CI indicates confidence interval; OR, odds ratio. To convert glucose to mmol/L, multiply by 0.0555.
Customize your JAMA Network experience by selecting one or more topics from the list below.
Kosiborod M, Inzucchi SE, Goyal A, et al. Relationship Between Spontaneous and Iatrogenic Hypoglycemia and Mortality in Patients Hospitalized With Acute Myocardial Infarction. JAMA. 2009;301(15):1556–1564. doi:10.1001/jama.2009.496
Context While glucose control is recommended by professional societies for patients with hyperglycemia hospitalized with acute myocardial infarction (AMI), enthusiasm for glucose lowering is tempered, in part, by concerns of inducing hypoglycemia. Yet, whether episodic hypoglycemia that occurs as a result of glucose-lowering therapy is harmful in patients with AMI is unknown.
Objective To determine whether the mortality risk associated with hypoglycemic events is similar in patients who develop hypoglycemia spontaneously and those who develop it as a result of insulin therapy.
Design, Setting, and Patients Retrospective cohort study using data from Health Facts, a contemporary database of patients hospitalized across the United States in 40 hospitals between January 1, 2000, and December 31, 2005. Of all the patients in the database, 7820 patients were hospitalized with AMI and were hyperglycemic on admission (glucose level ≥140 mg/dL). Patients were stratified based on whether they developed a hypoglycemic event (random glucose level <60 mg/dL) during subsequent hospitalization. Logistic regression models were used to evaluate the association between hypoglycemia and in-hospital mortality within subgroups of patients who were and were not treated with insulin therapy.
Main Outcome Measure All-cause in-hospital mortality.
Results Among patients treated or not treated with insulin, those with hypoglycemia were older and had more comorbidity. Hypoglycemia was associated with increased mortality in patients not treated with insulin (18.4% [25/136] mortality in patients with hypoglycemia vs 9.2% [425/4639] in those without hypoglycemia; P<.001), but not in those treated with insulin (10.4% [36/346] mortality in patients with hypoglycemia vs 10.2% [276/2699] in those without hypoglycemia; P = .92). After multivariable adjustment, there was a significant interaction between hypoglycemia and insulin therapy (P value for interaction = .01). Hypoglycemia was a predictor of higher mortality in patients who were not treated with insulin (odds ratio, 2.32 [95% confidence interval, 1.31-4.12] vs patients without hypoglycemia), but not in patients treated with insulin (odds ratio, 0.92 [95% confidence interval, 0.58-1.45] vs patients without hypoglycemia).
Conclusions While hypoglycemia was associated with increased mortality in patients with AMI, this risk was confined to patients who developed hypoglycemia spontaneously. In contrast, iatrogenic hypoglycemia after insulin therapy was not associated with higher mortality risk.
Elevated glucose levels in the setting of acute myocardial infarction (AMI) is a well-recognized risk factor for increased mortality,1-27 and limited clinical trial data suggest that better glucose control may improve patient outcomes after AMI.28 Based on these findings, professional societies currently endorse glucose control in patients with hyperglycemia hospitalized with AMI.29-31
Enthusiasm for glucose-lowering interventions has been tempered, in part, by concerns of inducing hypoglycemia because randomized controlled trials of intensive glucose control with insulin have shown higher rates of hypoglycemia than standard glucose management.32-35 The concerns generated by these findings were further heightened after several observational studies showed that hypoglycemia during a hospitalization for AMI was associated with higher mortality.22,26,36 However, these studies could not distinguish between hypoglycemia that occurs spontaneously (due to severe illness) and hypoglycemic events that occur as a consequence of glucose-lowering therapy.
These competing concerns surrounding the risk associated with hyperglycemia and the potential risk of hypoglycemia (an occasional, but inevitable consequence of intensive glucose management) highlight a critical unresolved issue. The issue is whether episodic hypoglycemia that occurs as a consequence of glucose-lowering therapy is harmful in patients hospitalized with AMI. Appropriately designed observational studies present the best opportunity to address this knowledge gap by determining whether hypoglycemia that occurs spontaneously and hypoglycemia that occurs following initiation of glucose-lowering therapy (such as insulin administration) carry similar prognostic implications during hospitalization for AMI. Specifically, if hypoglycemia was a direct cause of adverse events, it would be expected to be associated with poor prognosis regardless of its etiology.
To address this important knowledge gap, we analyzed data from Health Facts, a contemporary database of patients hospitalized with AMI in 40 hospitals across the United States between 2000 and 2005, to compare the association of iatrogenic and spontaneous hypoglycemia with in-hospital mortality. This database provides a unique opportunity to closely examine the relationship between hypoglycemia and patient outcomes during AMI because it contains detailed information regarding glucose measurements and insulin therapy in a large group of patients. We specifically sought to determine whether the prognostic implications of hypoglycemia differ in patients with AMI who develop hypoglycemia spontaneously and in those who experience hypoglycemia after the initiation of insulin therapy.
The details of the Health Facts database have been described.26 This database captures deidentified patient data from the contributor institutions' electronic health records. All 40 participating medical centers in the Health Facts consortium contributed deidentified information on consecutive patients treated between January 1, 2000, and December 31, 2005. Data included demographic variables (including race/ethnicity, all abstracted from the electronic medical records and documented during a patient's registration process), medical history and comorbidities (determined from International Classification of Diseases, Ninth Revision, Clinical Modification diagnostic codes), comprehensive laboratory data (including all venous and fingerstick blood glucose measurements during hospitalization), comprehensive pharmacy data, in-hospital procedures (including cardiac catheterization, percutaneous coronary intervention, coronary artery bypass graft surgery), in-hospital mortality, and hospital characteristics (such as geographic region, number of beds, presence of cardiac catheterization and angioplasty facilities, presence of cardiothoracic surgery facilities, teaching vs nonteaching status).
All patients hospitalized with a primary discharge diagnosis of AMI between January 2000 and December 2005 (using International Classification of Diseases, Ninth Revision, Clinical Modification codes 410.xx and excluding 410.x2, which represents readmission after AMI) who had a glucose measurement on admission and at least 1 documented abnormal troponin I or T or creatine kinase MB fraction were identified (n = 23 613). Subsequently, those patients who were transferred from or to other acute care facilities were excluded (n = 6742) because complete laboratory and medication administration details for the entire episode of AMI care were not available. Patients whose admission glucose level was less than 140 mg/dL (n = 8218) (to convert glucose to mmol/L, multiply by 0.0555) were excluded because these patients would typically not be considered candidates for pharmacological glucose lowering in this setting. In addition, 833 patients who had no additional glucose measurements following admission (in patients who were not treated with insulin), or after initiation of insulin (in insulin-treated patients) were excluded because their subsequent glucose levels could not be ascertained. The final cohort included 7820 patients with hyperglycemia on admission and biomarker-confirmed AMI.
The Health Facts database provided access to all of the patients' glucose levels (both capillary and plasma assessments), including their measurement time. Whole blood glucose specimens were converted by glucose meters into plasma glucose levels (in units of mg/dL) for all analyses (plasma glucose = whole blood glucose × 1.15). For clarity, all plasma and capillary glucose measurements are subsequently referred to as blood glucose.
Administration of any insulin during hospitalization (either subcutaneous or intravenous, short-acting or long-acting) was considered to constitute insulin treatment, and was used to stratify patients into those who were and were not treated with insulin during hospitalization. The time of insulin administration was recorded for all treated patients. These data were ascertained directly from the Health Facts database.
Hypoglycemia was defined as any random blood glucose level less than 60 mg/dL37,38 that occurred either following admission (in patients who were not treated with insulin), or following initiation of insulin therapy (in insulin-treated patients). The hypoglycemic events were ascertained by analyzing all of the patients' glucose levels during hospitalization (both capillary and plasma assessments). In a sensitivity analysis, a random blood glucose level of less than 70 mg/dL (hypoglycemia definition used by the American Diabetes Association) was also evaluated for its association with the outcome.39 The outcome for this study was all-cause in-hospital mortality as obtained directly from the Health Facts database.
Baseline demographic and clinical characteristics were compared between patients who did and did not develop hypoglycemia by use of the Pearson χ2 test for categorical variables and the t test for continuous variables.
To assess whether development of hypoglycemia is associated with in-hospital survival, mortality rates were compared between those patients who did and did not develop hypoglycemia. This was first done in the entire cohort and then within the subgroups of patients who were and were not treated with insulin, using the χ2 test.
Multivariable logistic regression models were subsequently developed to evaluate whether the association between hypoglycemia and mortality persisted after adjustment for other patient characteristics and potential confounders. These analyses were conducted separately within subgroups of patients who were and were not treated with insulin. Patient characteristics previously demonstrated to be prognostically significant or thought to be clinically important, and covariates identified in bivariate analyses as predictors of in-hospital mortality, were entered into the models. Covariates included demographic factors (age, sex, race); comorbidities (diabetes, heart failure, hypertension, cerebrovascular disease, peripheral vascular disease, chronic obstructive pulmonary disease, dementia); laboratory values on admission (creatinine, white blood cell count, hematocrit); peak troponin or creatine kinase MB level; procedures during hospitalization, including revascularization (cardiac catheterization, percutaneous coronary intervention, coronary artery bypass graft surgery); and medications during hospitalization (aspirin, clopidogrel, ticlopidine, β-blockers, calcium channel blockers, nitrates, diuretics, bronchodilators, statins, oral antihyperglycemic agents).
Models also were adjusted for the admission glucose level, given its known association with mortality in patients with AMI,1 and for the frequency of glucose testing because the intensity of testing could be related to both the presence and severity of hypoglycemia, as well as in-hospital mortality. Furthermore, analyses were adjusted for hospital length of stay and clustering by hospital site (using hierarchical logistic regression models with random effects introduced for hospital site). Nonlinear trends for all continuous variables were tested through the use of restricted cubic splines.
Several sensitivity analyses were conducted to support the robustness of the findings, and to minimize the possibility of residual confounding. First, all patients who were treated with oral antihyperglycemic agents during hospitalization were excluded from the cohort to eliminate any possible cases of hypoglycemia that occurred as a consequence of these medications. Second, patients who died within 24 hours of admission were excluded to reduce any possible survivor bias. Third, models were repeated using a different definition of hypoglycemia (<70 mg/dL).39 Fourth, a diabetes × hypoglycemia interaction term was introduced into the models to determine whether the association between hypoglycemia and mortality differed in patients with and without known diabetes.
A 2-sided P value of .05 was considered to be statistically significant. Analyses were conducted with SAS software version 8.02 (SAS Institute Inc, Cary, North Carolina). Use of the Health Facts database was approved by the Saint Luke's Mid America Heart Institute's institutional review board.
Baseline characteristics of 7820 patients with AMI across the subgroups of those who did and did not develop hypoglycemia are detailed in Table 1 and Table 2. Compared with patients who did not develop hypoglycemia, a greater proportion of those with hypoglycemic events were nonwhite, female, and had heart failure and diabetes. Patients who developed hypoglycemia were older, had higher admission levels of creatinine, higher white blood cell counts, lower hematocrit and peak troponin levels, were less likely to undergo coronary angiography and percutaneous intervention, and were treated less frequently with non–aspirin platelet inhibitors and more frequently with diuretics, bronchodilators, oral antihyperglycemic agents, and insulin. Patients who developed hypoglycemia also had higher rates of septic shock, respiratory failure, acute kidney injury, longer hospital lengths of stay, and higher glucose levels at admission.
Thirty-nine percent (n = 3045) of patients were treated with insulin during hospitalization. Median time to insulin administration in patients who were treated was 6.6 hours following admission (interquartile range, 3.2-16.3 hours). Patients treated with insulin received a mean of 3 insulin doses; 59% received only short-acting insulin, 2.7% received only long-acting insulin, and 38.3% received both. Seventeen percent of insulin-treated patients received insulin intravenously. The median number of glucose measurements following the admission assessment was 6 (interquartile range, 3-15) in insulin-treated patients and 3 (interquartile range, 2-7) in patients who were not treated with insulin.
Overall, 482 patients (6%) developed hypoglycemia. Patients treated with insulin experienced higher rates of hypoglycemia compared with those who were not treated with insulin therapy (11.4% [346/3045 patients] vs 2.9% [136/4775 patients], respectively; P < .001). Among patients treated with insulin, the rates of hypoglycemia were 6.5% (117/1796) for those who received only short-acting insulin, 17.1% (14/82) for those who received only long-acting insulin, and 18.4% (215/1167) for those who received both. The severity of hypoglycemic events was similar between those who developed hypoglycemia spontaneously, and those who developed hypoglycemia after insulin initiation (mean glucose level during hypoglycemic events, 46.6 vs 45.9 mg/dL, respectively). Mean glucose level during hospitalization was significantly lower in insulin-treated patients who developed hypoglycemia (vs insulin-treated patients without hypoglycemia); in contrast, in patients who were not treated with insulin, mean glucose level during hospitalization was similar among those who did and did not develop hypoglycemia (Table 2).
In the unadjusted analyses, hypoglycemia was associated with worse survival (mortality for patients with vs without hypoglycemia was 12.7% [61/482] vs 9.6% [701/7338], respectively; P = .03). However, the relationship between hypoglycemia and mortality was markedly different within subgroups of patients who developed it spontaneously and those who developed it following insulin therapy. Among patients who were not treated with insulin, those with hypoglycemia had much higher mortality compared with those who did not develop hypoglycemia (18.4% [25/136] vs 9.2% [425/4639], respectively; P<.001). In contrast, among patients who were treated with insulin, mortality rates were similar between those who did and did not develop hypoglycemia (10.4% [36/346] vs 10.2% [276/2699], respectively; P = .92 [P = .008 for hypoglycemia × insulin interaction]; Table 3).
After multivariable adjustment, the nature of association between hypoglycemia and mortality continued to differ markedly between those patients who developed spontaneous hypoglycemia, and those who had hypoglycemia following insulin administration. Hypoglycemia was associated with significantly higher mortality among patients who were not treated with insulin vs those without hypoglycemia (odds ratio [OR], 2.32; 95% confidence interval [CI], 1.31-4.12). However, among those patients who were treated with insulin, subsequent hypoglycemia was not associated with increased mortality risk vs those without hypoglycemia (OR, 0.92; 95% CI, 0.58-1.45; adjusted P = .01 for hypoglycemia × insulin interaction; Figure).
Exclusion of patients who were treated with oral antihyperglycemic agents during hospitalization did not alter the study findings. For patients who were not treated with insulin, the adjusted OR for hypoglycemia vs no hypoglycemia was 2.65 (95% CI, 1.48-4.79). For patients treated with insulin, the adjusted OR for hypoglycemia vs no hypoglycemia was 0.88 (95% CI, 0.55-1.44; adjusted P = .004 for hypoglycemia × insulin interaction). Excluding patients who died within 24 hours of admission, and using a different definition of hypoglycemia (glucose level <70 mg/dL) did not alter the study results (Table 3 and Figure).
The interaction test between diabetes and hypoglycemia was not significant (P = .21), suggesting that the relationship between hypoglycemia and mortality was similar in patients with and without known diabetes.
We observed more than 2 times greater risk of mortality associated with spontaneous hypoglycemia (OR, 2.32 [95% CI, 1.31-4.12] vs OR, 0.92 [95% CI, 0.58-1.45] among patients with iatrogenic hypoglycemia, P = .01). Based on the sample size and variation in outcomes observed in this study, we would have had more than 98% power to detect even a 50% greater risk associated with spontaneous vs iatrogenic hypoglycemia. Therefore, our study was well powered to demonstrate the interaction between hypoglycemia and treatment with insulin and their effect on mortality.
In this large, contemporary cohort of patients hospitalized with AMI, we found that while hypoglycemia was associated with increased mortality, this risk was confined to those who developed hypoglycemia spontaneously. In contrast, hypoglycemia that occurred after initiation of insulin therapy was not associated with higher mortality. These data suggest that hypoglycemia during hospitalization for AMI is a marker for more critical illness, rather than a direct cause of adverse outcomes and should provide some reassurance to clinicians in their efforts to manage glucose level after AMI.
Several prior observational studies have shown that hypoglycemia during hospitalization for AMI is associated with higher mortality risk.22,26,36 Because hypoglycemia occurs more commonly with the use of intensive glucose-control protocols (which almost universally use insulin),32-35 these findings have generated substantial concern, and hampered enthusiasm in regard to insulin-mediated glucose lowering among patients with AMI and hyperglycemia. However, these studies included patients with low and normal glucose levels at hospital admission, who would not normally be considered as candidates for glucose-lowering therapy.22,26,36 More importantly, the studies could not distinguish between patients who developed hypoglycemia spontaneously (likely as a result of severe systemic illness; eg, shock, sepsis, liver or multiorgan failure, malnutrition, or adrenal dysfunction), and those who developed it as a consequence of glucose-lowering therapy. These 2 etiologies of hypoglycemia may have drastically different pathophysiology and consequences, but the critical question of whether they carry different prognostic implications was not addressed until now.
By evaluating the largest patient sample to date that addresses the issue of hypoglycemia in AMI (both in terms of overall patient numbers as well as hypoglycemic events), focusing on patients who are potential candidates for strict glucose control, and analyzing the prognosis associated with hypoglycemic events separately in patients who were and were not treated with insulin, our study adds substantially to the prior findings and extends the current understanding of the relationship between hypoglycemia and mortality in AMI. Our results also support the findings of the recent report from the Diabetes Mellitus Insulin-Glucose Infusion in Acute Myocardial Infarction 2 (DIGAMI-2) study, in which episodic hypoglycemia that occurred as a consequence of insulin therapy in patients with AMI and known diabetes was not associated with increased mortality risk.41
Our findings have implications for the field of glucose control among AMI patients with hyperglycemia. Although the impact of glucose lowering on patient outcomes, as well as specific thresholds for insulin initiation and glucose treatment target levels remain subjects of debate, based on the information from previous observational studies1-27 and limited clinical trial data,28 the clinical practice guidelines and position documents of the American College of Cardiology and the American Heart Association currently recommend glucose control in patients with AMI and significant hyperglycemia.29-31 Insulin administration continues to be the most effective means of glucose control in the acute hospitalization setting.31 Even the best insulin protocols are likely to produce some hypoglycemic events.42 While rigorous efforts to avoid hypoglycemia are certainly warranted, our findings provide some degree of reassurance to clinicians that episodic hypoglycemic events, which occur in a setting of glucose control with insulin, do not appear to be associated with increased mortality risk.
The most likely explanation for our findings is that hypoglycemia is a marker, rather than a direct mediator of higher mortality in the setting of AMI. Certainly, patients who develop hypoglycemia spontaneously appear to have greater illness severity and comorbidity (as was seen in our study), and their hypoglycemic events are likely to be the result of severe concomitant conditions. Some of these factors may not be adequately captured in retrospective observational studies, and therefore could confound the previously reported association between hypoglycemia and higher mortality risk. The fact that hypoglycemic events were not associated with higher mortality risk in insulin-treated patients (despite a much higher rate of hypoglycemic events in this group) suggests that hypoglycemia itself is not a direct cause of adverse events following AMI.
An alternative explanation, however, is also possible. In our study, mean glucose levels during hospitalization were overall similar between patients who did and did not develop hypoglycemia in the absence of insulin therapy. However, in the insulin-treated group those who experienced hypoglycemia had a significantly lower mean glucose level during hospitalization than insulin-treated patients without hypoglycemia (probably as a result of more aggressive insulin therapy). The association between lower mean glucose level and better survival has been documented in prior studies.26,27 While our findings do not necessarily suggest that better glucose control with insulin is beneficial in a setting of AMI, it is possible that any detrimental effect of hypoglycemia (eg, higher levels of catecholamines)37 could be offset by a potential positive effect of improved glucose control in the insulin-treated patients—with these 2 competing processes producing a net neutral effect on mortality. This hypothesis is supported by data from the recent meta-analysis of intensive glucose-control clinical trials among critically ill patients, which observed a markedly higher rate of hypoglycemia in the intensive glucose-control group compared with usual care, but no significant difference in mortality.32 Therefore, vigilant efforts on the part of clinicians to minimize hypoglycemia (by using glucose-control protocols with known safety records) are certainly warranted.
The results of our study need to be interpreted in the context of several potential limitations. First, given the retrospective nature of the analyses, a possibility of residual confounding cannot be entirely excluded. While we rigorously attempted to account for baseline differences between patients with and without hypoglycemia in the insulin-treated and non–insulin-treated groups, and performed numerous sensitivity analyses, residual unmeasured differences may persist. Specifically, we were not able to control for certain clinical variables, such as the presence of ST-segment elevations. However, adjustment for the presence of ST-segment elevations did not significantly affect the prognostic effect of glucose levels in our prior studies,1 and we were able to control for other measures of infarct severity, including peak troponin level and multiple other clinical factors. Information on the duration of diabetes, body mass index, and on whether hypoglycemia was symptomatic also was not available, but is not likely to have affected our findings.
Second, the number of patients who experienced hypoglycemia was relatively small. Nevertheless, our study represents the largest one to date (both in terms of our patient sample and the number of hypoglycemic events) that specifically addresses the relationship between hypoglycemia and mortality in AMI. Moreover, the differences in the relationship of hypoglycemia and mortality between patients who were and were not treated with insulin were marked, and highly statistically significant. Third, we had limited data regarding the dose of insulin administered. However, given a paucity of firmly established protocols guiding insulin administration in patients with AMI and hyperglycemia, our study reflects routine clinical care. Fourth, the specific cause of hypoglycemia in patients who were not treated with insulin could not be definitively established. Finally, due to limited follow-up, we could not assess the effect of hypoglycemia on long-term outcomes; however, we believe that it is highly unlikely that an episode of insulin-induced in-hospital hypoglycemia would directly affect patient outcomes beyond the hospitalization period.
In conclusion, our findings demonstrate that while hypoglycemia is associated with increased short-term mortality in patients hospitalized with AMI, this risk is confined to patients who develop hypoglycemia spontaneously. In contrast, hypoglycemia that occurs after initiation of insulin therapy is not associated with higher mortality risk. While continuous efforts to avoid hypoglycemia are warranted, these data suggest that hypoglycemia is a marker of severe illness, rather than a direct cause of adverse outcomes.
Corresponding Author: Mikhail Kosiborod, MD, Mid America Heart Institute of Saint Luke's Hospital, 4401 Wornall Rd, Kansas City, MO 64111 (firstname.lastname@example.org).
Author Contributions: Drs Kosiborod and Spertus had full access to all of the data in the study and take responsibility for the integrity of the data and the accuracy of the data analysis.
Study concept and design: Kosiborod, Krumholz, Spertus.
Acquisition of data: Kosiborod, Spertus.
Analysis and interpretation of data: Kosiborod, Inzucchi, Goyal, Krumholz, Masoudi, Xiao, Spertus.
Drafting of the manuscript: Kosiborod.
Critical revision of the manuscript for important intellectual content: Kosiborod, Inzucchi, Goyal, Krumholz, Masoudi, Xiao, Spertus.
Statistical analysis: Kosiborod, Xiao.
Obtained funding: Kosiborod, Spertus.
Administrative, technical, or material support: Kosiborod, Spertus.
Study supervision: Kosiborod, Spertus.
Financial Disclosures: Dr Kosiborod reported that he has served on the advisory board of Sanofi-Aventis and has received speaking honoraria from the Vascular Biology Working Group and DiaVed, Inc. Dr Inzucchi reported that he received research grant support from Lilly. Dr Krumholz reported being a consultant or a subject matter expert for VHA Inc, having research contracts with the American College of Cardiology and the Centers for Medicare & Medicaid Services, being on an advisory board for UnitedHealthcare, Alere, and Amgen, providing testimony on behalf of plaintiffs in the Vioxx trial, serving as editor-in-chief of Circulation: Cardiovascular Quality and Outcomes and Journal Watch Cardiology of the Massachusetts Medical Society, and participating in early phase translational research activities for Centegen/LifeTech. Dr Masoudi reported that he has been a member of the advisory board for Takeda Pharmaceuticals. Dr Spertus reported that he receives research grant support from Sanofi-Aventis and Lilly. Drs Goyal and Xiao did not report any financial disclosures.
Funding/Support: The research for this article was supported by the American Heart Association Career Development Award in Implementation Research awarded to Dr Kosiborod and by the Cerner Corporation.
Role of Sponsors: The Cerner Corporation facilitated the collection of the deidentified Health Facts data and reviewed the manuscript, but had no role in the design or conduct of the study; the management, analysis, or interpretation of the data; or the preparation or approval of the manuscript. The American Heart Association had no role in the study.