Individualised versus conventional glucose control in critically-ill patients: the CONTROLING study—a randomized clinical trial

Bohé. Intensive Care Medicine 2021.

Clinical Question

  • In critically unwell adults does targeting individualised glycaemic control compared to glycaemia < 180 mg/dL improve 90 day mortality?


  • Hyperglycaemia has been shown to result in worse outcomes in the critically ill, especially in non-diabetic patients
  • In 2009, NICE-SUGAR showed that intensive glucose control was harmful with  higher rates of hypoglycaemia and mortality in the intensive control group
  • Some observational studies have shown that relationship between glycaemia in ICU and adverse events (hypoglycaemia and mortality) may depend on the patient’s pre-admission glycaemic control
  • HbA1c or glycated haemoglobin reflects the mean plasma glucose over the life of a red blood cell (~3 months), and values at ICU admission and prior to ICU admission correlate well


  • Multi-centre
  • Double blind randomised parallel group trial
  • Conducted in mixed, medical and surgical ICUs
  • Consent from patients or family obtained
  • If eligible then a blood sample drawn to ascertain HbA1c
    • If deemed eligible then needed to rule out presence of exclusion criteria within 96 hours of ICU admission
  • Glycaemic control was conducted in accordance with an online algorithm designed by two of the authors
    • Insulin was commenced prior to randomisation if > 180 mg/dL for all patients
  • Randomisation was 1:1 and stratified by centre
    • However, if numbers in each arm differed by >4% then next patient automatically included into group with lower numbers
  • Intervention maintained until patient discharged from ICU
  • All clinicians blinded to HbA1c result, and glycaemic and infusion history
    • For safety, the nurse in charge could access the last glycaemic value entered and insulin infusion rate and any history of hypoglycaemic episodes
  • Power calculation:
    • 2100 patients per group needed
    • 90% power to detect a 4% difference from a baseline mortality of 22% with a two-sided significance level of 5%
    • Based on NICE-SUGAR
  • Modified intention to treat basis
  • Trial stopped early after 49% of planned sample size recruited
    • Recommended by DSMB due to low likelihood of benefit and possibility of harm
  • Conversion from mg/dL to mmol/L
    • 217 = 12.1
    • 180 = 10
    • 111 = 6.2
    • 72 = 4.0
    • 40 = 2.2


  • 12 ICUs in France
  • May 2015 – July 2016


  • Inclusion:
    • > 18 yo
    • Admitted to ICU with the expectation not to be discharged within 2 days
    • Unable to feed orally
  • Exclusion:
    • Pregnancy
    • Prior enrolment
    • Limitations of therapy
    • Diabetics with a known history of transfusion > 3 red blood cell units in the last 3 months (may falsely lower HbA1c)
  • 5326 patients admitted → 2075 randomised of which 158 did not receive intervention → 942 in intervention arm and 975 in conventional arm
    • Reasons for not being included:
      • 690 expected to stay in ICU for < 48 hours
      • 1300 did not have exclusion criteria assessed
      • 423 met exclusion criteria
      • 827 did not have HbA1c level within 96 hrs
  • Comparing baseline characteristics of intervention (individualised control) vs. control (conventional) group
    • Age: 68 v 69
    • Female: 39 v 39%
    • BMI: 26 v 26
    • Medical admission: 82 v 82%
      • Respiratory: 37 v 36%
      • Cardiac: 14 v 16%
      • Endocrinology: 2 v 3%
    • Emergency surgery: 10 v 10%
    • Vasopressors: 31 v 31%
    • Invasive ventilation: 54 v 53%
    • Charlson score >/= 3: 44 v 46%
    • Diabetic patients: 34 v 32%
      • Insulin dependent: 8 v 8%
    • HbA1c: 5.8 v 5.8%
      • Non-diabetics: 5.6 v 5.6%
      • Diabetics: 6.9 v 6.9%
    • Insulin dose at randomisation: 0 v 0 IU/h
    • Interval from ICU admission to randomisation: 1.2 v 1.3 days
    • Days on algorithm: 4 v 4
    • No receipt of artificial nutrition: 39 v 41%
      • Calories administered similar in those receiving enteral, TPN or a mixture of both


  • Individualised glycaemic control
    • Glycaemic target calculated using HbA1c +15 mg/dL
    • Limits set at minimum of 111mg/dL and a maximum 217 mg/dL for safety reasons
    • The median HbA1c of 5.8% results in a calculated target of 120 mg/dL (+ 15 mg/dL)


  • Standard glycaemic control
    • Glycaemic target < 180mg/dL
    • Insulin infusion initiated when greater than this

Management common to both groups

  • Algorithm gave nurse instructions for insulin administration, sampling for glycaemia and correction for hypoglycaemia.
  • Sampling time between 1 – 6 hours depending on levels and previous stability
  • Following randomisation insulin commenced if above set target, it was reduced if 29mg/dL less than target
  • If levels reached 63 mg/dL then insulin stopped and 30% dextrose administered
  • All other care at the discretion of treating teams


  • Primary outcome: 90-day survival (intervention v control)
    • 67.2% [95% CI 64.2 – 70.3] v 69.6% [95% CI 66.7 – 72.5], p = 0.23
    • No difference when adjusted for multiple parameters including age, diabetic status, BMI and receipt of invasive ventilation
    • Post hoc subgroup analysis:
      • Significantly higher risk of 90 day mortality in intervention group for non-diabetics, surgical group and those with an HbA1c between 5 – 6%
  • Secondary outcomes:
    • No significant difference in 28d mortality, ICU length of stay, duration of vasoactive support, RRT requirement, mechanical ventilation and non-prophylactic anti-microbial use
  • Adverse Events:
    • Severe hypoglycaemia (<40mg/dL)
      • 3.9% v 2.5% (p = 0.09)
    • Any hypoglycaemia (<72 mg/dL)
      • 31.2% v 15.8% (p < 0.0001)
    • This was significantly higher in non diabetic patients, but not for diabetic patients

Authors’ Conclusions

  • Targeting a patient’s usual glycaemic level did not show a survival benefit compared to conventional control and targeting a glycaemic level of 180 mg/dL


  • Double blinded trial in an area of critical care with few randomised trials despite the frequency with which parameters for blood glucose control are set for ICU patients
  • Balanced baseline characteristics between groups
  • Minimal protocol violations (<1%)
  • Blinding methods minimise performance and detection bias
  • Minimal loss to follow up (n=9)


  • Trial ended early prior to enrolment of calculated sample size
  • Time lag bias as data collected in 2016 but not published until 2021
  • Large numbers not assessed for exclusion criteria may result in selection bias
  • The allowance of 96 hours following ICU admission for enrolment is a long time in the critically unwell
    • The intervention group received the same target as the conventional group until randomisation (<180mg/dL) results in contamination bias
    • Patients in the intervention group spent 26% of their ICU stay without an individualised target (and exposed to a target of 180 mg/dL)
  • The authors postulate that insulin resistance may change over critical illness
    • If this is true, does the use a value that reflects glycaemic control over the three months prior to critical illness provide the best method for targeting personalised glycaemic control whilst critically unwell?
  • The algorithm used was not validated prior to it’s use
  • The mean difference in time-weighted blood glucose was only 13 mg/dL between groups (0.7mmol/L) (eTable 3). It is unlikely this is clinically relevant
    • However if stratified by HbA1c there was statistical separation in time-weighted blood glucose for all HbA1c levels except between 7 and 8% (which was expected as an HbA1c of 7.4% = a target of 180mg/dL)

The Bottom Line

  • Until better evidence becomes available surrounding personalised glycaemic targets in ICU, I will continue to adopt a conventional target for all patients

External Links


Summary author: George Walker @hgmwalker89
Summary date: 7th December 2021
Peer-review editor: @davidslessor

Picture by: Pixabay


Leave a Reply

Your email address will not be published. Required fields are marked *

This site uses Akismet to reduce spam. Learn how your comment data is processed.