We screened 15 159 titles and collected 926 full text papers. Forty eight randomised controlled trials and 41 analyses of 26 cohort studies fulfilled all inclusion criteria (; for a complete set of references see bmj.com
). shows the main characteristics of the included studies.
The selection process for randomised controlled trials (RCTs) and cohort studies on omega 3 fatty acid and health outcomes
Characteristics of trials and cohort studies on effects of omega 3 fats on mortality, cardiovascular disease (CVD), and cancer
Intervention or exposure
Dietary supplements were given in 44 trials (36 as capsules, six as oil, one each as liquid emulsion and enriched margarine), advice on eating oily fish in three, and advice on diet and food supplements in one. Supplements were long chain omega 3 fats (usually whole or concentrated fish oil; one small trial used refined eicosapentaenoic acid and one used refined docosahexaenoic acid), and five studies provided shorter chain omega 3 fats. Doses of long chain omega 3 fats (summing eicosapentaenoic acid, docosahexaenoic acid, and docosapentaenoic acid) varied from 0.4 g to 7.0 g per day. Control groups received vegetable oils, other fats, “inert” or ill defined substances, different dietary advice, or nothing. The intervention lasted 6-11 months in 23 studies, 12-23 months in 16, 24-47 months in eight, and ≥ 48 months in one study.
Intake of omega 3 (varying combinations of eicosapentaenoic acid, docosahexaenoic acid, docosapentaenoic acid, along with α linolenic acid, supplemental fish oils, or dietary oily fish) was assessed by dietary and biochemical means in two cohorts, dietary means only in 18, and biochemical means only in 10. Groups with the lowest and highest intake of long chain omega 3 differed by 0.1-0.6 g omega 3 per day.
Twenty five RCTs were rated as having a low risk of bias (). Losses to follow-up were unclear in 16 cohort studies. In 15 cohort studies the outcome assessors were blinded to exposure, in two they were not blinded, and in nine blinding was unclear.
Quality assessment of randomised controlled trials of the effect of omega 3 fats on mortality, cardiovascular disease, and cancer
In the seven cohort studies that described omega 3 intake at baseline (five assessed long chain omega 3 fats only, one short chain omega 3 fats only, and one assessed both), the characteristics of participants with high and low intake of omega 3 fats differed. People who consumed most long chain omega 3 at baseline had an advantage with regard to lifestyle (smoking, diet, and exercise), interest in health, and social factors (education, living in town). Adjustment for these potential confounding factors may not have been adequate ().
Adjustment carried out in 10 cohorts that assessed effects of omega 3 fats on cancer
Deaths occurred in 15 RCTs (1995 deaths), and authors of 29 RCTs reported that no deaths occurred. Evidence that risk of death was reduced in participants randomised to omega 3 (relative risk 0.87, 95% confidence interval 0.73 to 1.03) was weak, and inconsistency was moderate (I2
= 42%) (). When analysis was restricted to studies at low risk of bias this effect was attenuated (0.98, 0.70 to 1.36; 138 deaths), and inconsistency between RCTs was low (I2
= 0%). This sensitivity analysis removed the RCT by Singh, whose studies have been questioned.18
Fig 2 Effect of omega 3 fatty acids on mortality. For references see bmj.com
Results were similar for long chain versus short chain omega 3 () and dietary advice versus supplements (data not shown). Meta-regression indicated that the risk of death increased as the length of the RCT increased (regression coefficient 0.008, 0.003 to 0.012). This is compatible with omega 3 fats having an early protective effect that later becomes harmful; however, the association was lost when we removed the large trial by Burr et al.14
Meta-regression did not suggest a relation between mortality and the dose of long chain omega 3. Cohort studies suggested that omega 3 protected against death (0.65, 0.48 to 0.88; I2
= 36%), but it was unclear whether adjustment for confounders was adequate.
Combined cardiovascular events
Eighteen RCTs provided data on cardiovascular events in 2628 participants. The meta-analysis showed no definite effect of omega 3 fats on cardiovascular events, but confidence intervals were wide (0.95, 0.82 to 1.12) and inconsistency was high (I2 = 65%) (). Removing studies at moderate or high risk of bias reduced but did not remove inconsistency (1.09, 0.87 to 1.37; 570 events; I2 = 49%).
Fig 3 Effect of omega 3 fatty acids on cardiovascular events. For references see bmj.com
Subgrouping by long chain versus short chain omega 3 or by advice to eat oily fish versus supplements did not generate robust effects of omega 3 fats on cardiovascular events. Cohort studies provided no strong evidence that omega 3 fats protect against cardiovascular events.
Ten RCTs reported the incidence of cancer; 391 diagnoses of cancer or death from cancer occurred in 17 433 participants (two of the trials reported no cancers). We found no evidence that omega 3 fats had an effect on the incidence of cancer (1.07, 0.88 to 1.30) and there was no inconsistency (I2 = 0%) (). Five trials and seven events remained on sensitivity analysis.
Fig 4 Effect of omega 3 fatty acids on the diagnosis of cancer or death from cancer. For references see bmj.com
Seven cohort studies provided data on cancer (832 events in the highest and lowest quantiles), and meta-analysis found no effect of high versus low intake of omega 3 (1.02, 0.87 to 1.19; I2 = 21%).
Outcomes related to bleeding
Nine RCTs reported at least one stroke (243 strokes in total), but little information was available specifically on haemorrhagic stroke. Omega 3 had no clear effect on the total numbers of strokes (1.17, 0.91 to 1.51; I2 = 0%), in sensitivity analysis (29 events), or in four cohort studies (0.87, 0.72 to 1.04) ().
Fig 5 Effect of omega 3 fatty acids on stroke. For references see bmj.com
Our meta-analysis of RCTs assessing the effects of increased omega 3 fats on total mortality found substantial variations between studies. Studies with stronger methodology had more consistent results, and the pooled relative risk of these studies was 0.98 (0.70 to 1.36; 138 events). We found no evidence from RCTs or cohort studies that omega 3 fats have an effect on combined cardiovascular events. Neither RCTs nor cohort studies showed significantly increased risks of cancer or stroke with higher intake of omega 3, but there were too few events to rule out important effects.
Strengths and weaknesses
The largest studies reviewed had greater potential for bias than some of the smaller ones. We hoped that pooling studies at low risk of bias might provide enough power to inform us of effects on health, but this was not the case (only 138 deaths and 570 cardiovascular events). Similarly, analysis of the effects of omega 3 on rarer outcomes such as stroke had insufficient power to detect clinically important effects. Unlike previous metaanalyses, we reviewed systematically the effects of omega 3 fats on mortality, cardiovascular disease, cancer, and bleeding events and analysed all relevant RCTs and prospective cohort studies. We also accounted for differences in study quality and examined the effects of long chain and short chain omega 3 fats in a wide group of participants; this provides high quality evidence to guide policy and practice.
Our findings differ from those of a recent systematic review by Bucher et al,13
which reviewed trials assessing the effects of long chain omega 3 fats over at least six months in patients with coronary heart disease and found significant protection from mortality (0.8, 0.7 to 0.9) and sudden death (0.7, 0.6 to 0.9). Bucher et al analysed 11 RCTs (nine included here, two excluded as multifactorial interventions) with 15 806 participants and 1335 deaths but did not include the recent study by Burr et al with 3114 participants included in our study.14
A systematic review from the United States collated RCTs and cohort studies to assess the effects of marine omega 3 fats, but this study did not perform meta-analysis, reported the results of the largest studies only, and did not investigate reasons for conflicting results.19
A recent review from the United Kingdom of the benefits and risks of consumption of fish on cardiovascular disease mentioned only four large trials and did not pool the results.20
Our inclusion of studies with short chain omega 3 as the active intervention or with participants who had a lower risk of cardiovascular disease does not explain the differences between our results and those of Bucher et al. When we pooled studies in which participants were given only long chain omega 3 fats or were at high risk of cardiovascular disease the risk of death was not significantly reduced. However, when we removed the study by Burr et al from our meta-analysis, risk of death was similar to that reported by Bucher et al (0.83, 0.75 to 0.91). The study by Burr et al (included in our review as an unpublished study, since published) is the second largest study in terms of deaths reported (525 deaths; the GISSI-P study reports 1031 deaths).21
Why does the study by Burr et al contradict the other large studies by not suggesting a benefit of omega 3?14
The first possible explanation is that this RCT had the longest follow-up of all RCTs and the harmful effects of methylmercury could be cumulative. Time course analysis of the GISSI-P trial was consistent with the risk of death rising over the 42 months of the trial, but the increase in mortality was seen within the first year in the study by Burr et al.14,22
A second explanation is that the study by Burr et al was the only RCT that specifically enrolled men treated for angina. Post hoc subgroup analysis of the GISSI-P trial indicates that people with heart failure benefited most from omega 3 supplements (R Marchioli, personal communication, 2004). As heart failure is more common after myocardial infarction this may explain an attenuation of effect but would not explain the increase in risk suggested in the study by Burr et al. A third possibility is that omega 3 from oily fish has a different effect to fish oil supplements, but this was investigated by Burr et al and found not to explain the differences. It is therefore not clear why the results of Burr et al differ from the other large studies on fish based omega 3. It is possible that performance bias due to differential care in the intervention and control arms occurred in trials where the intervention was not masked, including GISSI-P and the Burr study,14,21,23
but dietary intake, other relevant risk factors, and pharmacotherapy appeared similar in both arms of these trials. It may be that the effect of omega 3 fats on cardiovascular disease is smaller than previously thought, or that its beneficial effect is limited to a specific group (such as patients after myocardial infarction or with heart failure) not represented in the study by Burr et al.14
Cohort studies or RCTs?
A recent meta-analysis assessed the effects of consumption of fish on stroke in cohort studies and found that people who ate white or oily fish at least once a week had a significantly reduced risk of stroke.24
We excluded cohort studies that assessed only total fish intake (as this is not clearly related to omega 3 intake). The web of lifestyle, interest in health, and social factors (health patterning) seen in the cohort studies included in our review provides an advantage to people taking most long chain omega 3 fats, and this makes adequate adjustment for confounding difficult, if not impossible. Thus, we must rely on high quality RCTs to provide non-confounded answers about the effects of omega 3 fats on health. Some effects of fish on health may be due to components other than omega 3—for example, selenium or vitamin D.
It is not clear whether long chain or short chain omega 3 fats (together or separately) reduce or increase total mortality, cardiovascular events, cancer, or strokes. Our findings do not rule out an important effect of omega 3 fats on total mortality, as robust trials at low risk of bias reported few deaths. There is no evidence that the source (dietary or supplemental) and dose of omega 3 fats affected the effectiveness of long chain omega 3 fats.
What is already known on this topic
A systematic review of randomised controlled trials in coronary heart disease showed reduced mortality in patients taking supplemental long chain omega 3 fats
What this study adds
This systematic review assessed the health effects of long chain and shorter chain omega 3 fats (together or separately) on total mortality, cardiovascular events, cancer, and strokes in a wide group of participants and found no evidence of a clear benefit of omega 3 fats on health
UK guidelines encourage the general public to eat more oily fish, and higher amounts are advised after myocardial infarction (supported by trials after myocardial infarction).20,25,26
This advice should continue at present but the evidence should be reviewed regularly. It is probably not appropriate to recommend a high intake of omega 3 fats for people who have angina but have not had a myocardial infarction.
Adjustment for lifestyle factors appeared to be inadequate in the cohort studies, so policy and lifestyle decisions should be based on data from RCTs. To understand the effects of omega 3 fats on health, we need more high quality RCTs (with adequate concealment of allocation and masking of participants and health providers) of long duration that also report the associated harms.