Criteria abstracted from The Users' Guide to Medical Literature, from the Health Information Research Unit and Clinical Epidemiology and Biostatistics, McMaster University

Highlighted lines and questions below provide links to the pertinent description of criteria in The EBM User's Guide, now available at the Canadian Centres for Health Evidence

Article Reviewed:

Please visit the new Evidence Based Journal Club Reviews

Glucose control and mortality in critically ill patients.

Finney SJ, Zekveld C, Elia A, Evans TW.

JAMA. 2003 Oct 15;290(15):2041-7 [abstract]

Reviewed by Angie Chavez MD, University of Florida, Gainesville

Review posted January 6, 2004

I. What is being studied?

Study objective:

To determine whether blood glucose level or quantity of insulin administered is associated with reduced mortality in critically ill patients

Study design

Single-center, prospective, observational study

II. Are the results of the study valid?

Note: These questions follow from Randolph AG et al. Understanding articles describing clinical prediction tools. Crit Care Med 1998;26:1603-1612. [abstract]

1. Was a representative group of patients completely followed up? Was follow-up sufficiently long and complete?

The study population was predominantly male, > 60 years of age, over weight or obese, and 88% were post-op cardiac surgery patients. The investigators achieved 100% follow up of all patients in the cohort during their ICU stay and hospital stay.

Follow up was presumably complete for the 49 patients recruited retrospectively for the study, although this is not explicitly stated. There is no mention of whether mortality was ICU mortality or in-hospital mortality. These could be considered the derivation set. There was no validation set.

2. Were all potential predictors included?

Yes. The authors assessed the effect of insulin, the time spent in the individual glucose bands, Acute Physiology and Chronic Health Evaluation 2 score (APACHE II), Sequential Organ Failure Assessment Score (SOFA), age, sex, BMI, reason for admission, length of stay, and the effect of diabetes on ICU mortality.

The authors define 6 bands of glucose control: (1) glucose < 80 mg/dl, (2) glucose 80-110 mg/dl, (3) glucose 111-144 mg/dl, (4) glucose 145-180 mg/dl, (5) glucose 181-200 mg/dl, and (6) glucose > 201 mg/dl. Each patient had glucose levels that fell into several of these bands. For each patient, the proportion of the admission that he/she spent within each of the 6 bands was computed.

Insulin doses for each patient were calculated from the area under the time-insulin dose curve relative to the length of their admission.

3. Did the investigators test the independent contribution of each predictor variable?

Yes, the independent contribution of each predictor variable was tested. A separate model was generated to examine each of the six glucose bands. Each glucose band was divided into tertiles (least, intermediate, greatest) such that each tertile contained the same number of people. For example, the glucose band 80-110 mg/dl contained a value for each patient that represented the proportion of their stay in which their blood glucose was in that range. The lowest third of the values would be categorized in the "least" tertile, the middle third in the "intermediate tertile, and the highest third in the "greatest" tertile. The effect of the percentage of time spent in each glucose band was analyzed. Confounding variables such as APACHE II score, SOFA, age, sex, BMI, etc were scrutinized alone and in interaction with time in each glucose band. The models were then filtered to exclude those variables found to have no predictive power. The last modeling technique analyzed data gathered only from non-diabetic patients.

4. Were outcome variables clearly and objectively defined?

Yes. The primary end point of the study was ICU mortality. Secondary end points were hospital mortality, ICU and hospital length of stay. These outcomes are not open to subjective interpretation. The authors also interpreted their results to identify a threshold glucose level associated with increase risk of death.

III. What are the results?

1. What is(are) the prediction tool(s)?

Increased insulin administration is positively and significantly associated with ICU mortality, irrespective of glucose level. The data suggests that glycemic control rather than administration of exogenous insulin is responsible for the survival benefit.

In addition, the regression model suggests that maintaining blood glucose levels below 145 mg/dl confers reduced mortality.

2. How well does the model categorize patients into different levels of risk?

This model effectively differentiates between two groups of patients in the ICU, those with a higher risk of mortality from increased exogenous insulin exposure and those with a decreased risk of mortality from decreased insulin exposure. This relationship is significant in each of the glucose bands.

Blood glucose band	OR (95% CI)	P value
Glucose < 80 mg/dl Insulin administered	1.02 (0.99-1.04)	.06
Glucose 80-110 mg/dl Insulin administered	1.02 (1.01-1.04)	< .001
Glucose 111-144 mg/dl Insulin administered	1.02 (1.01-1.04)	< .001
Glucose 145-180 mg/dl Insulin administered	1.02 (1.01-1.03)	< .001
Glucose 181-200 mg/dl Insulin administered	1.02 (1.01-1.03)	< .001
Glucose ≥ 201 mg/dl Insulin administered	1.02 (1.003-1.03)	< .001

Odds ratios are for a 1 IU/day increase in insulin dose.

However, the odds ratio for the time of exposure to specific glycemic ranges did not reach statistical significance. Although the data significantly associates increased insulin administration to increased mortality, we do not know the threshold dose of insulin that places ICU patients at increased mortality risk. Moreover, the target upper limit blood glucose level of 145 mg/dl is an interpretation suggested by the regression model rather than statistically significant finding.

3. How confident are you in the estimates of the risk?

The 95% confidence interval reported for the odds ratio of death with insulin exposure is narrow and > 1.0 in 5 of the 6 glucose bands. However, the clinical significance is small.

IV. Will the results help me in caring for my patients?

1. Does the tool maintain its prediction power in a new sample of patients?

The tool was not tested in a new sample of patients. It was tested in a subset which excluded patients with diabetes and the results were the same. It is important to validate this model in a completely distinct cohort of patients.

2. Are your patients similar to those patients used in deriving and validating the tool(s)?

No. Our patients are children without significant coronary artery disease.

3. Does the tool improve your clinical decisions?

Not really because the study population (mostly overweight adult males s/p CT surgery) is completely different than my patient mix. However, the study is innovative and the implications compel us to consider studying the use of insulin to correct an aberrant and often disregarded physiologic parameter.

4. Are the results useful for reassuring or counseling patients?

This question is not very applicable to this particular study since the glycemic status of patients in the ICU is not controlled by the patients' own volition. In any case, I would not be able to reassure parents that their critically ill child's risk of death is changed because of insulin use or glycemic control because this study did not include children. However, the results of this study serve as a constructive reminder to health care providers that even the most altruistic attempts to improve the health status of our patients can often be met with unforeseen detrimental consequences, thus it is our duty to persist on finding innovative ways to maximize outcomes while minimizing the negative effects of such interventions.

 

Comments

Back to the EB Journal Club Index