Artificial intelligence -driven GVHD scoring and response tracking: Advancing FACT compliance, clinical trial readiness, and evidence-based therapy in a transplant program

Abstract Introduction : Graft-versus-host disease (GvHD) remains a major challenge in allogeneic transplantation. Despite standardized grading systems, real-world assessments often lack consistency. To improve compliance with FACT standards and support clinical trial enrollment, our BMT Unit collaborated with the AI Department to implement a quality improvement project using AI for structured GvHD documentation and tracking. Methods: We implemented a two-phase quality improvement initiative. Phase 1 focused on education and standardization: GvHD scoring workshops, eGVHD App training, and integration of NIH 2014/MAGIC-based acute and chronic GvHD templates into the EMR. Phase 2 leveraged in-house AI algorithms to extract GvHD grading and treatment response data from clinical notes in the Vista system into a real-time dashboard. Key performance indicators (KPIs) included accuracy of GvHD severity scoring and documentation of treatment response at predefined time points. Metrics were compared pre- and post-intervention to evaluate the impact on compliance, accuracy. Results: Structured documentation using NIH and MAGIC criteria improved consistency in GvHD classification. Scoring accuracy increased from 45% to 80% after introducing the eGVHD App and standardized templates. Response documentation at pre-defined time points rose from <30% to 70%. The AI-driven dashboard extracted all relevant data from EMR notes, enabling real-time tracking of GvHD severity and treatment response. This not only enhanced clinical oversight but also flagged high-risk or steroid-refractory patients, facilitating timely intervention and eligibility screening for clinical trials. The system allowed accurate, near-complete capture of key GVHD metrics essential for FACT accreditation and transplant outcomes reporting. Clinician adherence to documentation standards improved markedly, with measurable increases in grading accuracy, timeliness of intervention, and integration of care and research workflows. Conclusion: Applying AI-driven solutions with structured documentation significantly improved acute and chronic GvHD assessment, leading to better scoring accuracy and response tracking. This proactive approach enhanced the quality of GvHD data for patient care and research, and ensured compliance with guideline-defined standards and FACT accreditation requirements.