This study was a prospective, randomized, controlled trial conducted in an adult medical ICU of a tertiary medical center that has 1,235 beds in total, 59 of which are adult ICU beds. Trial and consent forms were approved by the Institutional Review Board of Kaohsiung Veterans General Hospital. The study was carried out between January 2006 and December 2006. Procedures were in accordance with the Helsinki Declaration.

Patients aged 18 or over who were admitted to the medical ICU with blood glucose concentrations over 180 mg/dl were eligible for inclusion. The criteria for exclusion included prior surgical treatment, pregnancy, participation in another study, patients with chronic renal loss, and patients who were expected to require treatment in the ICU for less than 4 days. Chronic renal loss was defined as persistent acute renal failure with the complete loss of kidney function for more than 4 weeks 22. The first author (C-WH) enrolled participants. A total of 283 patients were evaluated and 171 patients were excluded. The remaining 112 patients were eligible for the study (Figure 1).

After informed consent was obtained from patients or the next of kin, patients were randomly assigned to receive either MIT or CIT using software-generated simple randomization procedures (computerized random number) without blocking. Except for the interventionists (nurses and doctors), patients and other staff members were not informed of the group assignments. In the MIT group, continuous insulin infusion was started when the blood glucose concentrations exceeded 140 mg/dl in order to maintain a blood glucose concentration between 120 and 140 mg/dl. The insulin dose was adjusted using a neuro-fuzzy method; the insulin protocol is shown in Additional file 1. In the CIT group, continuous insulin infusion was delivered when blood glucose concentrations exceeded 200 mg/dl; the insulin dose was then adjusted to maintain a blood glucose concentration between 180 and 200 mg/dl.

All patients were fed either intravenously (total parenteral nutrition) or enterally starting the day after ICU admission. Enteral feeding was attempted as early as possible at the discretion of the attending physician. Once enteral feeding was started, patients were administered full-strength isotonic commercial formula Jevity (Abbott Laboratories, Brockville, ON, Canada) commencing at 20 ml/hour, and increased by 20 ml/hour every 4 hours to satisfy energy and protein requirements recommended by the clinical dietitian following the Canadian clinical practice guidelines for critically ill patients 23. All patients did not receive an albumin infusion during the study period.

At the time of randomization, demographic data and clinical characteristics were obtained. Blood glucose concentrations were measured upon admission and, subsequently, every 1 to 4 hours in all patients. Baseline serum albumin concentrations, prealbumin concentrations (Hitachi 7600; Hitachi, Tokyo, Japan), serum creatinine, and 24-hour urinary urea nitrogen (UUN) levels were collected and evaluated after beginning the study, and these data were collected and evaluated on days 3, 7, and 14 of the study. Additionally, we used full 24-hour urine collection rather than spot urine samples to determine the UUN. Blood cultures were obtained whenever the central body temperature exceeded 38.5°C or other clinical signs of sepsis were present. Blood transfusions were conducted according to the protocol when hemoglobin levels decreased to below 7 g/dl and were continued until target hemoglobin levels of 7 to 9 g/dl were attained. A higher hemoglobin level was required in patients with special circumstances; for example, myocardial ischemia, severe hypoxemia, acute hemorrhage, or lactic acidosis 24.

From the time of randomization to the time of discharge from the ICU, daily caloric and protein intakes, all blood glucose measurements, doses of insulin, red-blood-cell transfusions, blood cultures that were positive for pathogenic organisms, renal function, gastrointestinal bleeding, moderate hypoglycemia, and severe hypoglycemia were recorded.

Patients were classified as having diabetes on the basis of their medical history. Previous treatment with corticosteroids was defined as treatment with systemic corticosteroids for 72 hours or more immediately before randomization. Sepsis was defined by the presence of both infection and a systemic inflammatory response 25. Acute renal injury was defined when serum creatinine concentrations increased by 2.0 times or the glomerular filtration rate decreased by more than 50% 22. Creatinine clearance was calculated using the Cockcroft-Gault equation 26:

14 0 - age years × body weight kg × 0 . 85 if female / 72 × serum creatinine mg / dl

Bloodstream infection was diagnosed either when a recognized pathogen was isolated from a blood culture, or in the presence of one of fever (> 38°C), chills, or hypotension and any of the following: a common skin contaminant isolated from two blood cultures that were drawn on separate occasions; a common skin contaminant isolated from the blood culture of a patient with an intravascular access device, and the physician instituted appropriate antimicrobial therapy; or a positive antigen blood test 27.

Gastrointestinal bleeding was defined as the presence of hematemesis, melena, bright red blood per rectum, or a coffee grounds-like substance that was aspirated from the feeding tube. Patients with hypoglycemia were defined as those who developed at least one episode of hypoglycemia. A moderate hypoglycemic event was defined as a blood glucose concentrations ≤ 60 and > 40 mg/dl. A severe hypoglycemic event was defined as blood glucose concentration ≤ 40 mg/dl.

The 24-hour nitrogen balance was calculated using the following formula 28:

Nitrogen balance = protein intake / 6 . 25 - UUN + 4

The difference in daily insulin dose between the two groups was the total mean insulin dose of the MIT group minus the total mean insulin dose of the CIT group on the same day.

Primary outcomes included 24-hour UUN levels, nitrogen balance, serum albumin concentrations, and prealbumin concentrations. Secondary outcomes included ICU days, ventilator days, hospital days, acute renal injury, bloodstream infection, blood transfusions, gastrointestinal bleeding, moderate hypoglycemia, severe hypoglycemia, and hospital mortality rate.

All data were analyzed by SPSS version 12.0 (SPSS, Inc., Chicago, IL, USA). Data are presented as the mean ± standard deviation, median (interquartile range), or number (percentage). All data were analyzed according to the intention-to-treat principle. To detect differences of nitrogen balance using a two-sided 5% significance level and a power of 85% when the difference and standard deviation were equal to 0.80 and 1.97, a sample size of 55 patients per group was necessary. This value was based on previous studies involving nitrogen balance and different doses of insulin treatment in a burn ICU 21. Primary outcomes, such as 24-hour UUN levels, nitrogen balance, serum albumin concentrations, and prealbumin concentrations were analyzed with a generalized linear model for repeated measures using dummy variables. Student's t test was used to compare continuous variables with normally distributed data. Wilcoxon's rank-sum tests were used to compare continuous variables with non-normally distributed data. Chi-square tests were used to compare dichotomous variables. All P values were two-tailed. P < 0.05 was considered significant.

All participants were recruited from April 2006 to December 2006. This trial ended when it reach the sample size goal. Of the 112 patients randomized to the study, 55 patients received MIT and 57 patients received CIT. All patients were available for the intention-to-treat analysis (Figure 1). Table 1 shows the patient baseline characteristics of all the patients enrolled in the study. Demographic data at the time of randomization showed no significant differences in any of the parameters. Figure 2 shows serum creatinine, creatinine clearance, and urine output during the study period between the two groups; there were no significant differences. The two groups shared similar background characteristics.

Actual daily protein (grams per day and per kilogram of body weight) and caloric (calories per day and per kilogram of body weight) intake, blood glucose concentrations, and insulin dose (per hour corrected for caloric intake) are shown in Figure 3. Daily protein and caloric intake progressively increased until day 5, but then fluctuated after day 5 in both groups. The two treatment groups had similar protein and caloric intakes. The mean blood glucose concentration was 125.3 ± 2.4 mg/dl in the patients of the MIT group and 199.9 ± 4.0 mg/dl (P < 0.01) in the patients of the CIT group. The median daily insulin dose was 82 (69 to 111) units/day in the patients of the MIT group and 37 (26 to 43) units/day (P < 0.01) in the patients of the CIT group (Table 2). To maintain the low blood glucose concentrations, patients in the MIT group were administered a significantly higher insulin dose than patients in the CIT group. The insulin requirement per calorie decreased each day in both groups (Figure 3). The difference in mean daily insulin doses between the two groups ranged from 58 IU/day on day 1 to 43 IU/day on day 14. The difference in insulin requirements per day between the two groups decreased each day (Figure 4).

Both groups showed a trend of decreasing levels of excreted urinary nitrogen, better nitrogen balance, and increasing serum albumin and prealbumin concentrations during their hospital courses. A generalized linear model for repeated measurements using dummy variables revealed that there were significantly lower 24-hour urine nitrogen excretion (P = 0.027) (Figure 5) and higher serum albumin (P = 0.047) and prealbumin (P = 0.001) concentrations (Figure 6) during the study period in the MIT group. The MIT group exhibited a higher nitrogen balance than the CIT group; however, this difference was not statistically significant (P = 0.070). Differences in 24-hour urine nitrogen excretion, nitrogen balance, serum albumin concentrations, and prealbumin concentrations between the two groups were most significant on day 3 when all factors showed significant differences (P < 0.05). These differences decreased after day 3.

There were no statistically significant differences in ICU days, ventilator days, hospital days, rate of acute renal injury, bloodstream infection, red-blood-cell transfusion, and gastrointestinal bleeding in the two groups (Table 2).

Moderate hypoglycemia (blood glucose levels ≤ 60 mg/dl and > 40 mg/dl) occurred more often in the MIT group than in the CIT group, but this difference was not statistically significant (18.2% vs. 10.5%, P = 0.10; 2.4 episodes vs. 1.5 episodes per 100 treatment days, P = 0.18) (Table 2). The CIT group also exhibited a lower rate of severe hypoglycemia (blood glucose levels ≤ 40 mg/dl), but the difference was not significant (3.6% vs. 1.8%, P = 0.53; 0.3 episodes vs. 0.2 episodes per 100 treatment days, P = 0.44). No hemodynamic deterioration, convulsions, or neurological sequelae were noted in association with any hypoglycemic event in either group.

Patients with MIT had a trend for lower hospital mortality rate than that of patients with CIT, but they did not reach a statistically significant difference.