Model-based tight blood glucose control is possible with a validated patient-specific glucose-insulin regulatory system model that captures the fundamental dynamics. Chase and colleagues 212338 and Hann and colleagues 38 used a model that captured the rate of insulin utilisation, insulin losses, and saturation dynamics and that has been validated using retrospective data 383940, clamp data 41, and several short-term (not longer than 24 hours) clinical control trials 3637. The model thus captures the metabolic status of the highly dynamic ICU patient and uses it to provide tight control. However, computational resources are not available in some critical care units for effective computerised control methods, and their complexity can limit easy large-scale implementation required to test overall safety and efficacy. Hence, a simpler paper-based method was developed to mimic this protocol. SPRINT was implemented as a clinical practice change at the Christchurch Hospital Department of Intensive Care in August 2005. Further details on SPRINT, its development, and initial pilot study can be found in 272842.

The entry criterion for the SPRINT protocol was a blood glucose measurement of greater than 8 mmol/L on two occasions during standard patient monitoring, where the 8 mmol/L represents the upper limit of clinically desirable glycaemic control in the Christchurch ICU. Patients were occasionally put on SPRINT at the discretion of the clinician if the blood glucose levels were consistently greater than 7 mmol/L in severe critical illness. Patients were not put on the protocol if they were not expected to remain in the ICU for more than 24 hours. Data were collected for all blood glucose measurements, insulin administered, and nutrition given to the patient. The Upper South Regional Ethics Committee, New Zealand, granted ethics approval for the audit, analysis, and publication of these data.

Hourly blood glucose measurements are used to ensure tight control 27. Two-hourly measurements are used when the patient is stable, defined as three consecutive 1-hourly measurements in the 4.0 to 6.0 mmol/L band 2742, or when an arterial line is not present. SPRINT is stopped when the patient is adequately self-regulating, defined as 6 or more hours (three 2-hourly measurements) in the 4.0 to 6.0 mmol/L band with over 80% of the goal feed rate and a maximum of 2 U/hour of insulin 2742.

Total insulin prescribed by SPRINT is limited to 6 U/hour to minimise saturation and the administration of ineffective insulin 23293043. Insulin is given predominantly in bolus form for safety, avoiding infusions being left on at levels inappropriate for evolving patient condition. Occasionally, doctors prescribed a background insulin infusion rate of 0.5 to 2 U/hour, primarily for patients known to have type II diabetes, and the insulin bolus recommendations from SPRINT were added to this background rate. A background rate of 0.5 to 1.0 U/hour, to which SPRINT bolus insulin is added, is mandated in patients with type I diabetes.

Goal enteral nutrition rates are approximately 25 kcal/kg per day of RESOURCE Diabetic (Novartis Medical Nutrition, Minneapolis, MN, USA) or Glucerna (Abbott Laboratories, Abbott Park, IL, USA) with 34% to 36% of calories from carbohydrates 44. Minimum and maximum nutrition rates are 7.5 to 25 kcal/kg per day, with 2.7 to 9 kcal/kg per day from carbohydrates. Thus, an 80-kg male would receive a maximum of 2,000 kcal/day and a minimum of 600 kcal/day, with 216 to 640 kcal/day from carbohydrates, exceeding the minimum level below which there is an increased risk of bloodstream infections 45. These guidelines are detailed by Shaw and colleagues 26 and are approximately equivalent to the ACCP guidelines 35.

Baseline variables were compared using the two-tailed Mann-Whitney U test or chi-square test. Change in mortality was compared between the SPRINT and historical cohorts by means of the chi-square test. The Mann-Whitney and chi-square tests were used to compare blood glucose metrics between survivors and non-survivors. MINITAB^® Release 14.1 (Minitab Inc., State College, PA, USA) was used for statistical comparisons, and for all statistical tests, P values of less than 0.05 were considered significant.

Log-normal statistics were used to provide an accurate description of blood glucose control results as negative blood glucose concentrations are not possible and typical distributions of blood glucose measurements are asymmetric and show a skew toward higher concentrations. The design of the protocol was that, for periods outside the ideal target range, short periods of higher blood glucose levels were preferred over hypoglycaemic events. Thus, the distributions for blood glucose are right-skewed and log-normal.

SPRINT was implemented as a clinical practice change and thus was the sole method of treatment for hyperglycaemia. A retrospective cohort has been used to infer changes in patient outcome due to SPRINT. This cohort was extracted from all intensive care patients for the 20-month period of January 2003 to August 2005. Figure 1 shows the selection of patients into the SPRINT and retrospective patient cohorts. Entry criteria into the retrospective cohort were an ICU length of stay of at least 1 day and at least two blood glucose measurements of more than 8 mmol/L spaced not more than 24 hours apart. Patients were excluded where there were insufficient clinical data available to compute an Acute Physiology and Chronic Health Evaluation (APACHE) II score. There was no set protocol for treating hyperglycaemia in the Christchurch ICU during the retrospective period, and clinicians often used a variety of insulin sliding scales.

The retrospective patient pool had a larger proportion of operative cardiovascular patients, and the SPRINT patient pool had a larger proportion of gastrointestinal patients. Changes in the economics of health care caused changes in the types of patients admitted to the Christchurch ICU over the 4-year period encompassed by the SPRINT and retrospective data. The difference in cardiothoracic patients between the patient pools may have resulted from less case throughput and better pre-intensive care glycaemic control. Thus, to provide better-matched cohorts, retrospective operative cardiovascular patients and SPRINT gastrointestinal patients were randomly eliminated from the patient pools to create the cohorts used for analysis, as shown in Figure 1. The patient elimination procedure was repeated 100 times to create 100 cohorts. To present the data clearly, the median cohort results are presented based on mortality outcome for analysis in this article. The major results and outcomes were unaffected by the specific cohort iteration.

The clinical details of this retrospective cohort are compared with the SPRINT cohort by means of baseline variables, APACHE II scores, and APACHE III diagnosis codes in Table 1.

Table 2 presents a comparison of glycaemic control for the 371 SPRINT protocol patients against the 413 patients from the retrospective cohort. Measurements (27,664) were recorded for more than 44,769 hours of patient control on SPRINT compared with 13,162 measurements for 43,447 recorded hours of retrospective data. Patients on SPRINT had their blood glucose measured every hour during 24% of their time on the protocol and every 2 hours over the remaining 76% where there was improved glycaemic stability. Log-normal mean blood glucose levels in the SPRINT cohort for hourly and 2-hourly measurements were 6.3 mmol/L (standard deviation 1.6 mmol/L) and 5.6 mmol/L (standard deviation 1.1 mmol/L), respectively. The mean time between measurements in the SPRINT cohort was 1 hour 36 minutes compared with 3 hours 18 minutes for the retrospective cohort. The precision of the recordkeeping system in the Christchurch ICU is to the nearest hour, and nursing staff typically measured blood glucose and used the protocol on the hour.

The percentage time in the 4.4 to 6.1 mmol/L band defined by van den Berghe and colleagues 213 was 53.9% compared with 30.0% in the retrospective cohort. Hypoglycaemia was comparable to the retrospective cohort, with only 0.1% of measurements less than 2.2 mmol/L. SPRINT had a higher proportion of measurements below the 4.4 mmol/L limit; however, the two cohorts were comparable for measurements below the 4.0 mmol/L lower limit of the SPRINT target band. Per-patient results show that the mean and standard deviation of blood glucose for SPRINT are lower. Additionally, the interquartile range for both metrics amongst patients is tighter and thus there is less variability in glycaemic control performance between patients. Figure 2 shows a tightly controlled distribution of blood glucose measurements for all patients along with the 4.4–6.1 mmol/L range.

The mean overall hourly insulin usage on SPRINT was 2.8 U/hour, which is a level that avoids insulin saturation effects 293043. The median feed level recommended by SPRINT was 66.1% of the patient-specific goal feed 42. The mean overall nutrition rate was 1,283 kcal/day on SPRINT during periods when the patients was being fed, including via the parenteral route, compared with 1,599 kcal/day for the retrospective cohort. The mean nutrition rate over the entire length of stay, including periods in which feed was stopped for reasons outside glycaemic control, was 1,014 kcal/day on SPRINT. When no enteral or parenteral nutrition was recorded in the retrospective cohort data, it was not clear whether the nutrition administration was halted for clinical reasons or because the patient had begun eating meals. Thus, a nutrition comparison with the retrospective cohort was possible only for periods when the patient was receiving enteral or parenteral alimentation.

Figures 3 to 5 show the average percentage of measurements in the 4.4 to 6.1 mmol/L band, the average blood glucose concentration, and the average blood glucose standard deviation for patients grouped by starting blood glucose level and APACHE II score. The percentage of measurements in the target band was 66% to 203% higher and the blood glucose standard deviation was 6% to 30% lower on SPRINT compared with the retrospective cohort.

Figure 6 shows the box-and-whisker plot of hourly blood glucose concentration for all patients over first 48 hours on SPRINT. After approximately 7 hours, the blood glucose median and spread reach their average levels. This level of control is essentially maintained for the remainder of the period. Table 3 shows that 96% of SPRINT patients reached the 6.1 mmol/L band from the initial hyperglycaemic state compared with only 74% of the retrospective hyperglycaemic patients. SPRINT, therefore, brings a patient under control within 7 to 8 hours and maintains a constant level of performance.

Figure 7 shows the average nutrition intake and insulin administration rate for the first 7 days on the SPRINT protocol. The average nutrition intake is lower and the average insulin rate is higher during the initial phase of controlling hyperglycaemia. Once hyperglycaemia has been controlled, the average nutrition rate recommended by the protocol increases, generally as patient condition improves and carbohydrate tolerance increases, whilst average insulin administration rate remains relatively constant.

Figure 8 shows the percentage mortality for both the SPRINT and retrospective patients for both in-hospital and ICU mortality, grouped by length of ICU stay, for several iterations of the cohort selection procedure described in Figure 1.

Table 3 shows the change in mortality, both in-ICU and in-hospital, for patients with lengths of stay of at least 1 to 5 days, compared with the retrospective cohort using the chi-square test, for the median iteration of the cohort selection procedure. As length of ICU stay increases, the reduction in mortality becomes statistically stronger. Statistical significance (P < 0.05) is achieved for an ICU stay of 3 days or longer for in-hospital mortality.

Several recent studies have identified hyperglycaemia as a risk factor for mortality in critical care 1219464748. Table 4 compares average blood glucose, maximum blood glucose, and range of blood glucose between SPRINT ICU survivors and non-survivors by means of the Mann-Whitney test. There is no statistically significant difference between survivors and non-survivors for any of these glycaemic metrics.