Keywords: Optimal Design, Anesthesia, Personalized Medicine, Pharmacokinetics-Pharmacodynamics, Closed-loop Control
During the delivery of an anesthetic for a surgical procedure it is important to maintain a constant and stable level of sedation in the patient, with too little anesthetic resulting in intra-operative awareness and too much anesthetic potentially causing patient injury or death. However, heterogeneity in patient pharmacokinetics (PK) and pharmacodynamics (PD) make it difficult to predict how an individual patient will respond to a given quantity of drug. This has provided motivation for establishing closed-loop control of anesthetic delivery, in which the infusion of an anesthetic is controlled by a computerized syringe that receives signals indicating the patient’s Depth of Hypnosis (DoH) and adjusts the infusion rate accordingly. While several model-based controllers of DoH have been proposed, their performance is tied to the ability to learn patient-specific PK-PD parameters, which must occur in order to maintain the desired DoH and respond robustly to unanticipated surgical excitations. In our work, we consider ways in which to transition a patient from an awake state to a state of general anesthesia, referred to as the induction period, that are optimal for online learning about a patient’s specific PK-PD parameters. Published PK-PD models are used to build an empirically informed prior distribution that represents a hypothetical population and from which representative patients are simulated. This prior distribution is iteratively updated using Laplace Approximations to the posterior distribution while induction periods are optimized with respect to the Bayesian D- and I-optimality criteria. For each simulated patient, we measure the information gain associated with measurements from an optimal induction period relative to standard induction procedures. We further show how this additional information translates into more accurate prediction of a patient’s PK-PD and more robust closed-loop control of DoH in the presence of unanticipated external stimuli.