Tight blood glucose (BG) control has been shown to decrease morbidity and mortality in critically ill patients 1 but is difficult to achieve using standard insulin infusion protocols. We evaluated glucose control, using a software model predictive control (MPC) insulin administration algorithm, in a prospective randomized controlled multicenter comparison with standard care in three European hospitals (Royal Brompton Hospital [RBH], Medical University Graz [MUG], Charles University Hospital [CUP]).

Sixty ventilated patients (20 in each center) admitted to intensive care following elective cardiac surgery, with an arterial BG > 6.7 mmol/l within 4 hours of admission, were randomized to BG control by the standard insulin protocol of the participating ICU or MPC advised insulin infusion. All patients had BG measured hourly. Standard care (n = 30) involved insulin infusion in two centers (RBH, CUP) and insulin boluses in the third (MUG). The MPC algorithm was derived from software developed for closed loop glucose control in ambulatory diabetic patients 2. MPC, installed on a bedside computer, requires input of patient chronic insulin requirements, weight, carbohydrate intake and BG concentration. Insulin infusion rate advice for the next hour is displayed, targeted to maintain BG at 4.4–6.1 mmol/l. The study was continued for at least 24 hours with a maximum duration of 48 hours.

The percentage of glucose measurements in the target range were significantly greater in the MPC group over the first 24 hours compared with standard care: 52% (17–92) vs 19% (0–71), (median [min-max]), P < 0.01. Two hypoglycemic events (BG < 3 mmol/l) occurred in patients receiving standard care.

The MPC algorithm was safe and effective in controlling postoperative hyperglycaemia in this patient group.