Digital Twin–Enabled Robust MPC for Personalized Ultrafiltration Scheduling in Hemodialysis: A Proof-of-Concept Study

Abstract Background: Safe fluid removal during hemodialysis (HD) is challenging when the ultrafiltration rate (UFR) exceeds vascular refilling capacity, reducing circulating blood volume and increasing the risk of intradialytic hypotension (IDH).

Methods

The study develops a digital twin (DT)– enabled robust model predictive control (MPC) framework for personalized UFR scheduling. A mechanistic fluid-volume model and unscented Kalman filter estimate absolute blood volume (ABV) and vascular refilling from hematocrit measurements and propagate estimation uncertainty into tightened safety constraints. A residual machine-learning model is evaluated separately for longer-horizon hematocrit prediction.