Åke Tenerz, Centre for Clinical Research, Central Hospital, SE-721 89 Västerås, Sweden (fax: +46 21 17 37 33; e-mail: firstname.lastname@example.org).
Abstract. Tenerz Å, Nilsson G, Forberg R, Öhrvik J, Malmberg K, Berne C, Leppert J (Central Hospital, Västerås; Karolinska Hospital, Stockholm; and University Hospital, Uppsala; Sweden) Basal glucometabolic status has an impact on long-term prognosis following an acute myocardial infarction in non-diabetic patients. J Intern Med 2003; 254: 494–503.
Objectives. Patients with diabetes are known to have a worse prognosis after an acute myocardial infarction (AMI) compared with non-diabetic patients. The primary aim of this study was to investigate the effect of glucometabolic status on long-term prognosis in non-diabetic patients with an AMI. The second aim was to evaluate the extent to which blood glucose levels at admission depended on acute stress, assessed as serum cortisol, previous glucometabolic status, measured as haemoglobin A1c (HbA1c), or both.
Design. In a prospective study of patients with an AMI, blood glucose, HbA1c and cortisol were measured at admission. Fasting blood glucose was determined before discharge and also afterwards, if necessary, for classification. Patients were followed-up for 5.5 years.
Subjects. Of the 305 consecutive patients 24% were diagnosed as diabetic and 76% as non-diabetic.
Main outcome measures. Death or non-fatal myocardial re-infarction.
Results. In non-diabetic patients, a Cox regression model was used. With death or re-infarction as endpoint, the following prognostic factors had an impact on event-free survival: age (P < 0.001), HbA1c (P = 0.002), cortisol (P < 0.001) and thrombolytic treatment (P = 0.001). There was a correlation between cortisol and blood glucose at admission (r = 0.44, P < 0.001). Fasting blood glucose day 5 showed no association with event-free survival.
Conclusions. In non-diabetic patients with AMI, admission HbA1c and cortisol were predictors for 5.5-year survival without recurrent non-fatal myocardial infarction. The glucometabolic status of importance for prognosis was detected by HbA1c but not by fasting blood glucose or admission blood glucose, of which the latter was influenced by cortisol.
Elevated blood glucose at admission for an acute myocardial infarction (AMI) is associated with an increased risk of death in patients both with and without diabetes mellitus [1, 2]. Patients with diabetes have a worse prognosis after an AMI [3, 4] and a generally higher cardiovascular risk  compared with non-diabetic patients. Elevated haemoglobin A1c (HbA1c) has been shown to be a risk marker for 1-year mortality following AMI in non-diabetic patients . In the general population the relationship between plasma glucose concentration and increased risk of cardiovascular disease extends well below the glucose concentrations, which define both diabetes and impaired glucose tolerance . Similarly, in the general population, HbA1c is continuously related to cardiovascular mortality even in non-diabetic subjects .
In Sweden, 20–25% of patients with an AMI have known diabetes mellitus [9, 10] but a recent report using an oral glucose tolerance test (OGTT) for the diagnosis of diabetes shows that the true prevalence may be as high as 40–45% .
Stress hormones like cortisol and catecholamines are amongst the determinants of plasma glucose concentration at admission for an AMI [12–14]. Plasma cortisol rises early in response to AMI and falls within 28–72 h [13, 14]. In most studies [15, 16], but not in all , plasma cortisol is raised in proportion to the size of the infarction. The cortisol response to AMI is independent of age [18, 19]. In patients with AMI, an increased casual blood glucose at admission is not reliable for the diagnosis of diabetes mellitus, and a follow-up is necessary to establish the diagnosis . Consequently, blood glucose at admission for AMI may not accurately estimate the true prevalence of diabetes mellitus. HbA1c, as a measure of long-term glucose status, also has a limited value in discriminating diabetic from non-diabetic subjects [10, 20, 21]. The primary aim was to examine whether glucometabolic disturbances at admission could predict long-term prognosis after an AMI, in patients regarded as non-diabetic. The second aim was to evaluate the extent to which blood glucose levels at admission depended on acute stress, assessed as serum cortisol, previous glucometabolic status, measured as HbA1c, or both.
Material and methods
The Central Hospital in Västerås, located 120 km north-west of Stockholm, Sweden, serves a population of approximately 150 000 inhabitants. All patients admitted to the Central Hospital with suspected AMI are admitted to the Coronary Care Unit. During 1 year (October 1995–September 1996), 337 patients fulfilled the diagnostic criteria for AMI. Of these 337 patients, 13 were excluded because of their unwillingness to participate in the study. Nineteen patients, who had more than one AMI during the inclusion period, were only included at the first event.
The details of the protocol of the investigation have previously been described . In short, casual blood glucose was measured at admission, fasting blood glucose during hospital stay and, if necessary for classification of diabetes mellitus, 2–3 months after discharge. If at least one of two fasting blood glucose values at follow-up was 5.5–6.0 mmol L−1 a 75-g OGTT test was performed. After follow-up, 269 (88%) of the remaining 305 patients were classified as I (previously known diabetics), II (newly detected diabetics according to the World Health Organisation (WHO) criteria, 1998)  or as III (patients without diabetes). Thirty-six patients (12%) remained unclassified (IV) as follow-up was not possible, mainly because of early deaths. HbA1c values showed considerable overlap between the three groups of classified patients (I–III).
Thus, the present analysis was founded on 305 patients with AMI and amongst them 74 patients (24%) classified after follow-up as newly diagnosed or previously known diabetes (I + II), and 231 patients (76%) without known diabetes (III + IV) , denoted as non-diabetic patients (Fig. 1).
Samples for casual blood glucose, serum cortisol, HbA1c, total cholesterol and high-density lipoprotein (HDL) cholesterol were taken at admission and fasting blood glucose was also measured on the second and fifth day after admission. A specially trained nurse collected all medical data from each patient's medical records. Laboratory examinations were carried out according to the laboratory routines of the Central Hospital in Västerås, Sweden. There were some dropouts of blood samples for serum cortisol, HbA1c and particularly lipids at admission, partly because of early death but also depending on a nurse strike during the inclusion period.
Blood glucose. Venous sampling was carried out using tubes containing sodium fluoride and sodium heparin as additives. After treatment with a haemolysis reagent (Merck Diagnostica, Darmstedt, Germany) glucose was determined with glucose dehydrogenase on a Cobas Mira analyser (Roche Diagnostica, Basel, Switzerland). Blood samples were analysed immediately on arrival at the laboratory.
Serum cortisol. Cortisol concentration was measured with a radioimmunoassay, Cortisol 125 I reagent kit (Orion Diagnostica, Espoo, Finland). The reference ranges were, 250–750 nmol L−1 at 08.00 hours and 50–300 nmol L−1 at 22.00 hours.
Haemoglobin A1c. Haemoglobin A1c was measured with high-performance liquid chromatography using Mono S ion-exchange resin (Beckman System Gold; Beckman Instruments Inc., Fullerton, CA, USA). The reference range was 3.5–5.2%. Within-run precision expressed as coefficient of variation for 20 replicates at HbA1c levels of 4.5 and 8.3% was 0.4 and 0.3%, respectively. Total imprecision measured during a 5-month period was 2.0% at both HbA1c levels.
Creatine kinase-B. Creatine kinase-B (CK-B) was determined by an immunoinhibition method (Merck Diagnostica) on a Cobas Mira analyser (Roche Diagnostica, Basel, Switzerland). The method automatically compensated for adenylate kinase activity.
Total cholesterol and high-density lipoprotein cholesterol. These were measured by automatic enzymatic procedures on a Cobas Mira analyser (Roche Diagnostica). The reagents used were Monotest Cholesterol CHOD-PAP (no. 1442350) and HDL cholesterol precipitating reagent (no. 543004) from Boehringer, Mannheim, Germany. The total cholesterol to HDL cholesterol ratio was used in the analyses.
Mortality rate and cause of death. These were obtained from death certificates for the period October 1995 to September 2001 at the Epidemiological Centre, National Board of Health and Welfare.
Re-hospitalization for a non-fatal re-infarction. The data were taken from registers of hospital care in the County of Västmanland. Two patients who had moved from the area were contacted by telephone.
Event-free survival. This was calculated as time from inclusion with an AMI to re-hospitalization for a non-fatal re-infarction or death, the first to occur. The patients were followed until September 2001, i.e. for 5.5 years.
Diabetes mellitus. Manifest diabetes mellitus was established if the patient had been informed of the diagnosis by a physician before the admission or was undergoing treatment (diet, oral anti-diabetic agents or insulin). Diabetes mellitus was defined according to the WHO criteria, 1998 as fasting venous or capillary blood glucose ≥6.1 mmol L−1 on two occasions or, if performed, 75-g OGTT with a 2 h capillary blood glucose ≥11.1 mmol L−1 .
Non-diabetic. Non-diabetic was defined as a patient without known diabetes or an unclassified patient according to the protocol described above.
Myocardial infarction. A diagnosis of myocardial infarction was established if at least two of the following criteria were fulfilled: chest pain ≥15 min; serum CK-B value above the normal range 10–16 h after onset of symptoms; and development of new Q-waves or persistent ST-elevations typical of myocardial infarction in at least two of the 12 standard electrocardiography leads.
Myocardial re-infarction. Re-infarction was defined as a new hospitalization in the County of Västmanland for an AMI during the follow-up period.
Stroke. Stroke was defined as a hospitalization in the County of Västmanland for intra-cerebral infarction or intra-cerebral bleeding during follow-up period.
The main statistical comparisons were performed between patients with diabetes mellitus and non-diabetic patients and between patients who survived without events (death or non-fatal re-infarction) and those who did not.
All results are expressed as mean values and confidence intervals (CI) except for cortisol for which medians are given. Variables with skewly distribution [admission blood glucose, blood glucose day 5, serum cortisol, maximum CK-B (max. CK-B) and HDL cholesterol] were log-transformed before analysis.
Several baseline variables were compared. Differences between groups were tested with Student's t-test or chi-square. Association between variables was assessed using Pearson's correlation. Association between endpoint outcome and HbA1c level in Fig. 6 was assessed by using Kendall's τB.
Kaplan–Meier survival estimates were calculated for categorized tertiles of HbA1c and serum cortisol at admission and fasting blood glucose day 5. The generalized Wilcoxon test, which gives higher weights to early events, was used to test for differences between tertile groups. A two-tailed P-value <0.05 was considered statistically significant.
The main analysis used a Cox regression model. Both forward stepwise selection with inclusion of variables at the 5% level and removal at the 10% level in each step and backward stepwise elimination with the same levels were run to assure that the same model was selected with both methods. The following variables at admission were candidate covariates: blood glucose, serum cortisol, HbA1c, total cholesterol to HDL cholesterol ratio, max. CK-B, thrombolytic treatment, treatment for hypertension, previous myocardial infarction, smoking, sex and age. To check the proportional hazard assumption (PH-assumption) variables on a continuous scale were categorized into tertiles. Then the logarithm of minus the logarithm of the Kaplan–Meyer survival estimates in the different categories were plotted against time to check that they were reasonably parallel (PH-assumption). For each variable with a significant impact on event-free survival, the PH-assumption was also checked by including the interaction between the variable in question and time into the model. This test was originally proposed by Cox. If the coefficient for the interaction term differs significantly from zero, it indicates a violation of the PH-assumption. To check that there were no important interactions between factors, all pairwise interactions between the significant factors were tested for entering the model. Only variables with a significant effect on the event-free survival at 5% level were included in the model. Pseudo R2 value was calculated to estimate the extent to which the different parameters explain the variability in event-free survival .
The study complies with the Declaration of Helsinki and was approved by the Human Ethics Committee at Uppsala University, Sweden, and written consent has been obtained from the subjects.
Patient characteristics are given in Table 1. Previous myocardial infarction was more common and mortality rate was higher amongst patients with diabetes. Treatment at admission was for non-diabetic and diabetic patients, respectively; betablockers 27%/47%; aspirin 24%/47%; and angiotensin-converting enzyme inhibitors 10%/23%. Fasting blood glucose in non-diabetic as well as in diabetic patients was significantly higher day 2 compared with day 5 (P < 0.01 and P < 0.001, respectively).
Cardiovascular disease (ischaemic heart disease, stroke or peripheral artery disease) was the main cause of death in 84% of the patients with no significant difference between patients with or without diabetes.
Amongst patients with diabetes 11% were hospitalized for stroke during 5.5-year follow-up compared with 6% in the non-diabetic group, a non-significant difference.
Amongst the 231 non-diabetic patients, 36 could not be classified. The latter had a 5.5-year mortality rate of 86%, a mean age of 75.2 years (95% CI = 71.3–79.1), a mean casual blood glucose at admission of 9.0 mmol L−1 (95% CI = 8.1–9.8), a mean HbA1c of 5.0% (95% CI = 4.8–5.3) and a median serum cortisol at admission of 972 nmol L−1.
Cortisol and glucose parameters. Serum cortisol at admission had lost the normal circadian rhythm with a nadir during the night (Fig. 2). Serum cortisol and casual blood glucose at admission were positively correlated (r = 0.44, P < 0.001; Fig. 3a), in contrast to serum cortisol and fasting blood glucose day 5. Serum cortisol was also correlated positively with the max. CK-B (r = 0.34, P < 0.001; Fig. 3b).
No correlation was found between admission blood glucose and HbA1c. A weak positive correlation was found between admission blood glucose and fasting blood glucose day 5 (r = 0.15, P < 0.05). Finally there was a positive correlation between fasting blood glucose day 5 and HbA1c (r = 0.25, P < 0.01). The results were similar if the 36 patients who could not be classified were excluded from the analysis.
Event-free survival. During the 5.5 years of follow-up period, 38% of the patients died, 15% were re-admitted for non-fatal re-infarction; totally 45% had at least one of those two events.
Kaplan–Meier curves for event-free survival in the three tertiles of HbA1c showed that the prognosis was worse in the highest tertile of HbA1c (P < 0.01; Fig. 4a). High age and high HbA1c were associated with higher risk of death or non-fatal myocardial infarction during the 5.5-year follow-up (Fig. 5).
There was no correlation between tertiles of HbA1c and thrombolytics and no trend for drug treatment at discharge (Table 2).
Table 2. Treatment during hospital stay and at discharge for patients without diabetes divided into tertiles of haemoglobin A1c
HbA1c at admission (n = 164)
Calcium channal inhibitors
Lipid lowering agents
Kaplan–Meier curves for event-free survival in the three tertiles of serum cortisol showed that high serum cortisol was associated with poor event-free survival (P < 0.01, Fig. 4b). However, fasting blood glucose day 5, divided into tertiles, was not associated with event-free survival. If the 36 patients who could not be classified were excluded from analysis, the results were still significant for HbA1c but not for serum cortisol.
The result from the Cox regression showed that age (P < 0.001), HbA1c (P = 0.002) and serum cortisol (P < 0.001) had a negative effect, and thrombolytic treatment (P = 0.001) had positive effect on 5.5 years event-free survival. If the 36 patients who could not be classified were excluded from analysis the results were similar. The results were the same using either forward or backward elimination.
None of the pairwise interactions between the above main factors were significant. In testing the PH-assumption, the time serum cortisol and the time thrombolytic interactions were significant (P = 0.02 and P = 0.01, respectively) indicating a violation of the PH-assumption. As there were only small changes in the parameter estimates of the other factors when the interaction was included, the effect of the violation was minor and it was decided not to complicate the model with time-varying covariates.
According to pseudo R2 analyses, age, HbA1c, serum cortisol and thrombolytic treatment, altogether explains 31% of the total variability in event-free survival. Age, HbA1c, serum cortisol and thrombolytic treatment individually explains 19, 6, 5 and 6%, respectively.
If non-fatal stroke was added to death and non-fatal AMI as endpoint for event-free survival, the results were similar. Also using only death as endpoint gave similar results.
Diabetic patients: cortisol and glucose parameters
Casual blood glucose at admission and HbA1c showed a highly positive correlation (r = 0.60, P < 0.001). Serum cortisol at admission and max. CK-B showed a positive correlation (r = 0.47, P < 0.01) but no significant correlation was found between serum cortisol and casual blood glucose at admission (r = 0.28, P = 0.07).
All patients: death or non-fatal myocardial re-infarction
For all patients, there was a positive correlation between death or non-fatal myocardial re-infarction during 5.5 years and HbA1c level (Kendall τB = 0.195, P = 0.001; Fig. 6).
The main result from this prospective study was that the glucometabolic state measured by HbA1c was a major determinant of long-term prognosis in non-diabetic patients following an AMI. This study also confirmed the higher mortality rate in patients with diabetes [3, 4].
Amongst non-diabetic patients 5.5 years prognosis were not only related to age, as expected, but also to admission HbA1c, serum cortisol and thrombolytic treatment. HbA1c was a risk factor for myocardial re-infarction or death after 5.5 years amongst patients without known diabetes even if unclassified patients were excluded. In addition to the expected relationship between age and survival, glucometabolic status in this study was related to event-free survival which was not shown for traditional risk factors such as total cholesterol to HDL cholesterol ratio, hypertension, smoking and previous myocardial infarction. In the European Prospective Investigation of Cancer and Nutrition (EPIC) study, HbA1c was continuously related to cardiovascular mortality through the whole population distribution . This observation, also in the non-diabetic population, is in agreement with the findings in patients with myocardial infarction. Furthermore, this data is supported by a previous report showing that HbA1c was a risk factor for 1-year mortality following AMI in non-diabetic patients .
Age and thrombolytic treatment are important confounders, which influence the prognostic effect of HbA1c and serum cortisol. Unfortunately, no data are available on left ventricular function and heart failure during hospital stay. However, patients without diabetes and with HbA1c in the upper tertile have no medication expressing more severe morbidity.
In patients who were non-diabetic according to fasting blood glucose criterion, the glucometabolic disturbances, detected as an elevated HbA1c, were related to shorter event-free survival but this was not demostrated for fasting blood glucose day 5. In harmony with this observation, the DECODE study shows that fasting glucose concentrations alone do not identify individuals at increased risk of death associated with hyperglycaemia . Furthermore, fasting glucose may not be a reliable diagnostic test for diabetes, as there is a disagreement between the WHO criteria for diabetes mellitus based on OGTT  and the American Diabetic Association (ADA) criteria using only fasting blood glucose . Of patients diagnosed with diabetes according to 2-h blood glucose, 31% have normal fasting glucose concentrations and 20% have impaired fasting glucose . Also, in patients with an AMI, the prevalence of diabetes is almost doubled when an OGTT is performed .
In contrast to a previous study from Sweden , admission blood glucose in non-diabetic patients was not significantly related to long-term event-free survival, which may have been due to the lack of power to find such an association. A positive correlation between admission blood glucose and serum cortisol, which in turn was an independent predictor for event-free survival, could provide indirect evidence for such a relationship, which was undetected in the present study.
Blood glucose at admission was not correlated with long-term glycaemia, measured as HbA1c, but with the acute stress response induced by the myocardial infarction and assessed as serum cortisol . On the fifth day in hospital, the relationship between blood glucose and admission serum cortisol was lost, indicating that in the majority of patients the stress response has subsided at discharge from hospital. In contrast to the non-diabetic group, patients with diabetes showed the expected strong association between blood glucose at admission and HbA1c, whereas serum cortisol was less important as a determinant of the admission blood glucose, although there were no differences in serum cortisol levels between diabetic and non-diabetic patients. The total number of diabetic patients was low and a Cox regression analysis for prognosis was not possible. However, taking all patients together there is, as expected , a relationship between HbA1c level and long-term event-free survival. The weak positive correlation observed between the size of infarction determined from max. CK-B activity and serum cortisol at admission has possibly been diminished by thrombolytic treatment which, if successful, reduces myocardial damage and thereby max. CK-B. The admission serum cortisol which reflects the metabolic stress [15, 16] may thereby lose its relationship to the infarction size .
The normal circadian rhythm of cortisol, with the lowest levels between 22.00 and 04.00 hours  was lost in the acute phase of a myocardial infarction indicating that the stress imposed by the AMI overrides the diurnal rhythm of serum cortisol. In the Framingham study age, total cholesterol to HDL cholesterol ratio, and diabetes were risk factors for further manifestations of coronary heart disease or stroke in males, whereas for women systolic blood pressure and smoking were additional risk factors . In this study, total cholesterol to HDL cholesterol ratio at admission was not a predictor for prognosis which indicates that glucometabolic status is more important than serum lipids in non-diabetic patients. One reservation for this conclusion was the high number of missing lipid values.
Treatment for hypertension at admission did not emerge as an independent predictor of prognosis after an AMI in this study, in accordance with the EPIC study  which showed that the predictive value of HbA1c for total mortality was stronger than serum cholesterol and hypertension. Finally, smoking habits at admission were not a predictor of long-term prognosis. Thus, the importance of HbA1c for event-free survival seems to outweigh the traditional risk factors cholesterol, hypertension and smoking habits.
The glucometabolic abnormalities that is found to be of importance for the prognosis after an AMI in patients regarded as non-diabetics were detected by HbA1c but not by fasting blood glucose or admission blood glucose. A single HbA1c is an unreliable diagnostic test for diabetes mellitus [10, 11, 21]. Therefore, it would probably be appropriate to perform an OGTT after an AMI in order to identify patients with impaired glucose metabolism and the highest risk for cardiovascular death [7, 30]. Moreover, an OGTT is necessary to identify those with manifest diabetes mellitus, who consequently carry an increased risk of long-term microvascular complications.
In summary, the glucometabolic state measured by HbA1c is a major risk factor for prognosis in non-diabetic patients following an AMI.
Conflict of interest statement
K. Malmberg is a Medical Adviser for Glaxo SmithKline, Sweden, but affiliated to the Department of Cardiology at Karolinska Hospital as associate professor. There exists no conflict of interest for the other authors. This work was supported by grants from Centre for Clinical Research, Central Hospital, Västerås, Uppsala University, and the research foundation of Västmanland County Council.
We would like to acknowledge research assistants Petra Wahlèn and Kent Nilsson for data management and secretary Katarina Ringström for help with the manuscript, all at Centre for Clinical Research, Västerås, Uppsala University.