Intensive Care Med (2006) 32:867–874DOI 10.1007/s00134-006-0135-4 O R I G I N A L

Imogen MitchellSimon FinferRinaldo BellomoTracey HiglettANZICS Clinical Trials GroupGlucose ManagementInvestigators

Management of blood glucose in the criticallyill in Australia and New Zealand: a practicesurvey and inception cohort study

Received: 3 October 2004Accepted: 23 February 2006Published online: 19 April 2006© Springer-Verlag 2006

Electronic supplementary materialThe electronic reference of this article ishttp://dx.doi.org/10.1007/s00134-006-0135-4. The online full-text version of this articleincludes electronic supplementary material.This material is available to authorisedusers and can be accessed by means of theESM button beneath the abstract or in thestructured full-text article. To cite or link tothis article you can use the above reference.

I. Mitchell (�)The Canberra Hospital, Intensive Care Unit,Garran, 2605, ACT, Australiae-mail: imogen.mitchell@act.gov.auTel.: +61-2-62443305Fax: +61-2-62443507

S. FinferUniversity of Sydney Northern ClinicalSchool, Royal North Shore Hospital,Intensive Care Unit,St Leonards, 2065, NSW, Australia

R. BellomoAustin Hospital, Intensive Care Unit,Studley Road, Heidelberg, 3084, VIC,Australia

T. Higlett · ANZICS Clinical Trials GroupGlucose ManagementInvestigatorsANZICS Clinical Trials Group,ANZICS House, Level 3, 10 Ievers Terrace,Carlton, 3053, VIC, Australia

Abstract Objective: To documentcurrent management of blood glu-cose in Australian and New Zealandintensive care units (ICUs) and toinvestigate the association betweeninsulin administration, blood glucoseconcentration and hospital outcome.Design and setting: Practice sur-vey and inception cohort study inclosed multi-disciplinary ICUs inAustralia and New Zealand. Pa-tients: Twenty-nine ICU directorsand 939 consecutive admissions to29 ICUs during a 2-week period.Measurement and results: Datacollected included unit approachesto blood glucose management, pa-tient characteristics, blood glucoseconcentrations, insulin administrationand patient outcomes. Ten percentof the ICU directors reported usingan intensive insulin regimen in alltheir patients. In 861 patients (91.7%)blood glucose concentration wasgreater than 6.1 mmol/l, 287 (31.1%)received insulin, and the median bloodglucose concentration triggering in-sulin administration was 11.5 (IQR9.4–14) mmol/l. Univariate analysisdemonstrated that non-survivorshad a higher maximum daily bloodglucose concentration (12 mmol/l,9.4–14.8, vs. 9.5, 7.6–12.2) and were

more likely to receive insulin (47%vs. 28%). Multiple logistic regressionanalysis showed age (OR per 5-yeardecrease 0.93, 95% CI 0.87–1.00)and APACHE II (OR per point de-crease 0.87, 95% CI 0.84–0.90) tobe independently associated withhospital mortality. After controllingfor age and APACHE II both dailyhighest blood glucose (OR 0.95, 95%CI 0.90–1.00) and administration ofinsulin (OR 0.62, 95% CI 0.39–1.00)were independently associated whenadded to the model alone; neitherwas independently associated whenthey were simultaneously included inthe model. Conclusion: Few Aus-tralian and New Zealand ICUs haveadopted intensive insulin therapy. Inthis study, insulin administration andhighest daily blood glucose concen-tration could not be separated in theirassociation with hospital mortality.

Keywords Glycaemic control · Hy-perglycaemia · Hypoglycaemia ·Insulin · Hospital mortality ·Intensive care

Introduction

Hyperglycaemia is a common finding in acutely ill patients[1, 2] and is associated with an adverse outcome [1, 3, 4, 5].

It may worsen outcome through impairment of white cellfunction, greater susceptibility to infection [6, 7, 8], detri-mental effects on the cardiovascular system [9] and multi-organ failure [10]. Despite these observations, the concept

868

of normalising glucose concentrations in critically ill pa-tients has not been widely adopted as the fear of induc-ing occult hypoglycaemia has outweighed the perceivedpotential benefits of normalising blood glucose concentra-tions. A degree of hyperglycaemia has been regarded asan acceptable component of the normal stress response incritically ill patients [11].

More recently, however, a large, single centre, ran-domised, controlled trial found that an intensive insulinregimen maintaining a blood glucose concentration of4.4–6.1 mmol/l decreased hospital mortality in venti-lated surgical intensive care patients [12]. Although theauthors cautioned against extrapolating their findings topatient groups that were not represented in their study, ifthese findings could be reproduced in a large phase IIImulti-centre randomised controlled trial in an unselectedpopulation of medical and surgical critically ill patients,the intensive insulin regimen would likely be adoptedworldwide, and a significant number of deaths could beprevented.

As a prelude to such a study, the Australian and NewZealand Intensive Care Society Clinical Trials Group(ANZICS CTG) conducted a practice survey and aninception cohort study to determine current blood glucosemanagement practice in predominantly tertiary Australianand New Zealand intensive care units (ICUs) and todetermine the relationship between insulin administration,blood glucose concentration and in-hospital death.

Method

Practice survey

Forty-five ICUs affiliated to the ANZICS CTG weree-mailed a blood glucose practice survey (see ElectronicSupplementary Material) on 14 March 2003. Twenty-nineintensive care unit (ICU) directors answered the survey by21 March 2003. The survey enquired as to their familiaritywith the Van den Berghe et al. study [12], whether inten-sive insulin therapy (IIT) had been adopted in their ICU,and, if so, in which group of patients and their reasons, ifany, for not adopting IIT.

Inception cohort study

Forty-five ICUs affiliated with the ANZICS CTG wereinvited to take part in the study and 29 accepted the invi-tation. The 29 ICUs that accepted nominated an intensivecare specialist as a principal investigator. The protocol wasdeveloped by open discussion at ANZICS CTG meetingsand refined by e-mail correspondence using a closedmailing list. Of the 29 ICUs 25 were in public hospitals.Twenty-five ICUs were in Australia (from a possible totalof 158) and four in New Zealand (from a possible total of

25). Twenty-three were classified as tertiary level III ICUs(from a possible total of 74) by the Minimum Standardsof the Joint Faculty of Intensive Care Medicine of theAustralian and New Zealand College of Anaesthetists andthe Royal Australasian College of Physicians [13] and sixas level II ICUs (from a total of 71). (A level III unit isa tertiary referral unit capable of providing comprehensivecritical care including complex multi-system life supportfor an indefinite period; a level II unit should be capableof providing a high standard of general intensive care,including complex multi-system life support, whichsupports the hospital’s delineated responsibilities; a levelI unit should be capable of providing resuscitation andshort-term cardio-respiratory support for critically illpatients). All were closed multidisciplinary ICUs withpatient management being supervised by accredited ICUspecialists.

Ethical considerations

Twenty-nine institutional human research ethics commit-tees waived the need for patient consent, but three requiredapproval to collect data. Data were submitted to the coor-dinating centre in a de-identified format.

Patients

All consecutive admissions to participating ICUs duringany two consecutive weeks between 1 March and 31 Au-gust 2003 were included in the study. Only patients admit-ted to an ICU more than once during their hospital staywere excluded. In this instance only data from the firstICU admission were included. A total of 939 patients werestudied over 3,790 intensive care patient days; their char-acteristics are summarised in Table 1. The patient popula-tion was predominantly male, older than 60 years, over-weight (BMI > 25), and almost one-half were admittedfrom the operating theatre (45.9%). There were 165 pa-tients (17.6%) with a history of diabetes mellitus, includ-ing 46 (4.9% of overall cohort) who were taking insulin.

Data collection

For all patients admitted during the 2-week study periodbaseline data were collected. These included demographicdata, admission source, admission diagnosis, admissionbody mass index (BMI), Acute Physiology and ChronicHealth Evaluation (APACHE) II score [14], and co-morbidities. Specifically, data were collected detailing thepresence or absence of pre-existing diabetes mellitus andits management. During ICU stay daily data collectionincluded the highest and lowest blood glucose concen-tration, total daily intravenous glucose administration

869

Age (years; IQR) (n = 939) 63 (46–75)Gender: male (n = 936) 61.5%BMI (IQR) (n = 841) 25.7 (22.99–29.07)APACHE II score (IQR) (n = 939) 15 (10.5–22)Type of admission (IQR) (n = 939)

Medical 404 (43.0%)Elective surgery 328 (34.9%)Emergency surgery 149 (15.9%)Trauma 58 (6.2%)

ICU Admission Source (n = 938)Operating theatre 431 (45.9%)Emergency department 235 (25.1%)Ward 160 (17.1%)Other hospital 112 (11.9%)

Commonest diagnoses (n = 939)Cardiac surgery 114 (12.1%)Overdose 48 (5.1%)Pneumonia 43 (4.6%)Postoperative gastrointestinal neoplasm 36 (3.8%)Cardiac arrest 34 (3.6%)Head trauma, other 30 (3.2%)

ICU length of stay (days) (n = 939) 3 (2,5)Hospital length of stay (days) (n = 939) 11 (6, 21)ICU mortality (n = 939) 95 (10.1%)Hospital mortality (n = 933) 153 (16.3%)

Table 1 Patient characteristics ininception cohort study

(including glucose contained in parenteral nutrition),the initiation and method of administration of insulin tocontrol blood glucose and the blood glucose concentrationthat first triggered insulin administration. Patients werefollowed until death or hospital discharge; vital status wasrecorded at ICU and hospital discharge.

At the end of the study period copies of all data weresent to the coordinating centre at ANZICS House in Mel-bourne, Australia. Data were entered into a custom Ac-cess database (Microsoft, Seattle, Wash., USA) by a datamanager (T.H.) under the supervision of the study steeringcommittee (I.M., S.F. and R.B.). Data were checked for in-consistencies and logical errors on entry, and queries weresent to the source hospital for resolution. Where data aremissing, we report the number of observations and makeno assumptions about missing data.

Statistical analysis

Numeric data are presented as medians with interquartileranges (IQR). Categorical data are presented as counts andpercentages. Data were analysed using S-PLUS 6.1 forWindows (Insightful, Seattle, Wash., USA). Differencesbetween groups were assessed using Z tests for numericdata and χ2 tests for categorical data. The relationshipbetween hospital outcome, highest blood glucose con-centration during ICU stay and the administration ofintravenous insulin was determined using a multiplelogistic regression analysis model using hospital outcomeas the response variable. Administration of intravenousinsulin and the highest blood glucose during the ICUstay were the main exposure variables. The study ICU

number was included as a random effect to account forany differences between study ICUs. APACHE II score,age, sex, admission diagnosis, admission source and typeof ICU admission (medical, surgical or trauma) wereinitially included in the model as possible confoundingvariables. The model was refined by backward exclusionof the possible confounding variables; the multiple logisticregression analysis found that BMI was not significant,and because BMI was not available for a large number ofpatients (n = 98), it was excluded from further analysis.The degree of calibration of the models was tested usingthe Hosmer–Lemeshow goodness-of-fit statistic [15].

Results

Practice survey

The 29 ICU directors who responded to the practice surveyand took part in the subsequent audit were all familiar withthe Van den Berghe et al. study and its results (Table 2).Three (10.3%) had adopted IIT in all their patients. Nine(31.0%) used it in a selected patients, predominantly thosestaying in the ICU for more than 3 days, those diagnosedwith sepsis and those admitted from the operating theatre.The reasons given by 20 ICUs for not adopting ITT wereconcern about the risk of hypoglycaemia and the externalvalidity of the Van den Berghe et al. study, in particularthat the case mix and intensive care management differedgreatly from their own unit’s patient population and man-agement practices.

870

n

ICUs surveyed 29ICUs adopting IIT 12ICUs adopting IIT in all patients 3ICUs adopting IIT in selected patients 9Reasons for not implementing IIT

Practice not changed on basis of one single centre study 8ICU case mix not representative in single centre study 10ICU management not representative in single centre study 15Concern of risk of hypoglycaemia 9Do not believe the single-centre study 1

Table 2 Reported currentAustralasian blood glucosemanagement in the critically ill

Blood glucose concentration and insulin administration

Of the 939 patients 861 (91.7%) had a recorded bloodglucose concentration greater than 6.1 mmol/l and 340(36.2%) a concentration greater than 11.1 mmol/l at leastonce during their intensive care stay. The median valuefor a blood glucose concentration on ICU admission was7.2 mmol/l (IQR 5.8–9.3) while the median highest dailyblood glucose concentration during the ICU stay was9.9 mmol/l (7.8–12.6) and the median lowest concen-tration 4.9 mmol/l (2.5–6.2). Thirty patients (3.2%) hada lowest daily blood glucose concentration of 2.2 mmol/lor less. The median amount of intravenous glucoseadministered during the first 24 h of ICU admission was1.1 g (0–30) with a median daily dose for the whole studyperiod of 12.2 g (0–40).

Survived to hospital discharge Died in hospital

Age (years; IQR) 61 (44–74)*** 71 (60–79)Gender: male 60.4% 63.4%BMI (IQR) 25.8 (23.2–29.3)* 24.9 (22.5–28.7)APACHE II score (IQR) 14 (10–19)*** 25 (20–31)Type of admission 780*** 153

Medical 298 (38.2%) 105 (68.7%)Elective surgery 315 (40.4%) 9 (5.9%)Emergency surgery 116 (14.9%) 32 (20.9%)Trauma 51 (6.5%) 7 (4.6%)

ICU admission source 780*** 153Operating theatre 391 (50.1%) 35 (22.9%)Emergency department 190 (24.4%) 44 (28.6%)Ward 110 (14.1%) 50 (32.7%)Other hospital 89 (11.4%) 23 (15.3%)

First ICU blood glucose (mmol/l; IQR) 7.1 (5.9–9.1)** 7.6 (5.7–11.6)Highest blood glucose (mmol/l; IQR) 9.5 (7.6–12.2)*** 12 (9.4–14.8)Lowest glood glucose (mmol/l; IQR) 5.5 (4.5–6.5)*** 4.4 (2.3–6.1)Intravenous glucose first 24 h (IQR) 1.8 (0–32) 0 (0–22.4)Total intravenous glucose during ICU (IQR) 36 (0–109.3)* 33.8 (0–202.7)No. administered intravenous insulin 214 (28.0%)*** 71 (47.0%)ICU length of stay (days; IQR) 2 (2–4)** 3 (2–7)Hospital length of stay (days; IQR) 11 (6–21) 8 (2–21)Previous history of diabetes 137 (17.6%) 26 (17.0%)

* p < 0.05, ** p < 0.001, *** p < 0.0001

Table 3 Comparison of patientswho survived to hospitaldischarge with those who died

Intravenous insulin was given to 287 patients (31.1%)to control their blood glucose concentration during theirICU stay. The median blood glucose concentration whichtriggered the administration of intravenous insulin was11.5 mmol/l (9.4–14). The trigger blood glucose concen-tration did not differ significantly between the 29 ICUs,including the three ICUs that had purported that all theirpatients received “tight” glucose control, and the nine thatmanaged selected patients with “tight” glucose control.There was no significant difference in the number of dailyblood samples taken for glucose measurement among the29 hospitals (5.2 ± 3.8).

The patients who received intravenous insulin wereolder (67 years, 53–76, vs. 61, 43–75; p < 0.01), hada greater BMI (25.9 vs. 25.6, p < 0.01), higher APACHEII score (19 vs. 14, p < 0.0001), and were more likely

871

Table 4 Results of multiple logistic regression analysis expressed as odds ratio (OR) with 95% confidence intervals (CI) and significancelevels for all variables

OR 95% CI p

APACHE II score 0.87 0.84–0.90 < 0.001Per 5-year change in age 0.87 0.75–1.00 0.05Gender: male (reference: female) 0.67 0.42–1.09 0.10Source of admission: emergency department (reference: operating theatre) 1.31 0.50–3.39 0.58Source of admission: ward 0.74 0.30–1.84 0.51Source of admission: other hospital 1.10 0.43–2.86 0.84Type of admission: elective surgery (reference: medical) 5.38 1.53–18.95 0.01Emergency surgery 1.24 0.41–3.76 0.70Trauma 1.91 0.29–12.76 0.50Acknowledged tight glucose control protocol 0.94 0.57–1.55 0.79Highest daily blood glucose concentration 0.97 0.91–1.02 0.22Administration of intravenous insulin 0.74 0.43–1.27 0.27

to have pre-existing diabetes mellitus (40.8% vs. 7.4%,p < 0.0001). Patients who received intravenous insulin hada higher median daily highest blood glucose concentrationduring their ICU stay (13.6 vs. 8.7 mmol/l, p < 0.0001)received more intravenous glucose during their ICU staythan those who did not receive insulin (110 vs. 5.4 g,p < 0.0001) but did have a longer ICU length of stay(4 days, 2–9, vs. 2, 2–4; p < 0.001).

Patient outcomes

Ninety-five patients died in the ICU (ICU mortality10.1%) and a further 58 died following ICU discharge andprior to hospital discharge (hospital mortality 16.3%). Thepatients who died were older, had a higher APACHE IIscore and had a significantly lower BMI (Table 3). Theincidence of pre-existing diabetes mellitus did not differbetween survivors and non-survivors (17.6% vs. 17.0%,p = 0.96), although patients treated with intravenousinsulin during their intensive care stay had a higher unad-justed hospital mortality than those who did not receiveit (14.3% vs. 8.3% p < 0.001). Following adjustmentfor age, sex, APACHE II score and type of admissionthe hospital mortality rate of patients staying more than5 days in the three ICUs practising tight glucose controlwas similar to that of those in units not practising tightglucose control (without tight control 20.9%, and withtight control 29.9%; p = 0.29).

Data from the 889 patients were used for multiplelogistic regression analysis to determine whether therewas an association between highest daily blood glucoseconcentration, administration of intravenous insulinand hospital outcome. The effects of APACHE II score(p < 0.001), admission diagnosis (p < 0.001), and type ofadmission (p = 0.003) were statistically significant. Agewas significant at the 5% level. After controlling for theseeffects, both the highest daily blood glucose concentration(p = 0.043) and the administration of intravenous insulin(p = 0.043) were statistically significant when added sepa-rately to the model, but neither was significant when both

were added to the model simultaneously. Table 4 demon-strates odds ratio with 95% confidence levels for all factorsexcept admission diagnoses, when daily highest bloodglucose concentration and administration of intravenousinsulin are included in the model. Reliable estimates couldnot be calculated for individual diagnoses because of lackof sufficient data. Using the Hosmer–Lemeshow statistic,the goodness of fit of the three models was tested andlack of fit was not significant at the 5% level (model withhighest daily blood glucose: Hosmer–Lemeshow statistic10.1, p = 0.26, model with administration of intravenousinsulin: Hosmer–Lemeshow statistic 11.0, p = 0.20 andwith model with both: Hosmer–Lemeshow statistic 11.6,p = 0.17).

Discussion

Despite all the surveyed ICUs being familiar with theVan den Berghe et al. study [12] only 10% had adoptedIIT in all their patients. The remaining ICUs believedthat their patient population and clinical managementwere sufficiently different from that described by Vanden Berghe et al. to be concerned about the externalvalidity of the study’s results. This concern was borneout by our study, given the difference seen in our casemix and hospital outcome in similar diagnostic groups.Whilst these differences in mortality may well reflecta difference in patient selection, they highlight concernsabout the external validity of the Van den Berghe et al.results. The patient population in the Van den Berghestudy was also different from the Australian and NewZealand ICU population in a number of other ways,including having a substantially lower APACHE II score(9, 7–13, and 15, 10.5–22, respectively), lower incidenceof pre-existing diabetes mellitus (13% and 17.6%, re-spectively) and a higher tertiary referral rate (16.9% and11.9%, respectively).

Management of blood glucose in Australian and NewZealand ICUs in 2003 was similar to that of the conven-tional treatment group in the Van den Berghe et al. study.

872

Our finding that 31.1% of our patients received intravenousinsulin during their stay in ICU is somewhat less than the39.2% of patients in the conventional arm and substantiallyless than the 98.7% in those assigned to IIT group of theVan den Berghe et al. study. The lower percentage of ourpatients receiving insulin may be in part explained by thehigher median blood glucose concentration (11.5 mmol/l,9.4–14) triggering insulin administration than that used totrigger insulin administration in the conventional arm ofthe Van den Berghe et al. study (11.1 mmol/l). In addition,it is likely that fewer patients received insulin because ourpatients received a median only of 1.1 g intravenous glu-cose during their first day in ICU and one of only 12.2 gper day for the duration of their ICU stay. This amountof intravenous glucose administered is in sharp contrastto that used in the Van den Berghe et al. study. The highglucose load used in the latter would have required higherinsulin use, which in itself could have conferred a posi-tive mortality benefit. An alternative hypothesis is that thedetrimental effects of high-dose intravenous glucose wasresponsible for the findings in the Van den Berghe et al.study, whereby the administration of insulin offset theseeffects in the IIT group but not the control group. Whilstthe amount of enteral glucose administered in our patientsis potentially interesting, it was not collected because thefocus was primarily upon glucose control and the admin-istration of intravenous glucose. We sought to explore thehypothesis that the use of intravenous glucose in the Vanden Berghe et al. trial differed greatly from that used inAustralian and New Zealand ICUs.

In univariate analysis we found that patients who diedhad significantly higher blood glucose concentrationsthan those who survived, a finding that is in agreementwith other observational studies. Previous data have foundthat the level of admission blood glucose represents anindependent predictor of long-term prognosis followingacute myocardial infarction [16], cerebrovascular acci-dent [17], traumatic brain injury [18] and coronary arterybypass grafting [19]. However, in our multiple logisticregression analysis neither administration of intravenousinsulin nor highest daily blood glucose during ICU staywas an independent predictor of hospital mortality whenadded to the model simultaneously. Our findings are atodds with the results reported by Finney et al. [20] and arebest explained by increased blood glucose concentrationand administration of insulin being inextricably linked.The question of whether aggressive treatment of hyper-glycaemia with IIT improves outcome in a heterogeneousICU population, however, can only be answered with anappropriately designed randomised controlled trial with anadequate sample size [21]. Whilst a subsequent study usinga comparison with historical controls has suggested thatintensive blood glucose control benefits a heterogeneousICU patient population [22], no other group has reporteda randomised control trial demonstrating that intensiveglucose control reduces mortality. In a recent report from

the VISEP study [23] comparing IIT with conventionaltherapy in patients with severe sepsis and septic shock theGerman Competence Network Sepsis found no differencein 28-day (21.9% vs. 21.6%, p = 1.0) or 90-day mortality(32.8% vs. 29.5%, p = 0.43) but did find an increased riskof hypoglycaemia (12.1% vs. 2.1%) in the IIT group. Vanden Berghe et al. have recently published the results ofa randomised, controlled trial in medical ICU patients [24]which failed to demonstrate a significant mortality benefitusing IIT. As noted in the accompanying editorial [25], therole of IIT in heterogenous ICU patients remains unclear.

There are several potential limitations to our study, in-cluding the voluntary participation of ICUs. This raises theissue of the self-selected ICUs being non-representativeof the 183 Australian and New Zealand ICUs (74 levelIII, 71 level II and 38 level I), and it is possible that theresults are not representative of other ICUs in Australiaand New Zealand. However, those that did submit dataincluded both academic and non-academic institutions inboth metropolitan and non-metropolitan areas.

Another limitation to this study is the risk of theHawthorne effect during the 2-week period of data col-lection, the timing of which was self-selected. However,we believe that clinicians responsible for patient man-agement were largely unaware of the data collection, andthis reduced the likelihood that clinicians changed theirnormal day-to-day practice. The results of the study donot suggest changes in glucose management over time toa more aggressive control of blood glucose given that allICUs had similar blood glucose concentration triggers forthe use of intravenous insulin.

The initial survey identified ICUs that purported tobe performing tight glucose control, three ICUs in allpatients and nine in selected patients only. Potentially theobservations from these ICUs could influence the resultsfrom this inception cohort study. However, the patientsfrom the three ICUs performing tight control in all patientswere a small proportion of all the patients studied (n = 104,11%), suggesting that their overall influence on the resultswas small. Sub-group analysis of these patients did not re-veal any differences in hospital outcome. Even identifyingthose staying more than 5 days in the three ICUs perform-ing tight glucose control did not demonstrate a significantdifference in hospital mortality following adjustments forage, sex, APACHE II score and type of admission.

There were no appreciable differences in the bloodglucose concentration trigger for administration of insulinbetween ICUs, implying there was little difference in themanagement of blood glucose from one ICU to another.This demonstrates the importance of confirming actualbedside practice with observational studies and not relyingon surveys to describe current practice.

In conclusion, we demonstrate a very high awarenessof the issue of blood glucose control and of the Van denBerghe et al. study amongst ICU directors in Australiaand New Zealand. The practice of IIT, however, has not

873

been widely adopted because of concerns over the externalvalidity of the study’s findings. Our findings underscorethe need for a high quality, large-scale, multi-centrerandomised, controlled trial in a heterogeneous populationof intensive care unit patients.

Acknowledgements. Site investigators (all in Australia unlessotherwise specified) were: Auckland Hospital, New Zealand:C. McArthur, and L. Newby. Austin Hospital, R. Bellomo, D.Goldsmith, and S. Bates. Ballarat Base Hospital, T. Sutherland.Blacktown Hospital, G. Reece and M. Gopalakrisanan. The Can-berra Hospital, I. Mitchell and R. Tamhane. Concord Hospital, D.Milliss. Flinders Medical Centre, T. Hunt. Green Lane Hospital,New Zealand: J. Beca and T. Ginger. Gold Coast Hospital, B.Richards and M. Tallot. Hornsby Hospital, J. Fratzia. John Hunter

Hospital, B. McFadyen. Mater Misercordiae Hospital, P. Cook andL. Rudder. Monash Medical Centre, l. Lister. Nepean Hospital,I. Cole and L. Weisbrodt. North Shore Hospital, New Zealand: J.Liang. Port Macquarie Hospital, C. Hore. Prince Charles Hospital,D. Mullany. Prince of Wales Hospital, Y. Shehabi and H. Adamson.Royal Darwin Hospital, D. Stephens and J. Thomas. Royal HobartHospital, T. Bell and K. Marsden. Royal North Shore Hospital, S.Finfer and S. Dale. Royal Prince Alfred Hospital, R. Totaro and D.Rajbhandari. St George Hospital, J. Myburgh and M. Hodgetts. StVincents Hospital, Melbourne: J. Santamaria. Tauranga Hospital,New Zealand: T. Browne. The Queen Elizabeth Hospital, S. Peake.The Royal Melbourne Hospital, J. Cade and C. Boyce. WellingtonHospital, New Zealand: P. Hicks and D. Durham. Western Hospital,C. French and L. Little. Dr Jeff Wood, Statistical Consulting Unit,The Graduate School, The Australian National University, Canberra,ACT 0200, Australia

References

1. Umpierrez GE, Isaacs SD, Bazargan N,You X, Thaler LM, Kitabchi AE (2002)Hyperglycemia: an independent markerof in-hospital mortality in patients withundiagnosed diabetes. J Clin EndocrinolMeTable 87:978–982

2. Mizock BA (1995) Alterations incarbohydrate metabolism during stress:a review of the literature. Am J Med98:75–107

3. Capes SE, Hunt D, Malmberg K,Gerstein HC (2000) Stress hypergly-caemia and increases risk of deathafter myocardial infarction in patientswith and without diabetes: a systematicoverview. Lancet 355:773–778

4. Capes SE, Hunt D, Malmberg K,Pathak P, Gerstein HC (2001) Stresshyperglycaemia and prognosis ofstroke in non diabetic and diabeticpatients: a systematic overview. Stroke32:2426–2432

5. Knaus WA, Wagner DP, Draper EA,Zimmerman JE, Bergner M, Bastos PG,Sirio CA, Murphy DJ, Lotring T,Damiano A, Harrell FE (1991) TheAPACHE II Prognostic System. Riskprediction of hospital mortality forcritically ill hospitalized adults. Chest100:1619–1636

6. McMahon MM, Bistrian BR (1995)Host defenses and susceptibility toinfection inpatients with diabetesmellitus. Infect Dis Clin North Am9:1–9

7. Rassias AJ, Marrin CAS, Arruda J,Whalen PK, Beach M, Yeager MP(1999) Insulin infusion improvesneutrophil function in diabetic car-diac surgery patients. Anesth Analg88:1011–1016

8. Weekers F, Giulietti AP, Michalaki M,Coopmans W, Van Herck E, Mathieu C,Van den Berghe G (2003) Metabolic,endocrine, and immune effects ofstress hyperglycaemia in a rabbitmodel of prolonged critical illness.Endocrinology 144:5329–5338

9. Preiser JC, Devos P, Van den Ber-ghe G (2002) Tight control of gly-caemia in critically ill patients. CurOpin Clin Nutr Metab Care 5:533–537

10. Vincent JL, Bihari DJ, Suter PM,Bruining HA, White J, Nicolas-Chanoin MH (1995) The prevalenceof nosocomial infection in intensivecare units in Europe. Results of theEuropean Prevalence of Infection inIntensive Care (EPIC) Study. EPICInternational Advisory Committee.JAMA 274:639–644

11. Montori VM, Bistrian BR, McMahonMM (2002) Hyperglycaemia in acutelyill patients. JAMA 288:2167–2169

12. Van den Berghe G, Wouters P, Week-ers F, Verwaest C, Bruyninckx F,Schetz M, Vlasselaers D, Ferdinande P,Lauwers P, Bouillon R (2001) Intensiveinsulin therapy in critically ill patients.N Engl J Med 345:1359–1367

13. Joint Faculty of Intensive CareMedicine (2003) Minimum standardsfor intensive care units. Report no IC-1

14. Knaus WA, Draper EA, Wagner DP,Zimmerman JE (1985) APACHE II:a severity of disease classificationsystem. Crit Care Med 13:818–829

15. Hosmer DW, Lemeshow S (2000)Applied logistic regression. New York.Wiley

16. Malmberg K, Norhammar A, Wedel H,Ryden L (1999) Glycometabolic stateat admission: important risk marker ofmortality in conventionally treated pa-tients with diabetes mellitus and acutemyocardial infarction: long term resultsfrom the Diabetes and Insulin GlucoseInfusion in Acute Myocardial Infarc-tion (DIAGIMI) study. Circulation99:2626–2632

17. Wass CT, Lanier WL (1996) Glucosemodulation of ischemic brain injury:review and clinical recommendations.Mayo Clin Proc 71:801–812

18. Michaud LJ, Rivara FP, Longstreth WT,Grady MS (1991) Elevated initial bloodglucose levels and poor outcome fol-lowing severe brain injuries in children.J Trauma 31:1356–1362

19. Zindrou D, Taylor KM, Bagger JP(2001) Admission plasma glucose:an independent risk factor for longterm prognosis in non-diabetic womenafter coronary artery bypass grafting.Diabetes Care 24:1634–1639

874

20. Finney SJ, Zekveld C, Elia A,Evans TM (2003) Glucose controland mortality in critically ill patients.JAMA 290:2041–2047

21. Collins R, MacMahon S (2001) Reliableassessment of the effects of treatmenton mortality and major morbidity.I. Clinical trials. Lancet 357:373–380

22. Krinsley JS (2004) Effect of an intensiveglucose management protocol on themortality of critically ill adult patients.Mayo Clinic Proc 79:992–1000

23. Brunkhorst FM, Kuhnt E, Engel C,Meier-Hellmann A, Ragaller M,Quintel M, Weiler N, Grundling M, Op-pert M, Deufel T, Löffler M, Reinhart K,and the German Competence NetworkSepsis (SepNet) (2005) IntensiveInsulin therapy in patient with severesepsis and septic shock is associatedwith increased rate of hypoglycemia-results from a randomised multicenterstudy (VISEP). Infection 33:19

24. Van den Berghe G, Wilmer A, Her-mans G, Meersseman W, Wouters PJ,Milants I, Van Wijngaerden E, Bob-baers H, Bouillon MD (2006) Intensiveinsulin therapy in the Medical ICU.N Engl J Med 354:449–461

25. Malhotra A (2006) Intensive insulinin intensive care. N Engl J Med354:516–518