*2.1. Data Sources and Search Strategy*

Systematic search was performed on PubMed, Web of Science, Embase and MEDLINE from the establishment to May 2021. The search was limited to English literature, and the search keywords were "fish", "seafood", "fish products", "fish oil", "EPA", "eicosapentaenoic acid", "DHA", "docosahexaenoic acid", "DPA", "docosapentaenoic acid", "n-3 polyunsaturated fatty acid", "ω-3 polyunsaturated fatty acid", "n-3 PUFA", "ω-3 PUFA", "cardiovascular diseases", "CVD", "cardiovascular", "cohort", "follow-up", "prospective" and "longitudinal".

#### *2.2. Study Selection*

Two project members (L.J. and B.Z.) independently screened all titles and abstracts of the retrieved studies. Disagreements regarding the inclusion of the studies and the interpretation of the data were resolved by discussion among investigators. The studies were included in this meta-analysis if they met the following criteria: (1) study design: prospective cohort studies; (2) exposure: fish and marine n-3 PUFA; (3) source of n-3 PUFA: marine-derived n-3 PUFA (DHA, DPA, and EPA); and (4) outcomes: total CVD mortality which was reported as multivariate-adjusted relative risk (RR) and 95% confidence intervals (CI). The studies were excluded with the following criteria: (1) irrelevant; (2) not human studies; (3) not cohort studies; (4) not English studies.

#### *2.3. Data Extraction*

The following information was extracted from each eligible study: first author's surname; the year of publication; country; age; follow-up duration; the number of CVD deaths, sample size; gender; exposure levels; multivariate-adjusted RR with 95% CI for the highest versus the lowest category of fish or marine n-3 PUFA intake; adjusted covariates. Consumption of fish and marine n-3 PUFA was collected with adjusted RR (95% CI) to conduct dose-response analyses. Newcastle–Ottawa Quality Assessment Scale (NOS) was adopted to evaluate the quality of each included study [19]. The NOS score ranges from 0 (bad) to 9 (good).

The quality evaluation was performed independently by two project members (L.J. and B.Z.). The NOS quality score system assessed 3 items: population selection, comparability of the groups and outcome assessment. Any discrepancies in grading the quality were addressed by group discussion.

#### *2.4. Statistical Analyses*

All statistical analyses were performed using Stata (Version 15.1). RRs with 95% CI for all the exposure categories were extracted for the analysis. The main effect was RRs with 95% CI. A two-tailed *p* < 0.05 was considered as statistically significant. The summary estimation was conducted through the comparison of the highest and the lowest category. Heterogeneity was assessed using the I<sup>2</sup> statistic. In the case of heterogeneity for I2 > 50%, a random-effect model was adopted to pool the results. Otherwise, a fixed effect model was chosen.

Sensitivity analysis was implemented by deleting one study at a time. Subgroup analyses and meta-regression were performed to identify the possible sources of heterogeneity. In the subgroup analyses, the included studies were stratified by location (Asia, Europe plus America, Oceania and Five Continents), follow-up duration (<9 and ≥9 years), etc. In meta-regression, gender, country, dropout rate, follow-up duration, CVD history, adjustment for diabetes and adjustment for smoking were used as the covariates. Potential publication bias was accessed using funnel plots and Egger's test (*p* < 0.1 was considered statistically significant).

Non-linear dose-response analyses were performed to evaluate the relationship between fish, marine n-3 PUFA intake and CVD mortality risk [20]. Potential non-linear correlation was accessed by modeling the consumption level using restricted cubic splines. The distribution of four fixed knots were 5%, 35%, 65% and 95% [21]. Owing to the discrepancy of fish and marine n-3 PUFA intake categories, we selected studies with clear doses to perform dose-response analyses. Among each study, we used the median or mean consumption of fish and marine n-3 PUFA from each category. For open-ended categories, we set the lower boundary to zero in lowest category and the width of the category to be the same as the adjacent interval in the highest one [12,22].
