• Tom De Smet, Michel M R F Struys, Scott Greenwald, Eric P Mortier, and Steven L Shafer.
• Demed Engineering, Temse, Belgium.
• Anesth. Analg. 2007 Dec 1;105(6):1629-38, table of contents.

BackgroundImplementing Bayesian methods in a model-based closed-loop system requires the integration of a standard response model with a patient-specific response model. This process makes use of specific modeling weights, called Bayesian variances, which determine how the specific model can deviate from the standard model. In this study we applied simulations to select the Bayesian variances yielding the optimal controller for a Bayesian-based closed-loop system for propofol administration using the Bispectral Index (BIS) as a controlled variable.MethodsThe relevant Bayesian variances determining the modeling process were identified. Each set of such Bayesian variances represents a potential controller. The set, which will result in optimal control, was estimated using calculations on a simulated population. We selected 625 candidate sets. Similar to our previous closed-loop performance study, we applied a simulation protocol to evaluate controller performance. Our population consisted of 416 virtual patients, generated using population characteristics from previous work. A BIS offset trajectory similar to a surgical case was used.ResultsWe were able to develop, describe, and optimize the parameter setting for a patient-individualized model-based closed-loop controller using Bayesian optimization. Selection of the optimal set yields a controller performing with the following median absolute prediction errors at BIS targets 30, 50, and 70: 12.9 +/- 2.87, 7.59 +/- 0.74, and 5.76 +/- 1.03 respectively.ConclusionsWe believe this system can be introduced safely into clinical testing for both induction and maintenance of anesthesia under direct observation of an anesthesiologist.

16 keywords...

hide…