Robust closed-loop control of systemic oxygenation in acute lung injury

Tight control of systemic oxygenation is a key component of mechanical ventilation of critically ill patients. Yet, closed-loop control remains challenging due to the non-linear, uncertain, and time-varying relationship between the inspired oxygen and the resulting patient's oxygen saturation, especially in cases of acute lung injury. This paper uses a Hammerstein model to combine a clinically validated static model incorporating ventilation-to-perfusion mismatch with uncertain linear dynamics to synthesize a robust closed-loop controller for regulating systemic oxygenation in subjects with acute lung injury. Data from an experimental lung injury model in five pigs, in combination with data augmentation, allowed modeling of the uncertainty in the system. Subsequently, gain-scheduled proportional-integral controllers were designed using the mixed-sensitivity 7i infinity approach. The resulting closed-loop systems were analyzed for robust stability, and the performance was evaluated in a thousand synthetically generated subjects spanning the overall parameter space in various combinations. A mean rise time of less than three minutes was achieved for recovering from the severely hypoxemic state for simulated subjects with a range of ventilation-to-perfusion mismatch and arterial acid- base status. Furthermore, the results from a pilot animal experiment matched the simulated results well, and the system could respond to significant disturbances. The system was safe and reliable in both simulations and the pilot animal experiment. The methodological approach for uncertainty modeling and controller synthesis is an essential step toward the practical deployment of closed-loop control of systemic oxygenation in the acutely injured lung.