Volume 104, Issue 8 , Pages 1013-1017, 15 October 2009
Usefulness of Changes in Fasting Glucose During Hospitalization to Predict Long-Term Mortality in Patients With Acute Myocardial Infarction
Article Outline
Stress hyperglycemia is a complex phenomenon that incorporates the cumulative effects of multiple factors. Rapid changes in blood glucose may reflect neurohormonal and homodynamic events that affect patient outcome. We prospectively studied the relation between changes in fasting glucose (FG) during a hospital course and long-term mortality in 1,467 nondiabetic patients with acute myocardial infarction. FG was obtained at admission and later during the hospital course and classified at each time point as normal (<100 mg/dl), impaired (100 to 125 mg/dl), or diabetic range (≥126 mg/dl). The relation between measurements of FG and mortality (median follow-up 30 months) was assessed using Cox models. FG classification improved in 426 (29.0%) and worsened in 248 patients (16.9%) during hospitalization. Mean FG was a better predictor of mortality than baseline or final FG levels alone (C-index 0.670, 0.656, and 0.645, respectively). Changes in FG during hospitalization were strongly associated with changes in mortality risk. Compared to patients with persistent normal FG, the adjusted hazard ratio (HR) for mortality was 2.6 (95% confidence interval [CI] 1.0 to 7.2) for patients in whom FG increased to the diabetic range; the HR was 6.3 (95% CI 4.0 to 10.4) in patients with persistent FG in the diabetic range but decreased substantially when FG normalized during hospitalization (HR 2.7, 95% CI 1.3 to 5.1). In conclusion, persistent increase of FG during hospitalization for acute myocardial infarction has greater prognostic effect than baseline FG. Changes in FG during hospitalization are simple and sensitive indicators of dynamic changes in risk.
Stress hyperglycemia is a multifactorial and complex phenomenon that incorporates the cumulative effects of activation of multiple neurohormonal pathways that promote insulin resistance1 and respond to hemodynamic alterations and changes in the clinical status of the patient. Therefore, rapid fluctuations in blood glucose levels may occur, especially in the early phase of an acute event. In the present study we prospectively evaluated the long-term predictive value of changes in glucose levels in patients with acute myocardial infarction (AMI). Our aim was to determine whether changes in fasting glucose (FG) during hospital course could provide additional information with regard to the outcome of nondiabetic patients with stress hyperglycemia.
Methods
The study included patients presenting to the intensive coronary care unit of Rambam Medical Center (Haifa, Israel) with AMI from July 2001 to September 2007. AMI was diagnosed based on the European Society of Cardiology and American College of Cardiology criteria.2 Patients were included in the study if ≥3 glucose measurements were available: admission glucose and 2 FG measurements obtained during the hospital course. Exclusion criteria were known inflammatory disease and surgery or trauma within the previous month and use of intravenous glucose infusions during the hospital course. The institutional review committee on human research approved the study protocol.
Plasma glucose level was taken in all patients on admission. Baseline FG levels were obtained after an overnight fast of ≥8 hours, within 24 hours of admission. The second sample of FG (last FG) was obtained after an overnight fast of ≥8 hours during hospital stay, in most cases before planned hospital discharge. When several FG measurements were available, the last FG measurement was used for analysis.
Classification of normal FG at the 2 time points in patients who had not been diagnosed with diabetes was made according to criteria of the American Diabetes Association.3 FG was thus classified as normal (<100 mg/dl [<5.6 mmol/L]), impaired FG range (100 to 125 mg/dl [5.6 to 6.9 mmol/L]), or diabetic range (≥126 mg/dl [≥7.0 mmol/L]).
Patients were considered as having diabetes if they had been previously informed of the diagnosis by a physician, were taking oral antihyperglycemic agents or insulin, or receiving diet therapy. Patients without previously undiagnosed diabetes who required initiation of antihyperglycemic therapy during their hospital stay were also considered to have diabetes.4
The primary end point of the study was all-cause mortality. After hospital discharge, clinical end-point information was acquired by reviewing the national death registry and by contacting each patient individually.
Baseline characteristics of groups categorized by FG levels were compared using analysis of variance for continuous variables and by chi-square statistic for categorical variables. To assess the agreement between classification of FG at baseline and during hospital course, we used Cohen kappa.
Event-free survival curves were estimated by the Kaplan-Meier method and compared to log-rank test. Univariate and multivariate Cox proportional hazards models were used to calculate hazard ratios (HRs) and 95% confidence intervals (CIs) for various risk variables. In these models, we analyzed the effect of baseline FG levels, the last FG levels, and mean FG levels. In addition, changes in FG levels during the hospital course were modeled using an indicator variable, in which each patient was categorized based on baseline FG and last FG levels (normal, impaired FG, or diabetes at each time point, with 9 FG combinations).
In the multivariable Cox models, FG categories were adjusted for the Global Registry of Acute Coronary Events (GRACE) risk score.5 The GRACE risk score is a validated 9-variable prediction tool5 that can be used to estimate a patient's risk for all-cause mortality in the entire spectrum of patients with acute coronary syndromes.5, 6 Because changes in FG levels during hospitalization may be related to the time elapsing between the 2 FG easements (i.e., the difference between the first and second FG measurements is likely to be greater with longer intervals between the 2 measurements), we also adjusted for this variable.
We also assessed the likelihood ratio chi-square statistic as an indicator of the global goodness of fit of predictive models (a higher value indicates a better model fit). In addition, we estimated the Akaike information criterion,7 which is a statistical estimate of the trade-off between the likelihood of a model and its complexity, with a lower value indicating a better model.
The C-index derived from the multivariable models was used to assess the improvement in the prognostic model discrimination resulting from various measurements of FG and from the sequential addition of mean FG and FG change to a model including the GRACE risk score. Differences were considered statistically significant at the 2-sided p <0.05 level. Statistical analyses were performed using SPSS 15.0 (Chicago, Illinois) and STATA 10.0 (STATA Corp., College Station, Texas).
Results
From July 2001 to July 2007, 1,467 nondiabetic and 647 diabetic patients who presented with AMI were enrolled. Clinical characteristics of nondiabetic (according to baseline FG level) and diabetic patients are listed in Table 1. Increased levels of FG were associated with older age, female gender, and previous infarction, previous heart failure, and hypertension. Patients presenting with increased FG had higher creatinine and higher heart rates and Killip class on admission. They were less likely to receive primary angioplasty or coronary revascularization and received less angiotensin-converting enzyme inhibitors and β blockers during their hospital course.
Table 1. Baseline clinical characteristics according to baseline fasting glucose
| Characteristic | Baseline FG | Diabetes(n = 647) | p Value⁎ | ||
|---|---|---|---|---|---|
| <100 mg/dl(n = 591) | 100–125 mg/dl(n = 580) | ≥126 mg/dl(n = 296) | |||
| Age (years) | 58 ±13 | 60±13 | 65±13 | 64±12 | <0.0001 |
| Women | 97(16%) | 107(18%) | 79(27%) | 201(31%) | <0.0001 |
| Previous MI | 110(19%) | 109(19%) | 63(21%) | 185(29%) | <0.0001 |
| Previous heart failure | 33(6%) | 30(5%) | 17(6%) | 69(11%) | <0.0001 |
| Smoker | 330(56%) | 280(48%) | 130(44%) | 205(32%) | <0.0001 |
| Hypertension (history) | 252(43%) | 252(43%) | 170(57%) | 448(69%) | <0.0001 |
| Creatinine (mg/dl) | 0.9(0.8–1.1) | 0.9(0.8–1.2) | 1.0(0.8–1.2) | 0.9(0.8–1.2) | 0.005 |
| Systolic blood pressure at admission (mm Hg) | 130±23 | 131±26 | 130±29 | 133±28 | 0.10 |
| Heart rate at admission (beats/min) | 74±16 | 76±17 | 80±21 | 83±19 | <0.0001 |
| Killip class >I | 64(11%) | 108(19%) | 97(33%) | 219(34%) | <0.0001 |
| ST-elevation infarction | 433(73%) | 456(79%) | 241(81%) | 465(72%) | 0.002 |
| Anterior infarction | 240(41%) | 256(44%) | 133(45%) | 272(42%) | 0.52 |
| In-hospital medications | |||||
| Antiplatelet agents | 586(99%) | 569(98%) | 275(93%) | 624(96%) | <0.0001 |
| β blockers | 501(85%) | 503(87%) | 197(67%) | 552(81%) | <0.0001 |
| Angiotensin-converting enzyme inhibitors/angiotensin receptor blockers | 471(80%) | 470(81%) | 201(68%) | 516(80%) | <0.0001 |
| Statins | 441(77%) | 447(77%) | 193(65%) | 482(75%) | 0.002 |
| Reperfusion therapy | |||||
| Thrombolysis | 132(23%) | 138(24%) | 55(20%) | 106(16%) | 0.006 |
| Primary angioplasty | 196(33%) | 182(31%) | 109(39%) | 186(29%) | 0.08 |
| Coronary revascularization | 295(50%) | 294(51%) | 138(47%) | 264(41%) | 0.002 |
⁎Analysis of variance for continuous variables and by chi-square statistic for categorical variables. |
Median time between the first and second FG measurements was 4 days (interquartile range 2 to 5). Changes in classification of FG levels during the hospital course are listed in Table 2. Baseline FG classification remained unchanged in only 793 patients (54.1%). FG classification improved in 426 patients (29.0%) and worsened in 248 (16.9%). There was a moderate correlation between baseline and last FG levels (Spearman rho 0.47, p <0.0001), and a poor agreement between FG classification at baseline and before discharge (Table 2; Cohen kappa 0.27).
Table 2. Fasting glucose category in patients with acute myocardial infarction at admission and during hospital course (n = 1467)
| FG at Admission | Last FG | ||
|---|---|---|---|
| Normal | Impaired FG | Diabetes | |
| Normal (<100 mg/dl) | 408(58.5%) | 157(27.4%) | 26(13.3%) |
| Impaired FG (100–125 mg/dl) | 234(33.5%) | 281(49.0%) | 65(33.3%) |
| Diabetes (≥126 mg/dl) | 56(8.0%) | 136(23.7%) | 104(53.3%) |
During a median follow-up of 30 months (range 6 to 48), 215 patients died. In unadjusted analyses, higher baseline FG and last FG levels were strongly associated with increased risk for long-term mortality (Table 3). However, a model based on mean FG levels indicated that patients with a mean FG level in the diabetic range were at a particularly high risk for long-term mortality, with an adjusted HR of 5.6. The model based on mean FG levels had the highest C-statistics and a better goodness of fit based on the Akaike information criterion (Table 3). After adjustments for the GRACE risk score, compared to patients with a mean FG in the normal range, adjusted HRs for morality were 1.5 (95% CI 1.1 to 2.1, p = 0.02) in patients with a mean FG in the impaired FG range and 3.8 (95% CI 2.7 to 5.4, p <0.0001) in patients with a mean FG in the diabetic range. Using mean FG as a continuous variable, the adjusted HR for each 10-mg/dl increase was 1.26 (95% CI, 1.20 to 1.33, p <0.0001).
Table 3. Unadjusted Cox models for mortality
| Model | Patients | Events (%) | Unadjusted OR (95% CI) | p Value | Model C-Statistic | Akaike Information Criterion |
|---|---|---|---|---|---|---|
| Baseline FG | 0.656 | 2,958.3 | ||||
| Normal | 591 | 54(9.1) | 1.0(referent) | |||
| Impaired FG | 580 | 68(11.7) | 1.3(0.9–1.9) | 0.12 | ||
| Diabetes | 296 | 93(31.4) | 4.0(2.9–5.6) | <0.0001 | ||
| Last FG | 0.645 | 2,974.2 | ||||
| Normal | 698 | 64(9.2) | 1.0(referent) | |||
| Impaired FG | 574 | 90(15.7) | 1.8(1.3–2.5) | <0.0001 | ||
| Diabetes | 195 | 61(31.3) | 4.2(2.9–5.9) | <0.0001 | ||
| Mean FG⁎ | 0.676 | 2,933.6 | ||||
| Normal | 606 | 50(8.3) | 1.0(referent) | |||
| Impaired FG | 627 | 79(12.6) | 1.6(1.1–2.3) | 0.01 | ||
| Diabetes | 234 | 86(36.8) | 5.6(3.7–7.9) | <0.0001 | ||
| Mean admission and FG† | 0.659 | 2,947.6 | ||||
| Normal | 333 | 30(9.0) | 1.0(referent) | |||
| Impaired FG | 714 | 64(9.0) | 1.0(0.7–1.6) | 0.93 | ||
| Diabetes | 420 | 121(28.8) | 3.7(2.4–5.4) | <0.0001 |
⁎Mean of baseline and last FG levels. |
†Mean of admission glucose, baseline FG, and last FG. |
To better demonstrate the relation between changes in glycometabolic status during hospital course and outcome, changes in FG levels were modeled after categorizing patients based on their baseline and last FG levels. Figure 1 shows the effect of changes in FG levels on the GRACE score–adjusted HRs for mortality. For each baseline FG category, risk for long-term mortality followed changes in FG levels during the hospital course. In patients with normal FG levels at baseline, there was a graded increase in risk with increasing FG levels during the hospital course (adjusted HR 2.6 for patients in whom FG increased to the diabetic range during hospitalization compared to patients remaining with normal FG levels). Patients with baseline FG in the diabetic range were at highest risk for mortality if their FG levels remained in the diabetic range (adjusted HR 6.4). However, decreasing FG levels during hospitalization was associated with a decrease in mortality in these patients.
Figure 1.
Changes in mortality risk according to changes in classification of FG levels during hospitalization, with GRACE score–adjusted HRs (95% CIs) for each baseline and last FG category.
Notably, patients with increased FG at baseline had an increased mortality even if their FG decreased during their hospital stay. For example, the adjusted HR of patients with baseline FG levels in the diabetic range who normalized their FG during hospitalization was 2.7 compared to patients with normal FG at the 2 time points.
Adding the mean FG or FG change variables to the Cox model containing the GRACE risk score resulted in a significant increase in the discrimination of the model (C-index 0.784 vs 0.810, p = 0.024, and 0.784 vs 0.812, p = 0.028, respectively). Of note, the Akaike information criterion indicated a preference for the inclusion of mean FG or FG change in the GRACE-based model after adjustment for adding these variables (2,761.9 and 2,765.9 vs 2,818.4, respectively). This suggests that the model including FG change provided a better fit, even after adjustment for the increase in the number of predictors (model complexity).
To determine whether previously unrecognized diabetes could have contributed to the adverse outcome of nondiabetic patients with persistently increased FG level in the diabetic range, we compared the outcome of patients with previously known diabetes (n = 647) to that of patients without previously recognized diabetes and persistent increased FG in the diabetic range (n = 296). In a multivariable Cox model, compared to patients with known diabetes, the adjusted HR for mortality in patients without previously known diabetes but with persistent increased FG in the diabetic range was 2.2 (95% CI 1.6 to 3.0, p <0.0001).
Discussion
The results of the present study add significantly to the available data regarding the clinical significance of increased glucose levels in patients with AMI. Mean FG level, a simple measurement of glucose control over time, was a better predictor of outcome than baseline or last FG level alone. However, we observed wide fluctuations in FG levels during the first few days of hospitalization. An indicator variable that captured changes in FG level during the hospital course was also a strong predictor of outcome. Furthermore, changes in FG concentrations during hospitalization provided a robust indicator of individual changes in the long-term risk of mortality.
Previous studies have identified patients during the acute phase of the infarction, at admission, or during the first 24 hours of hospital stay.8, 9, 10, 11 A recent scientific statement from the American Heart Association highlighted the importance of establishing whether persistent hyperglycemia during hospitalization for acute coronary syndrome has a greater impact on prognosis than admission hyperglycemia alone.12 Using serial random glucose measurements, Kosiborod et al13 recently demonstrated that persistent hyperglycemia is a more important predictor of in-hospital mortality than admission hyperglycemia alone. In the present study, mean FG was also a better predictor of long-term outcome compared to a single measurement of FG at baseline or during hospitalization.
However, mean FG levels, and other measurements of glucose control over time,13 do not capture rapid changes in glycometabolic state that are associated with an increasing or decreasing risk in individual patients. Serial measurements of FG levels showed that a relatively large proportion of patients may proceed with rapid improvement or resolution of an abnormal glycometabolic state. For example, 18.9% of patients with baseline FG in the diabetic range showed reversion to normal FG levels during their hospital course, and this glycometabolic improvement was associated with a marked decrease in long-term mortality. Conversely, 30.0% of apparently low-risk patients with normal FG levels at admission exhibited an abnormal glycometabolic state when tested several days later. These new increases of FG occurring during hospital course shifted patients to a substantially higher risk. In these patients, intervening events such as unsuccessful reperfusion, recurrent infarction, and development of new heart failure or excessive neurohormonal activation may worsen the metabolic profile of the patient.
Although in-hospital improvement in FG was associated with improved outcome, long-term mortality remained higher even in patients in whom increased FG at admission returned to normal levels at discharge compared to patients with persistently normal FG levels. These results indicate that even a transient increase in FG portends increased long-term mortality. Thus, in contrast to measurements of glucose control over time, change in FG during hospitalization provides a simple and sensitive measurement that reflects individual changes in risk and can be used by clinicians to quickly reclassify a patient to a substantially lower or higher risk.
It has been suggested that glucose levels may be used as a biomarker in the setting of AMI,14 and our study supports this notion. In the setting of AMI, glucose levels satisfy 3 fundamental requirements for a potentially useful biomarker15: (1) analytic methods that allow reliable measurement are available to every clinician, (2) it has been consistently shown to provide incremental prognostic information when added to validated risk scores,8, 9, 10, 13 and (3) glucose levels may help the clinician to manage patients. In addition to the use of FG in risk stratification, results of the present study suggest that FG may be useful in monitoring the response to therapy during hospitalization for AMI.
Whether hyperglycemia, when present during the acute phase of AMI, should be treated with intensive insulin therapy even in nondiabetic patients remains controversial. Our observation that patients with increased FG at admission that normalized during the hospital course had a lower mortality compared to patients with persistent increased FG is intriguing and may affect the management of increased glucose levels during the acute phase of MI.
Although acute hyperglycemia is associated with numerous adverse effects that potentially contribute to a poor outcome in AMI,12 the present study cannot distinguish whether FG levels are merely risk markers or direct mediators of outcome. Thus, the observed changes in FG during hospitalization may represent the amelioration, aggravation, or persistence of the acute stress response.
In 2 recent studies, intensive insulin therapy (initiated when blood glucose levels exceeded 110 mg/dl and were adjusted to maintain glucose levels in the range of 80 to 110 mg/dl) was associated with a significantly increased rate of severe hypoglycemic events in patients in the intensive care unit, but provided no clear benefit.16, 17 The recent Intensive Care Evaluation–Survival Using Glucose Algorithm Regulation trial reported an increase in the rate of death with intensive glucose control compared to conventional control, with a marked increase in the frequency of severe hypoglycemic episodes in the intensive-control group.18 We observed a spontaneous improvement in some patients with increased FG at admission. Thus, patients initially perceived as appropriate candidates for initiation of intravenous insulin therapy may be at increased risk for hyperglycemia if a rapid simultaneous improvement in their metabolic status occurs.
Failure of stress-related increased glucose levels to decrease may represent pre-existing glycometabolic dysregulation, subclinical or frank previously undiagnosed diabetes, which is known to negatively affect prognosis.19 However, previously undiagnosed diabetes could not account for the increased mortality associated with persistent fasting hyperglycemia in patients without previously known diabetes because the mortality risk associated with persistent hyperglycemia in nondiabetic patients was considerably higher than that of patients with a previous diagnosis of diabetes. Mean FG level was obtained from only 2 measurements of FG. Mean FG derived from multiple daily determinations might provide superior prognostic information.
References
- . Determinants and importance of stress hyperglycaemia in non-diabetic patients with myocardial infarction. Br Med J Clin Res Ed. 1986;293:917–922
- . Myocardial infarction redefined—a consensus document of the Joint European Society of Cardiology/American College of Cardiology committee for the redefinition of myocardial infarction. J Am Coll Cardiol. 2000;36:959–969
- . Follow-up report on the diagnosis of diabetes mellitus. Diabetes Care. 2003;26:3160–3167
- . Newly diagnosed and previously known diabetes mellitus and 1-year outcomes of acute myocardial infarction: the VALsartan In Acute myocardial iNfarcTion (VALIANT) trial. Circulation. 2004;110:1572–1578
- . A validated prediction model for all forms of acute coronary syndrome: estimating the risk of 6-month postdischarge death in an international registry. JAMA. 2004;291:2727–2733
- . Global Registry of Acute Coronary Events (GRACE) hospital discharge risk score accurately predicts long-term mortality post acute coronary syndrome. Am Heart J. 2007;153:29–35
- . AIC model selection using Akaike weights. Psychon Bull Rev. 2004;11:192–196
- . Fasting glucose in acute myocardial infarction: incremental value for long-term mortality and relationship with left ventricular systolic function. Diabetes Care. 2007;30:960–966
- . Fasting glucose is an important independent risk factor for 30-day mortality in patients with acute myocardial infarction: a prospective study. Circulation. 2005;111:754–760
- . Impact of fasting glycemia on short-term prognosis after acute myocardial infarction. J Clin Endocrinol Metab. 2007;92:2136–2140
- . Prognostic significance of the change in glucose level in the first 24 h after acute myocardial infarction: results from the CARDINAL study. Eur Heart J. 2006;27:1289–1297
- . Hyperglycemia and acute coronary syndrome: a scientific statement from the American Heart Association diabetes committee of the council on nutrition, physical activity, and metabolism. Circulation. 2008;117:1610–1619
- . Glucometrics in patients hospitalized with acute myocardial infarction: defining the optimal outcomes-based measure of risk. Circulation. 2008;117:1018–1027
- . Glucose: a biomarker in acute myocardial infarction ready for prime time?. Circulation. 2008;117:990–992
- . Benchmarks for the assessment of novel cardiovascular biomarkers. Circulation. 2007;115:949–952
- . Intensive insulin therapy in the medical ICU. N Engl J Med. 2006;354:449–461
- . Intensive insulin therapy and pentastarch resuscitation in severe sepsis. N Engl J Med. 2008;358:125–139
- . Intensive versus conventional glucose control in critically ill patients. N Engl J Med. 2009;360:1283–1297
- . Mechanisms determining course and outcome of diabetic patients who have had acute myocardial infarction. Ann Intern Med. 1997;126:296–306
PII: S0002-9149(09)01173-4
doi:10.1016/j.amjcard.2009.05.053
© 2009 Elsevier Inc. All rights reserved.
Volume 104, Issue 8 , Pages 1013-1017, 15 October 2009
