Early versus late onset hematologic immune-related adverse events following immune checkpoint inhibition: Temporal patterns, clinical profiles, and risk stratification in faers reports (2014–2025)

Abstract Background: Hematologic immune-related adverse events (hem-irAEs) are rare but severe complications of immune checkpoint inhibitors (ICIs), with poorly characterized temporal dynamics. Aggregating early and late events may obscure clinically meaningful differences. We analyzed hem-irAEs from the FDA Adverse Event Reporting System (FAERS) to delineate early (<6 weeks) versus late (≥6 weeks) onset patterns, clinical phenotypes, and agent-specific risks. Methods: We linked FAERS DEMO, THER, REAC, and OUTC datasets (2014–2025) using R to identify reports with suspected ICI exposure and hematologic preferred terms (PTs). Onset latency was calculated as event date minus ICI start date. Events were classified as early or late using a 42-day cut-point; sensitivity analyses employed 28- and 56-day thresholds and excluded myelotoxic chemotherapy. Descriptive statistics, χ² tests, and logistic regression were used to characterize PT distributions and assess fatality and “seriousness” (death, life-threatening outcome, hospitalization, disability), adjusting for age, sex, and ICI class. Kaplan–Meier–style survival estimates and Cox models evaluated time to first hem-irAE. Trend tests assessed changes in early-onset proportions by year. Agent-level logistic models compared fatality odds (reference: nivolumab). Results: Among 216,408 hem-irAE reports, onset was nearly evenly split: 50.3% early and 49.7% late. Early events were enriched for acute cytopenias (neutropenia, thrombocytopenia), while late events included marrow-failure syndromes (aplastic anemia, hemophagocytic lymphohistiocytosis) (χ² = 2,667.9; df = 18; p < 0.001). Early-onset hem-irAEs increased from 30.8% in 2014 to 83.9% in 2025 (trend χ² = 1,017,677; p < 2.2 × 10⁻¹⁶). Crude fatality was slightly higher in early events (26.8% vs. 25.7%; p = 1.27 × 10⁻⁹). Adjusted models showed late onset was associated with lower odds of fatality (OR 0.93; 95% CI 0.91–0.95) but higher odds of serious outcome (OR 1.40; 95% CI 1.37–1.44). Median time to first hem-irAE was ~60 days. Compared to PD-L1 inhibitors, CTLA-4 inhibitors carried a 5.2% higher hazard (HR 1.05; 95% CI 1.003–1.103; p = 0.035), and PD-1 inhibitors a 2.9% higher hazard (HR 1.03; 95% CI 1.008–1.050; p = 0.006). Agent-level analysis showed significantly reduced fatality odds with pembrolizumab (OR 0.44; 95% CI 0.42–0.46), atezolizumab (OR 0.54; 95% CI 0.52–0.56), avelumab (OR 0.54; 95% CI 0.49–0.59), durvalumab (OR 0.53; 95% CI 0.46–0.60), and ipilimumab (OR 0.84; 95% CI 0.79–0.88) versus nivolumab. Combination regimens showed intermediate fatality risk (OR 0.69; 95% CI 0.67–0.71). Conclusions: Early and late hem-irAEs represent distinct clinical entities. Early events, driven by cytopenias, may reflect acute immune activation, while late events involve marrow failure with greater morbidity. Despite marginally lower fatality, late hem-irAEs were more often serious. Reporting trends indicate increased early detection, possibly due to greater awareness or earlier surveillance. Agent-specific risk profiles suggest potential for personalized monitoring. These findings support time-adapted surveillance strategies to optimize hematologic safety during ICI therapy.