. 2018 Nov 30;2018(11):CD003177. doi: 10.1002/14651858.CD003177.pub4

Summary of findings for the main comparison. High versus low LCn3 for preventing cardiovascular disease and mortality (primary outcomes).

High versus low LCn3 for preventing cardiovascular disease and mortality (primary outcomes)
Patient or population: adults with or without existing CVD  Setting: participants were living at home for most or all of the duration of their trials. Most studies were carried out in high‐income economies (World Bank 2018), but four trials were carried out in upper‐middle income countries (Argentina, Iran, Turkey and China). No studies took place in low‐ or low‐middle income countries.  Intervention: higher intake of long‐chain omega‐3 fats  Comparison: lower intake of long‐chain omega‐3 fats The intervention was dietary supplementation, a provided diet or advice on diet. Supplementation may have been in oil or capsule form or as foodstuffs provided, to be consumed by mouth (excluding enteral and parenteral feeds and enemas). The foodstuffs or supplements must have been: oily fish or fish oils as a food, oil, made into a spreading fat or supplementing another food (such as bread or eggs). Refined eicosapentaenoic acid (EPA), docosahexaenoic acid (DHA) or concentrated fish or algal oils, were also accepted.
Outcomes	*Anticipated absolute effects (95% CI)**		Relative effect  (95% CI)	№ of participants  (studies)	Certainty of the evidence  (GRADE)	Comments
Outcomes	Risk with lower LCn3	Risk with higher LCn3	Relative effect  (95% CI)	№ of participants  (studies)	Certainty of the evidence  (GRADE)	Comments
All‐cause mortality – deaths  Assessed with: number of participants dying of any cause, whether reported as an outcome or a reason for dropout Duration: range 12 to 72 months	90 per 1,000	88 per 1,000  (83 to 92)	RR 0.98  (0.93 to 1.03)	92,653  (39 RCTs)	⊕⊕⊕⊕  High^a	Meta‐analysis and indications of bias suggest risk reduction of less than 2%. Long‐chain omega‐3 fat intake makes little or no difference to all‐cause mortality.
Cardiovascular mortality – cardiovascular deaths  Assessed with: deaths from any cardiovascular cause. Where this was not available, cardiac death was used instead where known. Duration: range 12 to 72 months	69 per 1,000	66 per 1,000  (60 to 71)	RR 0.95  (0.87 to 1.03)	67,772  (25 RCTs)	⊕⊕⊕⊝  Moderate^b	Meta‐analysis and indications of bias suggest risk reduction of less than 5%. Long‐chain omega‐3 fat intake probably makes little or no difference to cardiovascular deaths.
Cardiovascular events – cardiovascular events Assessed with: number of participants experiencing any cardiovascular event Duration: range 12 to 72 months	165 per 1,000	164 per 1,000  (155 to 172)	RR 0.99  (0.94 to 1.04)	90,378  (38 RCTs)	⊕⊕⊕⊕  High^c	Meta‐analysis and indications of bias suggest risk reduction of less than 1%. Long‐chain omega‐3 fat intake makes little or no difference to risk of cardiovascular events.
Coronary heart disease mortality – CHD deaths  Assessed with: coronary deaths, or where these were not reported, IHD death, fatal MI or cardiac death (in that order) Duration: range 12 to 72 months	22 per 1,000	21 per 1,000  (18 to 24)	RR 0.93  (0.79 to 1.09)	73,491  (21 RCTs)	⊕⊕⊕⊝  Moderate^d	Meta‐analysis and indications of bias suggest risk reduction of less than 7%. Long‐chain omega‐3 fat intake probably makes little or no difference to coronary heart mortality.
Coronary heart disease events – CHD events  Assessed with: number of participants experiencing the first outcome in this list reported for each trial: CHD or coronary events; total MI; acute coronary syndrome; or angina (stable and unstable) Duration: range 12 to 72 months	68 per 1,000	63 per 1,000  (59 to 65)	RR 0.93  (0.88 to 0.97)	84,301  (28 RCTs)	⊕⊕⊕⊝  Moderate^e	Meta‐analysis and indications of bias suggest risk reduction of less than 7%. Long‐chain omega‐3 fat intake probably makes little or no difference to risk of coronary heart events.
Stroke Assessed with: number of participants experiencing at least one fatal or non‐fatal, ischaemic or haemorrhagic stroke Duration: range 12 to 72 months	20 per 1,000	21 per 1,000  (19 to 23)	RR 1.06  (0.96 to 1.16)	89,358  (28 RCTs)	⊕⊕⊕⊝  Moderate^f	Meta‐analysis and indications of bias suggest increased risk of less than 6%. Long‐chain omega‐3 fat intake probably makes little or no difference to risk of experiencing a stroke.
Arrhythmias Assessed with: number of participants experiencing fatal or nonfatal, new or recurrent arrhythmia, including atrial fibrillation, ventricular tachycardia and ventricular fibrillation. Duration: range 12 to 72 months	68 per 1,000	66 per 1,000  (62 to 72)	RR 0.97  (0.90 to 1.05)	53,796  (28 RCTs)	⊕⊕⊕⊝  Moderate^g	Meta‐analysis and indications of bias suggest risk reduction of less than 3%. Long‐chain omega‐3 fat intake probably makes little or no difference to risk of arrhythmia.
Harms: bleeding Assessed with: number of participants experiencing bleeding events. Duration: range 12 to 72 months	8 per 1,000	8 per 1,000  (5 to 11)	RR 1.06  (0.73 to 1.52)	45,562  (8 RCTs)	⊕⊝⊝⊝  Very low^h	The effect of long‐chain omega‐3 fat intake on bleeding is unclear as the evidence is of very low quality.
Harms: pulmonary embolus or DVT Assessed with: number of participants experiencing pulmonary embolus or deep vein thrombosis Duration: range 18 to 36 months	5 per 1,000	6 per 1,000  (2 to 18)	RR 1.25  (0.41 to 3.78)	3,011  (4 RCTs)	⊕⊝⊝⊝  Very lowⁱ	The effect of long‐chain omega‐3 fat intake on pulmonary embolus or DVT 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).  CHD: coronary heart disease; CI: confidence interval; DVT: deep vein thrombosis; IHD: ischaemic heart disease; MI: myocardial infarction; RCT: randomised controlled trial; RR: risk ratio.
GRADE Working Group grades of evidence  High certainty: we are very confident that the true effect lies close to that of the estimate of the effect.  Moderate certainty: we are moderately confident in the effect estimate: the true effect is likely to be close to the estimate of the effect, but there is a possibility that it is substantially different.  Low certainty: our confidence in the effect estimate is limited: the true effect may be substantially different from the estimate of the effect.  Very low certainty: we have very little confidence in the effect estimate: the true effect is likely to be substantially different from the estimate of effect.

^aAll‐cause mortality, LCn3

Risk of bias: effect size moved closer to no effect (RR 1.0) when analysis was limited to studies at low summary risk of bias and low risk of compliance bias (adding weight to the suggestion of little or no effect) but did not alter with fixed‐effect meta‐analysis or results in the analysis limited to larger studies. It was further noted by the WHO NUGAG Subgroup on Diet and Health that although many of the RCTs had issues with blinding, the tendency for lack of blinding is an overestimation of effect. This is less of a concern for this outcome, as the pooled effect was approaching null and not statistically significant. Not downgraded.
Inconsistency: I² was < 60% and I² reduced when analysis was limited to studies at low summary risk of bias. This adds weight to the suggestion of little or no effect. Not downgraded.
Indirectness: representative, generalisable adult population including men and women, including healthy participants and participants with CVD risk factors or previous CVD as well as non‐CVD health problems. Low‐ and middle‐income countries were represented but underrepresented. Not downgraded.
Imprecision: tight confidence intervals, very large numbers of participants have taken part in RCTs in long‐term studies with consistent results. Given the lack of a statistically significant effect in this very large set of participants, any effect appears too small to be individually relevant. Not downgraded.
Publication bias: the funnel plot suggested that some small studies with higher numbers of events in the intervention group might be missing. If such missing studies were added back in, the RR would rise. This adds weight to the suggestion of little or no effect. Not downgraded.

^bCardiovascular mortality, LCn3

Risk of bias: effect size moved closer to no effect (RR 1.0) when analysis was limited to studies at low summary risk of bias and low risk of compliance bias (adding weight to the suggestion of little or no effect) but did not alter with fixed‐effect meta‐analysis or results in the analysis limited to larger studies. It was further noted by the WHO NUGAG Subgroup on Diet and Health that although many of the RCTs had issues with blinding, the tendency for lack of blinding is an overestimation of effect. This is less of a concern for this outcome, as the pooled effect was approaching null and not statistically significant. Not downgraded.
Inconsistency: I² was < 60% and I² reduced when analysis was limited to studies at low summary risk of bias. This adds weight to the suggestion of little or no effect. Not downgraded.
Indirectness: representative, generalisable adult population including men and women, including healthy participants and participants with CVD risk factors or previous CVD as well as non‐CVD health problems. All studies were conducted in high‐income countries. Not downgraded.
Imprecision: Although very large numbers of participants have taken part in RCTs in long‐term studies, with consistent results, the 95% CI includes the null. Given the lack of a statistically significant effect in this very large set of participants, any effect appears too small to be individually relevant. However, as 95% confidence intervals do not exclude important benefits or harms. Downgraded once.
Publication bias: the funnel plot suggested that some small studies with higher numbers of events in the intervention group might be missing. If such missing studies were added back in, the RR would rise. This adds weight to the suggestion of little or no effect. Not downgraded.

^cCardiovascular events, LCn3

Risk of bias: effect size moved closer to no effect (RR 1.0) when analysis was limited to studies at low summary risk of bias and low risk of compliance bias (adding weight to the suggestion of little or no effect) but did not alter with fixed‐effect meta‐analysis or results in the analysis limited to larger studies. It was further noted by the WHO NUGAG Subgroup on Diet and Health that although many of the RCTs had issues with blinding, the tendency for lack of blinding is an overestimation of effect. This is less of a concern for this outcome, as the pooled effect was approaching null and not statistically significant. Not downgraded.
Inconsistency: I² was < 60% and I² reduced when analysis was limited to studies at low summary risk of bias. This adds weight to the suggestion of little or no effect. Not downgraded.
Indirectness: representative, generalisable adult population including men and women, including healthy participants and participants with CVD risk factors or previous CVD as well as non‐CVD health problems. Low‐ and middle‐income countries were represented but underrepresented. Not downgraded.
Imprecision: very large numbers of participants have taken part in RCTs in long‐term studies with consistent results. Given the lack of an important effect in this very large set of participants, any effect appears too small to be individually relevant. However, as 95% confidence intervals do not exclude important benefits or harms, we downgraded once.
Publication bias: the funnel plot suggested that some small studies with higher numbers of events in the intervention group might be missing. If such missing studies were added back in, the RR would rise. This adds weight to the suggestion of little or no effect. Not downgraded.

^dCoronary heart disease mortality, LCn3

Risk of bias: effect size moved closer to no effect (RR 1.0) when analysis was limited to studies at low summary risk of bias and low risk of compliance bias (adding weight to the suggestion of little or no effect) but did not alter with fixed‐effect meta‐analysis or results in the analysis limited to larger studies. It was further noted by the WHO NUGAG Subgroup on Diet and Health that although many of the RCTs had issues with blinding, the tendency for lack of blinding is an overestimation of effect. This is less of a concern for this outcome, as the pooled effect was approaching null and not statistically significant. Not downgraded.
Inconsistency: I² was < 60% and I² reduced when analysis was limited to studies at low summary risk of bias. This adds weight to the suggestion of little or no effect. Not downgraded.
Indirectness: representative, generalisable adult population including men and women, including healthy participants and participants with CVD risk factors or previous CVD as well as non‐CVD health problems. All studies were conducted in high‐income countries. Not downgraded.
Imprecision: very large numbers of participants have taken part in RCTs in long‐term studies with consistent results. Given the lack of a statistically significant effect in this very large set of participants, any effect appears too small to be individually relevant. However, as 95% confidence intervals do not exclude important benefits or harms we downgraded once.
Publication bias: the funnel plot suggested that some small studies with higher numbers of events in the intervention group might be missing. If such missing studies were added back in the RR would rise. This adds weight to the suggestion of little or no effect. Not downgraded.

^eCoronary heart disease events, LCn3

Risk of bias: effect size moved closer to no effect (RR 1.0) when was analysis limited to studies at low summary risk of bias. This adds weight to the suggestion of little or no effect. However, effect size did not alter with fixed‐effect meta‐analysis or limiting to studies at low risk of compliance bias or larger trials. It was further noted by the WHO NUGAG Subgroup on Diet and Health that there was a significant effect observed in main analysis but the effect moved closer to a non‐significant, null effect when analysis was limited to studies at low summary risk of bias. Downgraded once.
Inconsistency: I² was < 60% and I² reduced when analysis was limited to studies at low summary risk of bias. This adds weight to the suggestion of little or no effect. Not downgraded.
Indirectness: representative, generalisable adult population including men and women, including healthy participants and participants with CVD risk factors or previous CVD as well as non‐CVD health problems. Low‐ and middle‐income countries were represented but underrepresented. Not downgraded.
Imprecision: 95% CI did not include the null. Not downgraded.
Publication bias: the funnel plot suggested that some small studies with higher numbers of events in the intervention group might be missing. If such missing studies were added back in, the RR would rise. This adds weight to the suggestion of little or no effect. Not downgraded.

^fStroke, LCn3

Risk of bias: effect size moved closer to no effect (RR 1.0) when analysis limited to studies at low summary risk of bias (adding weight to the suggestion of little or no effect), but did not alter with fixed‐effect meta‐analysis or limiting to larger studies. Limiting to studies at low risk of compliance problems resulted in the suggestion of greater harm. It was further noted by the WHO NUGAG Subgroup on Diet and Health that although many of the RCTs had issues with blinding, the tendency for lack of blinding is an overestimation of effect. This is less of a concern for this outcome, as the pooled effect was approaching null and not statistically significant. Not downgraded.
Inconsistency: I² was < 60%. Not downgraded.
Indirectness: representative, generalisable adult population including men and women, including healthy participants and participants with CVD risk factors or previous CVD as well as non‐CVD health problems. Low‐ and middle‐income countries were represented but underrepresented. Not downgraded.
Imprecision: very large numbers of participants have taken part in RCTs in long‐term studies with consistent results. Given the lack of a statistically significant effect in this very large set of participants any effect appears too small to be individually relevant. However, as 95% confidence intervals do not exclude important benefits or harms, we downgraded once.
Publication bias: the funnel plot did not suggest any small study bias. Not downgraded.

^gArrhythmias, LCn3

Risk of bias: effect size remained similar in most sensitivity analyses, but moved closer to no effect (RR 1.01) when analysis used fixed‐effect meta‐analysis (adding weight to the suggestion of little or no effect) and suggested harm when limited to studies at low summary risk of bias. Not downgraded.
Inconsistency: I² was < 60% and I² reduced when analysis was limited to studies at low summary risk of bias. This adds weight to the suggestion of little or no effect. Not downgraded.
Indirectness: representative, generalisable adult population including men and women, including healthy participants and participants with CVD risk factors or previous CVD as well as non‐CVD health problems. Low‐ and middle‐income countries were represented but underrepresented. Not downgraded.
Imprecision: As 95% confidence intervals do not exclude important benefits we downgraded once.
Publication bias: funnel plot not interpretable as studies all of a similar size and weight. Not downgraded.

^hBleeding, LCn3

Risk of bias: effect size changed direction (from harmful to protective) when analysis limited to studies at low summary risk of bias. Downgraded once.
Inconsistency: I² was < 60%. Not downgraded.
Indirectness: representative, generalisable adult population including men and women, including healthy participants and participants with CVD risk factors or previous CVD as well as non‐CVD health problems. Low‐ and middle‐income countries not represented. Not downgraded.
Imprecision: 95% confidence intervals do not exclude large and important benefits or harms. Downgraded twice.
Publication bias: insufficient studies for funnel plot. Not downgraded.

ⁱPulmonary embolus or DVD, LCn3

Risk of bias: effect size suggested greater harm when analysis limited to studies at low summary risk of bias. Downgraded once.
Inconsistency: I² was < 60%. Not downgraded.
Indirectness: representative, generalisable adult population including men and women, including healthy participants and participants with CVD risk factors or previous CVD as well as non‐CVD health problems. Low‐ and middle‐income countries not represented. Not downgraded.
Imprecision: 95% confidence intervals do not exclude large benefits or large harms. Downgraded twice.
Publication bias: insufficient studies for funnel plot. Not downgraded.