In this paper, the design of an expert proportional-integral-derivative (PID) control system for blood glucose regulation in patients, such as the intensive care population, is described. The control system applied the concept of expert system, and proportional, integral and derivative control in clinical sliding table technique, to effect a control algorithm in the form of an "active" sliding table that is used to prescribe insulin infusion rates. This combination provided dynamic properties to the sliding table control. Clinical results have shown a comparable performance between the control system and routine clinical treatment, in terms of blood glucose level maintained. Nevertheless, the control system is sensitive to sensor reading artefact, particularly in the lower ranges of blood glucose level, mandating manual intervention.

A study was conducted to determine if continuous subcutaneous glucose monitoring (from MiniMed CGMS) could be used in real-time to control blood sugar level (BSL) in patients with critical illness. A closed-loop control system was constructed to use CGMS in a real-time manner, coupled with a proportional integral (PI) control algorithm based on a sliding scale approach, for automatic intravenous infusion of insulin to patients. A total of five subjects with high BSL (> 10 mmol/L) participated in formal studies of the closed-loop control system. Subjects were recruited from critically ill patients in the intensive care unit (ICU) after informed consent was obtained. Error grid analysis showed that 64.6% of the BSL readings as determined in real time using CGMS sensor, when compared to conventional BSL measurements on blood drawn from an arterial line, was clinically accurate (i.e., < 20% deviation from glucometer value). In the five patients who underwent closed-loop control, the controller managed to control only one patient's glycaemia without any manual intervention. Manual intervention was required due to the real-time sensor reading deviating more than 20% from the glucometer value, and also as a safety mechanism. Test on equality of mean and variance for BSL attained prior to, during, and post trial showed that the controller's performance was comparable to manual control. We conclude that the automatic sliding scale approach of closed-loop BSL control is feasible in patients in intensive care. More work is needed in the refinement of the algorithm and the improvement of real-time sensor accuracy.

The theory of H infinity optimal control has the feature of minimizing the worst-case gain of an unknown disturbance input. When appropriately modified, the theory can be used to design a "switching" controller that can be applied to insulin injection for blood glucose (BG) regulation. The "switching" controller is defined by a collection of basic insulin rates and a rule that switches the insulin rates from one value to another. The rule employed an estimation of BG from noisy measurements, and the subsequent optimization of a performance index that involves the solution of a "jump" Riccati differential equation and a discrete-time dynamic programming equation. With an appropriate patient model, simulation studies have shown that the controller could correct BG deviation using clinically acceptable insulin delivery rates.

A closed-loop control system was constructed for automatic intravenous infusion of insulin to control blood sugar levels (BSL) in critically ill patients. We describe the development of the system. A total of nine subjects were recruited to clinically test the control system. In the patients who underwent closed-loop control of BSL, the controller managed to control only one patient's glycaemia without any manual intervention. The average BSL attained during closedloop control approached the target range of 6-10 mmol/l, and had less deviation than when BSL had been maintained manually. We conclude that closed-loop BSL control using a sliding scale algorithm is feasible. The main deficiency in the current system is unreliability of the subcutaneous glucose sensor when used in this setting. This deficiency mandates high vigilance during use of the system as it is being developed.

No abstract