Ten RCTs (n=14,727) were included.
Five trials were considered to be of good quality as they satisfied all six quality criteria; four were rated poor because they either had an open design or did not mention intention-to-treat analysis; one trial was considered to be of moderate quality because the description of the double-blinding procedure was not sufficient. Seven studies reported no losses to follow-up; other studies reported losses from 0.1 to 13%.
All-cause mortality.
There was a statistically significant reduction of 16% in the risk of death for patients taking fish oils compared with placebo (6 studies; RR 0.84, 95% CI: 0.76, 0.94). There was no evidence of heterogeneity between these studies (P=0.68).
Fatal MI.
There was a statistically significant reduction of 24% in the risk of death due to MI for patients taking fish oils compared with placebo (4 studies; RR 0.76, 95% CI: 0.66, 0.89). There was no evidence of heterogeneity between these studies (P=0.59).
Other outcomes.
There were no statistically significant differences between the fish oil and placebo groups for any of the other outcomes: nonfatal MI (7 studies; RR 1.03, 95% CI: 0.87, 1.19, corresponding to a 3% increase in risk for fish oils), nonfatal stroke (3 studies; RR 1.36, 95% CI: 0.87, 2.29, corresponding to a 36% increase in risk for fish oils), and angina pectoris (7 studies; RR 1.03, 95% CI: 0.85, 1.20, corresponding to a 3% increase in risk for fish oils). There was no evidence of heterogeneity between the studies for any of these outcomes (P=0.35, P=0.39 and P=0.4 respectively).
Other analyses.
The sensitivity analyses showed that the two largest trials were having the most impact upon the results for all-cause mortality and fatal MI; deleting both these trials from the analysis meant the results were no longer statistically significant. Deleting other trials did not affect the results for these or any other outcomes. The test for publication bias showed that 11 or more trials showing no treatment effect would have to be added to produce a non significant result for fatal MI and all-cause mortality.