OUP user menu

The relationship between glycaemic variability and cardiovascular complications in patients with acute myocardial infarction and type 2 diabetes: a report from the DIGAMI 2 trial

Linda G. Mellbin, Klas Malmberg, Lars Rydén, Hans Wedel, Daniel Vestberg, Marcus Lind
DOI: http://dx.doi.org/10.1093/eurheartj/ehs384 374-379 First published online: 9 November 2012

Abstract

Aims Hyperglycaemia during hospitalization for acute myocardial infarction (AMI) is a risk predictor, but attempts to improve the prognosis by insulin-based glucose control have not been consistently successful. Increased glycaemic variability, a potential effect of insulin treatment, has been linked to a worse prognosis in critically ill patients. The present aim was to study the possibility of such a relation in patients with type 2 diabetes (T2DM) and AMI.

Method and results We studied 578 T2DM patients who had glucose levels measured hourly while receiving an insulin–glucose infusion during the first 48 h of hospitalization for AMI. Three measures of glycaemic variability: root mean square error (RMSE), range, and slope were studied in relation to a composite endpoint of mortality, stroke, and reinfarction and to mortality.

In unadjusted analyses, the mean level of glycaemic variability did not differ between patients who died during 12 months of follow-up compared with those who survived. In a Cox regression model adjusting for age and previous congestive heart failure, there was no increased risk for the composite endpoint associated with increased glycaemic variability; RMSE: hazard ratio (HR) 1.09 [95% confidence interval (CI) 0.93–1.27; P = 0.28], range: HR 1.01 (95% CI: 0.98–1.05; P = 0.47), and slope: HR 1.01 (95% CI: 0.99–1.04; P = 0.40). There was furthermore no increased risk in mortality; RMSE HR 1.14 (95% CI: 0.93–1.38; P = 0.21), range HR 1.03 (95% CI: 0.98–1.08; P = 0.28), and slope HR 1.01 (95% CI: 0.98–1.04; P = 0.55).

Conclusion The 1-year risk for death, reinfarction, or stroke did not relate to glycaemic variability in T2DM patients with AMI treated with insulin infusion.

  • Glucose variability
  • Diabetes mellitus Type 2
  • Myocardial infarction
  • Prognosis

Introduction

Diabetes mellitus is associated with an increased risk of micro- and macrovascular complications and an approximate two-fold greater risk of mortality compared with the general population.1 The prognosis following acute myocardial infarction (AMI) is worse among patients with than those without diabetes with greater rates of in- and post-hospital mortality.2,3 The impaired prognosis may, at least partially, be related to hyperglycaemia.

High glucose at admission and during hospitalization for an AMI is prognostically unfavourable46 and attempts have been made to improve the dismal outcome by means of intensive glucose control. The landmark trial in this context, the first Diabetes Mellitus Insulin-Glucose Infusion in Acute Myocardial Infarction (DIGAMI), showed a 11% absolute reduction in mortality after 3.4 years for patients randomly assigned to receive intensive insulin-based treatment in conjunction with and after an AMI compared with patients in the control group.7,8 A beneficial effect was, however, not confirmed in subsequent trials studying the concept of metabolic control by means of insulin,9,10 possibly due to a lack of difference in glucose control between treatment arms in the latter studies.

In addition to the negative impact of high blood glucose per se, it has been argued that glucose fluctuations may be deleterious as well.11,12 The latter relation has primarily been studied in outpatients with diabetes1319 and in critically ill patients admitted to intensive care units (ICU).2023 There are, however, indications that glycaemic variability may be less important in hospitalized patients with diabetes than in those without.21,24

The present investigation, an epidemiological report from the DIGAMI 2 trial, explores the prognostic implication of glycaemic variability during the first 48 h of hospitalization for an AMI in patients with type 2 diabetes.

Methods

Data source

DIGAMI 2, a prospective, randomized open trial with blinded evaluation, compared three different glucose management strategies in 1253 patients with type 2 diabetes and hospitalization due to a suspected AMI.9 In total 84% fulfilled the diagnosis of AMI while the remaining patients had coronary artery disease, mostly unstable angina pectoris. Patients were randomized to one of three study arms treated as follows: Group 1 received insulin–glucose infusion7 with the intention to decrease blood glucose as quickly as possible to a level of 7–10 mmol/L. The infusion lasted for at least 24 h or until the glucose levels were stabilized, and was thereafter followed by subcutaneous insulin-based glucose control, during the remaining study period, targeting a fasting glucose of 5–7 mmol/L and a postprandial level <10 mmol/L. Group 2 received the same initial treatment as Group 1, followed by glucose-lowering treatment according to local practice without protocol stated target glucose. Group 3 received glucose-lowering treatment according to local practice. The patients were followed during a median of 2.1 (inter-quartile range 1.03–3.00) years.

Blood glucose was obtained at randomization, i.e. as soon as possible after hospital admission. In patients in Groups 1 and 2, glucose was followed frequently during the first 24–48 h according to the infusion protocol. Glucose should be checked 1 h after the start of the infusion and the infusion rate was adjusted according to the protocol aiming for a glucose level of 7–10 mmol/L. A new check was prescribed 1 h after a change in the infusion rate and at least every second hour. Subsequently, fasting glucose was recorded daily until hospital discharge and at each follow-up. Glucose sampling in Group 3 was left to the discretion of the attending physician. Glucose is reported as the locally analysed value in whole blood expressed in mmol/L.

Participants

The present study comprises 578 patients who received insulin–glucose during the acute phase regardless of randomization group, had frequent glucose measurements performed, and who survived the first 48 h. These patients were distributed as follows: Group 1: 274 patients (47.4%); Group 2: 282 patients (48.8%); and Group 3, 22 patients (3.8%).

Measures of glucose variability and endpoints

Three measures of glycaemic variability, based on glucose values obtained during the initial 48 h, were applied: (i) the root mean square error (RMSE) for the model; (ii) the range of all values within the time frame of 48 h; and (iii) the best fitted regression line (slope/24 h). Each metric was studied in relation to two endpoints after 12 months of follow-up: (i) the primary endpoint which was a composite of total death and non-fatal MI or stroke, (ii) all-cause mortality. In addition, these endpoints were analysed after 3 and 18 months of follow-up. Furthermore, the impact of glucose at randomization and the mean glucose during the initial 48 h was assessed. All events were previously adjudicated by an independent committee composed of three experienced cardiologists.9

Statistical methods

Data are presented as n (%) for categorical and mean (SD) for continuous variables unless otherwise stated. The time of follow-up lasted from the time of randomization until a maximum of 18 months. Blood glucose at randomization, mean glucose during the initial 48 h, and the different measures of glucose variability were related to the endpoints with the Cox proportional hazards model as the basis. The results are presented as the hazard ratio (HR) and the 95% confidence interval (CI). The model was adjusted for age and previous congestive heart failure at the time of hospital admission. These covariates were selected as those remaining significant in a stepwise procedure, which initially included variables selected based on findings in previous reports from the DIGAMI 2 trial9,25,26: gender, age, smoking habits, previous MI or previous congestive heart failure recorded at the time of hospital admission, creatinine at randomization, and percutaneous transluminal coronary angioplasty or coronary artery bypass grafting during hospitalization. In a post hoc subgroup analysis, the corresponding analyses were performed separately for patients with glucose at randomization above and below the median of admission glucose levels.

The composite endpoint was defined as total mortality or non-fatal MI or stroke, and the number of composite endpoints observed during the follow-up was 150.

To test the proportionality, interaction terms (comprising the risk factors RMSE, age, heart failure × time) were analysed in relation to mortality. The achieved P-values were 0.77, 0.82, 0.49, respectively, and 0.30, 0.78, 0.31 when logarithmical transformation for the variable time was used. Thus, there is no evidence to reject the proportionality requirement in the Cox model. A quadratic term of glycaemic variability did not contribute significantly to the relationship between risk factors and endpoints.

The power has been calculated retrospectively. The gradient of risk per 1 SD (σ) of the variable is defined as EXP(β*σ), where β is the regression coefficient for the relationship to the endpoint. The power in this study is 75% for a gradient of risk per 1 SD of 1.20 and the power is >99% for a gradient of risk per 1 SD of 1.45. Most epidemiological relations between a continuous variable and a disease have a gradient of risk per 1 SD in the interval 1.2–2.6, which also was the case when relating updated mean HbA1c to diabetic complications in the United Kingdom Prospective Study.27 Two-tailed statistical tests were used at the 5% significance level. All analyses were performed with SAS version 9.2.

Ethical considerations

The study conformed to good clinical practice guidelines and followed the recommendations of the Helsinki Declaration. Local Ethics Review Boards approved the DIGAMI 2 protocol. Written informed consent was obtained from all patients prior to enrolment.

Results

Patient characteristics

Baseline characteristics of demographics, comorbidities, and medications at admission for the total DIGAMI 2 population in the original trial (n = 1253) and the present subset (n = 578) are described in Table 1. The mean age of the patients in the present study was 67.9 years, the body mass index was 28 kg/m2, and the majority was males (69%). Some patients were treated with insulin already at admission (32%) and 26% were smokers. In general, patient characteristics were similar between the complete DIGAMI 2 cohort and the current subgroup (Table 1).

View this table:
Table 1

Patients characteristics at admission

VariablePatient population
Present (n = 578)Total DIGAMI 2 (n = 1253)
Age (years)67.9 (10.7)68.4 (11.0)
Male (%)397 (69)837 (67)
BMI (kg/m2)28 (5)28(5)
Randomized treatment group (%)
 Group 1274 (47)474 (38)
 Group 2282 (49)473 (38)
 Group 322 (4)306 (24)
Diabetes duration (years)7.5 (8.2)7.9 (8.3)
Time from symptoms to hospitalization (h)4.5 (4.8)4.4 (4.6)
Time from symptoms to randomization (h)12.4 (7.3)13.0 (7.2)
Blood pressure (mmHg)
 Systolic135 (25)135 (25)
 Diastolic77 (15)76 (15)
Previous medical history (%)
 Myocardial infarction189 (33)423 (34)
 Angina pectoris266 (46)563 (45)
 Chronic congestive heart failure103 (18)220 (18)
 Hypertension277 (48)607 (49)
 Hyperlipidaemia172 (30)396 (32)
Current smoker147 (26)300 (24)
Previous CABG67 (12)136 (11)
Previous PTCA51 (9)100 (8)
Medication prior to admission (%)
 Insulin183 (32)390 (31)
 Metformin151 (26)309 (25)
 Glibenclamide126 (22)274 (22)
 Beta-blockers234 (41)512 (41)
 ASA276 (48)624 (50)
 ACE inhibitor163 (28)390 (31)
 Lipid lowering142 (25)349 (28)
Biochemistry at admission
 Blood glucose (mmol/L)12.7 (4.5)12.7 (4.5)
 HbA1C (%)a7.7 (1.8)7.7 (1.8)
 Serum creatinine (µmol/L)104 (50)104 (46)
 Total serum cholesterol (mmol/L)5.2 (1.3)5.2 (1.3)
 Serum triglycerides (mmol/L)2.3 (1.9)2.2 (2.0)
Killip class (%)
 1426 (74)895 (72)
 2116 (20)254 (20)
 330 (5)87 (7)
 45 (1)14 (1)
  • For categorical variables, n (%) and for continuous variables mean (SD) are presented.

  • aHbA1c was analysed by the mono-S standard with an upper normal limit of 5.3%.

Glucose, glycaemic variability during hospitalization, and prognosis

The total number of glucose measurements, within 48 h, were 11,530, distributed on a mean number of 20 ± 7 per patient (range 2–47). There was an association between blood glucose at randomization and 1-year mortality (HR: 1.06; 95% CI: 1.02–1.10; P = 0.01) but not with the combined endpoint (HR: 1.02; 95% CI: 0.99–1.06; P = 0.24).

Figure 1 outlines the distribution of the mean glucose during 48 h and the three measures of glycaemic variability (RMSE, range, and slope) for patients with and without any of the two endpoints during the 1-year follow-up and in Table 2 mean values and the 95% CI of the same set of variables are presented.

View this table:
Table 2

Mean glucose and glycaemic variability (root mean square error, slope, and range) while receiving an insulin-glucose infusion during the first 48 h of hospitalization for an acute myocardial infarction for all patients at the 1-year follow-up

Variable (48 h)Death/Stroke/Reinfarction (n = 150)No Death/Stroke/Reinfarction (n = 428)Death (n = 82)No Death (n = 496)
Mean95% CIMean95% CIMean95% CIMean95% CI
Mean blood glucose (48 h)8.538.29 to 8.778.508.34 to 8.668.608.24 to 8.968.508.35 to 8.65
RMSE (48 h)2.692.52 to 2.862.572.48 to 2.672.742.50 to 2.982.572.48 to 2.66
Range (48 h)10.19.46 to 10.749.769.40 to 10.1210.49.56 to 11.249.769.42 to 10.1
Slope per 24 h (48 h)−0.042−0.632 to 0.548−0.945−2.100 to 0.210−0.087−0.891 to 0.717−0.814−1.818 to 0.190
  • Mean and 95% confidence intervals (CI) are presented for the respective variable.

Figure 1

The distribution of the mean glucose (A) and the three measures of glycaemic variability [root mean square error (B), range (C), and slope (D)] for patients with and without any of the two endpoints during the 1-year follow-up. X = mean. Horizontal lines of the box = first quartile, median, and third quartile. Whiskers = 5th and 95th percentile.

Adjusted HRs estimated by Cox regression analysis for the two endpoints by a one-unit increase in each variable are presented in Table 3. There was no significant relative risk increase for any of the two endpoints related to more pronounced glycaemic variability. The HR for the composite endpoint of death/stroke/reinfarction for RMSE was 1.09 (95% CI: 0.93–1.27; P = 0.28), while the HR for all-cause mortality for RSME was 1.14 (95% CI: 0.93–1.38; P = 0.21). An analysis of the impact of glucose variability on early (≤3 months) and late (4–12 months) occurring events did not disclose any difference in the lack of prognostic importance (data not shown). Hazard ratios for the two endpoints were also computed for 3 and 18 months of follow-up without any significant relation between glucose variability and the composite endpoint of death/stroke/reinfarction or all-cause mortality (data not shown).

View this table:
Table 3

The relationship between death/stroke/reinfarction and all-cause mortality of the mean glucose and glycaemic variability while receiving an insulin-glucose infusion during the first 48 h of hospitalization for an acute myocardial infarction for all patients at the 1-year follow-up

VariableDeath/stroke/reinfarction (n = 150)Death (n = 82)
HR95% CIP-valueHR(95% CI)P-value
Mean blood glucose (48 h)1.03(0.95–1.13)P = 0.481.07(0.96–1.19)P = 0.24
RMSE (48 h)1.09(0.93–1.27)P = 0.281.14(0.93–1.38)P = 0.21
Range (48 h)1.01(0.98–1.05)P = 0.471.03(0.98–1.08)P = 0.28
Slope per 24 h (48 h)1.01(0.99–1.04)P = 0.401.01(0.98–1.04)P = 0.55
  • Cox regression hazard models adjusted for age and previous congestive heart failure.

Six of the patients, receiving insulin infusion died within 48 h, one of whom died within 24 h. In addition, one patient developed a non-fatal endpoint within 24 h. The five patients who died within 25–48 h were added to the studied population in an analysis looking at the impact of glycaemic variability during the first 24 h on events occurring thereafter. The outcome was consistent with the main results (data not shown).

Glycaemic variability was studied in relation to outcomes after 1 year in a subgroup analysis comparing patients with glucose above or below the median of 12.5 mmol/L at randomization. For patients with a glucose level above the median level the HR for the composite endpoint of death/stroke/reinfarction for RMSE was 1.19 (95% CI: 0.96–1.48; P = 0.11) and for mortality 1.24 (95% CI: 0.95–1.61 P = 0.12; Supplementary material online, Table S4). The corresponding results for patients with glucose levels below the median level were 0.83 (95% CI: 0.62–1.10; P = 0.19) and 0.67 (95% CI: 0.43–1.05; P = 0.08), respectively (Supplementary material online, Table S5). There was a significant interaction (P = 0.027) between the two groups with regard to mortality and a non-significant interaction for the combined endpoint (P = 0.05).

Discussion

This epidemiological report from the DIGAMI 2 trial did not reveal any association between glycaemic variability during the initial phase of an AMI and prognosis expressed as a composite endpoint of mortality and non-fatal cardiovascular events in patients with type 2 diabetes.

The current findings support previous indications, based on a registry report,28 that glucose variability does not predict mortality in patients with diabetes during the early phase of an AMI. Available results on the relation between glycaemic variability during hospitalization and cardiovascular prognosis are limited and have been performed in diverse populations. In a retrospective study of 858 patients treated in a surgical ICU, glycaemic variability was associated with an increased risk of mortality.20 Similar findings were seen in two different ICU settings with heterogeneous populations admitted for both surgical and medical indications.22,23 A different picture emerges when separating patients with or without diabetes. Thus, Egi et al.21 reported that glycaemic variability during an ICU stay was a significant predictor of hospital mortality in the overall population, a relation that did not exist in a subcohort of patients with diabetes. Likewise there was no association between glycaemic variability and hospital mortality in 942 patients with diabetes from one of the heterogeneous populations described above.24

It has been speculated17,24 that patients with diabetes may not react to glycaemic variability to the same extent because their cells have acquired an ability to adapt to the harmful effects of fluctuating glucose, thus leading to relative tolerance. It has also been proposed that hyperglycaemia is merely a marker of stress-related catecholamine and cytokine release and that the same degree of hyperglycaemia may reflect a larger degree of stress in patients without diabetes.24 Such explanations must, however, be interpreted with caution in the lack of prospective, randomized studies. Hitherto available observational analyses do not rule out glycaemic variability as a stronger marker for a more serious underlying disease in patients without diabetes than in those with such disease.

This study expands the available knowledge presenting data of a fairly large group of well-defined patients with type 2 diabetes, followed with frequent, prospectively collected glucose measurements according to a predefined algorithm, thereby limiting the risk that glucose measurements were triggered by factors such as hypo- or hyperglycaemia. Moreover, the endpoints were carefully adjudicated. The reason to base this report on patients surviving 48 h was the intention to cover the complete infusion period, thereby maximizing the availability of glucose measurements. The results were, however, consistent when the few patients who died between 25 and 48 h were included in an analysis covering glycaemic variability during the first 24 h. Still it should be acknowledged that the periods covered may be too short to reflect glycaemic variability during the complete hospital phase. Thus, the present study evaluates whether reducing glycaemic variability during the initial phase of an AMI is of prognostic importance. Furthermore, it may also be seen as a shortcoming that the study addressed the prognostic ability of glucose variability during 3–18 months rather than during the hospital period. The reason was the low number of events during this phase. A potential impact of glycaemic variability on short-term mortality cannot be excluded even if it seems unlikely. Further, it should be noted that the present patients, receiving insulin–glucose infusions with frequent measurements of glucose, may have a lower variability than patients not treated according to such a protocol and larger fluctuations, not captured in this study, may be seen in clinical practice. Even if the present lack of statistical significance does not support the notion that glycaemic variability is prognostically important, it does not prove that the scientific hypothesis is incorrect. An insufficient sample size and a relatively modest variability may be factors causing a false negative outcome. In this context, it is crucial to note that clinical studies of glycaemic variability to date have not generally evaluated continuous glucose curves for patients, and that prospective studies with predefined protocols of measurements are limited.

Finally, the post hoc nature of the subgroup analysis, and the relatively small numbers of patients in the subgroups, does not allow an explanation of the impact of glycaemic variability whether a patient has a higher or lower glucose at randomization. The observation that increased glucose variability seems to have an unfavourable relation to mortality in patients with high glucose at randomization, while the pattern is the opposite among patients with glucose levels at a lower range may be a coincidence but indicates that further studies in the field of glycaemic variability are of interest.

In conclusion, glycaemic variability during the initial phase of an AMI among patients with diabetes does seem to be a prognostic factor for reinfarction, stroke, or death. It should be underlined that the general importance of glycaemic variability for cardiovascular complications needs further elucidation in well-designed studies, e.g. using continuous glucose monitoring devices, preferentially without substantial time delay in glucose recordings, is of potential interest.

Funding

DIGAMI 2 was supported by the Swedish Heart-Lung Foundation, AFA Insurance and by unconditional research grants from Aventis Sweden and Novo Nordisk Denmark.

Conflict of interest: L.G.M. has received research grants from MSD and Sanofi Aventis and speakers' honorarium from MSD, Sanofi Aventis, Novartis, Bayer-Schering, Astra Zeneca and Lilly. K.M. is employed part-time by F. Hoffmann-La Roche, which was not involved in the trial and has no interest in this manuscript. L.R. has received research grants from Sanofi Aventis and Astra Zeneca, honoraria for advisory boards, steering committees and lectures from MSD, Sanofi Aventis, Novartis, Bayer-Schering, Astra Zeneca, Hoffmann-La Roche, Bristol Myers Squibb, and Lilly. H.W. has received honoraria from AstraZeneca, Pfizer, Hoffmann-La Roche, and Biotronik. M.L. has consultant or honoraria from Abbott Scandinavia, Bayer, Eli Lilly, Medtronic, Novartis, Novonordisk Scandinavia, Pfizer, Sanofi Aventis, participated in advisory board of Novonordisk Scandinavia and has research grants from Abbott Scandinavia, Astra Zeneca, and Novonordisk Scandinavia.

Acknowledgements

The authors wish to thank M. Molin, Statistiska Konsultgruppen, Sweden and R. Binisi, Karolinska Institutet, Sweden, for valuable support with the database.

References

View Abstract