Hyperglycemia is common in critically ill patients due to high levels of inflammatory cytokines and counter-regulatory hormones resulting in insulin resistance (IR), a condition characterized by reduced insulin-mediated glucose uptake by skeletal muscle and increased hepatic gluconeogenesis.1 However, critically ill patients may develop IR in the absence of overt hyperglycemia. Normal glucose levels may be preserved by increased endogenous production of insulin.

Biochemically, the gold standard for measuring IR is the insulin clamp test.2 This is a laborious and invasive technique using fixed insulin infusion and varying glucose infusion rates to keep serum glucose constant. Given these measurement difficulties, other means have been developed that approximate the insulin clamp technique. The homeostasis model assessment method (HOMA) is a simplified and established method for measuring insulin resistance.3 This method requires measurement of fasting serum glucose and insulin levels and their incorporation into a simple formula. The HOMA score has been shown to correlate closely with the insulin clamp technique in ambulatory patients, and a score of ≥4 represents significant insulin resistance.4'5

How or whether underlying IR affects critically ill patients is unclear. Hyperglycemia is associated with activation of inflammatory cytokines, coagulation abnormalities, immune suppression, and production of oxygen free radicals.6-9 Independent of hyperglycemia, non-overt IR or hyperinsulinemia by itself can cause these kinds of changes as well.10,11 These abnormalities are commonly seen in critically ill patients, and the degree of abnormality correlates with severity of illness.12,13 Because most critically ill patients are thought to be insulin resistant, it is difficult to determine to what degree insulin resistance and subsequent hyperglycemia augment inflammatory and coagulation abnormalities in these patients or are only associated with such changes.

The primary purpose of this study was to measure the prevalence, incidence, and pattern of resolution of IR in intensive care unit (ICU) patients. A secondary objective was to explore the relationship between IR and inflammatory cytokines, coagulation abnormalities, and clinical outcomes in critically ill patients. We hypothesized that IR would be a frequent occurrence and that severe underlying IR in critically ill patients would be associated with worsening inflammation and coagulation disturbances and would be a risk factor for increased morbidity and mortality independent of the severity of the underlying illness.

Methods

Patients and Setting

This prospective observational study was conducted in a tertiary care 21 bed mixed surgical- medical ICU from October 2002 to October 2003. Depending on the capacity of our laboratory to handle study specimens, we enrolled patients 18 years of age and older within 24 hours of admission to the ICU. We excluded patients who had elective surgery, overdoses, and patients who were expected to stay less than 24 hours. Our institutional review board approved the protocol for this study. Informed consent was obtained from next of kin before enrollment.

Clinical Management

As this was an observational study, the clinical management of patients was determined by the clinical team caring for the patient and the clinical protocols operational in our ICU. On average, patients were fed enterally within 24-48 hours of ICU admission, according to our local practice. Arterial or venous blood glucose levels were assessed daily in the morning and frequently throughout the day, and dose of insulin was prescribed by a glycemic control protocol to titrate a blood sugar between 4 and 9 mmol/L (between 72 and 162 mg/dL).

Clinical Data Collection

Baseline demographics, past medical history, and medications were obtained from patients or their charts. Necessary variables were recorded to calculate Acute Physiology and Chronic Health Evaluation II (APACHE II)14 on admission and Sequential Organ Failure Assessment (SOFA) scores15 daily until day 10, discharge from ICU, or death.

Laboratory Tests

Blood samples were collected in the morning after enrollment and each subsequent ICU day until discharge, death, or a maximum of 10 days. Serum glucose and hemoglobin A1c (HgbAlc; only day 1) plus routine blood work were measured daily. Insulin levels were measured using the IMMULITE automated immunometric analyzer using the chemiluminescent enzyme immunometric assay method (Diagnostic Products Corporation, distributed by Inter Medico® Markham, Ontario). Measured insulin included endogenous (produced by patient) and exogenous insulin (administered to the patient). In addition, plasma was analyzed for inflammatory and coagulation markers using the following assays: Protein C (PC; MDA® Protein C assay kit, Organon Teknika Corporation, Durham, NC); antithrombin (AT; MDA® Antithombin III assay kit, bioMerieux, Inc, Durham, NC); D-Dimer (MDA® D-Dimer assay kit, Organon Teknika Corporation, Durham, NC); plasminogen activator inhibitor (PAI; Spectrolyse®/pL PAI, Umea, Sweden); interleukin 6 (IL-6; Bender Medsystems ELISA kit-Cat BMS-213, Bender Med systems Inc, Burlingame, CA).

Outcome Measurements

Patients were divided into 3 groups according to their baseline HOMA and glucose measured (overt IR, nonovert IR, and insulin sensitive). Overt IR refers to patients who were hyperglycemic (glucose ≥ 7 mmol/L or 126 mg/dL) or required exogenous insulin to control their blood glucose at the time of enrollment into the study. Non-overt IR patients had normal glucose (ie, < 7 mmol/L or 126 mg/dL) but had IR according to HOMA criteria (HOMA ≥ 4). Insulin-sensitive (IS) patients had normal glucose (< 7 mmol/L or 126 mg/dL) and were insulin sensitive as defined by a HOMA < 4. Determinations of insulin resistance were made at baseline, determining the prevalence of IR, and throughout the course of ICU stay, determining the incidence of IR. For all patients, median (and interquartile ranges) HOMA scores were described over the 10-day study period.

Hyperglycemia was defined as a serum glucose ≥ 7 mmol/L or 126 mg/dL on 2 or more consecutive days, since glucose above this level has been shown to result in the activation of inflammation and coagulation systems.16 Three methods were used to represent a patient's glycemic control: Maximum admission glucose refers to highest day 1 serum glucose (Glucose^sub day1^^sup max^). Average daily peak glucose (Glucose^sub peak^^sup av^) refers to average of daily peak glucose values for the ICU stay. Hyperglycemia index (HI)17 refers to the average daily area under the curve above glucose level 7 mmol/L or 126 mg/dL. Thus, an HI value above zero would indicate hyperglycemia. IR was measured using HOMA. HOMA score was calculated by means of the following formula: serum insulin (µmol/L) × serum glucose (mmol/L)/22.5. HOMA value of ≥4 was used to diagnose IR, as previously described.3 Similar to HI, we developed an index for HOMA, IL-6, and PAI levels that incorporated measures above zero over the ICU stay. This was done by plotting daily measurements of these variables against time for up to day 10 of ICU stay, discharge from ICU, or death, calculating the area and dividing it by the number of days measured to develop an average daily index.

The clinical outcomes for this study included 14 days, 28 days, and hospital mortality; ICU-free days and ventilator-free days in the first 28 days; maximum and delta SOFA scores; and ICU-acquired infections (onset > 48 hours after ICU admission). The diagnosis of infection on admission or while in the ICU (onset > 48 hours after admission) was based on retrospective chart review by 2 independent blinded physicians using standardized definitions.18 Physicians were blinded to the results of blood tests specifically measured for this study, but not to routine tests. In case of discrepancy, the physicians would meet and resolve the issue by reanalyzing the chart.

Data Analysis

Based on limited funding, a convenient sample of 96 patients was enrolled in this study; no formal sample size calculation was performed. Descriptive statistics were used to show proportions, rates, means, and medians. Correlation and partial correlation (Spearman) were applied to measure the between- and within-patient associations with various measurements, including APACHE II score, inflammation and coagulation markers, and HOMA score. A nonparametric (Kruskal-Wallis) test was applied to compare groups of patients and relate these variables (including inflammation and coagulation markers) to insulin resistance and 28-day mortality. We further used Dunn's multiple comparison test to identify which groups were different when the P value was significant for across group differences. The χ^sup 2^ test was applied to compare percentage differences between groups and relate the variables to insulin resistance and 28-day mortality. Variables (continuous and categorical) significantly related to 28-day mortality in the above bivariate analyses were included in the forward stepwise selection procedure to develop the final logistic regression model. A .05 level of significance was used for all statistical tests, including the stepwise procedure to include/exclude variables in the model. The SAS (version 9. 1 ) statistical package was used for all the statistical analyses.

Results

Baseline Characteristics

A total of 96 patients were enrolled. Table 1 shows baseline demographic data of all the patients combined and for overt IR, non-overt IR, and IS groups. Thirty-eight patients received exogenous insulin during the study period; the average daily dose was 53.9 units (±47.6). All patients were enterally fed; 2 patients received short-term parenteral nutrition. The average caloric intake was 1350 vs 1116 calories/day in the overt IR group compared with non-overt IR (P = .61). Only 1 patient for 1 day was fed in the IS group.

Incidence, Prevalence, and Pattern of Resolution of IR

Upon admission to ICU, 64 (67%) patients had overt insulin resistance, 9 patients (9.4%) had non-overt IR, and 23 patients (24%) were insulin sensitive. Through the rest of the 10-day study period, 3 of 9 non-overt patients and 1 of 23 insulin-sensitive patients required exogenous insulin. An additional 12 patients who were initially IS developed overt hyperglycemia and/or developed an increased HOMA > 4. Overall, when considering the entire 10-day study period, 80 (83%) of patients manifested overt IR, 6 (6%) patients had non-overt IR, and 10 (10%) patients remained IS.

Baseline demographics, including baseline APACHE II and SOFA scores, were not significantly different across the 3 groups, except there were more patients with cardiac comorbidities in the IS group than the other groups (P = .02, see Table 1). Two insulin-sensitive patients (2/10, 20%) had a previous diagnosis of type 2 diabetes and yet a low calculated HOMA score. Figure 1 represents the pattern of average daily HOMA scores of the 3 groups over the 10-day study period. Patients who had overt IR remained so throughout the study period; their average daily insulin dose did not significantly change over this time period.

Glycemic Control and Insulin Resistance

Laboratory data related to glucose control and IR are shown in Table 2. Median Glucose^sub day1^^sup max^ (maximum day 1 glucose level), HI, and Glucose^sub peak^^sup av^ (average of daily peak glucose values) were elevated, indicating hyperglycemia in most patients. Median (and interquartile range) values for serum insulin level were 119 (61.5, 185.5) pmol/L. Average admission HOMA (day1) score was elevated at 5.5 (2.8, 9.5), indicating significant overall IR in patients admitted to the ICU. Measurements related to glucose control were significantly higher in patients with overt IR compared with the other 2 groups (see Table 2). Compared with insulin-sensitive patients, non-overt IR patients did not show a statistically significant difference in glucose control (Glucose^sub day1^^sup max^, Glucose^sub peak^^sup av^, HI, and HgbAlc levels; P value not significant); however, there was a statistically significant difference in HOMA index (4.1 vs 0; P = .004). Despite similar glucose control values, however, non-overt IR patients compared with insulin-sensitive patients tended to have higher admission of serum insulin (131.5 vs 68 pmol/L; P= .19), indicating high endogenous insulin secretion to overcome IR.

Levels of Inflammatory and Coagulation Markers

Measurements of inflammatory markers (IL-6 and C-reactive protein [CRP]) and coagulation factors (RAI, PC, AT, and Ddimer) are presented in Table 2. There was no statistically significant difference in these markers between the 3 groups, with the exception that the IL-6 index was higher in patients in the IS group (P = .02). Spearman correlations between the various measurements are shown in Table 3. There were no significant correlations between HOMA or baseline glucose levels and IL-6 or PAI. There was a weak but statistically significant relationship between insulin levels and PAI (r= 0.22; P < .05), but no relationship between insulin levels and IL-6.

Clinical Outcomes

This study population had an overall hospital mortality of 36.5% (Table 4), and there were no differences between the 3 groups (37.5% vs 16.7% vs 40%; P = .70). Other clinical outcomes such as 14- and 28-day mortality, ICU-acquired infections, ICU-free and ventilator-free days, SOFA(day 1), SOFA(max), and SOFA(delta) scores were not significantly different across the 3 groups.

Relationship of Inflammation, Coagulation, and IR With 28-Day Mortality

In Table 5, laboratory and clinical variables are compared in survivors and nonsurvivors. Nonsurvivors tended to be older, have higher SOFA scores and significantly higher APACHE II scores (see Table 5). Among parameters that represent glycemic control or IR, there was no significant difference with survivors vs nonsurvivors. Inflammatory markers were significantly higher in nonsurvivors compared with survivors. AT levels were lower (P = .03), PC levels were lower (P = .002), and D-dimer were higher (P = .002) in nonsurvivors compared with survivors. In a stepwise multiple logistic regression analysis incorporating age, APACHE II, maximal glucose, and HOMA index, HI, IL-6 index, and PAI index, only IL-6 index (odds ratio, 1.64; confidence interval, 1.12-2.64; P = .01) was significantly associated with 28-day mortality. Thus, the final logistic regression model included IL-6 index as the only independent variable in the model.

Discussion

In this prospective observational study conducted in critically ill patients, we measured daily plasma insulin values to calculate HOMA scores and measure IR. To elucidate the effects of isolated IR, for the first time, we differentiated patients who had overt IR (hyperglycemia) vs non-overt IR (normal glucose but elevated HOMA) from patients who were insulin sensitive. Patients with overt IR are easily recognized in any clinical setting because they are hyperglycemic or require exogenous insulin. However, patients with non-overt IR would be unrecognizable clinically and may be mistakenly categorized as insulin sensitive. In noncritically ill patients, isolated insulin resistance without overt hyperglycemia is also associated with increased inflammation and coagulation abnormalities.6'9 If underlying IR does have dire clinical consequences in critically ill patients, non-overt IR patients would be affected and their risk would be undetected.

We showed that, overall, 90% of patients were IR, with 67% manifesting IR upon admission to ICU and an additional 23% developing IR through the 10-day study period. Only 10% of study patients remained IS. For those patients who were IR, there appeared to be no significant resolution of IR over a 10-day course of ICU stay.

We analyzed the components of glycemic control using 3 different methods: Glucose^sub day1^^sup max^ (highest glucose measured on the day of admission), Glucose^sub peak^^sup av^ (represents episodes of severe hyperglycemia that may last only for short periods of time), and HI (which incorporates the chronic nature of elevated glucose levels during the ICU stay). It is unknown whether moderate prolonged hyperglycemia (HI) or intermittent spikes of more severe hyperglycemia (peak) result in worse clinical outcomes. As expected, patients with overt IR had higher glucose levels than the other 2 groups, which internally validates our approach to classifying patients by degree of IR (Table 2). We hypothesized that, in critically ill patients, coexistence of hyperglycemia or IR may independently augment inflammatory and coagulation abnormalities, and that patients with non-overt IR would differ from patients who are sensitive to insulin. Since these coagulation and inflammatory abnormalities have been shown to correlate with ICU mortality, we had expected to observe a higher mortality in patients with IR.

In our exploratory analysis of the relationship between IR and biochemical endpoints and clinical outcomes, we did not observe any differences in mortality or morbidity among patients with overt IR, non-overt IR, or patients who were insulin sensitive. The APACHE II and SOFA scores were similar between the 3 groups, suggesting similar degrees of illness. Coagulation and inflammatory markers were similar in all 3 groups. Of interest, contrary to what we expected to find, we observed an inverse correlation between HOMA score and IL-6 levels (as HOMA increased, IL-6 decreased). There was no association between insulin levels and inflammatory markers, consistent with prior observations that insulin's effect on cytokines is unlikely to explain the beneficial effect of tight glycemic control.21 We describe a strong relationship between IR and nutrition intake. All the patients who received continuous supplemental nutrition were insulin resistant; no patient in the IS group received nutrition for more than 1 day.

In the multivariate analysis, we did not show a significant contribution of IR or hyperglycemia to 28-day mortality when adjusting for underlying severity of illness. These findings are consistent with another recent report describing HOMA scores in critically ill patients with acute renal failure where investigators could not show a relationship between HOMA, length of stay in ICU, acidosis, and inflammatory markers.27 In the univariate analysis, the HOMA score was higher in nonsurvivors, but this relationship was not significant in the multivariate analysis.27

In this critically ill patient population with an overall mortality rate of almost 40%, the regression analysis identified only IL-6 as significantly correlating with 28-day mortality. Nonsurvivors had higher levels of IL-6 (peak), PAI (peak), D-dimer, but lower levels of PC and AT than survivors. These findings are consistent with previous studies and further internally validate our data.13

The results of our exploratory analysis suggest that the severity of illness and the underlying inflammatory response, not hyperglycemia or IR, may be the main contributor to the outcome of severely ill, critically ill patients. Our initial hypotheses that both overt and nonovert IR can further augment inflammatory and coagulation abnormalities and would be associated with worse clinical outcomes were not substantiated here. One possible explanation may be that in the presence of overwhelming activation of cytokines, the contribution of hyperglycemia or IR is less significant. This is perhaps best illustrated by 2 patients in the IS group who both presented with overwhelming sepsis and both died within 24 hours of admission to ICU. They each had several recordings of blood glucose, none of which were elevated; however, their IL-6 levels were the highest in the study cohort.

The discrepancies between our study and previous results may be due to differences in patient populations studied. Our study population included medical and surgical patients, the overall mortality was 36.7%, and the average APACHE II score was 20.1 ± 7.7. In other publications, only studies of surgical patients show an association of hyperglycemia and adverse clinical outcomes.17,24 Whitcomb et al23 stratified a large heterogeneous group of ICU patients based on admission glucose levels and observed that hyperglycemia increased mortality in cardiac and neurosurgical patients but not medical and general surgical ICU patients. In the single-center studies performed by Van den Berghe et al,19,20 a beneficial effect on mortality was observed in the surgical ICU study (average APACHE II score was 9 and mortality ranged from 4.6% to 8%) but not the medical ICU study (average APACHE II score was 23 and mortality rate about 25%). Furthermore, subsequent multicenter randomized trials of intensive insulin therapy have failed to demonstrate any beneficial effect of this approach in heterogeneous groups of critically ill patients.22,30 Moreover, other publications have also demonstrated that after adjustment for the underlying severity of illness (APACHE II), glycemic control factors no longer remain significantly associated with mortality.26 An alternative reason that we did not observe an association between hyperglycemia and mortality may be that we chose a more strict definition for hyperglycemia (glucose > 7 mmol/L or 126 mg/dL). The multiple clinical and biochemical side effects of hyperglycemia may not all have the same threshold of effect. In previous studies, a stronger association between hyperglycemia and mortality was observed at higher degrees of hyperglycemia.22,23

Overall, our study and those mentioned above suggest that critically ill patients may be metabolically different depending on their current illness. Perhaps not all will benefit from the same degree of modulation of IR. Medical patients may not respond the same as surgical patients, and further study is needed to clarify which type of patients would have the greatest benefits and which patients may be unaffected or even harmed. It may be that sicker patients with overwhelming inflammation may not benefit from intensive insulin therapy or tight glycemic control.

Limitation

The gold standard for measuring IR is the insulin-glucose clamp technique, which was developed and validated in ambulatory patients.3-5 However, HOMA measurement in critically ill patients has been reported by Basi et al.27 Based on APACHE II score, SOFA score, and mortality, their patient population was similar to ours. They were also unable to demonstrate a significant relationship between HOMA scores and mortality in the multivariable model. In our study, 38 patients received exogenous insulin, all belonging to the overt-IR group. None of the patients in the other 2 groups required further insulin, and as demonstrated by Table 2, the fact that our methods of categorizing degrees of IR based on HOMA scores resulted in groups of patients with significant differences in glycemic control adds some validity to our measurements.

Another limitation of this study is the small sample size. Tb better study the effects of isolated non-overt IR in critically ill patients, a larger sample size of this group would be needed. In addition, our measurements of insulin, cytokines, and coagulation factors show great variability, as seen in other studies. Given the small sample size and large variability, our study would be underpowered to demonstrate a small difference in clinical outcomes across the 3 groups. Thus, the results of our exploratory analysis have to be considered as preliminary or hypothesis-generating.

Conclusion

In this study, we have demonstrated a high prevalence of IR in critically ill patients that does not appear to significantly influence inflammatory and coagulation abnormalities. There may not be a strong relationship between IR and clinical outcomes in all critically ill patients.

Because our data demonstrate a remarkably high prevalence of IR in a heterogeneous ICU patient population, the clinical message is that limiting excessive exogenous energy intake to avoid hyperglycemia exposure is clinically warranted. This conclusion is supported by the fact that the patients who received the least energy intake were more likely to remain insulin sensitive. This approach has been recently recommended by the Canadian Clinical Practice Guidelines for nutrition in ICU patients.28 The time and resource commitments required to regulate glucose in ICU patients is enormous, intensive insulin therapy and the resultant hypoglycemia may even have detrimental effects in some patients,29 and subsequent multicenter trials have not be able to replicate the original findings.22,30 Further studies are needed to clarify which patient population benefits the most from this therapy, and to what degree plasma glucose should be regulated. Such studies are under way in Australia and Canada.30