Nonlinear model-based predictive control of non-depolarizing muscle relaxants using neural networks.

Neuromuscular blockade can be relatively easily measured in the clinical setting. Consequently, closed-loop control can be exercised by measuring the neuromuscular activity, calculating the dose of drug necessary to achieve a predefined degree of neuromuscular blockade and finally directing an infusion pump. Recently introduced short-acting blocking agents like mivacurium provide benefits for the clinical routine due to a small onset time and half life. In order to provide a stable blockade for different groups of patients a fast and highly adaptable control unit is needed. Furthermore its development should not imply costly investigations for determining a pharmacological model. The fulfilling of these requirements yield a self-adapting model-based predictive control system. The application of artificial neural networks allows an appropriate adjustment of specific parameters without the knowledge of inner pharmacodynamic processes. In a clinical study the EMG module within a Datex AS/3 monitor was used to measure the blockade and a Grasepy 3500 infusion pump for i.v. administration of mivacurium to 35 patients (ASA I-III). The performance of the novel system (mean of the T1 error: -0.32 +/- 1.7) compares favourably with closed-loop controllers demonstrated in the past. These promising results and the easy adaption to other blocking agents encourage to apply this technology even for delivering hypnotic drugs.