Evaluation of a long-range adaptive predictive controller for computerized drug delivery systems.

Abstract

A closed-loop adaptive control system, based on the generalized predictive control law with a terminal matching condition, has been developed for computerized drug delivery. The control law is a minimization of the squares of prediction errors over a small future prediction horizon plus weighted square of the prediction error at steady-state. A control-relevant, long-range identification algorithm is used for on-line parameter estimation. Since the control and identification are mutually compatible, the system truly satisfies the approximate dual control criterion. The system has been applied to the control of mean arterial pressure (MAP) by automatic infusion of sodium nitroprusside in the presence of physical and physiological constraints. Experimental evaluation on six mongrel dogs, in an ethics-approved manner, included setpoint tracking and regulation of MAP in the presence of unpredictable disturbances. The system was found to be capable of inducing hypotension in an average of 2.44 +/- 0.31 min (mean +/- standard error of mean) after probing without any overshoots in mean arterial pressure. The nitroprusside infusion was also free of any ringing. When the subjects were not disturbed, 96.2% of mean arterial pressure remained within 5 mm Hg of the target pressure. A series of disturbances introduced in the presence and absence of closed-loop control affirms the robustness and effectiveness of this control system.