NCBI Bookshelf. A service of the National Library of Medicine, National Institutes of Health.

Balk EM, Adam GP, Langberg V, et al. Omega-3 Fatty Acids and Cardiovascular Disease: An Updated Systematic Review. Rockville (MD): Agency for Healthcare Research and Quality (US); 2016 Aug. (Evidence Reports/Technology Assessments, No. 223.)

  • This publication is provided for historical reference only and the information may be out of date.

This publication is provided for historical reference only and the information may be out of date.

Cover of Omega-3 Fatty Acids and Cardiovascular Disease

Omega-3 Fatty Acids and Cardiovascular Disease: An Updated Systematic Review.

Show details


The present review evaluates the effects of, and the associations between, omega-3 fatty acids (n-3 FA)—including alphalinolenic acid (ALA), stearidonic acid (SDA), eicosapentaenoic acid (EPA), docosapentaenoic acid (DPA), and docosahexaenoic acid (DHA) and n-3 FA biomarkers— and cardiovascular disease (CVD) outcomes. The Brown Evidence-based Practice Center (EPC) conducted the review based on a systematic review of the published scientific literature using established methodologies as outlined in the Agency for Healthcare Research and Quality (AHRQ) Methods Guide for Effectiveness and Comparative Effectiveness Reviews.39

The review was conducted in parallel with a systematic review of n-3 FA and child and maternal health, conducted by another Evidence-based Practice Center (EPC). Several aspects of the review were coordinated, including eligibility criteria and search strategies regarding interventions and exposures structure of the reviews, and assessments of the studies’ risk of bias, strength of the bodies of evidence, and extraction of study characteristics needed to assess causality.

Topic Refinement and Review Protocol

We convened a Technical Expert Panel (TEP) to help refine the research questions and protocol. The TEP included five experts in nutrition, n-3 FA research specifically, CVD epidemiology, and cardiology. Also included in the discussions with the TEP were the Director of and a Senior Scientist at the Office of Dietary Supplements (ODS), and the AHRQ Task Order Officer. We discussed the Key Questions, analytic framework, study eligibility criteria, literature search, and analysis plans.

In regards to the populations of interest, we explicitly expanded the definition of the at risk for CVD population to include adults with cardiometabolic syndrome (and related conditions) and nondialysis dependent chronic kidney disease. Regarding the interventions of interest, we discussed the changes from the original AHRQ reports on n-3 FA, specifically that we included only studies that quantify n-3 FA content of the intervention, and that we added n-3 FA biomarkers as an exposure of interest. We also clarified that we excluded weight loss interventions that included n-3 FA as part of the intervention. Weight-loss studies, by definition need to create energy deficits; interventions generally aim to reduce total energy intake and/or increasing energy expenditure. Energy deficits trigger metabolic changes including altering lipid metabolism. Since CVD outcomes were the main outcome of interest, weight loss is a major confounder that may be in the causal pathway between exposures of interest (i.e., n-3 FA) and CVD outcomes. Regarding outcomes of interest, we refined the list of “major lipids” of interest to include only low density lipoprotein cholesterol (LDL-c), high density lipoprotein cholesterol (HDL-c), triglycerides (Tg), LDL-c to HDL-c ratio, and total cholesterol to HDL-c ratio. Compared to the original n-3 FA and CVD outcome report, we added peripheral vascular disease, arrhythmia events, congestive heart failure (CHF), and incident hypertension. We discussed a number of potential modifiers of interest to be searched for, including demographic features, weight, blood pressure (BP), source and type of n-3 FA, exposure duration, C reactive protein level, and specific co-interventions (i.e., statins, vitamin E).

It was agreed to maintain a minimum duration of followup of 1 month for intermediate outcomes (lipids and BP) and 1 year for all clinical outcomes. We agreed to include only randomized controlled trials (RCT) of specific comparisons of interventions and large, prospective, longitudinal observational studies of exposure (either baseline dietary intake or biomarker level). We also agreed to include the RCTs that are largest or report subgroup or factorial analyses, and the largest observational studies to constrain the total number of included studies to approximately 75 to 100. The search strategy was refined based on suggestions from the TEP. The TEP agreed that the primary literature search would be conducted for the period from 2002 to the present to capture studies published since the original EPC report, with older studies to come from existing systematic reviews including the original EPC report. For new topics (e.g., biomarkers, peripheral vascular disease), the TEP agreed that searches back to 2000 would be sufficient to capture relevant analyses.

In addition, in separate discussions with the Office of Dietary Supplements (ODS) representative and our Task Order Officer (TOO) we considered how and whether to assess the concept of causality, particularly for the observational studies. After discussion of the Bradford Hill criteria and related issues regarding causality,40 we agreed upon the creation of an appendix table (Appendix G) that provides the study-level data for items that may be pertinent for users of this report to assess causality.

Furthermore, we had joint discussions with the Southern California EPC—which conducted a parallel report of n-3 FA and maternal and child health—and our TOO and the ODS representative to coordinate our protocols and processes. The protocol was entered into the PROSPERO register (registry number CRD42014015602).

Literature Search

We conducted literature searches of studies in MEDLINE®, both the Cochrane Central Trials Registry® and Cochrane Database of Systematic Reviews®, Embase®, and CAB Abstracts® from 2002 to 8 June 2015 (to overlap with the last search run for the 2004 reviews). We searched earlier publications back to 2000 for the newly added outcomes (peripheral vascular disease, CHF, arrhythmias, hypertension) and for biomarkers of n-3 FA intake. We also rescreened and included all studies from the original reviews that met current eligibility criteria. We revised the search strategy used in the original reviews to capture new terms for n-3 FA, biomarkers, and additional outcomes. In electronic searches, we combined terms for n-3 FA (and biomarkers), CVD and risk factors (BP, plasma lipids, hypertension), limited to humans, English language, and relevant research designs. Titles and abstracts were screened to identify articles relevant to each Key Questions. We also reviewed reference lists of related systematic reviews for other potentially eligible studies. We invited TEP members to provide additional citations. In addition, a call for potentially relevant articles was posted on the Federal Register (in lieu of Scientific Information Packets), but yielded no additional studies. Appendix A displays the current complete search strategy.

Study Eligibility Criteria

The current eligibility criteria are mostly similar to the criteria used in the original 2004 review. The populations remain the same. The interventions and exposures have been expanded to include n-3 FA biomarkers. The list of CVD outcomes of interest has been expanded. Similar study designs were included.

For all Key Questions, the eligibility criteria are:


  • Healthy adults (≥18 years) without CVD or with low to intermediate risk for CVD
  • Adults at high risk for CVD (e.g., with diabetes, cardiometabolic syndrome, hypertension, dyslipidemia, nondialysis dependent chronic kidney disease)
  • Adults with clinical CVD (e.g., history of myocardial infarction [MI], angina, stroke, arrhythmia)
  • Exclude populations chosen for having a non-CVD or nondiabetes-related disease (e.g., cancer, gastrointestinal disease, rheumatic disease, dialysis)


  • n-3 FA supplements
  • n-3 FA supplemented foods (e.g., eggs)
  • n-3 FA content in diet
  • Biomarkers of n-3 FA intake
  • n-3 FA content of food or supplements must have been explicitly quantified (by any method). Therefore, studies such as those of fish diet where only servings per week were defined or Mediterranean diet studies without n-3 FA quantified were excluded. The n-3 FA quantification could be of total n-3 FA, of a specific n-3 FA (e.g., ALA) or of combined EPA+DHA (“marine oil”).
  • Exclude mixed interventions of n-3 FA and other dietary or supplement differences (e.g., n-3 FA and vitamin E versus placebo; n-3 FA as part of a low fat diet versus usual diet). However, factorial design (and other) studies that compared (for example) n-3 FA versus control, with or without another intervention (e.g., statins) were included.
  • Exclude n-3 FA dose ≥6 g/d, per the original review’s protocol based on the assessment that n-3 FA intake above this amount is impractical and has little relevance on health care recommendations.
  • Exclude weight loss interventions


  • Placebo or no n-3 FA intervention
  • Different n-3 FA source intervention
  • Different n-3 FA concentration intervention
  • Different n-3 FA dietary exposure (e.g., comparison of quantiles)
  • Different n-3 FA biomarker levels (e.g., comparison of quantiles)


  • All-cause death
  • Cardiovascular (CV), cerebrovascular, and peripheral vascular events:
    • Fatal vascular events (e.g., due to MI, stroke)
    • Total incident vascular events (e.g., MI, stroke, transient ischemic attack, unstable angina, major adverse CV events [MACE]; total events include fatal and nonfatal events; total stroke includes ischemic and hemorrhagic stroke)
    • Coronary heart disease (also known as coronary artery disease), new diagnosis
    • CHF, new diagnosis
    • Cerebrovascular disease, new diagnosis
    • Peripheral vascular disease, new diagnosis
    • Ventricular arrhythmia, new diagnosis, including sudden cardiac death
    • Supraventricular arrhythmia (including atrial fibrillation), new diagnosis
    • Major vascular interventions/procedures (e.g., revascularization, thrombolysis, lower extremity amputation, defibrillator placement)
  • Major CVD risk factors (intermediate outcomes):
    • BP (new-onset hypertension, systolic, diastolic, and mean arterial pressure)
    • Key plasma lipids (i.e., HDL-c, LDL-c, total/HDL-c ratio, LDL-c/HDL-c ratio, Tg)
  • Adverse events (e.g., bleeding, major gastrointestinal disturbance), only from intervention studies of supplements


  • Clinical outcomes, including new-onset hypertension (all study designs): ≥1 year followup (and intervention duration, as applicable)
  • Intermediate outcomes (BP and plasma lipids) (all study designs): ≥1 month followup
  • Adverse events (all study designs): no minimum followup


  • Community-dwelling (noninstitutionalized) individuals

Study Design

  • RCTs (all outcomes)
  • Randomized cross-over studies (BP and plasma lipids, adverse events), minimum washout period to be determined
  • Prospective nonrandomized comparative studies (clinical outcomes, adverse events)
  • Prospective cohort (single group) studies, where groups were compared based on n-3 FA intake or intake biomarker values (clinical outcomes). Observational studies must have reported multivariate analyses.
  • Exclude: Retrospective or case control studies or cross-sectional studies (but include prospective nested case control studies). Studies must have had measures of intake prior to outcome.
  • Minimum sample sizes
    Due to the very large number of potentially eligible studies (more than 400), we applied arbitrary thresholds based on sample size, followup duration, and whether subgroup or interaction analyses were reported. These were designed to give preference to larger studies with longer followup duration or that reported interaction analyses of interest.
    • RCTs
      • We aimed for a minimum of about 25 RCTs for each of the BP and plasma lipid outcomes. We preferentially included RCTs that reported relevant subgroup, interaction, or factorial analyses.
        • For RCTs with BP or lipid outcomes with subgroup, interaction, or factorial analyses, we included parallel design RCTs with a minimum of 30 participants per arm, factorial RCTs with a minimum of 30 participants per n-3 FA intervention, and crossover trials with a minimum of 20 participants.
        • For RCTs with lipid outcomes without subgroup analyses, we included parallel design RCTs with a minimum of 200 participants per arm, factorial RCTs with a minimum of 200 participants per n-3 FA intervention, and crossover trials with a minimum of 100 participants.
        • For RCTs with BP outcomes without subgroup analyses, if followup was ≥6 months, we included all RCTs; if followup was <6 months (≥1 month), we included parallel design RCTs with a minimum of 80 participants per arm, factorial RCTs with a minimum of 80 participants per n-3 FA intervention, and crossover trials with a minimum of 40 participants.
        • For RCTs with CVD event outcomes, we included all RCTs with at least 10 participants per arm.
    • Longitudinal observational studies
      • We aimed for a minimum of about 10 observational studies for each broad clinical outcome (see bullets below) and also for dietary marine oils, dietary ALA, marine oil biomarkers, and ALA biomarkers.
        • For cardiac event outcomes, we included observational studies with at least 10,000 participants.
        • For death outcomes, we included observational studies with at least 10,000 participants.
        • For stroke event outcomes, we included observational studies with at least 3000 participants.
        • For arrhythmia event outcomes, we included observational studies with at least 2000 participants.
        • For CHF event outcomes, we included observational studies with at least 700 participants.
        • For peripheral vascular disease event, incident hypertension, MACE, and revascularization outcomes, we included observational studies with at least 500 participants.
        • We screened smaller sample size observational studies (starting with the largest studies) to include additional studies of ALA biomarkers, regardless of the outcomes analyzed.
    • In all instances, if a study met eligibility criteria for any outcome, we extracted all outcomes of interest from that study; therefore, there are multiple instances of studies being included for an outcome even though the study might not have met study size criteria for that specific outcome.
  • English language publications
  • Peer reviewed publications

Study Selection

All citations found by literature searches or through other sources were independently screened by two researchers. Upon the start of citation screening, we implemented a training session where all researchers screen the same articles and conflicts were discussed. We iteratively continue training until we have reached agreement regarding the nuances of the eligibility criteria for screening. During double-screening, we resolved conflicts as a group. All screening of literature citations was done in the open-source, online software Abstrackr (

All potentially eligible abstracts (regardless of source) were entered into an “evidence map”. From each abstract, a single researcher extracted data on the study sample size (total), study design, study duration, the population category (healthy, at risk, CVD), the specific n-3 FA analyzed, whether biomarkers were reported, whether subgroup or factorial analyses were reported, and the outcomes mentioned in the abstract.

Based on the study descriptions in the evidence map, we selected the largest studies and those with subgroup or factorial analyses for full text review, with the goals of including a minimum of about 25 RCTs for each of the BP and plasma lipid outcomes, all RCTs with clinical outcomes, and a minimum of about 10 observational studies for each broad clinical outcome and also for dietary marine oils, dietary ALA, marine oil biomarkers, and ALA biomarkers.

Data Extraction

Each study was extracted by one methodologist. The extraction was reviewed and confirmed by at least one other experienced methodologist. Disagreements were resolved by discussion among the team, with the team leader, or between extractors. Data were extracted into customized forms in Systematic Review Data Repository (SRDR) online system ( and Excel spreadsheets, each designed to capture all elements relevant to the Key Questions. Upon completion of the review, the Excel spreadsheets (of observational study results data) were uploaded into SRDR and the database has made accessible to the general public (with capacity to read, download, and comment on data) (at The basic elements and design of these forms include elements that address population characteristics; descriptions of the interventions, exposures, or biomarker status (and comparators) analyzed; outcome definitions; enrolled and analyzed sample sizes; study design features; results; and risk of bias assessment. The form was developed off the forms used for the original review. We also included questions pertinent to issues related to causality. We tested the forms on several studies and revised them as necessary before full data extraction.

Quality (Risk of Bias) Assessment of Individual Studies

We assessed the methodological quality of each study based on predefined criteria. For RCTs, we used the Cochrane risk of bias tool,41 which asks about risk of selection bias, performance bias, detection bias, attrition bias, reporting bias, and other potential biases. For observational studies, we used relevant questions from the Newcastle Ottawa Scale.42 Additionally we included nutrition study specific risk of bias questions (e.g., related to uncertainty of dietary assessment measurements.13, 15, 43 Any quality issues pertinent to specific outcomes within a study were noted and applied to those outcomes. Any quality issues pertinent to specific outcomes within a study were noted and considered when determining the overall strength of evidence for conclusions related to those outcomes.

Data Synthesis

All included studies were summarized in narrative form and in summary tables that tabulate the important features of the study populations, design, intervention, outcomes, and results. Other study data are in Appendix tables.

We analyzed different study designs separately and compared and contrasted populations, exposures, and results across study designs. We examined any differences in findings between observational and intervention studies, and evaluated the risk of bias factors as possible explanations for any heterogeneity.

Statistical analyses were conducted in Stata version 13.1 (StataCorp, College Station, Texas). We conducted random effects model meta-analyses of comparative studies (i.e., RCTs) if, for each set of studies with the same outcome and intervention and comparator pair, there were at least six studies. We used the restricted maximum likelihood method (with the metareg command) to calculate the overall and population-specific (healthy, at risk, CVD) effect sizes. For trials that compared multiple n-3 FA doses to placebo, we included only the comparison of the highest dose of n-3 FA versus placebo in meta-analysis. Likewise, for trials that compared both purified EPA and DHA to placebo, we arbitrarily included only the EPA versus placebo comparison.

We summarized included observational studies both qualitatively and quantitatively. We looked at hazard ratios (HR) and their respective confidence intervals of categorical outcomes of interest for each quantile of n-3 FA exposure (intake or biomarker level) within a study versus its reference quantile. The HRs were plotted at the median dose within a quantile’s dose range (see below). Separate graphs were drawn for each combination of specific n-3 FA, measure type (e.g., intake, phospholipid level, percent FA), and outcome. We combined analyses of EPA+DHA and EPA+DHA+DPA. Within each graph, we plotted each reported cohort (i.e., from a given study, we plotted the analysis of the total cohort if that was reported, or we plotted both subgroup analyses—usually men and women—if only those were reported). We use unique symbols across graphs for all adults, men, women, and other subgroups.

When a study did not report the median doses for specific dose quantiles, we estimated them using the following rules. If the study provided the minimum and maximum dose within a quantile, we used the midpoint as the median dose. For the lowest and highest quantiles, if only one end of the range was reported (e.g., lowest quintile was <0.5 g/d), we estimated the median dose to be 20% less (or more) than that quantile’s upper (or lower) range.44 For studies that did not report the number of participants or person-years per quantile, we equally divided the total for the whole cohort to estimate the numbers per quantile.

We meta-analyzed multivariate observational cohorts when at least four cohorts analyzed the same n-3 FA, measure, and outcome. For each study cohort to be meta-analyzed, we used the STATA glst command to retrieve a set of coefficients and covariance matrices from generalized least squares trend estimation of splines with one knot each (exposure dose where the curve slope is allowed to change) across a range of knot points. Separately for ALA intake and EPA+DHA±DPA intake (the n-3 FA measured that had sufficient data for meta-analysis), we determined the range of knots for spline models by ordering the median values of all quantiles of all ALA or all EPA+DHA±DPA intake analyses being meta-analyzed (across outcomes) and selected a range from approximately the 5th lowest to 5th highest median values. Knot points were rounded to the nearest 0.1 g/d and stepped up in 0.1 g/d units to the highest knot point. We used the STATA glst command (generalized least squares) to estimate the splines for each cohort being meta-analyzed, across the range of knots. For a particular cohort, if a knot fell outside the cohort’s n-3 FA dose range, we generated a linear model without a knot. We then used the STATA mvmeta command to meta-analyze these spline models (at each knot). We captured the Akaike information criterion (AIC) for each meta-analyzed spline (at each knot). We tabulated all meta-analyzed spline models for each set of studies (within a range of knots that pertain to each set of studies). In the figures of the association of n-3 FA exposure versus risk of outcome, we included the meta-analysis spline with the best fit (the lowest AIC value).

Summary of Causality-Related Study Features

We compiled a pair of appendix tables (Appendix G) with data related to possible causality criteria. The list of items in this table was compiled based on discussions between the EPCs and ODS after discussion of the Bradford Hill criteria37 and other issues related to determining causality. The table includes a listing of included studies with their population category (healthy, at high CVD risk, with CVD), CVD risk type (e.g., diabetes, hypertension, chronic kidney disease, dyslipidemia), demographics (age, sex, race), CVD history, CVD risk factors (BP, plasma lipids, weight), baseline n-3 FA intake, n-3 FA source, n-3 FA type, how n-3 FA intake measured, study design (e.g., RCT, prospective or retrospective longitudinal cohort, or other design), exposure duration, followup duration, outcomes reported, whether outcomes were reported to be primary outcomes (vs. secondary), effect sizes, difference in n-3 FA intake (between low and high intake groups), and a dose-corrected effect size. The determination of primary outcomes was based on an explicit statement of the primary outcomes, the outcome used in reported power calculations, or if implied by focus of the original article. In addition, if the study was found in a registry and its primary outcome in this database differed from the stated primary outcome in the article, this information was included. The dose-corrected effect size is the effect size divided by the daily dose of tested n-3 FA.

Strength of the Body of Evidence

We graded the strength of the body of evidence as per the AHRQ Methods Guide on assessing the strength of evidence for each outcome.45 Following the standard AHRQ approach, for each intervention and comparison of intervention, and for each outcome, we assessed the number of studies, their study designs, the study limitations (i.e., risk of bias and overall methodological quality), the directness of the evidence to the Key Questions, the consistency of study results, the precision of any estimates of effect, the likelihood of reporting bias, and the overall findings across studies. Based on these assessments, we assigned the strength of evidence rating as being either high, moderate, or low, or there being insufficient evidence to estimate an effect.

We assigned “High” strength of evidence only if there were sufficient, consistent, precise RCTs without limitations and any association studies yielded similar conclusions. Without such RCT evidence, the highest possible strength of evidence was “Moderate”.

For outcomes with ≤2 RCTs providing evidence, the highest possible strength of evidence was “Low” under the presumption that observational studies (that analyzed the association between a one-time estimate of n-3 FA status and clinical outcomes ≥1 year in the future) cannot alone provide good evidence of an effect of n-3 FA intake. For outcomes with ≤2 RCTs, ≤2 observational studies of intake, and ≤2 observational studies of biomarkers, the strength of evidence grade was “Insufficient.” Where RCTs and observational studies yielded different conclusions about significance of effect/association, we assigned a low strength of evidence of the conclusion of the RCTs. For example, if RCTs found no significant effect, but observational studies found a significant association, we concluded low strength of evidence of no effect. If we were unable to conclude a finding of an association or effect, or no association or effect, (generally because of imprecision or inconsistency across studies), we determined that the evidence was “Insufficient” since it is not meaningful to state that there is a low strength of evidence of an unclear effect/association.

The strength-of-evidence dimensional rating are summarized in Evidence Profile tables detailing our reasoning for arriving at the overall strength of evidence rating. Study characteristics related to causality are tabulated in Appendixes G.1 and G.2.


We assessed the applicability within and across studies with reference to whether people in the studies were in the three populations of interest (healthy, at risk, and with CVD), and as pertains to n-3 FA source, type, and dose/exposure.

Peer Review and Public Commentary

A draft version of this report was reviewed by a panel of expert reviewers and the general public. The reviewers were either directly invited by the EPC or offered comments through a public review process. Revisions of the draft were made, where appropriate, based on their comments. The draft and final reports were also reviewed by the Task Order Officer and an Associate Editor from another EPC. However, the findings and conclusions are those of the authors, who are responsible for the contents of the report.


  • PubReader
  • Print View
  • Cite this Page
  • PDF version of this title (12M)

Recent Activity

Your browsing activity is empty.

Activity recording is turned off.

Turn recording back on

See more...