The aim of the study is to determine whether the use of the CGAOtm software is associated with a decrease in 90-day mortality when compared with the use of standard care methods for glucose control with target blood glucose levels inferior to 180 mg/dl. The CGAOtm software is designed to assist physicians and nurses in achieving tight glucose control (defined by a target for blood glucose levels between 80 and 110 mg/dl) in critically ill patients.

Hyperglycemia in response to critical illness has long been associated with adverse outcomes.

In 2001, the first "Leuven study", a randomized controlled trial conducted in surgical intensive care patients comparing a strategy based on a nurse-driven protocol for insulin therapy in order to maintain normal blood glucose levels [80 - 110 mg/dl] with standard care defined at the time as intravenous insulin started only when blood glucose level exceeded 215 mg/dl and then adjusted to keep blood glucose level between 180 and 200 mg/dl, showed a reduction in hospital mortality by one third.

The results of this trial have been enthusiastically received and rapidly incorporated into guidelines, such as the Surviving Sepsis Campaign in 2004, and now endorsed internationally by numerous professional societies.

However, subsequent randomized controlled trials have failed to confirm a mortality benefit with intensive insulin therapy among critically ill patients, in whom stress hypoglycemia is common. Moreover the Normoglycemia in Intensive Care Evaluation - Survival Using Glucose Algorithm Regulation (NICE-SUGAR) study, an international multicentre trial involving 6104 patients, the largest trial of insulin therapy to date, showed a lower 90-day mortality in the control group targeted blood glucose levels inferior to 180 mg/dl when compared to the intervention group with tight glucose control [80 - 110 mg/dl].

In addition, many studies and meta-analyses have reported high rates of hypoglycemia with tight glucose control. Consequently, considerable controversy has emerged as to whether tight glucose control is warranted in all critically ill patients especially as tight glucose control (without appropriate computer protocol) causes a significant increase in nurse workload.

The conflicting results between the first Leuven study and the NICE-SUGAR study could be explained by numerous differences between the two trials : the specific method (algorithms, compliance of nurses and physicians with recommendations, etc) used to achieve tight glucose control in each randomized control trial could be a major issue.

Several experimental and observational studies have highlighted the possible negative impact of glucose variability (large fluctuations in blood glucose possibly with undetected hypoglycemia and hypokalemia alternating with hyperglycemia) when implementing tight glucose control, be it due to the intrinsic properties of the algorithms used, technical factors (errors in measurements of the blood glucose level or lack of control over intravenous insulin therapy) or human factors (delay in performing glucose measurements or non respect of recommendations not based on clinical expertise but as a consequence of insufficient training inducing a lack of confidence in the algorithms by inexperienced nurses).

Therefore, remaining concerns about the best way to achieve glucose control in the ICU reduce the impact of conclusions of all of the recent randomized controlled trials on tight glucose control : are the negative results due to the concept, tight glucose control with intensive insulin therapy in critically ill patients in order to reduce the toxicity of high blood glucose levels, or are the negative results mainly due to specific methods used for achieving tight glucose control ? In most cases the methods used in clinical trials were never tested in numerical patients according to existing and validated models (in SILICO expertise) before implementing them in clinical practice on real patients.

Particularly, whether the use of a clinical computerized decision-support system (CDSS) designed for achieving tight glucose control in various ICU settings, and fine-tuned to reduce glucose variability, without increasing the incidence of severe hypoglycemia nor the nurse workload, has an impact on the outcome of patients staying at least three days in an ICU remains to be tested.

Among the different CDSS, the CGAOtm software has been developed to standardize different aspects of glucose control in an ICU setting based on 1) explicit replicable recommendations following each blood glucose level measurement concerning insulin rates and time to next measurement, 2) reminders and alerts and 3) various graphic tools, trends, and individual on-line data aiming to increase the confidence of the nursing staff in the computer protocol and therefore their adherence, to reduce necessary training time, and to give physicians and nurses a way to control the tight glucose control process during the whole ICU stay. Moreover, the CGAOtm software is designed to take into account irregular sampling, saturations, and some precision and stability issues.

The aim of the study is to evaluate the capability of the CGAOtm software to reduce 90-day mortality in a mixed ICU population of patients requiring intensive care for at least three days.

Sample size and power calculations. The expected all cause 90-day mortality in the control group is 25 % (identical to the observed all cause 90-day mortality in the control group of the NICE-SUGAR trial). Considering that all cause 90-day mortality in the experimental group (computer protocol group) is expected to be 22 % (absolute reduction of 3 %), considering an alpha risk and a beta risk respectively of 0.05 and 0.20 and three intermediate analyses performed according to the O'Brien-Fleming design, 3,211 patients per treatment arms are needed and will be recruited from the participating 60 centres, all located in France.

Inclusion Criteria:

• At time of the patient's admission to the ICU, the treating ICU specialist expects the patient will require treatment in the ICU that extends beyond the calendar day following the day of admission.

Exclusion Criteria:

• Age < 18 years or patient under guardianship.
• Pregnancy.
• Moribund patient or imminent death in the ICU (e.g. patient expected to die in the ICU within 24 hours).
• At time of the patient's admission, the treating physicians are not committed tu full supportive care.
• Patient admitted to the ICU for treatment of diabetic ketoacidosis or hyperosmolar state.
• Patient admitted to the ICU for hypoglycemia.
• Patient thought to be at abnormally high risk of suffering hypoglycemia (e.g. known insulin secreting tumor or history of unexplained or recurrent hypoglycemia or fulminant hepatic failure).
• Patient who have suffered hypoglycemia without documented full neurological recovery
• Patient is expected to be eating before the end of the day following admission.
• Patient previously enrolled in the CGAO-REA study.