Estimating cardiovascular effects of influenza vaccination in older adults: a target trial emulation using proximal causal inference

Summary

Background

The substantial burden of cardiovascular diseases highlights the urgent need for cost-effective interventions to mitigate their impact. While existing evidence on the cardioprotective effect of the influenza vaccine comes primarily from populations with cardiovascular comorbidities, these studies remain susceptible to several sources of bias, including immortal time bias and unmeasured confounding. To attenuate these limitations, our study aimed to assess the effect of influenza vaccination on cardiovascular events in an older population in China, utilizing a target trial emulation framework in conjunction with a proximal causal inference (PCI) approach.

Methods

This is a sequentially designed, propensity score (PS) matched, vaccine effectiveness study under a target trial emulation framework. We used data from the Yinzhou Regional Health Care Database and included permanent residents of Yinzhou, China, aged 65 years or older. We employed a sequential trial approach in which participants were categorized as influenza vaccinees or non-vaccinees based on their vaccination regimen during the one-week enrollment period of each sequential trial from 2020 to 2022. The outcomes of interest were major adverse cardiovascular events (MACE) and acute coronary syndromes (ACS) within one year of follow-up. To address measured and unmeasured confounding, PS matching was performed in conjunction with PCI using a two-stage Poisson regression to estimate incidence rate ratios (IRRs).

Findings

A total of 8,181,638 older adults were included across the 50 emulated trials between 2020 and 2022. Of these, 170,011 received influenza vaccination, while 8,011,627 remained unvaccinated. Vaccinated participants were generally frailer (severely frail: 19.1% vs. 14.7%) and had a higher prevalence of hypertension (83.0% vs. 74.9%). After PS matching, all measured characteristics were well-balanced among 339,976 matched participants. In conjunction with the PCI approach, we found influenza vaccination was associated with a decrease in one-year risk of MACE (IRR: 0.86 [95% CI: 0.83–0.89]) and one-year risk of ACS (IRR: 0.87 [95% CI: 0.83–0.91]) compared to non-vaccination. Results were consistent across strata of enrollment year, age, sex, current smoking status, hypertension, hyperlipidemia, prior influenza vaccination status, and numerous sensitivity analyses.

Interpretation

Influenza vaccination may reduce the risk of MACE and ACS among older adults. Aligned with the World Health Organization guidelines, our findings further support influenza vaccination as an effective public health strategy for potentially reducing cardiovascular disease burden.

Funding

National Natural Science Foundation of China; Science and Technology Project of Science and Technology Bureau of Yinzhou District, Ningbo City; Zhejiang Provincial Centre for Disease Control and Prevention Science and Technology Program; Bill & Melinda Gates Foundation.

Research in context.

Evidence before this study

We searched PubMed for English-language articles through March 1, 2025, that investigated the relationship between influenza vaccination and major adverse cardiovascular events (MACE), acute coronary syndromes (ACS), or myocardial infarction (MI), using the search terms “influenza vaccine” AND “major adverse cardiovascular events OR acute coronary syndromes OR myocardial infarction.”

Our review identified five randomized controlled trials (RCTs) evaluating the association of influenza vaccination with MACE (n = 5), ACS (n = 1) or MI (n = 4) among individuals with established cardiovascular disease. Three reported a cardioprotective effect of influenza vaccination, while two showed statistically non-significant risk reductions with trends favoring vaccination. Additionally, we identified 24 observational studies investigating MI (n = 15), MACE (n = 6), and ACS (n = 3), of which 17 reported significant protective effects. Furthermore, eight relevant meta-analyses of the RCTs or observational studies consistently supported these cardiovascular benefits.

However, the RCTs exclusively enrolled high-risk patients with existing cardiovascular diseases, limiting generalizability. Observational studies were often susceptible to immortal time bias (due to asynchronous eligibility assessment, treatment assignment, and follow-up initiation) and unmeasured confounding (e.g., health-seeking behavior, frailty), issues which our study mitigates using a target trial emulation framework combined with a proximal causal inference.

Added value of this study

Our sequentially designed target trial emulation included over 8 million older adults before PS matching and nearly 340,000 after matching. We found consistent association between influenza vaccination and reduced risk of MACE and ACS across primary, subgroup and sensitivity analyses. To our knowledge, this is among the first studies to incorporate the proximal causal inference within a target trial emulation, thereby attenuating risk of unmeasured confounding. While current guidelines from the American College of Cardiology highlight cardiovascular benefits of influenza vaccination in patients with established ACS, the World Health Organization's age-based recommendations for older adults largely prioritize preventing severe influenza infection. Our findings further support the WHO's guidelines from the perspective of cardiovascular protective effects among the general older population.

Implications of all the available evidence

Our findings indicate a reduced risk of MACE and ACS associated with influenza vaccination among older adults in a regional Chinese population, supporting influenza vaccination as a valuable public health strategy for cardiovascular disease prevention. Further studies employing advanced causal inference methods, such as quasi-experimental designs, are necessary to further validate these findings and guide clinical recommendations.

Introduction

The substantial burden of cardiovascular diseases (CVDs) highlights the urgent need for cost-effective preventive interventions. Respiratory infections, including influenza, may contribute to triggering cardiovascular events,^1,2 raising the possibility that influenza vaccination could reduce cardiovascular events by preventing influenza. Additionally, animal studies have suggested influenza vaccination may reduce atherosclerotic plaque size, enhance plaque stability, and lower levels of pro-inflammatory markers, even in the absence of exposure to influenza virus.³

Several randomized controlled trials (RCTs) have also provided robust evidence that influenza vaccination is associated with reduced risk of major adverse cardiovascular events (MACE) and acute coronary syndromes (ACS) (Table S1).⁴ However, their stringent inclusion and exclusion criteria may limit the generalizability of findings to broader populations. Existing observational studies,5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15 which potentially mitigate this limitation, are susceptible to various biases that challenge causal interpretation (Tables S2 and S3). For instance, in case–control studies, adjusting for covariates assessed at the time of cardiovascular outcomes (for cases) or at sampling (for controls) may result in biased estimates. In cohort studies, immortal time, selection, and prevalent user bias can arise when the timing of eligibility assessment, vaccination assignment, and initiation of follow-up are not properly synchronized. Moreover, few observational studies have employed statistical methods to mitigate the effects of unmeasured confounding derived from health-seeking behavior and frailty. This concern is heightened by previous reports of unmeasured confounding invaliding conclusions in real-world studies.^16,17

To attenuate these limitations, we aimed to assess the protective effect of influenza vaccination on the occurrence of MACE and ACS through a cohort of community-dwelling older adults in China.¹⁸ We integrated the target trial emulation (TTE) framework with the proximal causal inference (PCI) approach, a novel method proposed to correct unmeasured confounding through the use of negative controls as measured proxies of unmeasured confounders.¹⁹ This joint strategy could minimize the aforementioned common biases in conventional observational studies, thereby enhancing the validity of real-world evidence.

Methods

Data source

Participants were ascertained from the Yinzhou Regional Health Care Database (YRHCD), a database extensively used for safety and effectiveness studies of vaccines20, 21, 22 and drugs.23, 24, 25, 26 Established by the local Center for Disease Control and Prevention in 2006, the YRHCD now encompasses records from over 1.70 million residents in the Yinzhou District of Ningbo (city), Zhejiang (province), a developed region of eastern China.¹⁸ The YRHCD captures 99% of Yinzhou's permanent residents and continues to be updated.

YRHCD integrates longitudinal data from various sources, including population census, electronic medical records, death registry, and immunization registry linked through an anonymized unique identifier generated from the participant's national identity card number (Figure S1). Electronic medical records were obtained from a regionally representative network of healthcare providers, including five general hospitals, 24 township health centers, and 265 community health service stations of Yinzhou. In China, general hospitals are responsible for providing clinical care, delivering medical education, and conducting medical research, whereas community health service stations and township health centers primarily address basic medical needs (e.g., chronic disease management and medication refills).²⁷ Deaths within and outside the hospital are systematically recorded in the death reporting system, and all vaccinations are captured in the immunization registry. Moreover, healthcare services provided to patients who relocate to Yinzhou are also available in the YRHCD. Given the low emigration rate in the Yinzhou District (0.20% in 2022),²⁸ loss to follow-up is minimal in the YRHCD.

This study was approved by the Peking University Health Science Center Ethics Committee (IRB00001052-24067) and informed consent was waived, as the study involved de-identified data.

Specification and emulation of target trials

As outlined in Method S1, the target trial emulation approach is a structured framework to apply key design features from randomized controlled trials to the analysis of observational data, including pre-specifying and clearly defining the (1) eligibility criteria, (2) treatment strategies, (3) treatment assignment, (4) follow-up, (5) outcomes, (6) causal contrast, and (7) statistical analysis.²⁹ This framework attempts to mitigate biases inherent in observational studies while strengthening causal inference.³⁰

In our study, we emulate a sequence of hypothetical target trials conducted across three influenza seasons (2020–2023) to investigate influenza vaccination vs. non-vaccination for the prevention of MACE and ACS among older adults in the Yinzhou District. Table S4 compares the design components of our specified and emulated target trials, with emulation details of each component presented from “Study population and design” through “Statistical analysis.” Notably, conventional target trial emulation typically employs causal analysis methods—such as propensity score (PS) matching, inverse probability treatment weighting, or the parametric g-formula—to emulate the random treatment assignment in a hypothetical trial. However, these causal analysis methods rely on the assumption of “conditional exchangeability,” which assumes the absence of unmeasured confounders between treatment groups and outcome.³¹ Given the potential influence of health-seeking behavior and other unmeasurable confounders commonly observed in vaccine effectiveness studies,¹⁹ our study incorporates PCI techniques, leveraging negative control variables as proxies to correct unmeasured confounding.

Study population and design

In the first emulated trial, we designated the first week of September 2020 as the enrollment period, with September 1, 2020 defined as the index date. The Yinzhou government initiated a free influenza vaccination policy for local residents aged ≥70 years old in 2020. Therefore, the first trial included study participants who were: (1) aged 70–100 years; (2) permanent Yinzhou residents registered in the YRHCD for at least one year; and (3) had at least one medical record in the YRHCD within the past three years. Participants with missing data on sex, a history of acute infection or febrile illness within the past 14 days, or a history of Guillain-Barré syndrome were excluded from the study (Table S5 and Fig. 1). Participants with ACS, MACE, or died during the enrollment period were also excluded.

Fig. 1.

Illustration of Study Design for the Emulated Trial 1 (Enrollment period: September 1-September 7, 2020). Abbreviations: mFI, Multimorbidity frailty index; MACE, Major adverse cardiovascular events; ACS, Acute coronary syndromes.

Based on the eligibility criteria, the first trial enrolled a total of 101,615 participants, including 16 influenza-vaccinated participants and 101,599 unvaccinated participants within the one-week enrollment period. Given the small number of vaccinated participants in a single trial, meaningful statistical analysis was infeasible. Therefore, we implemented a sequential design,^32,33 emulating multiple trials on a weekly basis from 2020 to 2022 to accumulate a sufficient sample size.

Specifically, we designated each week between September 1 and December 28, 2020 (17 weeks), September 8 and December 28, 2021 (16 weeks), and September 1 and December 28, 2022 (17 weeks) as the enrollment period for each trial, with the first day of each enrollment period defined as the index date. This approach generated 50 weekly enrollment periods corresponding to a sequence of 50 emulated trials (Fig. 2). The same eligibility criteria were applied to all the emulated trials, with the exception that the age restriction for study participants was lowered from ≥70 years to ≥65 years starting in 2021, as the Yinzhou government extended the free influenza vaccine policy to population aged ≥65 years. Participants meeting the inclusion criteria at a given index date were included into the index date's corresponding trial. Notably, individuals could be included in multiple trials if they remained eligible at subsequent enrollment periods.

Fig. 2.

The Illustration of Sequential Design and Statistical Analysis. Note: A single individual may be included in multiple emulated trials. For example, consider Individual i, a 73-year-old male who received the influenza vaccination on September 17, 2020. During the first three enrollment periods in September 2020 (Trial 1: September 1–7, 2020; Trial 2: September 8–14, 2020; Trial 3: September 15–21, 2020), individual i met the eligibility criteria and was included in Emulated Trials 1 through 3. In Emulated Trials 1 and 2, Individual i was assigned to the non-vaccination group, whereas in Emulated Trial 3, he was assigned to the vaccination group.

Treatment strategy and assignment

The treatment strategies were influenza vaccination (ATC code: J07BB) vs. non-vaccination, within the one-week enrollment period of each emulated trial.

Outcomes

Our primary outcome of interest was MACE, defined by a primary or secondary diagnosis of ischemic stroke (I63–I64), intracranial hemorrhage (I60–I62), myocardial infarction (I21–I22), or heart failure (I11.0, I50, I97.1) based on International Classification of Diseases 10th edition (ICD-10) codes across ambulatory, emergency, and hospital care settings.34, 35, 36

Our secondary outcome was ACS, defined as a composite of acute myocardial infarction (I21–I22), unstable angina (I20.0), or other acute ischemic heart disease (I24.8, I24.9) in the ambulatory, emergency, and hospital care settings (Table S6).³⁷

Follow-up

We followed participants from the end of the enrollment period until the earliest occurrence of the following events: one year of follow-up, occurrence of MACE or ACS, or death (Fig. 1).

Covariates

Covariates were measured at the index date and included demographic characteristics (age, sex, education level); behavior and lifestyle (smoking, drinking); physical examination (body mass index [BMI], systolic blood pressure [SBP], diastolic blood pressure [DBP]); laboratory measurements (low-density lipoprotein cholesterol [LDL-C], high-density lipoprotein cholesterol [HDL-C], total triglyceride [TG]); multimorbidity frailty index (Method S2); comorbidity type (e.g., hypertension, heart failure, prior ACS, prior stroke); co-medication (e.g., renin-angiotensin-aldosterone system inhibitors, diuretics, beta-blocking agents); procedure history (percutaneous coronary intervention, coronary artery bypass graft); healthcare utilization (number of outpatient/inpatient visits, prior influenza vaccination). Detailed information for each covariate was reported in Table S7.

Negative controls

Under the PCI framework, the average treatment effect is non-parametrically identifiable using a pair of negative control exposure (NCE) and negative control outcome (NCO) as proxies for the known unmeasured confounder (detailed assumptions listed in Method S3).³⁸ This framework has been extended to a parametric two-stage regression approach to mitigate the challenge of solving ill-posed integral equations typically encountered in non-parametric methods.³⁹ Importantly, a NCE is an exposure variable known not to causally affect the outcome of interest, and a NCO is an outcome variable that is not expected to be causally associated with the exposure of interest. Both NCE and NCO should share a common confounding structure with the exposure and outcome of interest.³⁸ In this study, we employed a gastroesophageal reflux visit in the year prior to the index date as the NCE, and cataract visit during follow-up as the NCO, with an illustrative directed acyclic graph (DAG) shown in Fig. 3 and definitions reported in Table S8. The negative controls were prespecified based on clinical knowledge and have been used in prior studies on vaccine safety and effectiveness.40, 41, 42

Fig. 3.

Directed Acyclic Graph for Double Negative Control. Abbreviations: MACE, Major adverse cardiovascular events; ACS, Acute coronary syndromes; NCE, Negative control exposure; NCO, Negative control outcome; U, Unmeasured confounder (e.g. health-seeking behavior).

Statistical analysis

Given the low prevalence of influenza vaccination in each sequential trial, we employed 1:1 PS matching for each trial using a nearest-neighbor approach (Fig. 2), with a caliper of 0.2 times the standard deviation of the logit of the PS to balance the baseline characteristics between the treatment groups.^43,44 PS were estimated from logistic regression models within each trial, where influenza vaccination was the dependent variable and the covariates as independent variables, and then 50 matched datasets from the 50 emulated trials were generated and stacked together into a single analytical dataset. We assessed the balance of baseline covariates with standardized absolute mean differences (SAMDs), a value greater than 0.1 indicated meaningful imbalance.⁴⁵

We modeled study outcomes as count data following a Poisson distribution.^46,47 To investigate the effectiveness of influenza vaccination on MACE and ACS, we fitted a parametric two-stage Poisson regression model to further adjust for potential unmeasured confounders such as health-seeking behavior using the PCI method³⁹ as follows:

In the first stage, a Poisson regression was fitted for the NCO with the NCE, exposure of interest (A), and the natural logarithm of person-years (PY) as the offset. The expected value of the NCO (i.e., $\mathit{log}\left(\hat{E}\left[NCO|A,NCE\right]\right)$) was then obtained from this model:

$$\mathit{log}\left(E\left(NCO\right)\right)={\alpha }_0+{\alpha }_{1}A+{\alpha }_{2}NCE+\mathit{log}\left({PY}_{NCO}\right)$$

In the second-stage, appropriate assumptions (Method S4) would validate the expected value of the NCO in the first-stage regression as a surrogate variable for the unmeasured confounder. We then estimate the treatment effect $\left({\beta }_1\right)$ by including the expected value of NCO in the Poisson model:

$$\mathit{log}\left(E\left(Y\right)\right)={\beta }_0+{\beta }_{1}A+{\beta }_2\mathit{log}\left(\hat{E}\left[NCO|A,NCE\right]\right)+\mathit{log}\left({PY}_Y\right)$$

Covariates with less than 30% missing values, including education level, smoking, drinking, BMI, SBP, DBP, LDL-C, HDL-C and TG, were imputed using the multiple imputation by chained equation model in each emulated trial, conditional on vaccination status, covariates, outcome indicator and negative control variables.⁴⁸ The 95% confidence intervals (CIs) were derived from the multiple imputation bootstrap (pooled sample) method with 500 repetitions.⁴⁹

Subgroup and sensitivity analyses

Subgroup analyses were conducted to investigate potential effect modification by enrollment year, age, sex, current smoking status, hypertension, hyperlipidemia, and influenza vaccination status of the prior year. Within the subgroup data of each emulated trial, a PS was re-estimated and vaccinated and unvaccinated participants were matched based on the re-estimated PS.⁴³ A subgroup-specific incidence rate ratio (IRR) was then calculated by refitting the two-stage Poisson regression model (PCI method) on the PS-matched subpopulation of the subgroup analysis. P value for multiplicative interaction was calculated by inverting the corresponding bootstrapped CI of the difference in IRRs between these two subgroups.⁵⁰

We performed several sensitivity analyses to evaluate the robustness of our results. First, we assessed the sensitivity of our negative control selection by conducting the PCI analysis under two different scenarios: 1) changing the NCE from a gastroesophageal reflux visit to a constipation visit in the prior year, and 2) changing the NCO from a cataract visit to an osteoporosis visit. Second, we evaluated the sensitivity of missing data by performing a complete-case analysis. Third, we evaluated the sensitivity of the follow-up period by introducing a 14-day induction period for the influenza vaccine and restricting the analysis to events occurring between days 15 and days 365 after follow-up initiation. Fourth, we assessed the validity of the PCI methodology by employing an influenza visit during follow-up as a positive control outcome (Table S8), given the well-established evidence supporting the effectiveness of influenza vaccination in preventing influenza.⁵¹ Fifth, we assessed the influence of the number of emulated trials on the study results by extending or shortening the total enrollment periods by four weeks for each year. This yielded a total of 62 and 38 emulated trials, respectively.

All statistical analyses were performed with R Software (Version 4.3.1, R Foundation for Statistical Computing, Vienna, Austria).

Role of the funding source

The funder played no role in designing the study, collecting data, analyzing data, interpreting results, or writing of this manuscript.

Results

Study population

After applying the inclusion and exclusion criteria, a total of 8,181,638 participants (representing 214,173 unique individuals, one individual could be included into multiple emulated trials) were analyzed across the 50 emulated trials (Figure S2). Of these, 170,011 were vaccinated against influenza, while 8,011,627 remained unvaccinated. Key baseline characteristics are summarized in Table 1.

Table 1.

Baseline Characteristics of Patients From 50 Emulated Trials in n (%) or Median [IQR].

Characteristicsa	Unmatched			PS matched
	Non-vaccination (n = 8,011,627)	Flu-vaccination (n = 170,011)	SAMD	Non-vaccination (n = 169,988)	Flu-vaccination (n = 169,988)	SAMD
Demographics
Age, years	72.0 (68.0–77.0)	72.0 (69.0–77.0)	0.036	72.0 (69.0–77.0)	72.0 (69.0–77.0)	0.028
[65, 75)	5,109,110 (63.8)	108,915 (64.1)	0.145	110,327 (64.9)	108,897 (64.1)	0.018
[75, 85)	2,173,644 (27.1)	51,672 (30.4)		50,661 (29.8)	51,667 (30.4)
≥85	728,873 (9.1)	9424 (5.5)		9000 (5.3)	9424 (5.5)
Male	3,865,445 (48.2)	79,582 (46.8)	0.029	78,850 (46.4)	79,574 (46.8)	0.009
Education level			0.140			0.010
Senior high school or higher	842,808 (11.9)	13,922 (9.0)		13,472 (8.9)	13,919 (9.0)
Junior high school	1,776,112 (25.1)	33,301 (21.6)		33,280 (21.9)	33,296 (21.6)
Primary school	3,304,319 (46.6)	80,186 (51.9)		79,037 (51.9)	80,177 (51.9)
Semiliterate	1,165,880 (16.4)	27,113 (17.5)		26,388 (17.3)	27,110 (17.5)
Behavior and lifestyle
Smoking	802,148 (13.1)	20,563 (12.7)	0.011	18,274 (12.4)	20,560 (12.7)	0.009
Drinking	2,302,231 (38.8)	66,716 (41.7)	0.059	59,678 (41.6)	66,708 (41.7)	0.002
Physical examination
BMI, kg/m2	23.5 (21.5–25.6)	23.6 (21.6–25.7)	0.007	23.6 (21.6–25.7)	23.6 (21.6–25.7)	0.005
<18.5	252,758 (4.1)	6438 (4.0)	0.019	5577 (3.8)	6438 (4.0)	0.011
[18.5,24.9)	3,975,668 (64.6)	103,455 (63.9)		94,172 (63.9)	103,440 (63.9)
≥25.0	1,921,359 (31.2)	51,934 (32.1)		47,697 (32.3)	51,927 (32.1)
SBP, mmHg	130.0 (122.0–136.0)	130.0 (122.0–137.0)	0.033	130.0 (122.0–136.0)	130.0 (122.0–137.0)	0.045
<120	805,997 (13.3)	21,630 (13.4)	0.034	19,435 (13.3)	21,627 (13.4)	0.056
[120, 139)	4,059,161 (67.1)	106,566 (65.8)		99,561 (68.0)	106,550 (65.8)
[140, 159)	921,054 (15.2)	26,250 (16.2)		21,409 (14.6)	26,247 (16.2)
[160, 179)	225,261 (3.7)	6547 (4.0)		5174 (3.5)	6547 (4.0)
≥180	36,942 (0.6)	1018 (0.6)		787 (0.5)	1017 (0.6)
DBP, mmHg	76.0 (70.0–80.0)	76.0 (70.0–80.0)	0.004	76.0 (70.0–80.0)	76.0 (70.0–80.0)	0.004
<80	4,135,583 (68.4)	111,312 (68.7)	0.010	100,807 (68.9)	111,300 (68.7)	0.020
[80, 89)	1,628,830 (26.9)	43,037 (26.6)		39,237 (26.8)	43,028 (26.6)
[90, 99)	234,050 (3.9)	6360 (3.9)		5274 (3.6)	6358 (3.9)
[100, 109)	43,312 (0.7)	1146 (0.7)		922 (0.6)	1146 (0.7)
≥110	7628 (0.1)	174 (0.1)		153 (0.1)	174 (0.1)
Laboratory measurements
LDL-C, mmol/L	2.6 (2.1–3.2)	2.7 (2.1–3.2)	0.005	2.6 (2.1–3.2)	2.7 (2.1–3.2)	0.009
<3.4	5,353,751 (80.3)	130,940 (80.3)	0.010	123,952 (80.7)	130,926 (80.3)	0.012
[3.4, 4.1)	947,771 (14.2)	23,561 (14.5)		21,550 (14.0)	23,556 (14.5)
≥4.1	361,921 (5.4)	8536 (5.2)		8133 (5.3)	8532 (5.2)
HDL-C, mmol/L	1.3 (1.1–1.5)	1.3 (1.1–1.5)	0.028	1.3 (1.1–1.5)	1.3 (1.1–1.5)	0.023
<1.0	1,124,782 (16.9)	22,282 (13.7)	0.089	24,431 (15.9)	22,279 (13.7)	0.063
≥1.0	5,547,369 (83.1)	140,814 (86.3)		129,314 (84.1)	140,794 (86.3)
TG, mmol/L	1.3 (1.0–1.8)	1.3 (1.0–1.9)	0.004	1.3 (1.0–1.9)	1.3 (1.0–1.9)	0.012
<1.7	4,356,829 (69.7)	110,886 (69.0)	0.016	101,248 (68.6)	110,886 (69.0)	0.011
[1.7, 2.3)	1,052,388 (16.8)	27,714 (17.2)		25,446 (17.2)	27,714 (17.2)
≥2.3	841,306 (13.5)	22,151 (13.8)		20,893 (14.2)	22,149 (13.8)
Multimorbidity frailty index	0.1 (0.0–0.1)	0.1 (0.0–0.1)	0.247	0.1 (0.1–0.1)	0.1 (0.0–0.1)	0.005
Fit	3,161,769 (39.5)	44,255 (26.0)	0.293	42,323 (24.9)	44,255 (26.0)	0.035
Mild-frail	2,261,530 (28.2)	55,322 (32.5)		57,777 (34.0)	55,320 (32.5)
Moderate-frail	1,411,770 (17.6)	37,939 (22.3)		38,039 (22.4)	37,935 (22.3)
Severe-frail	1,176,558 (14.7)	32,495 (19.1)		31,849 (18.7)	32,478 (19.1)
Comorbidity type
Peripheral vascular disease	2,376,761 (29.7)	61,994 (36.5)	0.145	61,478 (36.2)	61,977 (36.5)	0.006
Hypertension	5,998,440 (74.9)	141,105 (83.0)	0.200	141,679 (83.3)	141,085 (83.0)	0.009
Hyperlipidemia	4,441,814 (55.4)	115,067 (67.7)	0.254	115,561 (68.0)	115,045 (67.7)	0.006
Heart failure	784,582 (9.8)	18,446 (10.8)	0.035	18,360 (10.8)	18,441 (10.8)	0.002
History of ACS	543,711 (6.8)	14,028 (8.3)	0.056	14,011 (8.2)	14,019 (8.2)	0.000
History of stroke	1,711,952 (21.4)	38,925 (22.9)	0.037	38,597 (22.7)	38,919 (22.9)	0.005
Medication history
RAASi	3,161,526 (39.5)	83,084 (48.9)	0.190	84,045 (49.4)	83,070 (48.9)	0.011
Diuretics	565,684 (7.1)	13,848 (8.1)	0.041	13,839 (8.1)	13,843 (8.1)	<0.001
β-blocking agents	1,198,000 (15.0)	29,851 (17.6)	0.071	29,848 (17.6)	29,840 (17.6)	<0.001
Calcium channel blockers	2,639,599 (32.9)	67,965 (40.0)	0.146	68,746 (40.4)	67,953 (40.0)	0.010
Lipid modifying agents	2,452,621 (30.6)	68,171 (40.1)	0.199	68,186 (40.1)	68,156 (40.1)	<0.001
Antidiabetic agents	1,375,179 (17.2)	33,664 (19.8)	0.068	34,215 (20.1)	33,658 (19.8)	0.008
Antiplatelet agents	1,537,087 (19.2)	40,035 (23.5)	0.107	39,841 (23.4)	40,027 (23.5)	0.003
Anticoagulants	103,975 (1.3)	2312 (1.4)	0.005	2304 (1.4)	2311 (1.4)	<0.001
Proton-pump inhibitors	1,819,676 (22.7)	49,362 (29.0)	0.145	49,310 (29.0)	49,347 (29.0)	<0.001
NSAIDs	1,967,970 (24.6)	53,611 (31.5)	0.156	53,945 (31.7)	53,594 (31.5)	0.004
Procedure history
Percutaneous coronary intervention	131,170 (1.6)	2732 (1.6)	0.002	2730 (1.6)	2729 (1.6)	<0.001
Coronary artery bypass graft	9431 (0.1)	186 (0.1)	0.002	213 (0.1)	186 (0.1)	0.005
Healthcare utilization
No. of primary care and outpatient visits	11.0 (3.0–21.0)	16.0 (9.0–26.0)	0.295	15.0 (8.0–26.0)	16.0 (9.0–26.0)	0.003
No. of inpatient visits	0.0 (0.0–0.0)	0.0 (0.0–0.0)	0.050	0.0 (0.0–0.0)	0.0 (0.0–0.0)	0.002
Prior influenza vaccination	2,396,280 (29.9)	65,333 (38.4)	0.180	65,714 (38.7)	65,324 (38.4)	0.005

Abbreviations: Flu, Influenza; SAMD, Standardized absolute mean difference; BMI, Body mass index; SBP, Systolic blood pressure; DBP, Diastolic blood pressure; LDL-C, Low-density lipoprotein cholesterol; HDL-C, High-density lipoprotein cholesterol; TG, Total triglyceride; RAASi, Renin–angiotensin–aldosterone system inhibitors; NSAID, Nonsteroidal anti-inflammatory drugs.

Missing rate: (1) Unmatched dataset, education level 11.5%, smoke 23.3%, drink 25.6%, BMI 22.9%, SBP 24.1%, DBP 24.1%, LDL-C 16.6%, HDL-C 16.5%, TG 21.6%. (2) Matched dataset, education level 9.8%, smoke 9.3%, drink 10.8%, BMI 9.0%, SBP 9.3%, DBP 9.3%, LDL-C 6.9%, HDL-C 6.8%, TG 9.3%.

^aTable 1 summarizes the descriptive statistics and SAMD for baseline characteristics, among which age, BMI, SBP, DBP, LDL-C, HDL-C, and TG were reported as both continuous and categorical variables. However, only the categorical forms of these variables were included in the association analysis.

Vaccinated participants had a higher multimorbidity frailty index than unvaccinated participants (severely frail: 19.1% vs. 14.7%; moderately frail: 22.3% vs. 17.6%; mildly frail: 32.5% vs. 28.2%; fit: 26.0% vs. 39.5%). Vaccinated participants also had a higher prevalence of peripheral vascular disease (36.5% vs. 29.7%), hypertension (83.0% vs. 74.9%), and hyperlipidemia (67.7% vs. 55.4%). Vaccinated participants were more likely to use renin-angiotensin-aldosterone system inhibitors (48.9% vs. 39.5%), calcium channel blockers (40.0% vs. 32.9%), lipid modifying agents (40.1% vs. 30.6%), antiplatelet agents (23.5% vs. 19.2%), proton-pump inhibitors (29.0% vs. 22.7%), and nonsteroidal anti-inflammatory drugs (31.5% vs. 24.6%). Additionally, the vaccination group had a higher frequency of outpatient visits (median [IQR]: 16 [9–26] vs. 11 [3–21]) and greater proportion of influenza vaccination in the past year (38.4% vs. 29.9%). All baseline covariates were balanced after PS matching with SAMDs less than 0.1 (Figure S3 and Tables S9 and S10).

Influenza vaccination and risk of MACE and ACS

Across our PS-matched treatment groups, 339,976 matched participants were followed for 330,567.2 person-years, and MACE occurred among 15,931 participants (Table 2). The incidence rate (IR) of MACE was 47.1 (95% CI: 46.1–48.2) per 1000 person-years among vaccinated participants and 49.2 (95% CI: 48.2–50.3) among unvaccinated participants. PS matching yielded an adjusted IRR of 0.96 (95% CI: 0.93–0.99). Further adjustment for unmeasured confounding with the PCI approach resulted in an IRR of 0.86 (95% CI: 0.83–0.89).

Table 2.

Incidence rates per 1000 person-years of study outcomes after PS matching^a.

Outcome	Exposure	No. of events (n/N)	Person-years	IR (95% CI)	IRRCrude (95% CI)	IRRPSmatching (95% CI)	IRRPCI (95% CI)
MACE
	Non-vaccination	8122/169,988	164,937.4	49.2 (48.2–50.3)	Reference
	Flu-vaccination	7809/169,988	165,629.8	47.1 (46.1–48.2)	1.08 (1.05–1.10)	0.96 (0.93–0.99)	0.86 (0.83–0.89)
ACS
	Non-vaccination	4389/169,988	166,733.0	26.3 (25.6–27.1)	Reference
	Flu-vaccination	4409/169,988	167,292.3	26.4 (25.6–27.1)	1.27 (1.24–1.31)	1.00 (0.96–1.04)	0.87 (0.83–0.91)

Abbreviation: PS, Propensity score; MACE, Major adverse cardiovascular events; ACS, Acute coronary syndromes; Flu, Influenza; IR, Incidence rate; CI, Confidence interval; IRR, Incidence rate ratio; PCI, Proximal causal inference.

^aThe IRR _Crude was derived from unmatched dataset by univariate Poisson regression; The IRR _PSmatching was derived from PS-matched dataset by univariate Poisson regression; The IRR _PCI was derived from PS-matched dataset by PCI method.

A total of 4409 ACS events were recorded among the vaccinated participants and 4389 among unvaccinated participants, corresponding to IRs of 26.4 (95% CI: 25.6–27.1) and 26.3 (95% CI: 25.6–27.1) per 1000 person-years, respectively. The adjusted IRR after PS matching was 1.00 (95% CI: 0.96–1.04). After subsequent adjustment for unmeasured confounding with the PCI approach, the IRR was 0.87 (95% CI: 0.83–0.91).

Subgroup and sensitivity analyses

Subgroup analyses revealed a lower risk of MACE and ACS among vaccinated participants across various strata of enrollment year, age, sex, current smoking status, hypertension, hyperlipidemia, and influenza vaccination status of the prior year (Fig. 4). These findings are consistent with the primary results, supporting the robustness of the observed associations between influenza vaccination and MACE and ACS.

Fig. 4.

Subgroup Analyses on Association of Influenza Vaccination with MACE and ACS (A: Subgroup effect estimates between influenza vaccination and MACE, B: Subgroup effect estimates between influenza vaccination and ACS). Abbreviations: MACE, Major adverse cardiovascular events; ACS, Acute coronary syndromes; N, Number of study participants; n, Number of outcome events; PY, Person-year; IRR, Incidence rate ratio; CI, Confidence interval; PCI, Proximal causal inference.

In the sensitivity analyses (Tables S11–S13), replacing the NCE with a constipation visit and NCO with an osteoporosis visit in the PCI models continued to support a protective effect of influenza vaccination against MACE and ACS, albeit with a slightly smaller magnitude. The complete-case analysis, based on 141,622 vaccinated participants and 141,622 unvaccinated participants, found an IRR estimate of 0.90 (95% CI: 0.87–0.94) for MACE and 0.93 (95% CI: 0.89–0.99) for ACS, which also aligns with our primary analysis. When employing a 14-day induction period for the influenza vaccine, the risk of MACE (IRR: 0.87 [95% CI: 0.84–0.90]) and ACS (IRR: 0.89 [95% CI: 0.85–0.93]) remained significantly lower among vaccinated participants. The positive control analysis with the PCI approach (IRR: 0.76 [95% CI: 0.69–0.84]) supported the validity of our PCI methodology. Lastly, varying the number of emulated trials did not change the results of our primary analysis.

Discussion

This sequential target trial emulation, involving 8,181,638 older adults before PS matching and 339,976 after matching, found a reduced risk of MACE and ACS among participants who received the influenza vaccine compared to non-vaccination participants. The reduced risk is consistent across many subgroups and sensitivity analyses. Incorporating the PCI approach within the target trial emulation framework further shifted the PS-adjusted treatment effect towards a greater magnitude of protection, aligning more closely with previous pooled analyses of RCTs.⁵² These findings contribute to the growing body of real-world evidence suggesting cardiovascular benefits of influenza vaccination, supporting the World Health Organization's recommendation for annual influenza vaccination in older adults.⁵³

The results of our primary analysis are consistent with a recent meta-analysis conducted by Omidi et al.,⁵² which pooled five RCTs published up to August 1, 2023, demonstrating influenza vaccination was protective against MACE (RR: 0.71 [95% CI: 0.55–0.92]). Furthermore, the meta-analysis by Yedlapati et al.⁴ incorporated three observational studies in addition to the four included RCTs, yielding a pooled RR estimate (RCTs: 0.57 [95% CI: 0.43–0.74]; observational studies: 0.90 [95% CI: 0.83–0.98]) that also supports this protective association. Prior observational studies using a variety of study designs—including case–control studies,⁶ cohort studies,^5,8 and self-controlled case-series¹⁰ (Table S2)—have similarly reported reduced MACE incidence with influenza vaccination; however, these studies remain susceptible to biases such as immortal time bias and the healthy vaccinee effect. Our current study, conducted within a Chinese population and employing rigorous causal inference methods, provides an updated contribution to the evolving evidence on this topic.

The results of our secondary outcome are consistent with a meta-analysis of four RCTs (RR: 0.63 [95% CI: 0.44–0.89])⁵⁴ and two cohort studies that assessed the association of influenza vaccination with ACS (HR: 0.46 [95% CI: 0.39–0.55] and 0.62 [95% CI: 0.52–0.81]).^11,12 However, these investigations have predominantly focused on high-risk populations with cardiovascular comorbidities. In particular, the four major RCTs55, 56, 57, 58 evaluating influenza vaccination in patients with established ischemic heart disease played a pivotal role in the recent upgrade of both the American College of Cardiology (2025) and European Society of Cardiology (2023) guidelines,^59,60 which endorse annual influenza vaccination with a Class I, Level A recommendation for ACS patients. Our study found the cardiovascular protective effects of influenza vaccination extend beyond those with preexisting cardiovascular comorbidities. This broader protective association is consistent with evidence presented by Streeter et al.¹³ and Davidson et al.,¹⁰ suggesting that more inclusive vaccination strategies may yield substantial benefits across an older population and such strategies could play a critical role in reducing the burden of cardiovascular diseases.

Although the free influenza vaccination policy minimizes the impact of ‘income or ability to pay’ as a confounder, and despite our statistical adjustments for various covariates through PS matching, unmeasured confounding remains inevitable in observational studies. For example, health-seeking behavior is a common unmeasured confounder that is challenging to quantify and adjust for in vaccine studies.⁶¹ It could introduce confounding by indication if patients with chronic diseases are channeled to be vaccinated and leads to an underestimation of vaccine effectiveness, since the less healthy population is at higher risk of adverse outcomes.⁶² Conversely, it is also plausible that individuals who actively pursue seasonal immunizations exhibit greater health literacy and adopt healthier lifestyles, which inherently lowering their risk of outcome of interest, a concept referred to as the healthy vaccinee effect.⁶³ Remschmidt et al. have noted both confounding by indication and healthy vaccinee bias are likely to operate simultaneously in observational studies of influenza vaccine, with overall direction of the bias varying across different studies.⁶⁴ In our study, vaccinated participants presented higher degree of frailty, along with greater prevalence of comorbidities and more extensive medication histories, suggesting that confounding by indication was the primary source of bias.

The methodological strength of our study is the application of the newly proposed regression-based PCI approach to attenuate unmeasured confounding, enabling us to more robustly establish the association between influenza vaccination and its protective effects against MACE and ACS. The PCI method requires correctly selecting negative control variables exhibiting null causal relationship with the exposure or outcome of interest. We conducted sensitivity analyses with various combinations of NCEs and NCOs and found results consistent with our primary, secondary, and subgroup analyses. The use of target trial emulation framework also reduced the risk of biases common to observational studies, such as immortal time bias.³⁰ To the best of our knowledge, this is the first pharmacoepidemiologic study that combined the target trial emulation framework with the regression-based PCI approach.

Our study also has several limitations. First, the data were derived from a region of China, which may limit the generalizability of our findings. Furthermore, distinct influenza vaccine utilization patterns in China—shaped by regional vaccination policies and clinical guidelines—as well as ethnic differences in cardiovascular disease susceptibility, may limit the transportability of our results.⁶⁵ Thus, caution should be warranted when extrapolating our results to populations outside China. Second, while our study provided insight into influenza vaccine effectiveness across three consecutive years and demonstrated a consistent protective effect, we could not account for the complex interplays between participants' serological history and the annual mismatches of the vaccine strains with evolving circulating strains.^66,67 Third, data on certain determinants of influenza vaccination, such as occupation, urbanization of residence and economic status were not available and could not be adjusted for in the analysis. Fourth, our study was not sufficiently powered to explore the associations between influenza vaccination and subtypes of MACE and ACS, nor was it powered for effectiveness evaluation before the COVID-19 outbreak to assess the impact of the pandemic, as the free influenza vaccination policy for older adults in Yinzhou district began in 2020, leading to a rapid increase in vaccination coverage. Fifth, although multiple biological mechanisms—such as anti-influenza effect and vaccine-induced cardioprotective immune responses—have been proposed to explain the cardiovascular protective effects of the influenza vaccine,⁶⁸ we were unable to determine which mechanism predominates. Thus, the heterogeneity observed in the association between influenza vaccination and MACE across different influenza seasons should be interpreted with caution. While this variability across influenza seasons may be attributable to differences in vaccine effectiveness against circulating influenza strains, it may also be influenced by fluctuations in the circulation and intensity of other respiratory pathogens capable of inducing systemic inflammation and triggering cardiovascular events.^69,70 Additionally, off-target effects of the influenza vaccine—such as cross-reactive immune modulation and unintended interactions with other pathogens—might further contribute to this heterogeneity.⁷¹

In conclusion, our target trial emulation with the PCI approach to address unmeasured confounding suggests that influenza vaccination in older adults is associated with a decreased risk of MACE and ACS compared to non-vaccination. These findings further validate influenza vaccination as an actionable public health measure for cardiovascular disease prevention. Additional studies using quasi-experimental designs with real-world data remain necessary to confirm and expand upon these findings.

Contributors

Jinxin Guo: writing—original draft, investigation, visualization, formal analysis, data curation. Tiansheng Wang: writing—review & editing, methodology, investigation. Zhike Liu: review & editing, data curation, investigation. Weihong Zeng: review & editing, data curation, investigation. Peng Shen: review & editing, resources, investigation. Yexiang Sun: review & editing, resources, investigation. Siyan Zhan: review & editing, conceptualization, funding acquisition, project administration. Yang Xu: review & editing, methodology, conceptualization, funding acquisition, project administration.

All authors approved the final version of the manuscript. Jinxin Guo, Siyan Zhan and Yang Xu accessed and verified the data.

Data sharing statement

Due to legislative restrictions from the “Data Security Law of the People's Republic of China”, person-level data of this study cannot be shared but aggregated data is available from the corresponding author on reasonable request, following approval from the relevant data custodians.

Declaration of interests

Tiansheng Wang receives research funding from American Diabetes Association. All other authors declare no financial or non-financial competing interests.