Hyperglycemia is common in critically ill patients and leads to various severe complications and even death. Keeping blood glucose within the range of 80–110 mg/dL (4.4–6.1 mmol/L) has been shown to reduce mortality and morbidity in intensive care units (ICU). Many studies on BG control systems for ICU patients have been reported. However, it is not easy to maintain blood glucose within the desired range because of the time variability of insulin sensitivity in critically ill patients. In this study, to improve the prediction accuracy of blood glucose level in patients, we modified a glycometabolism model developed in our previous study, by identifying parameter values from clinical ICU data. Then, we modified insulin sensitivity online identification algorithm to avoid a sudden change in insulin sensitivity during online identification that updates insulin sensitivity value at intervals of 30 min. Finally, since hypoglycemia prevention as important, we designed a glycemic control system using nonlinear model predictive control based on the modified model and the online identification algorithm of insulin sensitivity. The new glycemic control system achieved 71% of blood glucose measurements within the range of 80–110 mg/dL and 1.5% of measurements below 80 mg/dL, which indicated effectiveness and safety.