Abstract: TH-PO0875
Near-Real-Time Surveillance of US 5008x High-Volume Hemodiafiltration: An Artificial Intelligence (AI)-Accelerated Real-World Evidence Platform
Session Information
- Dialysis: Prescription, Physiology, and Treatment Optimization
October 22, 2026 | Location: Exhibit Hall A, Convention Center
Abstract Time: 10:00 AM - 12:00 PM
Category: Dialysis
- 801 Dialysis: Hemodialysis and Frequent Dialysis
Authors
- Blankenship, Derek M., Renal Research Institute, New York, New York, United States
- Torgbenu, Moses, Renal Research Institute, New York, New York, United States
- Lax, Julianne, Renal Research Institute, New York, New York, United States
- Ficociello, Linda, Renal Research Institute, New York, New York, United States
- Chatoth, Dinesh K., Fresenius Medical Care (GMO), Waltham, Massachusetts, United States
Background
The US rollout of the 5008x machine and HighVolumeHDF (HVHDF) requires robust monitoring of real-world efficacy, specifically tracking mortality, hospitalizations, and missed treatment-related outcomes. Because traditional retrospective analyses are too slow for an active national device rollout, this platform optimizes analytical speed and statistical rigor. It includes a broad, organizational-level view of the entire patient population as well as pre-specified sub-cohort analyses. The purpose is to obtain a comprehensive reporting platform driven by rigorous methods that minimizes time to actionable clinical insights.
Methods
The system relies on a robust backend data pipeline harmonizing longitudinal electronic medical record data. The SAS-based data processing and analysis workflow and an R-Shiny reporting graphical user interface (GUI) were developed using a disciplined "human-in-the-loop" approach, combining biostatistical and programming expertise with code development and optimization via the Gemini large language model. This pipeline embeds core epidemiological techniques, including propensity score generation and adjusted statistical models, to update the reporting platform monthly.
Results
Bridging the SAS backend to the R-Shiny frontend, the platform ingests the pipeline's de-identified patient and analysis output and guides users from a high-level view to granular clinical details via a sequential tab structure. The platform visualizes broad cohort accrual and exclusion patient flow, followed by baseline balance evaluations assessing demographic and clinical comparability between HVHDF patients and standard high-flux controls. An outcome analysis dynamically visualizes hazard ratios, incidence rate ratios, and their detailed data for the efficacy endpoints. A dedicated module performs sample size calculations based on continually accruing real-world data.
Conclusion
The semi-automated data and analysis pipeline with the interactive reporting platform allows for rapid understanding of complex clinical data. This AI-accelerated framework delivers clinical rigor with rapid translation to practice, empowering clinical decision-makers to make immediate, evidence-based decisions regarding patient outcomes during the national 5008x rollout and is generalizable to other quality improvement initiatives.
Funding
- Commercial Support – Fresenius Medical Care