Publication

• Title: Intensive Insulin Therapy in the Medical ICU
• Acronym: Leuven 2 (informal)
• Year: 2006
• Journal published in: The New England Journal of Medicine
• Citation: Van den Berghe G, Wilmer A, Hermans G, Meersseman W, Wouters PJ, Milants I, et al. Intensive insulin therapy in the medical ICU. N Engl J Med. 2006;354(5):449-461.

Context & Rationale

• Background

  • Stress hyperglycaemia is common in critical illness and is biologically plausible as a mediator of immune dysfunction, infection risk, catabolism, and organ injury.
  • The Leuven surgical ICU trial reported reduced morbidity and mortality with intensive insulin therapy targeting normoglycaemia (80–110 mg/dL) compared with a more permissive strategy.¹
  • Whether the surgical-ICU findings generalised to medical ICU populations (different case-mix, baseline risks, feeding patterns, and competing causes of death) remained uncertain.
  • Potential harms of tight glycaemic targets—particularly severe hypoglycaemia—were an important safety concern for broader implementation.
• Research Question/Hypothesis

  • Does intensive insulin therapy to achieve strict normoglycaemia (80–110 mg/dL) reduce in-hospital mortality and major morbidity compared with conventional insulin therapy (target 180–200 mg/dL) in adult medical ICU patients expected to require at least 3 days of intensive care?
• Why This Matters

  • This question was practice-shaping because the Leuven surgical ICU result drove rapid uptake of “tight” glucose control internationally, despite uncertainty about reproducibility and safety across settings.
  • Medical ICUs represent a large proportion of critical care admissions, and even modest absolute changes in mortality or major complications (renal failure, ventilation duration, infection) would have substantial system-level effects.
  • Operationally, the intervention requires high-frequency monitoring and protocolised nurse-driven titration; external validity and safety are tightly linked to ICU staffing and measurement infrastructure.

Design & Methods

• Research Question: In adult medical ICU patients expected to remain in ICU ≥3 days, does intensive insulin therapy targeting 80–110 mg/dL reduce in-hospital mortality and morbidity versus conventional insulin therapy targeting 180–200 mg/dL?
• Study Type: Prospective, randomised, controlled, single-centre (tertiary university hospital), nurse-driven insulin infusion protocol; allocation stratified by diagnostic category with permuted blocks of 10; unblinded intervention delivery with blinding of selected outcome assessments (e.g., bacteremia adjudication; ICU death cause).
• Population:
  • Setting: Medical ICU, Leuven, Belgium; enrolment from 14 March 2002 onward.
  • Inclusion: Adult patients admitted to the medical ICU who were assumed on admission to require at least a third day of intensive care.
  • Key exclusions: Expected ICU stay <3 days (already eating); do-not-resuscitate orders on admission; postoperative patients; participation in another study; other ineligibility; lack of consent.
  • Screening/randomisation: 2110 evaluated; 863 excluded; 1200 randomised (605 conventional; 595 intensive).
• Intervention:
  • Insulin strategy: Continuous IV insulin (Actrapid HM 50 IU in 50 mL 0.9% NaCl) started when blood glucose >110 mg/dL (6.1 mmol/L).
  • Target: 80–110 mg/dL (4.4–6.1 mmol/L).
  • Cap: Maximum infusion rate set at 50 IU/hour.
  • Delivery: Nurse-adjusted titration guidelines adapted from prior surgical-ICU protocol; whole-blood glucose measured at 1–4 hourly intervals for titration (arterial blood preferred; capillary if no arterial line; point-of-care glucometer used when needed).
  • Post-ICU: At ICU discharge, a conventional approach was used (maintenance ≤200 mg/dL).
• Comparison:
  • Insulin strategy: Continuous IV insulin started only when blood glucose >215 mg/dL (12 mmol/L).
  • Target: 180–200 mg/dL (10–11 mmol/L); infusion tapered and stopped when glucose <180 mg/dL.
  • Co-interventions: Usual ICU care including nutritional management aiming for 22–30 kcal/kg/day with balanced macronutrient composition; enteral feeding attempted as early as possible once haemodynamically stable.
• Blinding: Not feasible for bedside staff (insulin titration based on glucose values); bacteremia interpretation and ICU cause-of-death determination were performed by personnel blinded to allocation; ward physicians after ICU discharge had no access to glucose test results and were unaware of group assignment.
• Statistics: Power calculation: 1200 patients required to detect a 7% absolute reduction in risk of death in the subgroup staying in ICU ≥3 days, with two-sided alpha <0.05 and beta 0.2 (80% power); primary analysis by intention-to-treat; proportional hazards regression used for mortality (with correction for baseline risk factors); predefined subgroup analysis for ICU stay ≥3 days.
• Follow-Up Period: In-hospital outcomes through discharge (primary endpoint); 90-day mortality also assessed.

Key Results

This trial was not stopped early.

• Intensive insulin therapy achieved substantially lower glycaemia (mean morning blood glucose 111 ± 29 mg/dL vs 153 ± 31 mg/dL; P<0.001) with substantially more insulin (mean daily insulin 71 ± 77 IU vs 46 ± 71 IU; P<0.001), demonstrating clear treatment separation.
• The primary endpoint (in-hospital mortality) was not significantly different overall (HR 0.94; 95% CI 0.84 to 1.06; P=0.31), while the predefined ICU ≥3-day subgroup showed lower in-hospital mortality (HR 0.84; 95% CI 0.73 to 0.97; P=0.02).
• Severe hypoglycaemia was markedly more frequent with intensive therapy (18.7% vs 2.8%; P<0.001), a key safety trade-off alongside morbidity signals (renal injury reduction; faster ventilation weaning and discharge in time-to-event analyses).

Internal Validity

• Randomisation and allocation concealment
  • Randomisation used sealed envelopes, stratified by diagnostic category, balanced with permuted blocks of 10.
  • Envelope-based allocation has theoretical susceptibility to subversion, but no direct evidence of allocation compromise was reported.
• Drop-out / exclusions and analysis population
  • 1200 patients were randomised and analysed by intention-to-treat for the primary outcome.
  • Exclusions occurred pre-randomisation (e.g., expected ICU stay <3 days; DNR on admission; postoperative; other trial participation; consent refusal), influencing representativeness more than internal validity.
  • A post hoc exploratory mortality analysis censored 65 patients with early limitation/withdrawal of intensive care; this was not the primary analysis.
• Performance and detection bias
  • Intervention was unblinded at the bedside because insulin titration required access to glucose values.
  • Several key endpoints were objective (mortality, dialysis initiation), mitigating detection bias for those outcomes.
  • Bacteremia interpretation and ICU cause-of-death assignment were performed by blinded assessors; ward physicians post-ICU were unaware of allocation and did not have access to glucose test results.
• Protocol adherence and separation of the variable of interest
  • Mean morning blood glucose: 111 ± 29 mg/dL (intensive) vs 153 ± 31 mg/dL (conventional); P<0.001.
  • Distribution of morning glucose values: 80–110 mg/dL 53.3% vs 21.3%; 150–200 mg/dL 2.0% vs 23.6%; >200 mg/dL 0.3% vs 6.1% (intensive vs conventional).
  • Insulin exposure: insulin administered in 98.2% vs 69.8% of patients (P<0.001); median insulin per ICU day 59 (IQR 37–86) IU vs 10 (IQR 0–38) IU (intensive vs conventional).
  • Severe hypoglycaemia (≤40 mg/dL): 18.7% vs 2.8% overall; and 23.1% vs 3.5% among ICU ≥3-day patients (intensive vs conventional).
• Baseline characteristics and illness severity
  • Groups were broadly similar in age (64.0 vs 64.4 years), sex (male 62.5% vs 61.3%), APACHE II (median 23 vs 23), and diabetes prevalence (15.8% vs 16.0%) (intensive vs conventional).
  • One measured imbalance: higher proportion with baseline TISS-28 >33 in the intensive group (26.6% vs 20.6%; P=0.02).
• Heterogeneity and subgroup structure
  • Medical ICU case-mix was heterogeneous; randomisation was stratified by diagnostic category to reduce imbalance.
  • The ICU ≥3-day subgroup was predefined, but ICU length of stay is determined after randomisation and is related to prognosis, complicating causal interpretation of subgroup effects.
• Timing and dose
  • Insulin was commenced based on early glucose thresholds (>110 mg/dL intensive; >215 mg/dL conventional) with 1–4 hourly titration (as clinically required).
  • The 80–110 mg/dL target achieved strong separation but at the cost of frequent severe hypoglycaemia, suggesting the “dose” of glycaemic lowering may be near the safety boundary for many medical ICU patients.
• Outcome assessment and statistical rigour
  • Primary endpoint was all-cause in-hospital mortality; mortality modelling used proportional hazards with correction for baseline risk factors.
  • P values were not adjusted for multiple comparisons across numerous secondary outcomes; several secondary results should be interpreted as exploratory within the trial’s multiplicity context.

Conclusion on Internal Validity: Moderate. Randomisation and protocol separation were strong, with objective primary outcomes and some blinded adjudication; however, unblinded delivery, single-centre conduct, multiplicity across secondary endpoints, and the interpretability of length-of-stay-defined subgroup effects limit causal certainty for some findings.

External Validity

• Population representativeness
  • Eligibility required clinicians to judge on admission that ICU care would be required for ≥3 days; in practice, 433/1200 (36%) stayed <3 days.
  • Exclusion of patients with DNR orders on admission, postoperative patients, and those expected to eat early may reduce applicability to contemporary mixed medical ICUs where admission trajectories and ceilings-of-care vary.
• Setting and co-interventions
  • Single-centre, high-resource academic ICU with stable nurse staffing and established nurse-driven titration protocols.
  • Nutritional practices and the balance of enteral versus parenteral feeding can materially influence insulin requirements, hypoglycaemia risk, and the external transportability of “tight” targets.
• Modern applicability
  • Subsequent multicentre trials and guidelines shifted practice towards more moderate targets; thus, the 80–110 mg/dL strategy is now uncommon outside selected contexts.
  • Many ICUs now use different glucose measurement technologies, insulin protocols, and feeding practices than those used in Leuven in the early 2000s.

Conclusion on External Validity: Limited-to-moderate. The biological question is broadly relevant, but the single-centre environment, specific co-interventions, and the intensive monitoring required mean results may not generalise to settings with different nutrition strategies, staffing ratios, or glycaemic monitoring infrastructure.

Strengths & Limitations

• Strengths:
  • Large prospective randomised trial (n=1200) in a medical ICU population with prespecified hypotheses informed by prior work.
  • Clear physiological separation in glucose exposure and insulin dosing, supporting an interpretable “intervention signal”.
  • Clinically meaningful secondary endpoints (renal injury, ventilation duration, ICU/hospital discharge) with time-to-event analyses.
  • Structured monitoring and nurse-driven protocol implementation with high measurement frequency.
• Limitations:
  • Single-centre design limits generalisability and raises the possibility that system factors (staffing, nutrition practice, protocol expertise) are effect modifiers.
  • Unblinded intervention delivery introduces potential performance bias for non-mortality outcomes, despite objective endpoints and partial blinding of adjudication.
  • High rate of severe hypoglycaemia in the intensive arm (18.7%) is a major safety concern and complicates benefit–harm interpretation.
  • Multiple secondary outcomes without multiplicity adjustment increase the risk of chance findings.
  • Interpretation of subgroup results based on ICU length of stay is complex because length of stay is influenced by post-randomisation prognosis and care decisions.

Interpretation & Why It Matters

• Clinical signal

  Overall in-hospital mortality was not significantly reduced, despite robust glucose separation; the main “benefit” signal emerged in the predefined ICU ≥3-day subgroup and in morbidity endpoints (renal injury and resource use proxies).
• Safety boundary

  The intervention materially increased severe hypoglycaemia (18.7% vs 2.8%), highlighting that tight targets can impose substantial iatrogenic risk even in a highly protocolised environment.
• Mechanistic implications

  The pattern of apparent benefit in longer-stay patients is consistent with a hypothesis that metabolic modulation may matter most when the dominant determinants of outcome are ICU-acquired complications rather than early irreversible disease burden.

Controversies & Subsequent Evidence

• Editorial perspective contemporaneous with Leuven 2
  • The accompanying editorial highlighted that the mortality benefit observed in surgical ICU patients was not reproduced as an overall effect in the medical ICU cohort, emphasising the tension between morbidity signals and a null primary endpoint alongside substantial hypoglycaemia risk.²
  • The editorial underscored that the subgroup signals (by ICU length of stay) should be interpreted cautiously because length of stay is not knowable at baseline and is influenced by illness trajectory and care decisions.²
• Multicentre RCTs after Leuven 2
  • VISEP (severe sepsis): stopped early due to increased rates of severe hypoglycaemia in the intensive insulin group (12.1% vs 2.1%); no improvement in 28-day mortality (24.7% vs 26.0%).³
  • NICE-SUGAR: intensive glucose control was associated with higher 90-day mortality compared with conventional control (a pivotal reversal of the Leuven hypothesis in a large, multicentre population) and increased severe hypoglycaemia.⁴
• Meta-analytic syntheses and evolving consensus
  • A large meta-analysis identified no mortality benefit from tight glucose control across ICU trials while confirming a substantial increase in hypoglycaemia risk.⁵
  • More recent individual participant data meta-analysis further evaluated heterogeneity and benefit–harm trade-offs across patient strata, reinforcing that tight targets do not confer consistent mortality benefit and highlighting context dependence of harms and effects.⁶
• Guideline impact
  • Modern sepsis guidelines recommend insulin strategies targeting moderate glycaemic ranges rather than strict normoglycaemia in critically ill adults, reflecting the aggregate trial evidence and safety profile of hypoglycaemia with tight targets.⁷

Summary

• In a 1200-patient single-centre medical ICU RCT, intensive insulin therapy (target 80–110 mg/dL) did not significantly reduce in-hospital mortality overall (37.3% vs 40.0%; HR 0.94; 95% CI 0.84 to 1.06; P=0.31).
• The predefined ICU ≥3-day subgroup showed lower in-hospital mortality with intensive therapy (43.0% vs 52.5%; HR 0.84; 95% CI 0.73 to 0.97; P=0.02) alongside favourable morbidity signals.
• Intensive therapy produced strong glucose separation (mean morning glucose 111 ± 29 vs 153 ± 31 mg/dL; P<0.001) but substantially increased severe hypoglycaemia (18.7% vs 2.8%; P<0.001).
• Morbidity endpoints favoured intensive therapy (e.g., reduced new kidney injury among ICU ≥3-day patients: 53.1% vs 60.7%; P=0.03; and faster weaning/discharge in time-to-event analyses).
• Subsequent multicentre trials and meta-analyses did not reproduce an overall mortality benefit from tight targets and reinforced hypoglycaemia risk, contributing to guideline shifts toward moderate glucose targets.

Notes

• Glucose thresholds were reported in mg/dL with mmol/L equivalents: 110 mg/dL ≈ 6.1 mmol/L; 215 mg/dL ≈ 12 mmol/L; 40 mg/dL ≈ 2.2 mmol/L.

Overall Takeaway

Leuven 2 demonstrated that reproducing tight glucose control in a medical ICU is feasible and can yield morbidity signals, but it failed to show an overall mortality benefit and produced a high burden of severe hypoglycaemia. Together with subsequent multicentre trials and meta-analyses, it helped reframe strict normoglycaemia as a high-risk strategy rather than a universal ICU standard, accelerating the shift toward safer moderate targets.

Bibliography

• 1Van den Berghe G, Wouters P, Weekers F, Verwaest C, Bruyninckx F, Schetz M, et al. Intensive insulin therapy in critically ill patients. N Engl J Med. 2001;345(19):1359-1367.
• 2Malhotra A. Intensive insulin in intensive care. N Engl J Med. 2006;354(5):516-518.
• 3Brunkhorst FM, Engel C, Bloos F, Meier-Hellmann A, Ragaller M, Weiler N, et al. Intensive insulin therapy and pentastarch resuscitation in severe sepsis. N Engl J Med. 2008;358(2):125-139.
• 4Finfer S, Chittock DR, Su SY, Blair D, Foster D, Dhingra V, et al. Intensive versus conventional glucose control in critically ill patients. N Engl J Med. 2009;360(13):1283-1297.
• 5Wiener RS, Wiener DC, Larson RJ. Benefits and risks of tight glucose control in critically ill adults: a meta-analysis. JAMA. 2008;300(8):933-944.
• 6Adigbli D, Finfer S, Udy AA, et al. Tight glucose control and mortality in critically ill patients: an individual participant data meta-analysis. NEJM Evid. 2024;EVIDoa2400082.
• 7Evans L, Rhodes A, Alhazzani W, Antonelli M, Coopersmith CM, French C, et al. Surviving sepsis campaign: international guidelines for management of sepsis and septic shock 2021. Intensive Care Med. 2021;47(11):1181-1247.