A comprehensive literature search was performed in August 2025 across PubMed, Embase, Scopus, and the Cochrane Library. The main search strategy aimed to broadly identify studies evaluating the effects of influenza vaccination on cardiovascular outcomes in patients with ischemic heart disease or heart failure. For all databases, we used a consistent approach that combined controlled vocabulary and free-text terms for key concepts such as ischemic heart disease, coronary artery disease, myocardial infarction, heart failure, and influenza vaccination. This ensured systematic and inclusive coverage of the literature from January 2005 to August 2025. Detailed, database-specific search strings are fully provided in the Supplementary Materials. We used the EndNote software and Rayyan to compile records and remove duplicates. Two reviewers (Y.D., P.D.) independently screened article titles and abstracts against predefined eligibility criteria. A third author (M.M.) resolved disagreements. We manually examined the reference lists of included articles to find any missed studies. We excluded trial registration databases, preprint servers, case reports, case series, letters, editorials, review articles, conference abstracts, and nonpeer-reviewed studies.

Studies were included if they met the following PIC structure: (1) Participants: Adults with ischemic heart disease or heart failure; (2) Intervention: Influenza vaccination (high- or standard-dose); (3) Comparison: Placebo or no vaccination; (4) Outcomes: Studies reporting clinical outcomes related to cardiovascular disease with at least 3 months of follow-up. Studies were included if they reported at least 1 relevant clinical outcome, such as all-cause mortality, cardiovascular mortality, or specific cardiovascular events (like MI or MACE), regardless of whether these were designated as primary, secondary, or exploratory endpoints. We excluded studies where cardiovascular events were only reported as incidental adverse events in the general population. This approach ensured our analysis focused on the outcomes most relevant to the high-risk IHD and HF population while maintaining a comprehensive data set. We excluded secondary analyses using data from other trials to avoid result duplication. We also excluded nonpeer-reviewed studies, abstracts only, and non-English publications to ensure data quality and consistency across included studies. Although initial searches were language-agnostic, we later limited inclusion to English due to practical constraints.

Two authors (A.A., P.R.) independently screened article full texts against eligibility criteria that were defined in advance and documented in a prespecified spreadsheet. Any disagreements were resolved by consensus with a third author (S.S.). We extracted the following variables from the included studies: study characteristics (e.g., author, year, sample size), patient characteristics (e.g., age, sex, comorbidities), and outcome data (e.g., hazard ratios, risk ratios, and 95% confidence intervals [CIs]). We organized the data in a Google spreadsheet. Further analysis was conducted using the merged data extraction file. Two authors (R.Y., Y.D.) independently assessed the study’s risk of bias. Differences were resolved by discussion with a third author (A.A.). We used the Robins I for observational studies and the Cochrane Collaboration’s Risk of Bias‐2 (RoB‐2) for clinical trials to assess risk of bias.

The primary outcome was all-cause mortality. Secondary outcomes included cardiovascular (CV) mortality, Major Adverse Cardiovascular Events (MACE) (as defined by each included study), myocardial infarction (MI), stroke, CV-related hospitalization, and all-cause hospitalization. To further explore the data, we conducted subgroup analyses based on study design (randomized controlled trials versus observational studies), patient population (ischemic heart disease versus heart failure), age (<70 years versus ≥70 years), and follow-up duration (short-term 3 to <12 months, midterm 12 to 36 months, and long-term >36 months).

Two complementary meta-analytic approaches were used. First, an IPD meta-analysis was conducted for the primary outcome. Second, conventional study-level meta-analyses were performed for both the primary and secondary outcomes.

For the IPD analysis, time-to-event data were reconstructed from published Kaplan-Meier (KM) curves using the method of Liu et al. KM plots were digitized to extract survival probabilities and numbers at risk. Reconstructed IPD were generated using the IPDfromKM package in R. Reconstruction accuracy was verified using root mean square error (RMSE), mean absolute error (MAE), and maximum absolute error.

Pooled reconstructed datasets were used to estimate survival probability and hazard ratios (HRs). Cox proportional hazards models were applied to obtain pooled HRs with 95% CIs. Proportional hazards assumptions were assessed using Schoenfeld residuals. In case of violation, time-varying Cox models were fitted. Landmark analyses at 2 and 4 months assessed conditional survival among patients alive at those time points. Time-restricted analyses for 0 to 2 and 0 to 4 months were also performed to evaluate early hazard dynamics.

For all outcomes, hazard ratios (HRs) from individual studies were extracted and pooled using a random-effects model (Paule-Mandel estimator for between-study variance with Hartung-Knapp adjustment for CIs). Between-study heterogeneity was assessed using the I² statistic, interpreted as low (0% to 25%), moderate (25% to 50%), substantial (50% to 75%), and high (>75%). Sensitivity analyses were performed using leave-one-out (LOO) procedures to evaluate the stability of pooled estimates. To further explore sources of heterogeneity, subgroup analyses were conducted based on age group (<70 vs ≥70 years), study design (RCT vs observational), study population (IHD vs HF), and follow-up duration. In addition, meta-regression analyses were performed using study-level covariates including mean age, prevalence of diabetes and hypertension, and proportion of male participants. Publication bias was planned to be evaluated using funnel plots and Egger’s test when ≥10 studies were available for an outcome. All analyses were conducted using R software and RStudio. ^, Two-sided p-values < 0.05 were considered statistically significant.