.79 [0.55, 1.13]
2.79.3 >48 months	2	49247	Risk Ratio (M‐H, Random, 95% CI)	0.85 [0.63, 1.15]
2.79.4 unclear duration	2	278	Risk Ratio (M‐H, Random, 95% CI)	0.60 [0.10, 3.58]
2.80 CHD events, subgroup by baseline SFA Show forest plot	11		Risk Ratio (M‐H, Random, 95% CI)	Subtotals only

2.80.1 up to 12%E SFA baseline	0	0	Risk Ratio (M‐H, Random, 95% CI)	Not estimable
2.80.2 >12 to 15%E SFA baseline	4	51311	Risk Ratio (M‐H, Random, 95% CI)	0.96 [0.88, 1.06]
2.80.3 >15 to 18%E SFA baseline	1	55	Risk Ratio (M‐H, Random, 95% CI)	0.31 [0.10, 1.01]
2.80.4 >18%E SFA baseline	1	846	Risk Ratio (M‐H, Random, 95% CI)	0.77 [0.56, 1.04]
2.80.5 unclear	5	987	Risk Ratio (M‐H, Random, 95% CI)	0.78 [0.49, 1.26]
2.81 CHD events, subgroup by SFA change Show forest plot	11		Risk Ratio (M‐H, Random, 95% CI)	Subtotals only

2.81.1 up to 4%E difference	4	51311	Risk Ratio (M‐H, Random, 95% CI)	0.96 [0.88, 1.06]
2.81.2 >4 to 8%E difference	1	55	Risk Ratio (M‐H, Random, 95% CI)	0.31 [0.10, 1.01]
2.81.3 >8%E difference	1	846	Risk Ratio (M‐H, Random, 95% CI)	0.77 [0.56, 1.04]
2.81.4 unclear	5	987	Risk Ratio (M‐H, Random, 95% CI)	0.78 [0.49, 1.26]
2.82 CHD events, subgroup by sex Show forest plot	11		Risk Ratio (M‐H, Random, 95% CI)	Subtotals only

2.82.1 Men	7	3819	Risk Ratio (M‐H, Random, 95% CI)	0.84 [0.70, 1.02]
2.82.2 Women	1	48835	Risk Ratio (M‐H, Random, 95% CI)	0.97 [0.87, 1.07]
2.82.3 Mixed, men and women	3	545	Risk Ratio (M‐H, Random, 95% CI)	0.88 [0.18, 4.36]
2.83 CHD events, subgroup by CVD risk Show forest plot	11		Risk Ratio (M‐H, Random, 95% CI)	Subtotals only

2.83.1 Low CVD risk	2	47404	Risk Ratio (M‐H, Random, 95% CI)	0.90 [0.76, 1.05]
2.83.2 Moderate CVD risk	3	545	Risk Ratio (M‐H, Random, 95% CI)	0.88 [0.18, 4.36]
2.83.3 Existing CVD disease	7	5250	Risk Ratio (M‐H, Random, 95% CI)	0.94 [0.75, 1.16]
2.84 CHD events, subgroup by TC reduction Show forest plot	11		Risk Ratio (M‐H, Random, 95% CI)	Subtotals only

2.84.1 serum chol reduced by at least 0.2mmol/L	8	4149	Risk Ratio (M‐H, Random, 95% CI)	0.76 [0.58, 0.99]
2.84.2 serum chol reduced by <0.2mmol/L	3	49050	Risk Ratio (M‐H, Random, 95% CI)	0.97 [0.88, 1.08]
2.84.3 serum chol reduction unclear	0	0	Risk Ratio (M‐H, Random, 95% CI)	Not estimable
2.85 CHD events, subgroup decade of publication Show forest plot	11	53201	Risk Ratio (M‐H, Random, 95% CI)	0.83 [0.68, 1.01]

2.85.1 1960s	5	1731	Risk Ratio (M‐H, Random, 95% CI)	0.84 [0.68, 1.05]
2.85.2 1970s	1	102	Risk Ratio (M‐H, Random, 95% CI)	0.27 [0.14, 0.52]
2.85.3 1980s	1	2033	Risk Ratio (M‐H, Random, 95% CI)	0.91 [0.73, 1.14]
2.85.4 1990s	1	57	Risk Ratio (M‐H, Random, 95% CI)	0.33 [0.10, 1.09]
2.85.5 2000s	3	49278	Risk Ratio (M‐H, Random, 95% CI)	0.97 [0.88, 1.08]
2.86 DIABETES DIAGNOSES Show forest plot	1		Risk Ratio (M‐H, Random, 95% CI)	Subtotals only

Comparison 2. SFA reduction vs usual diet ‐ secondary health events

Comparison 3. SFA reduction vs usual diet ‐ secondary blood outcomes

Outcome or subgroup title	No. of studies	No. of participants	Statistical method	Effect size
3.1 Total cholesterol, mmol/L Show forest plot	14	7115	Mean Difference (IV, Random, 95% CI)	‐0.24 [‐0.36, ‐0.13]

3.2 TC, mmol/L, subgroup by any replacement Show forest plot	14		Mean Difference (IV, Random, 95% CI)	Subtotals only

3.2.1 replaced by PUFA	9	3888	Mean Difference (IV, Random, 95% CI)	‐0.33 [‐0.47, ‐0.19]
3.2.2 replace by MUFA	1	24	Mean Difference (IV, Random, 95% CI)	0.30 [‐0.93, 1.53]
3.2.3 replace by CHO	6	5094	Mean Difference (IV, Random, 95% CI)	‐0.18 [‐0.32, ‐0.04]
3.2.4 replace by protein	4	4986	Mean Difference (IV, Random, 95% CI)	‐0.15 [‐0.27, ‐0.04]
3.2.5 replacement unclear	1	72	Mean Difference (IV, Random, 95% CI)	‐0.34 [‐0.64, ‐0.04]
3.3 TC, mmol/L, subgroup by main replacement Show forest plot	14		Mean Difference (IV, Random, 95% CI)	Subtotals only

3.3.1 replaced by PUFA	8	3838	Mean Difference (IV, Random, 95% CI)	‐0.28 [‐0.37, ‐0.19]
3.3.2 replace by MUFA	1	24	Mean Difference (IV, Random, 95% CI)	0.30 [‐0.93, 1.53]
3.3.3 replace by CHO	4	3181	Mean Difference (IV, Random, 95% CI)	‐0.19 [‐0.40, 0.01]
3.3.4 replace by protein	0	0	Mean Difference (IV, Random, 95% CI)	Not estimable
3.3.5 replacement unclear	1	72	Mean Difference (IV, Random, 95% CI)	‐0.34 [‐0.64, ‐0.04]
3.4 LDL cholesterol, mmol/L Show forest plot	5	3291	Mean Difference (IV, Random, 95% CI)	‐0.19 [‐0.33, ‐0.05]

3.5 LDL, mmol/L, subgroup by any replacement Show forest plot	5		Mean Difference (IV, Random, 95% CI)	Subtotals only

3.5.1 replaced by PUFA	1	50	Mean Difference (IV, Random, 95% CI)	‐0.48 [‐0.90, ‐0.06]
3.5.2 replace by MUFA	0	0	Mean Difference (IV, Random, 95% CI)	Not estimable
3.5.3 replace by CHO	3	2985	Mean Difference (IV, Random, 95% CI)	‐0.16 [‐0.35, 0.02]
3.5.4 replace by protein	2	2935	Mean Difference (IV, Random, 95% CI)	‐0.09 [‐0.15, ‐0.04]
3.5.5 replacement unclear	2	306	Mean Difference (IV, Random, 95% CI)	‐0.29 [‐0.51, ‐0.08]
3.6 LDL, mmol/L, subgroup by main replacement Show forest plot	5		Mean Difference (IV, Random, 95% CI)	Subtotals only

3.6.1 replaced by PUFA	0	0	Mean Difference (IV, Random, 95% CI)	Not estimable
3.6.2 replace by MUFA	0	0	Mean Difference (IV, Random, 95% CI)	Not estimable
3.6.3 replace by CHO	3	2985	Mean Difference (IV, Random, 95% CI)	‐0.16 [‐0.35, 0.02]
3.6.4 replace by protein	0	0	Mean Difference (IV, Random, 95% CI)	Not estimable
3.6.5 replacement unclear	2	306	Mean Difference (IV, Random, 95% CI)	‐0.29 [‐0.51, ‐0.08]
3.7 HDL cholesterol, mmol/L Show forest plot	6	5147	Mean Difference (IV, Random, 95% CI)	‐0.01 [‐0.02, 0.01]

3.8 HDL, mmol/L, subgroup by any replacement Show forest plot	6		Mean Difference (IV, Random, 95% CI)	Subtotals only

3.8.1 replaced by PUFA	2	1905	Mean Difference (IV, Random, 95% CI)	‐0.01 [‐0.04, 0.01]
3.8.2 replace by MUFA	0	0	Mean Difference (IV, Random, 95% CI)	Not estimable
3.8.3 replace by CHO	4	4840	Mean Difference (IV, Random, 95% CI)	‐0.01 [‐0.03, 0.00]
3.8.4 replace by protein	3	4790	Mean Difference (IV, Random, 95% CI)	‐0.01 [‐0.03, 0.00]
3.8.5 replacement unclear	2	307	Mean Difference (IV, Random, 95% CI)	0.01 [‐0.10, 0.12]
3.9 HDL, mmol/L, subgroup by main replacement Show forest plot	6		Mean Difference (IV, Random, 95% CI)	Subtotals only

3.9.1 replaced by PUFA	1	1855	Mean Difference (IV, Random, 95% CI)	‐0.01 [‐0.04, 0.02]
3.9.2 replace by MUFA	0	0	Mean Difference (IV, Random, 95% CI)	Not estimable
3.9.3 replace by CHO	3	2985	Mean Difference (IV, Random, 95% CI)	‐0.01 [‐0.03, 0.01]
3.9.4 replace by protein	0	0	Mean Difference (IV, Random, 95% CI)	Not estimable
3.9.5 replacement unclear	2	307	Mean Difference (IV, Random, 95% CI)	0.01 [‐0.10, 0.12]
3.10 Triglycerides, mmol/L Show forest plot	7	3845	Mean Difference (IV, Random, 95% CI)	‐0.08 [‐0.21, 0.04]

3.11 TG, mmol/L, subgroup by any replacement Show forest plot	7		Mean Difference (IV, Random, 95% CI)	Subtotals only

3.11.1 replaced by PUFA	3	604	Mean Difference (IV, Random, 95% CI)	‐0.19 [‐0.35, ‐0.02]
3.11.2 replace by MUFA	0	0	Mean Difference (IV, Random, 95% CI)	Not estimable
3.11.3 replace by CHO	3	2985	Mean Difference (IV, Random, 95% CI)	‐0.04 [‐0.32, 0.25]
3.11.4 replace by protein	2	2935	Mean Difference (IV, Random, 95% CI)	0.01 [‐0.08, 0.09]
3.11.5 replacement unclear	2	306	Mean Difference (IV, Random, 95% CI)	‐0.09 [‐0.52, 0.33]
3.12 TG, mmol/L, subgroup by main replacement Show forest plot	7		Mean Difference (IV, Random, 95% CI)	Subtotals only

3.12.1 replaced by PUFA	2	554	Mean Difference (IV, Random, 95% CI)	‐0.16 [‐0.30, ‐0.01]
3.12.2 replace by MUFA	0	0	Mean Difference (IV, Random, 95% CI)	Not estimable
3.12.3 replace by CHO	3	2985	Mean Difference (IV, Random, 95% CI)	‐0.04 [‐0.32, 0.25]
3.12.4 replace by protein	0	0	Mean Difference (IV, Random, 95% CI)	Not estimable
3.12.5 replacement unclear	2	306	Mean Difference (IV, Random, 95% CI)	‐0.09 [‐0.52, 0.33]
3.13 total cholesterol /HDL ratio Show forest plot	3	2985	Mean Difference (IV, Random, 95% CI)	‐0.10 [‐0.33, 0.13]

3.14 TC /HDL ratio, subgroup by any replacement Show forest plot	3		Mean Difference (IV, Random, 95% CI)	Subtotals only

3.14.1 replaced by PUFA	1	50	Mean Difference (IV, Random, 95% CI)	‐0.58 [‐1.33, 0.17]
3.14.2 replace by MUFA	0	0	Mean Difference (IV, Random, 95% CI)	Not estimable
3.14.3 replace by CHO	3	2985	Mean Difference (IV, Random, 95% CI)	‐0.10 [‐0.33, 0.13]
3.14.4 replace by protein	2	2935	Mean Difference (IV, Random, 95% CI)	‐0.09 [‐0.21, 0.04]
3.14.5 replacement unclear	0	0	Mean Difference (IV, Random, 95% CI)	Not estimable
3.15 TC /HDL ratio, subgroup by main replacement Show forest plot	3		Mean Difference (IV, Random, 95% CI)	Subtotals only

3.15.1 replaced by PUFA	0	0	Mean Difference (IV, Random, 95% CI)	Not estimable
3.15.2 replace by MUFA	0	0	Mean Difference (IV, Random, 95% CI)	Not estimable
3.15.3 replace by CHO	3	2985	Mean Difference (IV, Random, 95% CI)	‐0.10 [‐0.33, 0.13]
3.15.4 replace by protein	0	0	Mean Difference (IV, Random, 95% CI)	Not estimable
3.15.5 replacement unclear	0	0	Mean Difference (IV, Random, 95% CI)	Not estimable
3.16 LDL /HDL ratio Show forest plot	1		Mean Difference (IV, Random, 95% CI)	Subtotals only

3.17 Lp(a), mmol/L Show forest plot	2	2882	Mean Difference (IV, Random, 95% CI)	0.00 [‐0.00, 0.00]

3.18 Lp(a), mmol/L, subgroup by any replacement Show forest plot	2		Mean Difference (IV, Random, 95% CI)	Subtotals only

3.18.1 replaced by PUFA	1	50	Mean Difference (IV, Random, 95% CI)	0.00 [‐1.37, 1.37]
3.18.2 replace by MUFA	0	0	Mean Difference (IV, Random, 95% CI)	Not estimable
3.18.3 replace by CHO	2	2882	Mean Difference (IV, Random, 95% CI)	0.00 [‐0.00, 0.00]
3.18.4 replace by protein	1	2832	Mean Difference (IV, Random, 95% CI)	0.00 [‐0.00, 0.00]
3.18.5 replacement unclear	0	0	Mean Difference (IV, Random, 95% CI)	Not estimable
3.19 Lp(a), mmol/L, subgroup by main replacement Show forest plot	2		Mean Difference (IV, Random, 95% CI)	Subtotals only

3.19.1 replaced by PUFA	0	0	Mean Difference (IV, Random, 95% CI)	Not estimable
3.19.2 replace by MUFA	0	0	Mean Difference (IV, Random, 95% CI)	Not estimable
3.19.3 replace by CHO	2	2882	Mean Difference (IV, Random, 95% CI)	0.00 [‐0.00, 0.00]
3.19.4 replace by protein	0	0	Mean Difference (IV, Random, 95% CI)	Not estimable
3.19.5 replacement unclear	0	0	Mean Difference (IV, Random, 95% CI)	Not estimable
3.20 Insulin sensitivity Show forest plot	4		Mean Difference (IV, Random, 95% CI)	Subtotals only

3.20.1 HbA1c (glycosylated haemoglobin), %	0	0	Mean Difference (IV, Random, 95% CI)	Not estimable
3.20.2 GTT (glucose tolerance test), glucose at 2 hours, mmol/L	3	249	Mean Difference (IV, Random, 95% CI)	‐1.69 [‐2.55, ‐0.82]
3.20.3 HOMA	1	2832	Mean Difference (IV, Random, 95% CI)	0.00 [‐0.04, 0.04]

Comparison 3. SFA reduction vs usual diet ‐ secondary blood outcomes

Comparison 4. SFA reduction vs usual diet ‐ secondary outcomes including potential adverse effects

Outcome or subgroup title	No. of studies	No. of participants	Statistical method	Effect size
4.1 Cancer diagnoses Show forest plot	4	52294	Risk Ratio (M‐H, Random, 95% CI)	0.94 [0.83, 1.07]

4.2 Cancer deaths Show forest plot	5	52283	Risk Ratio (M‐H, Random, 95% CI)	1.00 [0.61, 1.64]

4.3 Weight, kg Show forest plot	6	43062	Mean Difference (IV, Random, 95% CI)	‐1.77 [‐3.54, ‐0.01]

4.4 BMI, kg/m2 Show forest plot	6	43894	Mean Difference (IV, Random, 95% CI)	‐0.42 [‐0.72, ‐0.12]

4.5 Systolic Blood Pressure, mmHg Show forest plot	5	3812	Mean Difference (IV, Random, 95% CI)	‐0.19 [‐1.36, 0.97]

4.6 Diastolic Blood Pressure, mmHg Show forest plot	5	3812	Mean Difference (IV, Random, 95% CI)	‐0.36 [‐1.03, 0.32]

4.7 Quality of Life Show forest plot	1	40130	Mean Difference (IV, Random, 95% CI)	0.04 [0.01, 0.07]

Comparison 4. SFA reduction vs usual diet ‐ secondary outcomes including potential adverse effects