Early detection of emerging safety signals in GVHD prophylaxis agents using a 12-year faers “slope-watch” approach

Abstract Background: Over the past decade, the prevention of graft-versus-host disease (GVHD) after allogeneic hematopoietic stem cell transplantation has expanded beyond calcineurin inhibitors to include a diverse array of agents. In addition to tacrolimus and cyclosporine, contemporary prophylaxis regimens employ sirolimus and everolimus, mycophenolate mofetil, methotrexate, post-transplant cyclophosphamide (PTCy), JAK inhibitors (ruxolitinib, baricitinib, tofacitinib), co-stimulation blockade with abatacept, gut-homing integrin antagonism via vedolizumab, and the ROCK2 inhibitor belumosudil. While these therapies have reduced GVHD incidence, they carry novel and sometimes delayed toxicities, ranging from opportunistic infections to endothelial injury, immune dysregulation, and neuro-metabolic syndromes, that may not become apparent until months after widespread use. Traditional pharmacovigilance in the FDA Adverse Event Reporting System (FAERS) relies on cumulative disproportionality measures, which often detect safety signals only after they are well established. To provide earlier warning, we developed a “slope-watch” method that tracks year-to-year changes in the Information Component (IC), a Bayesian disproportionality metric. By focusing on the annual difference in IC (IC-Δ), we aimed to identify accelerated reporting of drug–event pairs 12–24 months before formal regulatory actions. Method: We retrieved all FAERS case reports from January 2014 through June 2025, isolating ~17.5 million “primary suspect” entries for thirteen GVHD prophylaxis agents. Reaction terms were mapped to eleven clinically relevant MedDRA Preferred Terms and grouped into four categories: infections (sepsis, cytomegalovirus infection, BK virus infection, fungal pneumonia), immune dysregulation (post-transplant lymphoproliferative disorder, hemophagocytic lymphohistiocytosis), endothelial syndromes (veno-occlusive disease, capillary-leak syndrome, thrombotic microangiopathy), and neuro-metabolic events (posterior reversible encephalopathy syndrome, hyperlipidemia). These mappings yielded 26.3 million reaction records. For each drug–event pair in each calendar year, we counted observed co-reports (aₜ) and calculated expected co-reports (Eₜ) under independence using annual drug and event marginals. Annual IC was computed as log₂[(aₜ + 0.5)/(Eₜ + 0.5)], with slope IC-Δ as ICₜ – ICₜ₋₁. A drug–event was flagged as an emerging signal if ICₜ > 0 and IC-Δ > 0.5 for two consecutive years. A lower-bound IC₂₅ was estimated using standard error under normal approximation. Results: Sequential IC-Δ identified 24 emerging signals between 2016–2025 (median inflection: 2023). Infections dominated (58%), followed by immune dysregulation (17%), endothelial (13%), and neuro-metabolic events (12%). Examples include:PTCy →PRES: IC-Δ rose +0.82 in 2019 (aₜ = 8; ICₜ = 2.45; IC₂₅ ≈ 1.21), preceding the March 2021 FDA label update by ~18 months.Ruxolitinib → CMV infection: Consecutive IC-Δ of +0.59 (2019) and +0.72 (2020) led to ICₜ = 2.02 (IC₂₅ ≈ 0.98), identifying risk over a year before the February 2022 warning.Abatacept →PTLD: IC rose from –0.28 (2017) to +0.89 (2023), with sustained IC-Δ ≥ 0.5 and IC₂₅ ≈ 2.16 by 2022; no warning exists as of mid-2025. Additional signals included:Mycophenolate → BK virus infection (2019; ICₜ = 4.16)Methotrexate → EBV-associated lymphoproliferation (2022; ICₜ = 5.14)Tacrolimus → Pneumocystis pneumonia (2025; ICₜ = 3.63) Conclusion: Sequential IC-Δ offers a robust, forward-looking pharmacovigilance framework that identifies adverse event acceleration 12–24 months before regulatory action. By emphasizing directional change and incorporating a simple confidence approximation, this approach enhances early signal detection in large safety datasets. Integration with electronic health records, regulatory pipelines, and interactive dashboards could further streamline horizon scanning and improve patient safety in GVHD prophylaxis.