We reviewed databases of Pubmed, EMBASE, Cochrane Library and Web of Science for eligible studies from the inception to Aug-15-2022. The combined search strategy of relevant keywords and Medical Subject Headings (MeSH) terms used in current study are: “Omega-3 fatty acids”, “docosahexaenoic acid”, “DHA”, “Eicosapentaenoic acid”, “EPA”, “cardiovascular disease”, “cardiovascular events”, “coronary heart disease”, “myocardial infarction”, “stroke” and “randomized controlled trial”. A detailed search strategy has been given in Table 1. No special restrictions were applied to language. Reference lists of the retrieved literature were also searched manually.

All searched articles went through a two-step review process. They were initially screened for titles and abstracts. Then, the full texts of possibly eligible studies were reviewed by two independent authors (Xue Qi and Hao Huang). Any disagreements were resolved by a discussion in a group panel with another author (Ru Ya), who is familiar with cardiology and evidence-based medicine.

The eligible criteria following the PICOS principles were listed as:

Populations: Adult populations ($\ge$18 yr) with CVD or high-risk factors (e.g., smoking, obesity, lack of physical activity, etc.) for CVD; and no restrictions on their gender, race, nationality and CV-related comorbidities (e.g., diabetes, hypertension, kidney circulation dysfunction).

Intervention/comparison: Omega-3 PUFA from dietary supplements, capsules or drug prescriptions was used. Considering the difficulty in quantifying n-3 PUFA intake from marine fish food sources, Omega-3 PUFA directly derived from these resources was considered not eligible.

Outcomes: At least one of the following outcomes was reported with available data for calculating: major adverse cardiovascular events (MACE), myocardial infarction (MI), CHD, revascularization, stroke, sudden cardiac death (SCD), CV mortality, all-cause mortality, hospitalization, hospitalization for all heart disease, hospitalization for heart failure, and atrial fibrillation (AF).

Study design: Randomized controlled trial (RCT).

The same trial with a longer follow-up period could be included to avoid duplication. Eligible RCTs should have a registered protocol, and provide ethics approval and consent of individuals. Observational studies, reviews, case reports, conference abstracts and experimental studies were excluded. Studies without essential data were also excluded.

Data extraction was performed by two independent authors (Xue Qi and Ru Ya) following a pre-ruled protocol from included studies. The extracted information included characteristics of eligible studies (year of publication, first author name, performed country, trial name, follow-up period, etc.), characteristics of the populations (gender (proportion of male), mean age (SD) and sample size (in experimental and control groups), etc.), and the characteristics of the program (interventions in two groups (n-3 PUFA or placebo or other dietary supplements), dose of n-3 PUFA (1–4 g/d), type of n-3 PUFA (EPA + DHA and EPA alone), prevention type (secondary and mixed), registration number, etc.). The risk estimates of hazard ratio (HR) relative risk (RR) and odds ratio (OR) would be evaluated in fully adjusted models if available. If not, the unadjusted models would be evaluated, and special descriptions would be given. Intentional-to-treat (ITT) principles would be applied if available. The authors would contact the primary authors for some missing data to facilitate the current analysis, and the current study would still have been taken without these data if no response was received.

Herein, outcomes including MACE, MI, CHD, revascularization, stroke, SCD, CV mortality, all-cause mortality, hospitalization, hospitalization for all heart disease, and hospitalization for heart failure and AF were analyzed. Details about the definitions on these outcomes were summarized in Supplementary Table 3. Briefly, MACE indicated a composite of MI, stroke, cardiac death or any revascularization; MI included fatal and no-fatal MI; stroke included fatal and no-fatal stroke; and AF meant new AF events.

For evaluating the quality of included studies, we applied the Cochrane Risk of Bias Tool, which has been widely used for assessing the methodological quality of RCTs in meta-analyses [34]. Seven specific bars in the Cochrane Risk of Bias Tool were objectively evaluated by two independent authors (Xue Qi and Ru Ya) including the generation of randomized sequences, concealment of allocation protocols, blinding of study participants and related persons, blinding of outcome evaluators, incomplete data on study results, selective reporting of results and other sources of bias. If each bar from the Cochrane Risk of Bias Tool was not available or wrongly conducted, assessment on the bar would be high risk.

Fully adjusted HR and the corresponding 95% confidence intervals (95% CIs) for the outcomes of interests obtained from Cox-Hazard regression analysis were mainly estimated with DerSimonian-Laird (D-L) random effects model because the assumptions might be attributed to the presence of whining-study and between-study heterogeneity. The adjusted/unadjusted RR and OR in primarily included studies were approximately considered as HR. HRs and standard errors (SEs) originating from the correspondence 95% CIs were logarithmically transformed to stabilized variance, and the distribution then was normalized. Between-study heterogeneity was determined with the Cochran Q chi-square test and the I${}^{\mathrm{2}}$. An I${}^{\mathrm{2}}$ $>$50% or a p value for the Q test 0.1 was regarded as equal to significant heterogeneity [35].

Sensitivity analysis would be performed by removing one study each turn to reduce and elaborate the causes of the heterogeneity in the case of significant heterogeneity. Post-subgroup analyses were also conducted to ascertain the influence of other risk factors on the outcome results on MACE, CV mortality and all-cause mortality, since there were abundantly included studies on those outcomes. According to the main characteristics of the populations and trial, the subgroups were identified as follows: the proportion of statin use populations (50% vs. $\ge$50%) in each trial, proportion of antiplatelet drug use populations (50% vs. $\ge$50%) in each trial, n-3 PUFA formulations (EPA + DHA vs. EPA) in each trial, actual amount of n-3 PUFA intake (2 g/d vs. $\ge$2 g/d) in each trial and prevention type (primary prevention vs. secondary prevention vs. mixed prevention) in each trial. The analyses results of the subgroup were visualized by forest plots.

Publication bias was estimated using Begg’s correlation test and Egger’s linear regression test at p 0.10 indicating significant publication bias [36]. All analyses were performed using Stata software version 12.0 (https://www.stata.com/); two-sided p 0.05 was considered statistically significant. When 95% CIs of HR were on 1.00, the p value for HR would be checked, with p 0.05 indicating statistical significance.

Of 4772 studies searched from databases, 1865 came from PubMed, 1424 from EMBASE, 515 from Cochrane Library, and 968 from Web of Science. Additionally, 33 were further achieved from other literature available. Then, 4634 records were excluded after initial screening, 12 new records were complemented by reviewing reference list when making the initial screening, and 18 were excluded after full-text consideration due to no outcome of interest or definition, duplicated study, no useful data or no n-3 PUFA intake. Finally, a total of 19 studies were eligible for systematic review and meta-analysis, and the selection process and exclusion reasons could be found in Fig. 1.

Fig. 1.

The flow chart for study screening and selection.

Totally, 19 RCTs incorporating 116,498 populations were considered eligible and included in current systematic review and meta-analysis (Table 2, Ref. [6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24]). Ten studies [6, 7, 9, 11, 12, 13, 15, 18, 21, 24] were conducted in Europe, 1 [19] in the USA, 2 [10, 20] in Asia, and the other 6 [8, 14, 16, 22, 23, 24] were performed on multicenter individuals. Only one study [12] was conducted on all male populations. Among all 19 clinical trials, proportion of statin use $\ge$50% among individuals was observed in 10 studies [10, 13, 14, 15, 16, 20, 21, 22, 23, 24], and 50% in 8 studies [6, 7, 8, 9, 11, 12, 18, 19]; proportion of antiplatelet drug use $\ge$50% among individuals was observed in 11 studies [6, 9, 11, 13, 14, 15, 16, 17, 20, 23, 24], and 50% in 4 studies [7, 10, 18, 21]. Almost all included studies [6, 7, 8, 9, 11, 12, 13, 15, 16, 17, 18, 19, 21, 23, 24] used combined EPA + DHA for n-3 PUFA supplementation, three [10, 20, 22] used only EPA for n-3 PUFA intake, and one [14] used EPA + DHA + ALA for n-3 PUFA intake. Six studies [7, 8, 12, 22, 23, 24] provided $\ge$2 g/d n-3 PUFA for the included populations, and 9 [6, 9, 10, 11, 15, 16, 17, 18, 21] provided 2 g/d n-3 PUFA for the included populations. As for the control group prescription, 16 studies [7, 8, 9, 11, 12, 13, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24] used placebo only, one [6] did not provide any treatment for participants in the control group, one [10] provided standard of care, and one [14] provided placebo and ALA. Eight studies [6, 7, 9, 13, 14, 15, 20, 24] were designed for the secondary prevention of CVD, 10 [8, 10, 11, 12, 16, 17, 18, 19, 22, 23] for mixed (primary and secondary) prevention for CVD, and only one [21] for the primary prevention for CVD. The mean follow-up period was 3.4 yr. Besides, three studies [15, 18, 19] reported unadjusted results, while the rest reported fully adjusted results based on adjustment on variables such as age, serum glucose, body mass index, systolic blood pressure, and smoking status at baseline.

In terms of the study methodological quality, both Nilsen et al. [7] and Einvik et al. [12] failed to provide detailed descriptions on the blinding methods. Nilsen et al. [7] did not provide evidence to support the process of randomization and allocation concealment; Einvik et al. [12] did not provide any information about allocation concealment as well. Bowman et al. [21] carried out an open-label study without taking blinding for participants and outcome assessments. Brouwer et al. [8], Kromhout et al. [14] and Bonds et al. [19] provided incomplete outcome data. Other sources of bias remained unclear in Marchioli et al. [6], Svensson et al. [9], Einvik et al. [12], Macchia et al. [17], Bonds et al. [19] and Nosaka et al. [20](Supplementary Table 4).

Thirteen studies [7, 9, 12, 13, 14, 15, 16, 17, 18, 21, 22, 23, 24] with 75,611 participants (37,804 in the n-3 PUFA group and 37,807 in the control group) on MACE outcome showed risks for MACE could not be significantly reduced by n-3 PUFA (HR: 0.98, 95% CI: 0.91–1.06; p = 0.592) with significant heterogeneity (I${}^{\mathrm{2}}$ = 62.7%; p = 0.001) (Fig. 2A). After removing one heterogeneous study [22] (8179 participants), the HR turned to 1.00 (95% CI: 0.96–1.05; p = 0.872) with little heterogeneity (I${}^{\mathrm{2}}$ = 0.0%; p = 0.955). In subgroup analysis, n-3 PUFA could not significantly reduce MACE on the prevention type (secondary prevention (6 [6, 7, 13, 14, 15, 24]): 1.07, 95% CI (0.97–1.18); mixed prevention (n = 6 [12, 16, 17, 18, 22, 23]): 0.92, 95% CI (0.82–1.04); only one study [21] for primary prevention: (0.97, 95% CI: 0.87–1.08), statin use proportion in trial (50% (n = 4 [7, 9, 12, 18]): 0.99, 95% CI: 0.90–1.09; $\ge$50% (n = 8 [13, 14, 15, 16, 21, 22, 23, 24]): 0.98, 95% CI: 0.89–1.09), antiplatelet drug use proportion in trial (50% (n = 3 [7, 18, 21]): 0.98, 95% CI: 0.91–1.05; $\ge$50% (n = 8 [9, 13, 14, 15, 16, 17, 23, 24]): 1.02, 95% CI: 0.96–1.07) or actual n-3 PUFA intake amount (2 g/d (n = 6 [9, 15, 16, 17, 18, 21]): 1.00, 95% CI: 0.96–1.05; $\ge$2 g/d (n = 5 [7, 12, 22, 23, 24]): 0.93, 95% CI: 0.78–1.12). Only one study [22] was included for EPA use analysis (HR: 0.75, 95% CI: 0.68–0.83). Therefore, more evidence was still required for this result (Fig. 2B). Thank you.

Fig. 2.

Forrest plots and subgroup analyses for MACE. (A) Forest plot of main result on MACE. Each bar and the middle diamond represented HR and 95% CI for each included study with the detailed value marked on right. The bottom diamond represented the synthesized result, if the whole diamond located on the left of vertical full line (on 1.00), the risk of MACE was significantly reduced, if the whole diamond located on the right of vertical full line, the risk of MACE was significantly improved, otherwise there was no statistically significant association. Vertical dashed line in red represented location of synthesized HR from which we could presume the trend of the association. (B) Forest plot of subgroup analysis. Each outcome included a couple of subgroups, the reference list of studies in each subgroup has been marked in the forest plot, synthesized result with 95% CI for each subgroup has been visualized on the right. MACE, major adverse cardiovascular events; EPA, eicosapentaenoic acids; DHA, docosahexaenoic acids; HR, hazard ratio; 95% CI, 95% confidence interval.

Eight studies [7, 8, 9, 10, 11, 16, 22, 24] involving 48,401 participants (24,221 in the n-3 PUFA group and 24,180 in the control group) reported MI outcome. The results indicated that the existence of risks for MI could not be significantly reduced by n-3 PUFA (HR: 0.86, 95% CI: 0.70–1.05; p = 0.137) with significant heterogeneity (I${}^{\mathrm{2}}$ = 70.5%; p = 0.001) (Fig. 3). After removing 3 heterogeneous studies [9, 16, 22] (20,921 participants), the HR kept stable from 0.86 (95% CI: 0.70–1.05; p = 0.137) to 0.86 (95% CI: 0.72–1.03; p = 0.112) with a shortened confidence interval and little heterogeneity (I${}^{\mathrm{2}}$ = 15.5%; p = 0.316). Main heterogeneity across analysis on MI was found among the three studies [9, 16, 22].

Fig. 3.

Forest plot for MI. Forest plot of main result on MI. Each bar and the middle diamond represented HR and 95% CI for each included study with the detailed value marked on right. The bottom diamond represented the synthesized result, if the whole diamond located on the left of vertical full line (on 1.00), the risk of MI was significantly reduced, if the whole diamond located on the right of vertical full line, the risk of MI was significantly improved, otherwise there was no statistically significant association. Vertical dashed line in red represented location of synthesized HR from which we could presume the trend of the association. MI, myocardial infarction; HR, hazard ratio; 95% CI, 95% confidence interval.

Six studies [7, 9, 10, 13, 18, 23] involving 47,243 participants (23,615 in the n-3 PUFA group and 23,628 in the control group) reported results on CHD, and n-3 PUFA had the trend to reduce the incidence of CHD, but the statistic was not significant (HR: 0.90, 95% CI: 0.80–1.01; p = 0.079) with little heterogeneity (I${}^{\mathrm{2}}$ = 37.3%; p = 0.158) (Fig. 4). Little intra-study heterogeneity restricted the implementation of sensitivity analysis.

Fig. 4.

Forest plot for CHD. Forest plot of main result on CHD. Each bar and the middle diamond represented HR and 95% CI for each included study with the detailed value marked on right. The bottom diamond represented the synthesized result, if the whole diamond located on the left of vertical full line (on 1.00), the risk of CHD was significantly reduced, if the whole diamond located on the right of vertical full line, the risk of CHD was significantly improved, otherwise there was no statistically significant association. Vertical dashed line in red represented location of synthesized HR from which we could presume the trend of the association. CHD, coronary heart disease; HR, hazard ratio; 95% CI, 95% confidence interval.

Nine studies [7, 13, 15, 16, 20, 21, 22, 23, 24] were included for analysis on revascularization with a total of 57,144 participants analyzed (28,601 in the n-3 PUFA group and 28,543 in the control group). N-3 PUFA could significantly reduce the incidence of revascularization (HR: 0.90, 95% CI: 0.81–1.00; p = 0.006), although the upper 95% CI was on 1.00, and the p value for HR was 0.006 (0.05). Significant heterogeneity was observed (I${}^{\mathrm{2}}$ = 62.6%; p = 0.006) (Fig. 5). After removing three heterogeneous studies [20, 22, 24] (3431 participants), the HR changed from 0.90 (95% CI: 0.81–1.00; p = 0.006) to 0.96 (95% CI: 0.91–1.02; p = 0.221) with little heterogeneity observed (I${}^{\mathrm{2}}$ = 0.0%; p = 0.914). Synthesized results on revascularization were not robust since the results changed after heterogeneous studies were removed. In this case, more relevant studies are still needed to confirm the controversial results.

Fig. 5.

Forest plot for revascularization. Forest plot of main result on revascularization. Each bar and the middle diamond represented HR and 95% CI for each included study with the detailed value marked on right. The bottom diamond represented the synthesized result, if the whole diamond located on the left of vertical full line (on 1.00), the risk of revascularization was significantly reduced, if the whole diamond located on the right of vertical full line, the risk of revascularization was significantly improved, otherwise there was no statistically significant association. Vertical dashed line in red represented location of synthesized HR from which we could presume the trend of the association. HR, hazard ratio; 95% CI, 95% confidence interval.

Nine trials [6, 9, 10, 11, 13, 16, 21, 22, 24] involving 76,860 participants (38,457 and 38,403 in n-3 PUFA and control groups, respectively) were eligible for stroke outcome analysis, and n-3 PUFA exerted little effect on reducing stroke incidence (HR: 1.00, 95% CI: 0.91–1.10; p = 0.967) with little heterogeneity (I${}^{\mathrm{2}}$ = 34.6%; p = 0.141) (Fig. 6).

Fig. 6.

Forest plot for stroke. Forest plot of main result on stroke. Each bar and the middle diamond represented HR and 95% CI for each included study with the detailed value marked on right. The bottom diamond represented the synthesized result, if the whole diamond located on the left of vertical full line (on 1.00), the risk of stroke was significantly reduced, if the whole diamond located on the right of vertical full line, the risk of stroke was significantly improved, otherwise there was no statistically significant association. Vertical dashed line in red represented location of synthesized HR from which we could presume the trend of the association. HR, hazard ratio; 95% CI, 95% confidence interval.

Six studies [6, 7, 10, 11, 15, 18] involving 53,575 participants were analyzed for SCD (26,805 and 26,770 in the n-3 PUFA group and the control group, respectively). It was found that n-3 PUFA could not improve the outcome of SCD (HR: 0.90, 95% CI: 0.80–1.02; p = 0.111) with little heterogeneity (I${}^{\mathrm{2}}$ = 3.2%; p = 0.396) (Fig. 7).

Fig. 7.

Forest plot for SCD. Forest plot of main result on SCD. Each bar and the middle diamond represented HR and 95% CI for each included study with the detailed value marked on right. The bottom diamond represented the synthesized result, if the whole diamond located on the left of vertical full line (on 1.00), the risk of SCD was significantly reduced, if the whole diamond located on the right of vertical full line, the risk of SCD was significantly improved, otherwise there was no statistically significant association. Vertical dashed line in red represented location of synthesized HR from which we could presume the trend of the association. SCD, sudden cardiac death; HR, hazard ratio; 95% CI, 95% confidence interval.

Thirteen studies [6, 7, 10, 11, 14, 15, 16, 18, 19, 20, 21, 22, 23] were selected to calculate CV mortality. Totally, 111,082 participants were included, among which, 56,051 were in the n-3 PUFA group and 55,031 in the control group. The synthesized results showed that n-3 PUFA intake could significantly reduce CV mortality (HR: 0.91, 95% CI: 0.85–0.97; p = 0.003) with little heterogeneity (I${}^{\mathrm{2}}$ = 20.5%; p = 0.236) (Fig. 8A). In subgroups in terms of secondary prevention (HR: 0.85, 95% CI: 0.75–0.97; n = 5 [6, 7, 14, 15, 20]), 50% (HR: 0.90, 95% CI: 0.84–0.97; n = 5 [6, 7, 11, 18, 19]) and $\ge$50% (HR: 0.90, 95% CI: 0.81–1.00; n = 8 [10, 14, 15, 16, 20, 21, 23]) population with statin use, 50% population with antiplatelet use (HR: 0.87, 95% CI: 0.77–1.00; n = 4 [7, 10, 18, 21]), EPA + DHA intake (HR: 0.93, 95% CI: 0.88–0.98; n = 10 [6, 7, 11, 14, 15, 16, 18, 19, 21, 23]), only EPA intake (HR: 0.78, 95% CI: 0.66–0.91; n = 3 [10, 20, 22]), and 2 g/d n-3 PUFA intake (HR: 0.90, 95% CI: 0.85–0.96; n = 7 [6, 10, 11, 15, 16, 18, 21]), n-3 PUFA could significantly reduce CV mortality. No statistical significance was observed in other subgroups (Fig. 8B), and most results of subgroup analyses were consistent with the total result on CV mortality.

Fig. 8.

Forest plot for CV mortality. (A) Forest plot of main result on CV mortality. Each bar and the middle diamond represented HR and 95% CI for each included study with the detailed value marked on right. The bottom diamond represented the synthesized result, if the whole diamond located on the left of vertical full line (on 1.00), the risk of CV mortality was significantly reduced, if the whole diamond located on the right of vertical full line, the risk of CV mortality was significantly improved, otherwise there was no statistically significant association. Vertical dashed line in red represented location of synthesized HR from which we could presume the trend of the association. (B) Forest plot of subgroup analysis. Each outcome included a couple of subgroups, the reference list of studies in each subgroup has been marked in the forest plot, synthesized result with 95% CI for each subgroup has been visualized on the right. EPA, eicosapentaenoic acids; DHA, docosahexaenoic acids; HR, hazard ratio; 95% CI, 95% confidence interval.

There were 15 studies [6, 7, 8, 9, 10, 12, 13, 14, 15, 16, 17, 21, 22, 23, 24] involving 93,613 participants (46,825 in the n-3 PUFA group and 46,788 in the control group) included for the meta-analysis for all-cause mortality, and n-3 PUFA could not significantly reduce the risk for all-cause mortality (HR: 0.96, 95% CI: 0.89–1.04; p = 0.339). Significant heterogeneity was found (I${}^{\mathrm{2}}$ = 54.2%; p = 0.006) (Fig. 9A). After removing three heterogeneous studies [12, 22, 24] (9756 participants), the HR turned to 0.98 (95% CI: 0.94–1.04; p = 0.544) with little heterogeneity (I${}^{\mathrm{2}}$ = 8.4%; p = 0.363) and a shortened confidence interval. Results on all-cause mortality kept robust through sensitivity analysis. As for subgroup analysis, on trials of 50% population with statin use (HR: 0.78, 95% CI: 0.66–0.91; n = 5 [6, 7, 8, 9, 12]), n-3 PUFA could reduce the risk for all-cause mortality. However, statistics was not significant in other groups (Fig. 9B).

Fig. 9.

Forest plot for all-cause mortality. (A) Forest plot of main result on all-cause mortality. Each bar and the middle diamond represented HR and 95% CI for each included study with the detailed value marked on right. The bottom diamond represented the synthesized result, if the whole diamond located on the left of vertical full line (on 1.00), the risk of all-cause mortality was significantly reduced, if the whole diamond located on the right of vertical full line, the risk of all-cause mortality was significantly improved, otherwise there was no statistically significant association. Vertical dashed line in red represented location of synthesized HR from which we could presume the trend of the association. (B) Forest plot of subgroup analysis. Each outcome included a couple of subgroups, the reference list of studies in each subgroup has been marked in the forest plot, synthesized result with 95% CI for each subgroup has been visualized on the right. EPA, eicosapentaenoic acids; DHA, docosahexaenoic acids; HR, hazard ratio; 95% CI, 95% confidence interval.

Two studies [11, 17] involving a total of 7571 participants were included in analysis on hospitalization (3783 in the n-3 PUFA group and 3788 in the control group), and n-3 PUFA could not significantly reduce the hospitalization incidence (HR: 0.99, 95% CI: 0.81–1.20; p = 0.884) with little heterogeneity found (I${}^{\mathrm{2}}$ = 33.1%; p = 0.221) (Supplementary Fig. 1).

Five studies [11, 16, 18, 20, 22] were obtained to synthesize the results on hospitalization for all heart diseases. In detail, 40,441 participants were totally included, with 20,227 in the n-3 PUFA group and 20,214 in the control group. N-3 PUFA presented the signal to reduce the risk for hospitalization for all heart diseases (HR: 0.91, 95% CI: 0.83–1.00; p = 0.059) with significant heterogeneity (I${}^{\mathrm{2}}$ = 70.9%; p = 0.008), but the statistic was not significant (Supplementary Fig. 2). After removing two heterogeneous studies [20, 22] (8417 participants), the HR turned to 0.96 (95% CI: 0.92–1.00; p = 0.048) with little heterogeneity (I${}^{\mathrm{2}}$ = 0.0%; p = 0.475). More evidence on the outcome of hospitalization for all heart diseases is still needed, because the upper 95% CIs are on 1.00 and the p values for HR are close to 0.05.

Six studies [11, 16, 17, 20, 23, 24] reported outcomes on hospitalization for heart failure. A total of 34,427 participants, with 17,227 in the n-3 PUFA group and 17,200 in the control group, were analyzed, and the integrated results showed that n-3 PUFA could not decrease the incidence for hospitalization for heart failure (HR: 0.97, 95% CI: 0.91–1.04; p = 0.450) with little heterogeneity (I${}^{\mathrm{2}}$ = 0.0%; p = 0.715) (Supplementary Fig. 3).

Three studies [6, 23, 24] with 14,678 participants (7333 in the n-3 PUFA group and 7345 in the control group) reported the AF results, and the incidence of AF was significantly increased with n-3 PUFA intake (HR: 1.56, 95% CI: 1.27–1.91; p 0.001) with little heterogeneity (I${}^{\mathrm{2}}$ = 0.0%; p = 0.407) (Supplementary Fig. 4).

No publication biases were found across analyses on MACE (Begg’s test p = 0.428, Egger’s test p = 0.427), MI (Begg’s test p = 1.000, Egger’s test p = 0.776), CHD (Begg’s test p = 1.000, Egger’s test p = 0.858), revascularization (Begg’s test p = 0.048, Egger’s test p = 0.282), stroke (Begg’s test p = 0.348, Egger’s test p = 0.451), SCD (Begg’s test p = 1.000, Egger’s test p = 0.510), CV mortality (Begg’s test p = 0.760, Egger’s test p = 0.532), and all-cause mortality (Begg’s test p = 0.428, Egger’s test p = 0.598).