In-hospital glycemic variability and all-cause mortality among patients hospitalized for acute heart failure

Continuous variables with normal distributions are presented as mean ± SD and were compared using the Student’s t-test or the Mann–Whitney U test when group distributions were skewed. Categorical variables were compared using the Chi-square test. We used the Youden method to obtain the optimal cutoff point for GV, seeking equilibrium between sensitivity and specificity [23]. The cumulative probability of events was analyzed using the Kaplan–Meier method, and the log‐rank test was used to compare the probability of event curves between the groups. We used the cutoff point for GV to divide the study population into two groups and investigated the effect of GV on clinical outcomes. Cox proportional hazard regression was used to determine the independent effect of high GV in the study population on the clinical outcomes. Variables found to be statistically significant (P < 0.05) in the univariate analysis and variables that could directly affect blood glucose level (e.g., insulin treatment) were included in the multivariate model, except for variables that were closely related to the other variables. All statistical tests were two-tailed, and a P value < 0.05 was considered significant. All analyses were performed using R version 4.0.3 (R Foundation for Statistical Computing, Vienna, Austria) using the “pROC” package and “survival” package.

The data sets generated and analyzed during the current study are available from the corresponding author upon reasonable request.

The baseline characteristics of the study population are shown in Table 1. In this study, 2,617 patients were enrolled. Additional file 1: Table S1 shows the characteristics of those excluded from the study because of a lack of glucose measurements compared to the study population. Compared to the excluded group, the subjects in the study group were older (72 vs 71 years, P = 0.003), had a higher prevalence of type 2 diabetes (42% vs 35%, P < 0.001) and chronic kidney disease (CKD) (16% vs 12%, P < 0.001), a higher proportion of New York Heart Association (NYHA) class III or IV patients (88% vs 81%, P < 0.001), and a longer index hospitalization (10 vs 7 days, P < 0.001). Overall, the study population had more unfavorable prognostic factors for HF than the excluded group.

The median age of the study population was 72 years; 53% were male, 42% had type 2 diabetes, and 60% had hypertension. During the follow-up period, 398 (15.2%) patients died within 6 months, 583 (22.3%) patients died within 1 year, and 576 (22.0%) patients experienced HF readmission (Additional file 1: Table S2). We compared the two patient groups according to the 1-year all-cause mortality (Table 1). Compared to those who survived at 1-year, the patients who died were older (76 vs 71 years, P < 0.001) and had a higher prevalence of hypertension (68% vs 57%, P < 0.001), type 2 diabetes (46% vs 41%, P = 0.035), previous HF admission (42% vs 31%, P < 0.001), and CKD (26% vs 13%, P < 0.001), and they were treated with renal replacement therapy (13% vs 6%, P < 0.001) more frequently.

Additional file 1: Table S3 shows the parameters related to glycemic status according to the 1-year all-cause mortality. Compared to those who survived at 1-year, the patients who died had a significantly higher mean glucose level (133 vs 125 mg/dL, P < 0.001), SD of glucose level (42.0 vs 34.6 mg/dL, P < 0.001), and CoV of glucose level (31.0% vs 27.5%, P < 0.001). Although the prevalence of type 2 diabetes was higher in the patients who died, the HbA_1c level was not significantly different between the two groups (6.6% vs 6.4% [49 vs. 46 mmol/mol], P = 0.155). There was no significant difference in the rate of oral hypoglycemic agent use, including sulfonylureas, and insulin treatment between the two groups. The distributions of blood glucose levels were shown in Additional file 2: Fig. S1, Additional file 3: Fig. S2, Additional file 4: Fig. S3. Scatter plot of the blood glucose level of all subjects (at the time of admission, minimum, maximum values during hospitalization, and at discharge, respectively) (Additional file 2: Fig. S1) and the distribution of blood glucose levels according to the presence of diabetes (Additional file 3: Fig. S2) or insulin use (Additional file 4: Fig. S3) were presented.

Additional file 5: Fig. S4 shows the spline curves for the 1-year all-cause mortality according to the SD (Additional file 5: Fig. S4A) and CoV of glucose level (Additional file 5: Fig. S4B). These figures show that the higher the SD and CoV, the higher the hazard ratio (HR) of all-cause mortality (P < 0.001, for both). Thereafter, we attempted to obtain the optimal cutoff point of each parameter to maximize the log-rank test statistics for mortality. In the study population, the cutoff value for the SD and CoV were 24.5 mg/dL and 20.9%, respectively (Additional file 1: Table S4). In patients with and without type 2 diabetes, the cutoff value for the SD and CoV were 68.3 mg/dL and 21.8%, and 24.5 mg/dL and 20.5%, respectively. Since the SD cutoff difference between patients with and without type 2 diabetes was larger, we decided to use the CoV as an indicator of GV instead of the SD and divided the study population into two groups based on the CoV. Thereafter, the cutoff value was set as 21%, which was the most approximate integer value obtained from the total study group and the type 2 diabetes/non-diabetes subgroups.

Then we analyzed the glycemic parameters and clinical events of the study population according to GV (cutoff CoV = 21%). The mean glucose level, type 2 diabetes prevalence, and HbA_1c values were significantly higher (6.6% vs 6.2%, P < 0.001) in the high GV group (CoV > 21%, n = 1693) than in the low GV group (CoV ≤ 21%, n = 924) (Table 2). Regarding clinical events, 1-year all-cause mortality was more prevalent in the high GV group than in the low GV group, but HF readmission was not significantly different between the two groups. In terms of arrhythmic events regarding the GV, among the patients without previous history of atrial fibrillation at baseline (n = 1711), patients with the high GV were more frequently detected to have new atrial fibrillation on the electrocardiogram during the index hospitalization compared to those with the low GV (13% vs. 6%, P = 0.02). Other clinical characteristics (Additional file 1: Table S5) showed that the high GV group had a higher prevalence of patients with CKD (19.0% vs 11.1%, P < 0.001) and ischemic heart disease (30.0% vs 24.8%, P = 0.005), higher proportion of NYHA III or IV patients (88.5% vs 85.6%, P = 0.035), and higher in-hospital treatment rate with mechanical ventilation (28.9% vs 7.3%, P < 0.001) and mechanical circulatory support devices (8.0% vs 3.2%, P < 0.001), suggesting that the high GV group had a more severe clinical course of HF.

We evaluated the association of GV with the 1-year all-cause mortality via Kaplan–Meier survival analysis. Among the total study population, the high GV group had a higher risk of all-cause mortality (log-rank P < 0.001) than the low GV group (Fig. 1). When analyzing the study population according to diabetic status, this association was still significant at both 6-month and 1-year follow-up in non-diabetic patients (log-rank P < 0.001, both). However, in diabetic patients, the association was significant at 6-month follow-up period (log-rank P = 0.00024), but this association was attenuated at 1-year follow-up (log-rank P = 0.093). When analyzing by quartile, we determined that the higher the GV, the higher the risk of 1-year all-cause mortality among all study subjects and in patients without type 2 diabetes (Fig. 2). Baseline characteristics by dividing dichotomized GV groups and diabetes is shown in Additional file 1: Table S6.

Based on these results, Cox proportional-hazard regression was used to determine the association of high GV with the 1-year all-cause mortality. In the univariate analysis (Additional file 1: Table S7), older age, male sex, low body mass index, low systolic and diastolic blood pressure, low estimated glomerular filtration rate, high natriuretic peptide levels (brain natriuretic peptide ≥ 500 pg/ml or N-terminal prohormone of brain natriuretic peptide ≥ 1000 pg/ml) at admission, hypertension, CKD, previous HF admission, ischemic heart disease, and high GV (CoV > 21%) were significant predictors of all-cause mortality. Also, inotropic therapy, renal replacement therapy, mechanical circulatory support device therapy during hospitalization, and longer duration of hospitalization were associated with increased risk for all-cause mortality.

When all these significant variables were integrated into multivariate Cox regression analysis (Table 3), high GV still remained a significant predictor of 1-year all-cause mortality (HR 1.59, 95% CI 1.29–1.96, P < 0.001), and this association was also significant in the 6-month all-cause mortality (HR 1.97, 95% CI 1.50–2.57, P < 0.001) (Table 4). In addition, we analyzed the Cox proportional-hazard regression according to diabetic status. In non-diabetic patients, high GV was significantly associated with both 6-month (HR 2.00, 95% CI 1.44–2.78, P < 0.001) and 1-year all-cause mortality (HR 1.92, 95% CI, 1.47–2.51, P < 0.001). In diabetic patients, however, high GV was only associated with 6-month all-cause mortality (HR 2.04, 95% CI, 1.26–3.30, P = 0.004), but not with 1-year all-cause mortality (HR 1.23, 95% CI, 0.89–1.70, P = 0.215) (Table 4).

Figure 3 illustrates the subgroup analysis based on the classification of each clinical component. Regarding the association of high GV and 1-year all-cause mortality, we found a significant difference between the subgroups with and without type 2 diabetes (P for interaction = 0.018) as shown above. In other subgroups, including those who used medications such as renin–angiotensin–aldosterone inhibitors, beta-blockers, mineralocorticoid receptor antagonists, loop diuretics, or sulfonylureas, there was no significant interaction.

Many studies have indicated that dysregulation of glucose levels is associated with mortality. Most of these studies included patients with acute or severe conditions such as those in the ICU [1‐3], and some reports showed its association with cardiovascular events in patients with ischemic heart disease [10, 11]. Another large-scale prospective cohort study which analyzed more than 4,800 participants with or without type 2 diabetes, a post hoc analysis of data from the Antihypertensive and Lipid-Lowering Treatment to Prevent Heart Attack Trial (ALLHAT), demonstrated that visit-to-visit GV was associated with all-cause mortality [26]. A recent meta-analysis by Chen et al. showed that long-term GV leads to a higher risk of mortality and various diabetes-related outcomes in patients with diabetes [27]. However, there have been few studies on the impact of GV in acute HF.

Dungan et al. investigated 748 hospitalized patients with HF and revealed that the GV index could predict in-hospital mortality [28]. However, this study focused only on short-term (< 60 days) outcomes, whereas the present study analyzed the 1-year mortality outcomes. Lazzeri et al. later observed that the GV on the first day of admission in the ICU was associated with mortality in 247 patients with acute HF [14]. They revealed that GV was an independent predictor of mortality within a median follow-up period of 10 months, but the mean glucose levels were not. They used two methods to measure GV: the SD of glucose levels and the mean absolute glucose change per hour. However, the number of participants was small, and the study focused on acute coronary syndrome (72% of total study population). In our study, we analyzed patients with ischemic (28%) and non-ischemic (72%) HF, and subgroup analysis showed consistent results regardless of the presence of either ischemic heart disease or HF with reduced ejection fraction (≤ 40%). In addition, a recent study showed that elevated GV in patients with acute HF with type 2 diabetes (n = 214) predicted the 6-month mortality (HR 2.34, P = 0.03) [15]. However, the authors only focused on diabetic patients, and the study population was small.

In addition, we showed that atrial fibrillation was more commonly detected during hospitalization in patients with high GV than those with low GV among the patients without previously reported AF. This finding is in line with the previous studies that revealed cardiac arrhythmias were common in insulin-treated patients with type 2 diabetes and are associated with GV [29], which was also independently associated with the development of new-onset atrial fibrillation in these diabetic patients [30].

Glycemic dysregulation might reflect poor pancreatic β-cell function and the inadequate release of endogenous insulin; therefore, hyperglycemia and high GV would be more pronounced in patients with type 2 diabetes with worse glucose control and longer disease duration [15]. However, in acute HF, SNS activation is one of the important neurohormonal mechanisms related to disease progression [31]. Thus, high GV may reflect an acute stressful state and increased sympathetic tone even in non-diabetic patients with acute HF. This may mean that the HR of high GV for mortality at 6 months may be higher than that at 1 year, which is consistent with our findings (Table 4). The use of HF medications, such as renin–angiotensin–aldosterone inhibitors, beta-blockers, mineralocorticoid receptor antagonists, or loop diuretics, may affect the neurohormonal system. However, the present results showed that the effect of GV on mortality was significant regardless of the use of these drugs.

There may be several explanations that high GV is associated with higher mortality in acute HF patients, particularly non-diabetic patients. In acute HF patients, there is a period wherein the readmission rate is high early after discharge, which is often referred to as a “vulnerable phase” because it determines the prognosis in post-discharge period [32]. The present results showed that a high GV during hospitalization was associated with early mortality within 6 months and 1 year after discharge. This is in line with another study of the KorAHF registry that showed an association between hyperglycemia and a high 1-year mortality rate in non-diabetic patients [8]. It may be assumed that GV could be affected by SNS activation followed by decompensation rather than glucose metabolism itself in non-diabetic patients. Also, a previous study suggested that oscillating glucose levels can have more deleterious effects than constant high glucose on endothelial function and oxidative stress in both subjects with normoglycemia and with type 2 diabetes [33]. A recent study showed that greater short-term GV was associated with increased risk for all-cause mortality even in type 2 diabetic patients with well-controlled glucose profile by continuous glucose monitoring [34]. Moreover, in another recent study that analyzed echocardiographic parameters in non-diabetic subjects, visit-to-visit GV was also associated with left ventricular diastolic dysfunction, also supporting our findings [35]. In the above-mentioned post hoc analysis, the results suggested that greater visit-to-visit GV is associated with an increased risk for all-cause mortality, especially among people without type 2 diabetes [26]. These previous results support our findings that GV has a significant association with mortality even in patients without diabetes. Although the measurement of GV in our study was short-term and an in-hospital parameter, which is different from the visit-to-visit GV, we showed that high GV was associated with higher mortality in non-diabetic patients. Even in patients with type 2 diabetes, our results showed that high GV was associated with higher mortality only at 6 months, not 1-year after discharge, which can also be considered the effect of acute stress in the post-discharge period of acute HF.

In the present study, the mean glucose level, prevalence of type 2 diabetes, and use of exogenous insulin treatment were higher in the high GV group than in the low GV group. It can be expected that insulin treatment and GV are closely related because hyperglycemia requires insulin treatment, which may cause hypoglycemic side effects. In a study of the KorAHF registry patients with type 2 diabetes, the mortality rate was higher in patients treated with insulin than in patients treated with oral hypoglycemic agents alone [25]. However, our study analyzed patients with and without type 2 diabetes, and we found that insulin use was not significantly associated with mortality in the Cox analysis. There may be two reasons for this difference. First, the previous study was a result of comparing insulin treatment and oral hypoglycemic agents in diabetic patients; and second, we enrolled subjects whose glucose levels were measured at least three times during hospitalization, and patients who died during hospitalization were excluded from our study population.

A recent randomized clinical trial showed that several indicators, including HbA_1c level, were more significantly improved when using continuous glucose monitoring compared to conventional glucose measurement in diabetic patients receiving insulin treatment [36]. In acute HF patients, the more severe the HF, the higher the GV tends to be due to the influence of increased neurohormonal activity or inflammation with deteriorated hemodynamic conditions. However, this does not mean that any interventions to reduce GV improve the outcomes of acute HF patients, so further research is needed to determine whether GV may be a therapeutic target in this population. Considering the increasing use of wearable devices and telemonitoring in clinical practice, it is expected that more studies will be warranted to reveal the clinical usefulness of GV monitoring using continuous glucose monitoring in outpatient care and impact of intervening to reduce GV in HF patients.