Meta Analysis
Insulin therapy for critically ill hospitalised patients
This review may be edited
Posted By: Laurie Hall
Posted Date: 01/09/2005
Title: Insulin therapy for critically ill hospitalised patients
Authors: Pittas AG, Siegel RD, Lau J.
Reference: Arch Intern Med 2004. 164(18): 2005
Link: Click here for a direct link to the paper. A password may be required for access to fulltext.
Abstract: BACKGROUND: Hyperglycemia is common in critically ill hospitalized patients, and it is associated with adverse outcomes, including increased mortality. The objective of this meta-analysis was to determine the effect of insulin therapy initiated during hospitalization on mortality in adult patients with a critical illness.
METHODS: An electronic search in the English-language articles of MEDLINE and the Cochrane Controlled Clinical Trials Register and a hand search of key journals and relevant review articles were performed. Randomized controlled trials that reported mortality data on critically ill hospitalized adult patients who were treated with insulin were selected. Data on patient demographics, hospital setting, intervention (formulation and dosage of insulin, delivery method, and duration of therapy), mortality outcomes, adverse events, and methodological quality were extracted.
RESULTS: Thirty-five trials met the inclusion criteria. Combining data from all trials using a random-effects model showed that insulin therapy decreases short-term mortality by 15% (relative risk [RR], 0.85; 95% confidence interval [CI], 0.75-0.97). In subgroup analyses, insulin therapy decreased mortality in the surgical intensive care unit (RR, 0.58; 95% CI, 0.22-0.62), when the aim of therapy was glucose control (RR, 0.71; 95% CI, 0.54-0.93), and in patients with diabetes mellitus (RR, 0.73; 95% CI, 0.58-0.90). A near-significant trend toward decreasing mortality was seen in patients with acute myocardial infarction who did not receive reperfusion therapy (RR, 0.84; 95% CI, 0.71-1.00). No randomized trials of insulin in the medical intensive care unit were identified.
CONCLUSION: Insulin therapy initiated in the hospital in critically ill patients has a beneficial effect on short-term mortality in different clinical settings.
Are the Results Valid?
1. Did the overview address a focused clinical question?
Yes. The objective stated in the abstract was to determine the effect of insulin therapy initiated during hospitalisation on mortality in adult patients with a critical illness.
'Critical illness' was broadly defined as 'acute myocardial infarction, stroke, coronary artery bypass grafting or an illness requiring admission to an ICU'. More focus would have been maintained if each of the individual groups were the subject of the review.
2. Were the criteria used to select articles for inclusion appropriate?
Yes. 'An abstract was judged relevant if it included original clinical trial data of critically ill hospitalised adult patients who were treated with insulin (regardless of type and form) while hospitalised in which mortality outcomes were reported in relation to insulin therapy.'
3. Is it unlikely that important, relevant studies were missed?
Yes/Probably. They searched MEDLINE and the Cochrane Clinical trials REgister but not EMBASE or any other bibliographical database. They did mention hand searching the reference lists of key papers and review papers. They also attempted to seach conference proceedings for published abstracts. They did not contact field experts or industry.
With respect to search terms, they searched using ‘intensive care unit’, but this is a lower branch of ‘critical care’ which I thought would have been better. [PubMed maps the term 'intensive care unit' to the MeSH heading 'Intensive Care Units'. The hierarchy for this term can be viewed , here. In the Evidence-based Links section of this site, under Medline etc. we provide two sets of terms that may be used to search the Critical Care literature more efficiently. -Gord]
4. Was the validity of the included studies appraised?
Yes. They assessed randomisation sequence generation, allocation concealment, blinding status and intention-to-treat analysis. They found that only two studies address the most important criteria.
Although these aspects of the trials were appraised, there is no reporting of each of these criteria for each trial. For example, Table 1 reports 'Properly Randomized' not 'sequence generation' and 'whether allocation concealment was maintained'.
Doesn't seem to be any estimate of effect for studies of high vs low methodological quality.
5. Were assessments of studies reproducible?
Yes. Two reviewers independently screened abstracts according to the inclusion criteria. Full-text articles were retrieved and reviewed if they couldn't decide to include or exclude based on the abstract.
The outline explicitly criteria they used to assess study quality and also the data they extracted. They don't explain what they did in the event the two reviewers couldn't agree on something. (They also don't say what the third reviewer did!)
6. Were the results similar from study to study?
No. A measure of heterogeneity was not reported. Many authors incorrectly believe that heterogeneity is not important when a 'random effects model' is used for the analysis. Current research suggests that heterogeneity is VERY important, even when using a random effects model.
Of the 35 studies included in the meta-analysis, roughly half favoured insulin (not statistically significant), a fifth favoured control (NS), and another fifth showed no difference between groups. There was no report of a test for heterogeneity but the authors did concede that the studies shared the use of insulin in critically ill hospitalised patients, but that they differed in many other ways (e.g. hospital setting, type of patient, type of insulin). To cope with this they did a number of subgroup analyses. Heterogeneity was not reported for these subgroup analyses either.
What are the Results?
1. What are the overall results of the overview?
Caution should be used when interpreting these overall estimates. No measures of heterogeneity were provided.
Insulin therapy in critically ill hospitalised adult patients was associated with a 15% reduction in death relative to controls (RR 0.85).
  • In trials that targeted glucose control a 29% reduction in mortality was seen in patients randomised to insulin compared with controls (RR 0.71).
  • Analysis of glucose-insulin-potassium (GIK) indulin vs non-GIK insulin found that there was a trend for GIK to reduce mortality that was not significant (RR 0.9) whereas 5 studies that used non-GIK insulin did show a significant reduction in mortality of 27% (RR 0.73).
  • Patients receiving insulin were at three times higher risk of hypoglycemia compared with controls (RR 3.4, 10 studies).
  • 2. How precise were the results?
    The confidence interval for overall mortality was 0.75 to 0.97.
  • The confidence interval for trials targeting euglycemia was 0.54 to 0.93. The confidence interval for GIK studies was (0.77 to 1.04) in contrast with 5 non-GIK studies (0.56 to 0.95).
  • The confidence intervals for hypoglycaemia were 1.9 to 6.3.
  • Will the Results Help Me In Caring For My Patients?
    1. Can the results be applied to my patient care
    Unsure. Considering that the overall measure of effect was drawn from studies encompassing differing types of insulin administration and critically ill patients with differing admitting diagnoses, it may be applicable to most types of critically ill patients however caution should be exercised given the lack of a formal measure of heterogeneity.
    2. Were all clinically important outcomes considered?
    No, only 10 of the 35 studies reported the incidence of hypoglycaemia.
    3. Are the benefits worth the harms and costs?
    Not sure. Episodes of hypoglycaemia were more common in studies that used a protocol aiming at maintaining euglycamia, however, it was these studies that showed the greatest reduction in mortality.

    What other people had to say about:
    Insulin therapy for critically ill hospitalised patients

    Add Your Comment
    [ Back ]

    Any questions or comments please contact Gordon S. Doig
    Implemented and designed by John Soer and Gordon Doig
    Page last modified on Friday August 10, 2001