A Closed-Loop Digital Twin Framework for Automated Control of Extracorporeal Membrane Oxygenation: Toward Physiology-Aware Decision Support in Critical Care

Extracorporeal membrane oxygenation (ECMO) is a life-saving therapy for severe cardiopulmonary failure, yet its management remains highly manual, experience-dependent, and vulnerable to delayed or suboptimal adjustments in rapidly evolving clinical states. Clinicians must continuously balance oxygen delivery, carbon dioxide removal, and hemodynamic stability using sparse, intermittently sampled physiologic data, creating substantial cognitive and operational burden in high-acuity intensive care settings. Here, we present a closed-loop digital twin framework for physiology-aware ECMO control that integrates a mechanistic cardiopulmonary model with constrained model predictive control to enable real-time, adaptive decision support. The digital twin explicitly represents patient-specific interactions between cardiac function, pulmonary gas exchange, and extracorporeal support, allowing continuous estimation of physiologic state and forward prediction under changing clinical conditions. The framework was evaluated in silico across representative clinical scenarios, including acute respiratory distress syndrome and cardiogenic shock, capturing distinct pathophysiologic regimes encountered in ECMO practice. Simulation results demonstrate that the proposed approach maintains arterial oxygenation and carbon dioxide targets while preserving hemodynamic stability and respecting clinically meaningful safety constraints. Compared with static or heuristic control strategies, the digital twin-driven controller exhibits improved robustness to disturbances and parameter uncertainty, supporting consistent performance across heterogeneous disease states. This study establishes a foundation for digital twin-based ECMO decision-support systems that augment clinician oversight with transparent, physiology-grounded intelligence. By enabling predictive, interpretable, and adaptable control in silico, this work advances the broader vision of digital medicine in critical care and provides a scalable pathway toward safer, data-informed extracorporeal support. O_FIG O_LINKSMALLFIG WIDTH=200 HEIGHT=102 SRC="FIGDIR/small/26343307v1_ufig1.gif" ALT="Figure 1"> View larger version (33K): org.highwire.dtl.DTLVardef@14cff90org.highwire.dtl.DTLVardef@1e6661org.highwire.dtl.DTLVardef@1d36f77org.highwire.dtl.DTLVardef@3b1799_HPS_FORMAT_FIGEXP M_FIG "Digital Twin-based Model Predictive Control for Extracorporeal Membrane Oxygenation (ECMO)." A mechanistic 0D cardiopulmonary digital twin integrates patient physiology with oxygenator diffusion dynamics. This model enables an MPC controller to autonomously titrate ECMO settings (Qb, sweep gas, FiO2) to maintain precise physiological targets in simulated ARDS and Cardiogenic Shock cohorts. C_FIG