Signal-rich and signal-poor: Divergent detection of early bacterial sepsis and latent viral complications after hematopoietic stem cell transplantation in the FDA adverse event reporting system

Abstract

Abstract Background Bacterial sepsis, cytomegalovirus (CMV) reactivation, Epstein–Barr virus post-transplant lymphoproliferative disorder (EBV-PTLD), and invasive fungal infection (IFI) are canonical threats after allogeneic hematopoietic stem cell transplantation (allo-HSCT). Pharmacovigilance databases such as the FDA Adverse Event Reporting System (FAERS) provide population-scale insights but are limited by under-reporting and variable clinical detail. Whether FAERS differentially captures acute, high-severity events (e.g., septic shock) versus protocol-managed viral complications (e.g., pre-emptively treated CMV DNAemia) remains uncertain and has implications for safety signal interpretation in transplantation. Methods We analyzed FAERS reports from January 2014 to June 2025. Allo-HSCT cases were identified using MedDRA indication terms, with a drug-based heuristic classifying records listing a graft-versus-host disease (GVHD) prophylactic agent (e.g., tacrolimus, cyclosporine, sirolimus, or post-transplant cyclophosphamide) as allogeneic. Autologous HSCT (auto-HSCT) served as a biologic comparator. Valid transplant and event dates yielded 2,208 allo and 178 auto recipients. Four infection clusters, bacterial sepsis, CMV, EBV-PTLD, and IFI, were captured using exact and substring MedDRA searches. Events were anchored to transplant date and classified as Early (0–100 days) or Late (>100 days). We calculated incidence proportions, crude risk ratios (RR), reporting odds ratios (ROR), and Bayesian Information Components (IC₂₅), using 0.5 continuity corrections. Results Early bacterial sepsis occurred in 118 of 2,208 allo recipients (5.3%) versus 2 of 178 auto recipients (1.1%) (RR 4.8, 95% CI 1.1–13.9; ROR 3.9, 95% CI 1.11–13.9). Late sepsis rates were similar (1.9% vs 2.8%; ROR 0.4, 95% CI 0.19–1.01). CMV reactivation was infrequent: Early CMV occurred in 46 allo (2.1%) and 3 auto (1.7%) cases (ROR 1.0, 95% CI 0.34–3.03); Late CMV was reported in 10 allo (0.5%) and 3 auto (1.7%) recipients (ROR 0.17, 95% CI 0.05–0.61). Contemporary meta-analyses report 10–20% clinically significant CMV incidence despite letermovir prophylaxis, underscoring under-ascertainment in FAERS. EBV-PTLD (six Early, five Late) and IFI (two Early) were too sparse for stable ROR estimation, despite known registry-reported rates of 1–3%. Only Early sepsis yielded a positive IC₂₅; all others were neutral or negative, mirroring ROR trends. Conclusions FAERS functions as a “signal-rich” detector for sudden, high-severity complications such as early bacterial sepsis but is “signal-poor” for protocol-managed or indolent infections like CMV or EBV-PTLD. The clear sepsis signal aligns with prospective bloodstream infection studies, validating FAERS for acute event surveillance. Conversely, the lack of allo-specific CMV or EBV-PTLD signal likely reflects spontaneous-report limitations rather than biological equivalence. Regulators and investigators should prioritize FAERS for severe, early toxicities, while relying on transplant registries and EHR linkages to track prophylaxis-modulated or slow-onset infections. Future integration of FAERS with registry denominators may help calibrate pharmacovigilance sensitivity across the post-HSCT infection spectrum.