Bleed-versus-clot paradox under immune-checkpoint inhibition: A decade-long US faers analysis reveals divergent class-specific hemostatic phenotypes

Abstract Background: Immune-checkpoint inhibitors (ICIs) have transformed cancer therapy but can destabilize coagulation, leading to both thrombosis and immune-mediated bleeding. Current evidence is fragmented and rarely compares bleeding and thrombotic signals directly across ICI classes. We conducted a comprehensive analysis of hemostatic adverse events across all approved ICIs using the U.S. FDA Adverse Event Reporting System (FAERS) over a ten-year period. Methods: We analyzed FAERS data from January 2014 through March 2025 using R 4.3.2. ICIs were grouped as PD-1, PD-L1, or CTLA-4 agents. Bleeding and thrombosis events were defined using pre-specified MedDRA preferred terms. Reporting odds ratios (RORs) were calculated versus all FAERS reports. Multivariable logistic regression adjusted for age, sex, and anticoagulant use. Sensitivity analyses excluded VEGF agents, restricted to primary-suspect reports, and stratified by era (pre-2020 vs. 2020–2025). Time-to-event and 30-day fatality were also assessed. Results: Among 84,285 ICI-exposed reports, 1,327 (1.6 %) involved bleeding and 2,164 (2.6 %) thrombosis; 96 reports had both. Atezolizumab had the strongest thrombotic signal (ROR 1.60, 95 % CI 1.44–1.78), while toripalimab and atezolizumab had the highest bleeding signals (2.08, 1.32–3.27; and 2.60, 2.31–2.93, respectively). Toripalimab's bleed-to-clot ratio (BCR) was 6.3, contrasting with BCR < 1.0 for legacy agents. Median onset was 58 days for bleeding and 62 days for thrombosis. Thirty-day fatality was 0.08 % (113/147,999) for bleeding and 0.07 % (221/294,517) for thrombosis. In adjusted models, toripalimab was associated with increased bleeding (aOR 19.2, p = 0.005) and reduced thrombosis (aOR 0.05, p = 0.003). Anticoagulation decreased bleeding risk (aOR 0.68, p = 0.019) but increased thrombotic risk (aOR 1.45, p = 0.029). Age modestly increased bleeding risk (aOR per decade 1.06, p = 0.051); sex was not predictive. Sensitivity analyses confirmed robustness. Post-2020, toripalimab's bleeding signal intensified (aOR ≈ 21). Conclusion: FAERS data reveal divergent hemostatic profiles under ICI therapy: PD-L1 blockade favors thrombosis, while toripalimab, a novel PD-1 antibody, drives immune-mediated bleeding. These class-specific toxicities appear mechanistically distinct and temporally early, often manifesting within the first 60 days of treatment. This suggests that current approaches treating all ICIs as a uniform risk group may obscure critical safety signals. Toripalimab warrants early and proactive bleeding surveillance, including platelet and coagulation monitoring, while PD-L1 therapies may require delayed-onset thrombo-prophylaxis tailored to patient-specific risk. These findings should inform future immuno-oncology guidelines, support mechanistic research into immune-driven hemostatic dysfunction, and guide rational risk-adapted management