COVID-19 mRNA vaccination and immune-related adverse events in patients with cancer receiving immune checkpoint inhibitors: a target trial emulation study

Abstract

Background

COVID-19 messenger RNA (mRNA) vaccines enhance immune checkpoint inhibitor (ICI) efficacy through type I interferon responses, but may theoretically increase the burden of immune-related adverse events (irAEs). This study aimed to determine whether vaccination increases irAE risk in patients with ICI-treated cancer and to quantify the balance between toxicity and survival benefits.

Methods

We conducted a retrospective cohort study using target trial emulation methodology within the TriNetX Research Global Network (January 2021 to December 2025). We identified 7,218 adult patients with metastatic cancer initiating ICI therapy. Patients who received COVID-19 mRNA vaccination within 100 days before ICI initiation (n=3,609) were matched 1:1 to unvaccinated controls (n=3,609) using propensity scores. The primary outcome was the composite incidence of irAEs at 36 months. Secondary outcomes included severe irAEs, organ-specific irAEs, all-cause mortality, and intensive care unit (ICU) admission. Probabilistic quantitative bias analysis (VanderWeele-Ding formulation) and a composite ocular negative-control outcome (incident age-related cataract and glaucoma) were used to assess robustness to unmeasured confounding and differential ophthalmological surveillance.

Results

Vaccinated patients had a significantly increased risk of overall irAEs compared with unvaccinated controls (HR 1.22, 95% CI 1.17 to 1.28; E-value 1.74). This increase included clinically significant severe irAEs (HR 1.11, 95% CI 1.04 to 1.18). Organ-specific analyses revealed that the overall increase was primarily driven by ocular irAEs (HR 1.48, 95% CI 1.17 to 1.87), with non-significant trends for dermatologic and rheumatologic irAEs. However, vaccination was associated with substantial survival benefits, including 14% lower all-cause mortality (HR 0.86, 95% CI 0.80 to 0.93) and 57% fewer ICU admissions (HR 0.43, 95% CI 0.25 to 0.73). Point estimates were numerically higher in females and patients aged ≥50 years, although tests for interaction were not statistically significant.

Conclusions

COVID-19 mRNA vaccination is associated with a modest but clinically meaningful 22% increase in irAEs, driven largely by ocular manifestations. However, the substantial survival benefits clearly outweigh these toxicity risks. Vaccination should continue in all eligible patients, with enhanced monitoring recommended for irAEs, particularly ophthalmological surveillance for high-risk groups.

WHAT IS ALREADY KNOWN ON THIS TOPIC

• COVID-19 messenger RNA vaccines enhance immune checkpoint inhibitor efficacy through type I interferon responses, with improved survival demonstrated in patients with advanced non-small cell lung cancer.

• Dual immune activation from vaccination and immunotherapy raises theoretical concerns about increased overall immune-related adverse event burden.

• Previous studies were limited by small sample sizes (≤400 patients), lack of propensity score matching to address confounding by indication, and absence of quantitative bias analysis.

WHAT THIS STUDY ADDS

• COVID-19 vaccination significantly increases overall immune-related adverse event risk by 22% in patients with cancer receiving immune checkpoint inhibitors, including an 11% increase in severe immune-related adverse events (irAEs) requiring intensive management.

• The overall irAE increase is primarily driven by ocular manifestations (48% increase), with non-significant trends for dermatologic and rheumatologic involvement.

• Survival benefits (14% mortality reduction, 57% fewer intensive care unit admissions) substantially outweigh the increased overall irAE burden across all patient subgroups.

• Female patients and those aged ≥50 years have higher overall irAE risk requiring enhanced monitoring.

• Quantitative bias analysis (VanderWeele-Ding formulation) demonstrates partial robustness to unmeasured confounding, with a 62% probability that the adjusted HR remains above 1.0 and an E-value of 1.74 (lower bound 1.62).

• Multiple sensitivity analyses confirm robustness of the overall irAE association across different analytical approaches, population restrictions, and vaccination timing windows.

HOW THIS STUDY MIGHT AFFECT RESEARCH, PRACTICE OR POLICY

• Vaccination should continue in all patients with cancer receiving immunotherapy, with the survival benefit clearly outweighing the 22% increase in overall irAE incidence.

• Findings support integration of COVID-19 vaccination into standard oncology care protocols with proactive monitoring strategies for overall irAE burden.

• Clinicians should counsel vaccinated patients about increased overall irAE risk and maintain heightened vigilance for early detection across all organ systems, particularly ophthalmological manifestations.

• Risk stratification based on sex and age enables personalized monitoring strategies, with baseline ophthalmological examinations recommended for high-risk patients (females, age ≥50 years).

• Multidisciplinary care coordination between oncology, ophthalmology, dermatology, and rheumatology is essential for optimal management of the increased irAE burden.

Introduction

Immune checkpoint inhibitors (ICIs) targeting programmed cell death protein 1, programmed death-ligand 1, and cytotoxic T-lymphocyte associated protein 4 have transformed cancer treatment, but immune-related adverse events (irAEs) occur in 60–70% of patients, potentially affecting any organ system.¹ Recent evidence demonstrates that COVID-19 messenger RNA (mRNA) vaccines can enhance ICI efficacy by stimulating type I interferon responses and resetting the tumor immune microenvironment.² Grippin and colleagues showed that patients with advanced non-small cell lung cancer who received COVID-19 mRNA vaccination within 100 days of ICI initiation had significantly improved overall survival over a follow-up period extending to approximately 40 months compared with unvaccinated patients (median Overall survival (OS) 37.3 vs 20.6 months; adjusted HR 0.51, 95% CI 0.37 to 0.71, p<0.001).² However, this dual immune activation raises a critical safety question: does COVID-19 vaccination increase the overall burden of irAEs when combined with ICI therapy?

The biological plausibility for increased irAE risk stems from mRNA vaccines inducing robust type I interferon responses and T-cell activation.^{2 3} When combined with ICIs that remove inhibitory checkpoints, this dual activation could theoretically lower the threshold for autoimmune reactions across multiple organ systems. Understanding the overall irAE burden, its severity, and organ-specific patterns is essential for risk-benefit assessment. Conversely, the proven survival benefits of COVID-19 vaccination in patients with cancer and the enhanced antitumor efficacy demonstrated by Grippin et al suggest potential overall clinical benefit despite any increased irAE risk.

Previous studies examining COVID-19 vaccine safety in patients with ICI-treated cancer reported no increased irAE risk but were limited by small sample sizes (≤400 patients),^{4 5} lack of propensity score matching to address confounding by indication (healthier patients more likely to be vaccinated), and absence of quantitative bias analyses to assess robustness to unmeasured confounders such as performance status. Furthermore, comprehensive evaluations of overall irAE burden and severity remain scarce.

We therefore conducted a target trial emulation study using propensity score matching and quantitative bias analysis to rigorously estimate the causal effect of COVID-19 mRNA vaccination on overall irAE incidence in patients with cancer initiating ICI therapy. We assessed both the overall irAE burden and its clinical significance through severity grading and organ-specific characterization. Our approach addresses key methodological limitations of previous research and provides clinically actionable evidence for vaccination decision-making in this vulnerable population.

Methods

Study design and data source

We conducted a retrospective cohort study using target trial emulation methodology^{6 7} with data from the TriNetX Research Global Network, a federated health research platform aggregating de-identified electronic health records from over 120 million patients across 69 healthcare organizations worldwide. The study period extended from 1 January 2021 through 31 December 2025. This study followed Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) guidelines⁸ and was deemed exempt from institutional review board approval as it used de-identified data compliant with relevant data protection regulations.

Target trial specification

We applied target trial emulation principles^{6 7 9} to define the hypothetical randomized trial we aimed to emulate:

• Eligibility criteria: Adult patients with cancer (≥18 years) initiating ICI therapy without prior ICI exposure, active severe comorbidities within 30 days, or immunosuppressive status.

• Treatment strategies: COVID-19 mRNA vaccination within 100 days before ICI initiation versus no vaccination.

• Assignment: At ICI initiation date (index date, time zero).

• Follow-up: From index date through first irAE, death, loss to follow-up, or 36 months.

• Outcomes: Overall composite irAE incidence; severe irAEs; organ-specific irAEs; mortality; intensive care unit (ICU) admission; pneumonia.

• Causal contrast: Intention-to-treat effect.

• Analysis: Cox proportional hazards regression with propensity score matching stratified by matched pairs.

Study population

We identified adult patients (≥18 years) with metastatic cancer who initiated ICI therapy during the study period. Inclusion criteria comprised: (1) diagnosis of malignant neoplasm (International Classification of Diseases, 10th Revision (ICD-10): C00–C96); (2) first ICI prescription (pembrolizumab, nivolumab, cemiplimab, atezolizumab, durvalumab, avelumab, ipilimumab, or tremelimumab); (3) minimum 6 months follow-up. Exclusion criteria included: (1) prior ICI therapy; (2) baseline autoimmune or inflammatory conditions corresponding to irAE-defining ICD-10 codes (eg, pre-existing uveitis, thyroiditis, colitis) within 30 days before ICI initiation, to avoid confounding of irAE outcome ascertainment; (3) immunosuppressive conditions or therapies.

Exposure definition

The primary exposure was receipt of at least one dose of COVID-19 mRNA vaccine (BNT162b2 (Pfizer-BioNTech) or mRNA-1273 (Moderna), including all formulations: original monovalent, bivalent, and updated variants) within 100 days before ICI initiation. Vaccination status was determined from CPT codes (91300–91303), CVX codes, and medication records. The comparison group comprised patients with no documented COVID-19 mRNA vaccination within 100 days before ICI initiation.

Patients vaccinated more than 100 days before ICI initiation were classified in the unvaccinated group, as their vaccination fell outside the prespecified exposure window. An exploratory dose-response analysis compared patients who received two to three mRNA vaccine doses within the 100-day window versus unvaccinated controls.

Outcome definitions

The primary outcome was overall composite irAE incidence within 36 months of ICI initiation, defined using ICD-10 codes for: cardiac (myocarditis, pericarditis), endocrine (thyroid disorders, adrenal insufficiency, diabetes), gastrointestinal (colitis, hepatitis), hematologic (anemia, thrombocytopenia, neutropenia), ocular manifestations, including uveitis (ICD-10: H20.0 (primary iridocyclitis), H20.9 (unspecified iridocyclitis)) and optic neuritis (ICD-10: H46.9 (unspecified optic neuritis)), pulmonary (pneumonitis, interstitial lung disease), dermatologic (dermatitis, Stevens-Johnson syndrome), and rheumatologic (arthritis, myositis) manifestations. Complete code lists are provided in the online supplemental eTable S1.

Severe irAEs were operationally defined as grade 3–4 equivalent events based on hospitalization, ICU admission, or specific high-severity diagnoses including myocarditis, pneumonitis, hepatic failure, colitis requiring surgery, and severe dermatologic reactions (eg, Stevens-Johnson syndrome). This operational definition approximates but does not directly map to Common Terminology Criteria for Adverse Events (CTCAE) grades, as CTCAE grading requires clinical assessment data not available in electronic health record (EHR-based databases. Other secondary outcomes included organ-specific irAEs analyzed individually, all-cause mortality, pneumonia, and ICU admission.

Negative control outcomes (inguinal hernia, gallbladder stone, traumatic fractures) were included to validate the study design and assess for residual confounding or systematic bias.¹⁰

Statistical analysis

We performed 1:1 propensity score matching¹¹ using greedy nearest-neighbor matching without replacement (caliper 0.1 SDs). The propensity score model included: age, sex, race/ethnicity, cancer type, ICI type, comorbidities (diabetes, hypertension, heart disease, kidney disease, liver disease, cerebrovascular disease), prior cytotoxic chemotherapy (ATC codes L01A-L01D, excluding L01FF), prior radiation, and prior COVID-19 infection. Balance was assessed using standardized mean differences (SMD <0.1 indicating adequate balance).

We estimated HRs and 95% CIs using Cox proportional hazards models stratified by matched pairs. Kaplan-Meier curves illustrated time-to-event distributions. We calculated E-values¹² to quantify the minimum strength of association an unmeasured confounder would need with both treatment and outcome to explain away observed associations.

For organ-specific irAE analyses, we applied the Benjamini-Hochberg false discovery rate (FDR) correction within each time point to account for multiplicity arising from simultaneous testing across 10 organ systems. We selected the Benjamini-Hochberg procedure over Bonferroni correction because organ-specific irAEs likely share common immunological mechanisms (a single underlying antitumor and autoimmune T-cell activation), and the independence assumption underlying Bonferroni correction is therefore inappropriate. Overall irAEs, as the prespecified primary outcome, were not subject to multiplicity adjustment. A corrected p value<0.05 was considered statistically significant for organ-specific comparisons.

Quantitative bias analysis¹³ was performed using both deterministic scenarios (four confounding strengths) and probabilistic Monte Carlo simulation (n=50,000 iterations, with log-normal priors for RRₘ_CD and RRₘ_CE having median 1.7 and SD on the log scale 0.25) under the VanderWeele-Ding formulation of the bias factor (BF = (RR_CD × RR_CE) / (RR_CD + RR_CE − 1)), focusing on performance status (Eastern Cooperative Oncology Group (ECOG) score) as a key unmeasured confounder. We additionally assessed the possibility of differential ophthalmological surveillance using a composite ocular negative-control outcome (incident age-related cataract (ICD-10 H25.x) and glaucoma (H40.x)), restricted to incident events within the same 1:1 propensity-score-matched cohort and analyzed using stratified Cox regression with matched pairs as the stratum.^{14 15}

To address potential confounding by unmeasured performance status, we performed a sensitivity analysis incorporating EHR-derived proxy variables into the propensity score model: palliative care encounters (ICD-10: Z51.5), weakness (R53.1), fatigue (R53.83), and gait abnormalities (R26). These variables have been validated as performance status surrogates in EHR-based oncology research.¹⁶ To evaluate the potential impact of competing risks from mortality on irAE estimates, we estimated cumulative incidence functions (CIF) using the Aalen-Johansen estimator with death as the competing event for each cohort separately. We conducted fixed follow-up sensitivity analyses restricting observation to 1-year and 2-year windows, ensuring identical follow-up duration for both groups.

We adopted a cause-specific hazard approach in the primary analysis, which is recommended when the research question concerns disease etiology rather than clinical prediction.^{17 18} To evaluate the potential impact of competing risks from mortality on irAE estimates, we estimated CIF using the Aalen-Johansen estimator with death as the competing event for each cohort separately. Fine-Gray subdistribution hazard models¹⁹ could not be calculated within the TriNetX federated platform, as individual-level time-to-event data are not exportable. Fixed follow-up sensitivity analyses restricting observation to 1-year and 2-year windows ensured identical follow-up duration for both groups.

Subgroup analyses examined effect modification by age (<50, 50–64, 65–74, ≥75 years), sex, race/ethnicity, cancer type, ICI type, vaccine timing (≤30 vs 31–100 days), vaccine brand (Pfizer vs Moderna), and vaccine formulation (original monovalent vs bivalent). Interaction terms were tested using likelihood ratio tests.

Sensitivity analyses included: (1) on-treatment cohort (≥2 ICI administrations); (2) US population only; (3) vaccination within 50 days; (4) landmark analyses excluding events in the first 1, 2, or 3 months. All analyses were performed using the TriNetX platform and R software V.4.3.0. Two-sided p values<0.05 were considered statistically significant.

Primary statistical analyses were performed using the TriNetX analytics platform. Post hoc visualization and RMST analysis code is provided as online supplemental code.

We performed restricted mean survival time (RMST) analysis, which does not require the proportional hazards assumption, to complement the Cox regression findings. Formal Schoenfeld residuals testing and time-varying coefficient Cox models could not be implemented within the TriNetX federated analytics platform, as these procedures require individual-level time-to-event data that are not exportable from the platform; the time-stratified Cox estimates at eight prespecified time points (table 2) and the RMST analysis together provide a parsimonious description of the time-dependent treatment effect.

To assess temporal confounding, we performed a calendar-time stratified analysis dividing the cohort by ICI initiation date: 2021–2022 (predominantly original and early bivalent vaccines, alpha/delta/early Omicron) and 2023–2025 (updated vaccine formulations, later Omicron subvariants). Propensity score matching was performed separately within each stratum.

To disentangle potential vaccine-related ocular events from ICI-related ocular irAEs, we performed a sensitivity analysis excluding patients with any ocular event (ICD-10: H20.0, H20.9, H46.9) within 90 days of ICI initiation, followed by re-matching. This 90-day window encompasses the typical onset period for vaccine-associated ocular adverse events (mean 6.5–9.4 days post-vaccination)^{20 21} with a conservative margin.

Results

Study population and baseline characteristics

From the TriNetX Research Global Network (n=124,733,401), we identified 155,300 patients with cancer initiating ICI therapy during the study period: 3,609 (2.3%) received COVID-19 mRNA vaccination within 100 days (approximately 3.3 months) before ICI initiation, and 151,691 (97.7%) did not. After 1:1 propensity score matching, 3,609 patients per group were included in the primary analysis (figure 1). After matching, all baseline characteristics were well-balanced (all SMD <0.1) (table 1).

Figure 1. Study flow diagram showing patient selection, exclusion criteria, propensity score matching, and follow-up. From 124,733,401 patients in the TriNetX Research Global Network, 235,748 patients with cancer initiating ICI therapy after 1 January 2021 were identified. After applying inclusion and exclusion criteria, 220,045 patients remained eligible, of whom 155,300 had adequate follow-up (≥2 ICI administrations) for the primary analysis. Following 1:1 propensity score matching, 7,218 patients (3,609 vaccinated, 3,609 unvaccinated) with well-balanced baseline characteristics (all SMD<0.1) were analyzed. Study period: 1 January 2021 to 31 December 2025. Data source: TriNetX Research Global Network. HCOs, healthcare organizations; ICI, immune checkpoint inhibitor; irAE, immune-related adverse event; mRNA, messenger RNA; PSM, propensity score matching; SMD, standardized mean differences.

Table 1. Baseline characteristics before and after propensity score matching.

Characteristics	Before propensity score matching			After propensity score matching
	Vaccinated (n=3609)	Unvaccinated (n=151 691)	SMD	Vaccinated (n=3609)	Unvaccinated (n=3609)	SMD
Demographics
Age at index, mean (SD)	67.2 (12.0)	64.9 (12.9)	0.183	67.2 (12.0)	67.0 (12.1)	0.009
Sex, n (%)
Male	1,919 (53.2)	83,447 (55.0)	0.037	1,919 (53.2)	1,931 (53.5)	0.007
Female	1,690 (46.8)	68,199 (45.0)	0.037	1,690 (46.8)	1,678 (46.5)	0.007
Race/ethnicity, n (%)
White	2,775 (76.9)	107,733 (71.0)	0.134	2,775 (76.9)	2,875 (79.7)	0.067
Black or African American	390 (10.8)	12,486 (8.2)	0.088	390 (10.8)	336 (9.3)	0.050
Asian	186 (5.2)	8,456 (5.6)	0.019	186 (5.2)	174 (4.8)	0.015
Unknown	258 (7.1)	23,016 (15.2)	0.250	258 (7.1)	224 (6.2)	0.020
Primary cancer site, n (%)
Respiratory and intrathoracic organs	1,237 (34.3)	55,238 (36.4)	0.045	1,237 (34.3)	1,191 (33.0)	0.027
Digestive organs	779 (21.6)	27,049 (17.8)	0.094	779 (21.6)	797 (22.1)	0.012
Melanoma and other skin cancers	670 (18.6)	28,681 (18.9)	0.009	670 (18.6)	636 (17.6)	0.024
Urinary tract	565 (15.7)	23,817 (15.7)	0.001	565 (15.7)	546 (15.1)	0.015
Breast	372 (10.3)	12,200 (8.0)	0.079	372 (10.3)	381 (10.6)	0.008
Female genital organs	311 (8.6)	11,733 (7.7)	0.032	311 (8.6)	316 (8.8)	0.005
Male genital organs	216 (6.0)	6,767 (4.5)	0.069	216 (6.0)	179 (5.0)	0.045
Lymphoid and hematopoietic tissue	206 (5.7)	7,260 (4.8)	0.041	206 (5.7)	180 (5.0)	0.032
Lip, oral cavity and pharynx	191 (5.3)	9,695 (6.4)	0.047	191 (5.3)	175 (4.8)	0.020
Bone and articular cartilage	62 (1.7)	2,652 (1.7)	0.002	62 (1.7)	59 (1.6)	0.006
Comorbidities, n (%)
Hypertensive diseases	2,137 (59.2)	72,091 (47.5)	0.236	2,137 (59.2)	2,123 (58.8)	0.008
Ischemic heart diseases	869 (24.1)	27,665 (18.2)	0.143	869 (24.1)	798 (22.1)	0.047
Type 2 diabetes mellitus	837 (23.2)	27,934 (18.4)	0.118	837 (23.2)	790 (21.9)	0.031
Liver diseases	693 (19.2)	25,202 (16.6)	0.068	693 (19.2)	677 (18.8)	0.011
Chronic kidney disease	541 (15.0)	17,573 (11.6)	0.100	541 (15.0)	501 (13.9)	0.032
Cerebrovascular diseases	362 (10.0)	12,930 (8.5)	0.052	362 (10.0)	301 (8.3)	0.059
Heart failure	326 (9.0)	10,421 (6.9)	0.080	326 (9.0)	283 (7.8)	0.043
Procedures, n (%)
Chemotherapy	1,929 (53.4)	78,889 (52.0)	0.030	1,928 (53.4)	1,925 (53.3)	0.002
Radiation	968 (26.8)	36,479 (24.0)	0.064	968 (26.8)	966 (26.8)	0.001
ICI agents, n (%)
Pembrolizumab	2,059 (57.1)	78,387 (51.7)	0.108	2,059 (57.1)	2,020 (56.0)	0.022
Nivolumab	792 (21.9)	40,112 (26.4)	0.105	792 (21.9)	797 (22.1)	0.003
Durvalumab	295 (8.2)	12,543 (8.3)	0.003	295 (8.2)	336 (9.3)	0.040
Atezolizumab	333 (9.2)	13,239 (8.7)	0.017	333 (9.2)	321 (8.9)	0.012
Ipilimumab	234 (6.5)	12,465 (8.2)	0.066	234 (6.5)	210 (5.8)	0.028
Cemiplimab	71 (2.0)	2,791 (1.8)	0.009	71 (2.0)	70 (1.9)	0.002
Avelumab	38 (1.1)	1,257 (0.8)	0.023	38 (1.1)	48 (1.3)	0.026
Tremelimumab	10 (0.3)	673 (0.4)	0.028	10 (0.3)	10 (0.3)	<0.001

Data are presented as mean (SD) or n (%). Categories within each section are ordered by descending frequency in the matched cohort.

ICI, immune checkpoint inhibitor; SMD, standardized mean difference.

The mean age was 67.2 years (SD 12.0), 53.2% were male, and 76.9% were white. The most common cancer types were respiratory/intrathoracic (34.3%), digestive organs (21.6%), and melanoma/skin (18.6%). The most frequently prescribed ICIs were pembrolizumab (57.1%), nivolumab (21.9%), and durvalumab (8.2%). Mean follow-up duration was 23.0 months (SD 12.2) in the vaccinated cohort and 21.0 months (SD 12.1) in the unvaccinated cohort.

From 124,733,401 patients in the TriNetX Research Global Network, 235,748 patients with cancer initiating ICI therapy after 1 January 2021 were identified. After applying inclusion and exclusion criteria, 220,045 patients remained eligible, of whom 155,300 had adequate follow-up (≥2 ICI administrations) for the primary analysis. Following 1:1 propensity score matching, 7,218 patients (3,609 vaccinated, 3,609 unvaccinated) with well-balanced baseline characteristics (all SMD <0.1) were analyzed.

Primary outcome: overall irAE incidence

During 36-month follow-up, 1,986 irAE events occurred in the vaccinated cohort (55.0%) compared with 1629 events in the unvaccinated cohort (45.1%). Vaccinated patients had a significantly increased overall irAE risk (HR 1.22, 95% CI 1.17 to 1.28, p<0.001; E-value 1.74) (figure 2, table 2). The association emerged gradually over time, reaching statistical significance at 12 months (HR 1.09, 95% CI 1.01 to 1.16, p=0.023) and strengthening through 36 months.

Figure 2. Kaplan-Meier survival curves for time to first immune-related adverse event in vaccinated versus unvaccinated cohorts. Vaccinated patients had significantly increased overall irAE risk (HR 1.22, 95% CI 1.17 to 1.28, log-rank p<0.001). The association emerged gradually over time, becoming statistically significant at 12 months (HR 1.09, p=0.023) and strengthening through 36 months, demonstrating a sustained increase in overall irAE burden. irAE, immune-related adverse event.

Table 2. HRs for immune-related adverse events at different time points.

Outcome	1 month	3 months	6 months	12 months	18 months	24 months	30 months	36 months
Overall irAEs	1.03 (0.886 to 1.187) p=0.734	1.07 (0.969 to 1.175) p=0.189	1.06 (0.982 to 1.152) p=0.129	1.09 (1.01 to 1.16) p=0.023	1.11 (1.03 to 1.18) p=0.004	1.14 (1.08 to 1.20) p=0.001	1.18 (1.12 to 1.24) p=0.001	1.22 (1.17 to 1.28) p<0.001
Cardiac irAEs	0.97 (0.749 to 1.265) p=0.839	1.06 (0.88 to 1.281) p=0.532	1.13 (0.967 to 1.325) p=0.122	1.10 (0.959 to 1.26) p=0.175	1.09 (0.957 to 1.241) p=0.193	1.09 (0.966 to 1.24) p=0.158	1.08 (0.959 to 1.224) p=0.196	1.08 (0.959 to 1.218) p=0.204
Endocrine irAEs	0.98 (0.821 to 1.171) p=0.823	1.02 (0.907 to 1.158) p=0.691	1.03 (0.937 to 1.141) p=0.501	1.05 (0.966 to 1.151) p=0.235	1.06 (0.971 to 1.149) p=0.204	1.07 (0.989 to 1.168) p=0.087	1.07 (0.987 to 1.162) p=0.099	1.07 (0.991 to 1.166) p=0.082
Gastrointestinal irAEs	0.82 (0.623 to 1.071) p=0.142	0.93 (0.782 to 1.112) p=0.436	1.11 (0.955 to 1.286) p=0.176	1.04 (0.921 to 1.184) p=0.500	1.07 (0.953 to 1.205) p=0.249	1.07 (0.958 to 1.202) p=0.224	1.05 (0.941 to 1.177) p=0.369	1.05 (0.943 to 1.176) p=0.361
Hematologic irAEs	1.10 (0.717 to 1.688) p=0.663	1.10 (0.843 to 1.431) p=0.488	1.28 (1.02 to 1.614) p=0.033*	1.27 (1.043 to 1.537) p=0.017*	1.15 (0.96 to 1.366) p=0.12*	1.09 (0.923 to 1.298) p=0.296	1.12 (0.944 to 1.319) p=0.199	1.12 (0.955 to 1.325) p=0.160
Dermatologic irAEs	0.92 (0.700 to 1.199) p=0.522	1.04 (0.877 to 1.242) p=0.632	1.04 (0.901 to 1.195) p=0.610	1.03 (0.913 to 1.161) p=0.635	1.06 (0.949 to 1.191) p=0.288	1.13 (1.011 to 1.264) p=0.031*	1.07 (0.959 to 1.192) p=0.231	1.08 (0.973 to 1.207) p=0.143
Ocular irAEs	1.29 (0.639 to 2.585) p=0.480	1.63 (1.03 to 2.589) p=0.035*	1.59 (1.11 to 2.274) p=0.011*	1.31 (0.982 to 1.742) p=0.066	1.61 (1.225 to 2.11) p=0.001*†	1.56 (1.209 to 2.014) p=0.001*†	1.54 (1.202 to 1.966) p=0.001*†	1.48 (1.169 to 1.874) p=0.001*†
Rheumatologic irAEs	0.91 (0.595 to 1.386) p=0.655	1.12 (0.857 to 1.452) p=0.415	1.15 (0.920 to 1.425) p=0.223	1.18 (0.982 to 1.419) p=0.076	1.14 (0.962 to 1.352) p=0.130	1.11 (0.944 to 1.312) p=0.201	1.10 (0.934 to 1.287) p=0.261	1.13 (0.962 to 1.318) p=0.138
Renal irAEs	0.90 (0.704 to 1.138) p=0.366	0.98 (0.838 to 1.151) p=0.822	0.98 (0.863 to 1.112) p=0.745	1.03 (0.925 to 1.149) p=0.581	1.04 (0.945 to 1.157) p=0.391	1.11 (1.007 to 1.225) p=0.036*	1.07 (0.969 to 1.172) p=0.192	1.06 (0.965 to 1.165) p=0.221
Neurological irAEs	1.40 (1.048 to 1.881) p=0.022*	1.28 (1.060 to 1.558) p=0.011*	1.12 (0.961 to 1.317) p=0.142	1.07 (0.937 to 1.230) p=0.304	1.10 (0.966 to 1.247) p=0.152	1.12 (0.990 to 1.267) p=0.072	1.07 (0.946 to 1.203) p=0.289	1.07 (0.954 to 1.209) p=0.238
Pulmonary irAEs	---	---	1.34 (0.69 to 2.596) p=0.387	1.27 (0.804 to 2.018) p=0.301	1.37 (0.872 to 2.146) p=0.170	1.39 (0.912 to 2.112) p=0.125	1.35 (0.905 to 2.028) p=0.139	1.40 (0.943 to 2.076) p=0.094

Data are presented as HR (95% CI).

--- Indicates insufficient follow-up or no events in either group.

Bold values indicate statistical significance at p < 0.05.

^*Indicates statistical significance (p<0.05). n=3,609 per group for all time points.

^†Significant after Benjamini-Hochberg FDR correction (adjusted p<0.05). Overall irAEs (prespecified primary outcome) not subject to multiplicity adjustment.

FDR, false discovery rate; irAE, immune-related adverse event.

Vaccinated cohort (blue) and unvaccinated cohort (orange). The lower number-at-risk in the vaccinated group at later time points (678 vs 1,221 at 36 months) reflects higher cumulative irAE incidence (more patients experienced the event and exited the at-risk set), not differential loss to follow-up. Vaccinated patients had significantly increased overall irAE risk (HR 1.22, 95% CI 1.17 to 1.28, log-rank p<0.001). The association emerged gradually over time, becoming statistically significant at 12 months (HR 1.09, p=0.023) and strengthening through 36 months, demonstrating a sustained increase in overall irAE burden.

Secondary outcomes

Severe irAEs

Vaccinated patients also demonstrated an 11% increased risk of severe irAEs requiring hospitalization or intensive management (HR 1.11, 95% CI 1.04 to 1.18, p=0.001; E-value 1.45), with 398 severe events in the vaccinated group versus 105 events in the unvaccinated group. This confirms that the overall irAE increase includes clinically significant events, not merely mild manifestations.

Organ-specific irAEs

To characterize which organ systems contributed to the overall irAE increase, we performed organ-specific analyses. The overall irAE increase was primarily driven by ocular manifestations, with vaccinated patients demonstrating markedly increased ocular irAE risk (HR 1.48, 95% CI 1.17 to 1.87, p<0.001; E-value 2.33), accounting for 194 events (5.4%) versus 108 events (3.0%). Non-significant trends toward increased risks were also observed for dermatologic irAEs (HR 1.08, 95% CI 0.97 to 1.21, p=0.143) and rheumatologic irAEs (HR 1.12, 95% CI 0.96 to 1.32, p=0.138). After Benjamini-Hochberg FDR correction, ocular irAE was the only organ-specific outcome to achieve corrected significance (FDR-adjusted p=0.006), achieving corrected significance from 18 months through 36 months (table 2). No statistically significant differences were observed for cardiac, endocrine, gastrointestinal, hematologic, or pulmonary irAEs after FDR correction (figure 3).

Figure 3. Secondary outcomes at 36-month follow-up. Forest plot showing HRs and 95% CIs for overall irAEs, severe irAEs, organ-specific irAEs, all-cause mortality, pneumonia, and ICU admission. The overall irAE increase (HR 1.22) includes clinically significant severe irAEs (HR 1.11) and is primarily driven by ocular manifestations (HR 1.48, 95% CI 1.17 to 1.87, p<0.001). Substantial survival benefits were observed: mortality HR 0.86 (95% CI 0.80 to 0.93, p<0.001) and ICU admission HR 0.43 (95% CI 0.25 to 0.73, p<0.001). ICU, intensive care unit; irAE, immune-related adverse event.

The ocular irAE signal was robust to exclusion of potentially vaccine-related early events. Sensitivity analysis excluding patients with ocular events within 90 days of ICI initiation (n=3,110 per group after re-matching) yielded a higher HR (HR 1.608, 95% CI 1.239 to 2.085; see online supplemental eTable S19), indicating that early ocular events attenuated rather than drove the association. Additionally, a composite ocular negative-control outcome assessing incident age-related cataract (ICD-10 H25.x) and glaucoma (H40.x) within the matched cohort showed no excess risk in vaccinated patients (HR 1.03, 95% CI 0.91 to 1.17; see online supplemental eTable S24), arguing against differential ophthalmological surveillance as an explanation.

Clinical outcomes

Despite the increased irAE burden, vaccinated patients had significantly lower all-cause mortality (HR 0.86, 95% CI 0.80 to 0.93, p<0.001), corresponding to 14% relative risk reduction, and dramatically lower ICU admission rates (HR 0.43, 95% CI 0.25 to 0.73, p<0.001), corresponding to 57% relative risk reduction (figure 3).

In an exploratory analysis, vaccinated patients had significantly lower documented COVID-19 infection rates compared with unvaccinated patients (5.8% vs 13.3%; HR 0.50, 95% CI 0.42 to 0.60, p<0.001), corresponding to an associated 50% relative reduction in documented COVID-19 infection in this ICI-treated cancer population. Complete clinical outcomes including COVID-19 pneumonia, ICU admission, and mortality are presented in the online supplemental eTable S23.

Competing risk analysis confirmed cumulative mortality incidence at 36 months of 11.8% in the vaccinated group versus 17.5% in the unvaccinated group (difference: 5.7 percentage points; see online supplemental eFigure S5). The median overall survival was not reached in either group during the 36-month follow-up period.

Subgroup analyses

Subgroup analyses revealed stronger overall irAE associations in females (HR 1.15, 95% CI 1.05 to 1.26) than males (HR 1.07, 95% CI 0.98 to 1.18), and stronger effects in patients aged ≥50 years (figure 4).

Figure 4. Subgroup analyses for overall immune-related adverse events at 36 months. Forest plot showing HRs and 95% CIs stratified by age, sex, race/ethnicity, cancer type, ICI type, vaccine timing, vaccine brand, and vaccine formulation. Stronger overall irAE associations were observed in females (HR 1.15, 95% CI 1.05 to 1.26) than males (HR 1.07, 95% CI 0.98 to 1.18), and in patients aged ≥50 years. ICI, immune checkpoint inhibitor; irAE, immune-related adverse event; PD-1, programmed cell death protein 1; PD-L1, programmed death-ligand 1.

Sensitivity analyses

Multiple sensitivity analyses confirmed the robustness of our primary findings regarding overall irAE risk across different analytical approaches and population restrictions.

Incorporating performance status proxy variables into the propensity score model yielded an overall irAE HR of 1.211 (95% CI 1.158 to 1.267; see online supplemental eTable S16), virtually identical to the primary analysis (HR 1.22), indicating that unmeasured confounding by performance status does not materially affect the observed association.

Competing risk analysis using Aalen-Johansen cumulative incidence functions with death as the competing event confirmed that the irAE excess in vaccinated patients persisted after accounting for mortality. The CIF for overall irAEs at 36 months was 77.5% in the vaccinated cohort versus 64.6% in the unvaccinated cohort (difference: 12.9 percentage points; see online supplemental eTable S17), while the mortality CIF was lower in the vaccinated group (11.8% vs 17.5%; see online supplemental eFigure S5). Fixed follow-up sensitivity analyses with identical observation windows confirmed the primary findings: 1-year fixed window HR 1.056 (95% CI 0.982 to 1.136), 2-year fixed window HR 1.086 (95% CI 1.013 to 1.163, p=0.019; see online supplemental eTable S18).

On-treatment cohort analysis

When restricting analysis to patients who received at least two ICI administrations (on-treatment cohort, n=2403 per group), the association between vaccination and overall irAE risk remained significant (HR 1.12, 95% CI 1.04 to 1.21, p=0.002). This analysis demonstrated that the overall irAE association persisted among patients with documented ICI exposure and reduced the potential for surveillance bias from differential follow-up (see online supplemental eTable S12).

Geographic restriction to US population

Analysis restricted to US healthcare organizations only (n=3,544 per group) yielded consistent results for overall irAE risk (HR 1.07, 95% CI 1.00 to 1.14, p=0.042), though the point estimate was slightly attenuated, likely reflecting differences in healthcare delivery patterns and irAE surveillance intensity (see online supplemental eTable S13).

Vaccination timing window restriction

When restricting the vaccination window to 50 days before ICI initiation (n=2,045 per group), the overall irAE association remained directionally consistent (HR 1.09, 95% CI 1.00 to 1.18, p=0.050), though with reduced precision at this narrower window reflecting the smaller cohort (see online supplemental eTable S14).

In an exploratory dose-response analysis, patients who received two to three vaccine doses within the exposure window (n=545 per group) showed a numerically higher overall irAE risk (HR 1.284, 95% CI 1.089 to 1.513) compared with the primary analysis (HR 1.22), suggesting a possible dose-dependent effect (see online supplemental eTable S21).

Landmark analyses

Landmark analyses excluding patients with events in the first 1, 2, or 3 months after ICI initiation demonstrated consistent overall irAE associations: 1-month landmark HR 1.08 (95% CI 1.02 to 1.16, p=0.016); 2-month landmark HR 1.11 (95% CI 1.04 to 1.18, p=0.003); 3-month landmark HR 1.12 (95% CI 1.05 to 1.20, p=0.001). The consistency across all landmark time points strengthens confidence that the overall irAE increase persists beyond the immediate peri-initiation period (see online supplemental eTable S15).

Calendar-time stratified analysis demonstrated consistent overall irAE associations: HR 1.194 (95% CI 1.106 to 1.290) for 2021–2022 (n=2397/group) and HR 1.213 (95% CI 1.099 to 1.338) for 2023–2025 (n=1503/group; see online supplemental eTable S20). The mortality benefit was more pronounced in 2023–2025 (HR 0.827 vs 1.021).

RMST analysis confirmed that vaccinated patients experienced progressively reduced irAE-free time: the RMST difference was 6 days (0.2 months) at 12 months, 24 days (0.8 months) at 24 months, and 39 days (1.3 months) at 36 months (see online supplemental eFigure S4).

Negative control outcomes

None of the three negative control outcomes showed significant associations: inguinal hernia (HR 0.94, 95% CI 0.72 to 1.24, p=0.676), gallbladder stone (HR 1.16, 95% CI 0.98 to 1.38, p=0.081), and traumatic fractures (HR 0.92, 95% CI 0.80 to 1.05, p=0.202). The null findings for these mechanistically unrelated outcomes support the validity of our propensity score matching approach and suggest minimal residual confounding or differential surveillance bias (see online supplemental eTable S8).

Quantitative bias analysis

Under the VanderWeele-Ding formulation of the bias factor (BF = (RR_CD × RR_CE) / (RR_CD + RR_CE − 1)), a probabilistic Monte Carlo analysis (n=50,000 iterations; log-normal priors for RR_CD and RR_CE with median 1.7 and SD on the log scale 0.25) yielded a median adjusted HR of 1.034 (95% simulation interval 0.813–1.241), with the probability of adjusted HR >1.0, HR >1.1, and HR >1.2 equal to 61.7%, 27.4%, and 6.0%, respectively. Deterministic scenarios under weak (RR_CD 1.3, RR_CE 1.5), moderate (1.8, 1.6), strong (2.0, 2.0), and very strong (2.5, 2.3) confounding yielded adjusted HRs of 1.126, 1.017, 0.915, and 0.806, respectively. The E-value for the point estimate was 1.74 (lower 95% CI bound 1.62), indicating that an unmeasured confounder would need to exhibit moderately strong associations with both vaccination and irAE outcomes to fully explain away the observed association (see online supplemental eTable S11).

Discussion

Principal findings

In this target trial emulation study of 7,218 propensity score-matched patients with cancer, COVID-19 mRNA vaccination within 100 days before ICI initiation was associated with a 22% increased risk of overall immune-related adverse events at 36 months (HR 1.22, 95% CI 1.17 to 1.28; E-value 1.74). Importantly, this overall increase included clinically significant severe irAEs (11% increase, HR 1.11, 95% CI 1.04 to 1.18), demonstrating that the elevated irAE burden comprises meaningful clinical events requiring intensive management. Organ-specific analyses revealed that ocular manifestations (48% increase) were the primary driver of this overall increase, with non-significant trends for dermatologic and rheumatologic involvement. However, vaccination was also associated with substantial survival benefits: 14% lower mortality and 57% fewer ICU admissions. Multiple sensitivity analyses confirmed the robustness of the overall irAE association across different analytical approaches, population restrictions, and assumptions about unmeasured confounding.

Comparison with other studies

Our findings extend recent evidence that COVID-19 mRNA vaccines enhance ICI efficacy through type I interferon responses.² While Grippin et al demonstrated improved survival in vaccinated patients, our comprehensive safety analysis reveals the trade-off: a modest but significant increase in overall irAE burden, balanced against substantial survival benefits. Previous studies examining COVID-19 vaccine safety in patients with ICI-treated cancer reported no increased irAE risk but were limited by small sample sizes (≤400 patients) and lacked propensity score matching or rigorous sensitivity analyses.^{4 5 22} A recent systematic review and meta-analysis of vaccination outcomes in ICI-treated patients similarly reported no signal for major irAE increases, but included only studies with limited sample sizes and short follow-up.²³

The 2.3% vaccination rate within the 100-day pre-ICI window warrants discussion. Benchmarking analysis across multiple vaccine types and vaccine eras confirmed that this low rate reflects the restrictive temporal window and post-outbreak booster-era dynamics rather than inadequate EHR capture. When extended to a 365-day window, COVID-19 vaccination rose to 7.3% and influenza vaccination reached 35.7%, consistent with published influenza vaccination rates in oncology populations (30–40%).²⁴ All vaccine types showed similarly low rates at the 100-day window (influenza 3.8%, pneumococcal 1.1%, varicella/zoster <0.1%), confirming that the narrow window, rather than database capture limitations, drives the low ascertainment. Critically, the appropriate population benchmark for COVID-19 is era-dependent: whereas primary-series uptake reached 45–75% of US adults during 2021–early 2022 (Saelee et al, MMWR 2022),²⁵ bivalent booster uptake fell to 17–20% in the general population and 38% in patients with cancer during 2022–2023 (Skarbinski et al, JAMA Oncol 2025),²⁶ and updated 2023–2024 and 2024–2025 COVID-19 vaccine uptake declined further to 14–22% of US adults (CDC COVIDVaxView; CDC MMWR Fall 2024).^{27 28} The observed 7.3% rate in our cohort is therefore consistent with post-outbreak booster-era uptake rather than the peak-pandemic primary-series benchmark. Cancer patient-specific barriers, including concerns about immune interactions with ongoing immunotherapy and physician-advised delays around treatment cycles,²⁹ further contribute to the low vaccination rate within this restricted window. Importantly, vaccination rates for influenza, pneumococcal, and varicella/zoster vaccines were virtually identical between matched groups at the 365-day window (all SMDs <0.02), arguing against differential healthy vaccinee bias (see online supplemental eTable S22). Additionally, COVID-19 vaccines administered at pharmacies or community vaccination sites may not be linked to the patient’s EHR record,³⁰ leading to potential underascertainment that would bias results toward the null.

The significant overall irAE increase coexisting with individually non-significant organ-specific results (except ocular) reflects the statistical properties of composite endpoints. The HRs for most organ-specific irAEs were consistently above 1.0 (range 1.05–1.12 at 36 months), indicating a uniform trend toward increased risk across organ systems. However, each individual analysis was underpowered to detect these modest effect sizes. The composite endpoint aggregates these small, directionally consistent effects across all organ systems, achieving sufficient statistical power. After Benjamini-Hochberg FDR correction, ocular irAE emerged as the only organ-specific outcome with adjusted significance (FDR-adjusted p=0.006), identifying it as the single robust organ-specific signal warranting targeted clinical surveillance. The non-significant organ-specific trends for hematologic, endocrine, and dermatologic irAEs, while directionally consistent with the composite signal, did not survive multiplicity correction and should be interpreted as hypothesis-generating rather than confirmatory.

The distinction between immune-mediated pulmonary irAEs and infection-related pulmonary complications is clinically important. Pulmonary irAEs (pneumonitis, interstitial lung disease) showed a non-significant trend toward higher rates in vaccinated patients (HR 1.40, 95% CI 0.94 to 2.08, p=0.094), consistent with the overall pattern of modestly elevated organ-specific irAEs and not surviving FDR correction. In contrast, COVID-19 pneumonia trended lower in vaccinated patients (HR 0.93, p=0.131), and documented COVID-19 infection was significantly reduced (HR 0.50, p<0.001). The net pulmonary risk-benefit profile thus favors vaccination.

The 22% overall irAE increase represents a clinically meaningful elevation in toxicity burden. The concurrent 11% increase in severe irAEs confirms that this is not simply detection bias of mild events, but rather includes clinically significant toxicities requiring hospitalization and intensive management. The predominance of ocular manifestations is particularly notable given the eye’s immune-privileged status. Vaccine-induced systemic immune activation combined with checkpoint blockade may overcome ocular immune tolerance mechanisms, leading to uveitis and optic neuritis, which substantially contribute to the overall irAE burden.

The slightly attenuated HR in the US-only analysis (HR 1.07 vs 1.22 in the full cohort) primarily reflects re-matching with a new propensity score model within the US-restricted eligible pool, rather than a small change in cohort composition. Although the US-only analysis retained 98% of the vaccinated cohort (3,544 of 3,609), the overall eligible pool differed by approximately 20% from the full cohort (124,255 vs 155,300 ICI-treated patients) because of the large number of non-US unvaccinated controls. This meaningful reduction in the eligible pool, rather than the apparent 2% reduction in vaccinated patients, drives the re-matching. Propensity score matching is sensitive to the composition of the eligible pool, and different matching runs produce different matched pairs. Importantly, the direction of the association remained consistent across all sensitivity analyses (HR range 1.056–1.284), supporting the robustness of the primary finding.

Mechanisms and biological plausibility

The increased overall irAE burden likely stems from synergistic immune activation. COVID-19 mRNA vaccines induce robust type I interferon responses and activate innate and adaptive immunity.^{2 3} Preclinical work demonstrates that mRNA platforms efficiently target dendritic cells and prime cross-reactive T-cell responses, providing a mechanistic basis for synergy with checkpoint blockade.^{31 32} When combined with ICIs that remove inhibitory checkpoints on T cells, this dual activation may lower the threshold for breaking tolerance to self-antigens across multiple organ systems, manifesting as increased overall irAE incidence. This mechanism is consistent with reports of immune-mediated phenomena and disease flares in the general population following mRNA vaccination.^{33 34} The fact that ocular tissues are particularly susceptible suggests that immune-privileged sites may be especially vulnerable to this combined immune stimulation.

The novel finding of increased ocular irAEs (HR 1.48, 95% CI 1.17 to 1.87) warrants careful interpretation, as COVID-19 mRNA vaccines have been independently associated with uveitis and optic neuritis in the general population.^{35 36} Systematic reviews report a mean onset of 6.5–9.4 days post-vaccination,^{20 21} and a Korean nationwide study demonstrated modestly elevated uveitis risk up to 180 days post-vaccination.³⁷ However, population-level reporting rates for vaccine-associated uveitis are extremely low (0.57 per million doses for BNT162b2),³⁸ making it implausible that vaccine-related events account for the magnitude of the observed excess (194 vs 108 events). The sensitivity analysis excluding early ocular events strengthened rather than weakened the association (HR 1.608), and the sustained temporal accumulation pattern supports ICI-mediated immune activation. The biological plausibility of enhanced ICI-induced ocular irAEs is supported by the eye’s status as an immune-privileged organ. Ocular immune privilege is maintained by the blood-ocular barrier, absence of lymphatic drainage, and local production of immunosuppressive factors including transforming growth factor-beta (TGF-beta), alpha-melanocyte-stimulating hormone (alpha-MSH), and Fas ligand (FasL).39,41 ICI therapy may breach this immune privilege, and prior mRNA vaccination may further lower the threshold for breaking ocular immune tolerance through type I interferon pathway sensitization.⁴²

Clinical implications

Continue COVID-19 vaccination

The substantial survival benefit (14% mortality reduction) clearly outweighs the 22% increase in overall irAE incidence. All patients with cancer initiating ICI therapy should continue to receive recommended COVID-19 vaccinations. The benefits of infection protection combined with potential antitumor synergy demonstrated by Grippin et al² strongly support vaccination despite the increased irAE burden.

Enhanced monitoring for overall irAE burden

Given the significant increase in overall irAE incidence including severe manifestations, we recommend: (1) proactive counseling of vaccinated patients about increased irAE risk before ICI initiation; (2) heightened clinical vigilance for early irAE detection across all organ systems; (3) particular attention to ophthalmological symptoms given the predominant ocular contribution; (4) lower threshold for irAE evaluation and specialist referral; (5) patient education emphasizing the importance of promptly reporting new symptoms. When irAEs are identified, management should follow established clinical practice guidelines43,45 and current evidence on immunosuppressive treatment strategies for severe immune toxicity, including consideration of steroid-sparing biologic agents in selected refractory cases.46,48

The apparent paradox of increased severe irAEs coupled with reduced ICU admissions warrants careful interpretation. Our performance status proxy sensitivity analysis provides empirical evidence against residual confounding by functional status: incorporating palliative care encounters, weakness, fatigue, and gait abnormality into the propensity score model yielded an overall irAE HR of 1.211 (95% CI 1.158 to 1.267), a change of less than 1% from the primary analysis (HR 1.22). If the ICU reduction reflected vaccinated patients being systematically healthier, this sensitivity analysis would be expected to attenuate the irAE signal substantially; its near-complete stability argues against this explanation. Furthermore, the ICU outcome captures all-cause admissions; the 50% reduction in documented COVID-19 infection in vaccinated patients (HR 0.50, 95% CI 0.42 to 0.60, p<0.001) is expected to translate into substantial reductions in infection-driven ICU admissions, which likely accounts for a large proportion of the 57% ICU reduction rather than reflecting attenuated irAE severity.²⁶ More fundamentally, the irAE increase and survival benefit should not be viewed as independent opposing outcomes. Accumulating evidence demonstrates that irAE development during ICI therapy correlates with enhanced antitumor responses and improved survival.49,51 We propose that mRNA COVID-19 vaccination, through induction of robust type I interferon responses and activation of cross-reactive T-cell populations,² may lower the threshold for both autoimmune irAEs and antitumor immune responses when combined with ICI therapy. Through this lens, increased irAEs may partly mediate the survival benefit via enhanced antitumor immunity, rather than representing an independent cost. A formal mediation analysis would be needed to quantify this relationship, but is not feasible within the TriNetX platform.

The irAE association was consistent across multiple vaccine-related dimensions. Subgroup analyses showed comparable HRs between vaccination ≤30 days (HR 1.12) and 31–100 days (HR 1.14) before ICI, indicating no temporal gradient within the exposure window. The association was also consistent across vaccine brands (Pfizer HR 1.15, Moderna HR 1.16) and formulations (monovalent HR 1.16, bivalent HR 1.12), supporting a class effect of the mRNA vaccine platform. The attenuated HR in the 50-day window restriction (HR 1.09 vs 1.22 in the primary analysis) primarily reflects reduced statistical power from a 43% reduction in sample size (n=2,045 vs 3,609 per group), rather than a true dose-timing relationship. An exploratory dose-response analysis suggested a possible dose-dependent effect (two to three doses HR 1.284, 95% CI 1.089 to 1.513), though the limited sample size precludes definitive conclusions. Patients vaccinated more than 100 days before ICI were classified as unvaccinated; residual vaccine-induced immune memory in this subset may attenuate the observed association, rendering our estimates conservative.

Risk stratification

Female patients and those aged ≥50 years warrant particularly intensive monitoring given their higher overall irAE risk. However, the survival benefits observed in all subgroups support continued vaccination across all demographic groups.

Strengths and limitations

Strengths include: (1) large sample size (n=7,218) providing adequate power to detect overall irAE differences and characterize severity; (2) target trial emulation methodology with propensity score matching to minimize confounding; (3) quantitative bias analysis to assess robustness to unmeasured confounding; (4) negative control outcomes validating study design; (5) comprehensive assessment of overall irAE burden, severity, and organ-specific patterns; (6) extended 36-month follow-up capturing late irAEs; (7) rigorous subgroup analyses to identify effect modifiers; and (8) multiple sensitivity analyses confirming robustness of the overall irAE association.

Limitations include: (1) retrospective observational design limiting causal inference despite target trial emulation approach; (2) potential for unmeasured confounders, particularly performance status (ECOG score); the quantitative bias analysis (VanderWeele-Ding formulation, 50,000 Monte Carlo draws) yields a median adjusted HR of 1.034 with a 61.7% probability that the true HR exceeds 1.0 under plausible priors (RR_CD and RR_CE sampled from log-normal distributions with median 1.7), and an E-value of 1.74 (lower bound 1.62). This indicates only partial robustness and represents the principal caveat to our findings. However, convergent evidence from a propensity-score model incorporating performance-status proxy variables (see online supplemental eTable S16; HR 1.211, 95% CI 1.158 to 1.267), null negative-control outcomes (see online supplemental eTable S8), and an ocular-specific negative-control outcome (incident cataract and glaucoma; see online supplemental eTable S24, HR 1.03, 95% CI 0.91 to 1.17) together argue against performance-status confounding or differential ophthalmological surveillance as dominant explanations; (3) incomplete capture of mild irAEs managed exclusively in outpatient settings; (4) lack of CTCAE severity grading in administrative data; (5) variation in irAE monitoring practices across healthcare organizations; and (6) 36-month follow-up may miss very late irAEs (>3 years).¹

Although differential follow-up between groups (mean 23.0 vs 21.0 months) could theoretically introduce surveillance time bias, multiple complementary analyses argue against this explanation. Competing risk analysis with death as the competing event demonstrated persistent irAE excess (CIF difference: 12.9 percentage points at 36 months). Fixed follow-up sensitivity analyses with identical observation windows confirmed the association (HR 1.086 at 2 years, p=0.019). Landmark analyses excluding early events yielded consistent results (HR 1.08–1.12). The stability of the irAE association after incorporating performance status proxy variables (HR 1.211 vs 1.22, <1% change) provides empirical evidence that performance status confounding does not substantively explain the vaccination-irAE relationship.¹⁶

The vaccinated subpopulation (2.3% of ICI-treated patients within the 100-day window) is inherently selected. Patients who receive vaccination proximate to ICI may differ systematically from the broader ICI-treated population in unmeasured ways, limiting generalizability of absolute effect estimates. However, the biological mechanism of mRNA vaccine-induced immune priming modifying irAE risk is unlikely to be population-specific.

Calendar-time stratification confirmed temporal consistency of the irAE association (HR 1.194 for 2021–2022 and HR 1.213 for 2023–2025), arguing against confounding from evolving vaccination practices, viral variant transitions, or monitoring intensity. By contrast, the mortality benefit was more pronounced in 2023–2025 (HR 0.827) than in 2021–2022 (HR 1.021), possibly reflecting reduced COVID-19-related mortality in the later period that allowed a clearer signal of vaccine-associated survival benefit to manifest, or improved real-world vaccine effectiveness against contemporary Omicron subvariants.

The gradual increase in HR from 1.03 (1 month) to 1.22 (36 months) is more consistent with a cumulative biological mechanism (progressive immune sensitization from vaccine-primed immune memory interacting with ongoing ICI therapy) than with acute confounding, which would produce its strongest effect early and attenuate over time.

Future research

Future studies should: (1) prospectively validate the overall irAE increase in randomized trials if ethically feasible; (2) investigate mechanisms underlying the overall irAE burden and organ-specific patterns; (3) develop risk prediction models for overall irAE risk incorporating vaccination status, genetic markers, and baseline autoantibody profiles; (4) evaluate optimal timing and dosing strategies to minimize overall irAE burden while maintaining efficacy; and (5) assess whether specific vaccine formulations (original vs bivalent) differentially affect overall irAE incidence.

Conclusions

COVID-19 mRNA vaccination is associated with a clinically meaningful 22% increase in overall immune-related adverse events in patients with cancer receiving ICIs, including an 11% increase in severe irAEs. This overall increase is primarily driven by ocular manifestations (48% increase), with non-significant trends for dermatologic and rheumatologic involvement. Multiple sensitivity analyses confirmed the robustness of the overall irAE association across different analytical approaches, though moderate sensitivity to unmeasured confounding was demonstrated through quantitative bias analysis. However, substantial survival benefits (14% mortality reduction, 57% fewer ICU admissions) clearly outweigh the increased irAE burden. Vaccination should continue in all eligible patients with enhanced monitoring for overall irAE burden, particularly ophthalmological surveillance for high-risk groups (females, patients aged ≥50 years). Our findings support integration of COVID-19 vaccination into standard oncology care protocols with proactive multidisciplinary monitoring strategies to facilitate early detection and management of irAEs.

The funding source had no role in the design and conduct of the study; collection, management, analysis, and interpretation of the data; preparation, review, or approval of the manuscript; and decision to submit the manuscript for publication.

Data availability statement

Data may be obtained from a third party and are not publicly available.