Heart failure and the prognostic impact and incidence of new-onset of diabetes mellitus: a nationwide cohort study

In Denmark, every resident is assigned a unique personal identification number enabling individual-level linkage between nationwide health care registries. The Danish National Patient Registry entails information on all hospital admissions from 1978 and forward. Each hospital contact is coded with a primary diagnosis and several secondary diagnoses according to The International Classification of Disease, Eighth Revision (ICD-8) until 1993, and The International Classification of Disease, Tenth Revision (ICD-10) from 1994 onwards. The Danish National Prescription Registry holds information (dosage, dates, and Anatomical Therapeutic Chemical (ATC) codes on all prescriptions dispensed from a pharmacy since 1995 and the Danish Civil Registration System records vital-status.

We included adults (older than 18 years) with a first-time diagnosis of HF in hospital discharge records in the period 2003–2014. The index date (date of inclusion) was 30 days after discharge from the hospital.

Prevalent diabetes was defined by at least one prescription of a glucose-lowering drug and/or a previous ICD code of diabetes 6 months prior to the index date. New-onset diabetes was defined by a first claimed prescription of a glucose-lowering drug and/or an ICD code of diabetes after index date in patients with no prior history of diabetes. Combining the use of ICD codes and prescriptions to assess diabetes status have been validated previously with a positive predictive value of 97% and 95%, respectively [8‐10].

Comorbidities were identified through ICD codes from hospital records up to 10 years prior to the index date, and continuously updated throughout the follow-up period (see Appendix: Table 2 for details and ICD codes). Information on concomitant medical therapy was obtained from dispensed prescriptions as listed in the Danish National Prescription Registry and defined by at least one redeemed prescription of the drug 6 months prior to the index date. The following drugs were recorded at inclusion baseline: angiotensin-converting enzyme inhibitors (ACE), angiotensin II receptor blockers (ARB), calcium channel blockers, loop diuretics, thiazides, digoxin, platelet inhibitors (acetylic acid and adenosine diphosphate receptor inhibitor), mineralocorticoid receptor antagonists (MRA), statins, beta-blockers, insulin, metformin, sulfonylurea, thiazolidinedione (TZD), dipeptidyl peptidase-4 (DPP-4) inhibitor, glucagon-like peptide-1 (GLP-1) receptor agonist, sodium-glucose cotransporter-2 (SGLT2) inhibitor and patients who treated with a combination of two anti-diabetic drugs. DPP-4 inhibitors, GLP-1 receptor agonists and SGLT2 inhibitors were combined in one group defined as newer antidiabetic medication for further statistical analysis (see Appendix: Table 2 on details regarding details and ATC codes).

The outcomes of the study were new-onset diabetes and all-cause death. Patients were followed to new-onset diabetes, death, emigration or end of study (31 December 2015).

Baseline characteristics were described by the use of proportions for categorical variables with means and standard deviations (SD) or medians and interquartile ranges (IQR) for continuous variables. Differences between groups were tested by use of the Chi square test for categorical variables, non-parametric test for non-normally distributed continuous variables and parametric for normally distributed continuous variables. Annual incidence rates of new-onset diabetes were calculated per 100 person-years. In analyses of all-cause death, we treated new-onset diabetes as a time-dependent variable and compared the risk with prevalent diabetes patients and patients free of diabetes as reference. To account for the longer duration of HF in patients with new-onset diabetes, follow-up time was split into 1-year intervals from inclusion date, and according to calendar year in 3-year intervals. These variables were included in the adjusted model. Comorbidity and antidiabetic medication were continuously updated throughout follow-up, and age was updated at the beginning of each interval. A multivariable Cox proportional hazards analysis was performed to compare hazard ratios of death according to diabetes status. The model was adjusted for age, sex, duration of HF, education level and each individual comorbidity (ischemic heart disease, cancer, atrial fibrillation, chronic obstructive pulmonary disease, chronic kidney disease, hypertension, and stroke). To compare new-onset diabetes to prevalent diabetes, we repeated the analysis, but used prevalent diabetes status as the reference and included antidiabetic medication (metformin, insulin, TZD, sulfonylurea and newer anti-diabetic drugs consisting of DPP-4 inhibitors, SGLT2-inhibitors, and GLP-1 receptor agonists) in the model. Tests for interactions of diabetes status and sex in relation to risk of death were performed. Analyses were performed using SAS (version 9.4 for Windows, SAS Institute, North Carolina) and R (version 3.5.0 for Windows, R Foundation for Statistical Computing) [11].

During a mean follow-up of 3.9 years, 8164 (10%) developed new-onset diabetes yielding an event rate of 2.5 per 100 person-years. Over time, annual incidence rates of new-onset diabetes increased from around 2 per 100 person-years in the first years following HF hospitalization up to 3 per 100 person-years after 5 years of follow-up (Fig. 2).

A total of 61,424 (59%) patients died during the study period, with an event rate of 15.0 per 100 person-years. Rates were lowest among patients without diabetes (13.9 per 100 person-years), intermediate in those with prevalent diabetes (17.9 per 100 person-years) and highest among patients with new-onset diabetes (21.5 per 100 person-years). In age- and sex-adjusted analyses this yielded hazard ratios (HR) of 1.46 (95% CI 1.43–1.49) for prevalent diabetes and HR 1.86 (95% CI 1.78–1.91) for new-onset diabetes with patients with HF and no diabetes as reference. In adjusted analyses additionally including education level, and continuously updated duration of HF and comorbidity, the risk estimates were somewhat lower with HR 1.19 (95% CI 1.16–1.21) for prevalent diabetes and HR 1.47 (95% CI 1.42–1.52) for new-onset diabetes (Fig. 3). Including antidiabetic medication (metformin, insulin, sulfonylurea, and newer anti-diabetic drugs) and comparing prevalent diabetes patients directly with new-onset diabetes patients in the adjusted model, we found a significantly higher risk estimate for patients with new-onset diabetes HR 1.24 (95% CI 1.20–1.29). We found no interaction between diabetes status and sex in relation to the risk of all-cause death (p = 0.229).

The association between diabetes and HF has been well established and reproduced in several observational cohorts. Diabetes is associated with an up to four times increased risk of HF when compared to a general population without diabetes [12]. It is not clear whether this is due to shared risk factors for the two conditions or the presence of a specific diabetic cardiomyopathy, where pathophysiological mechanisms related to abnormal cardiac handling of glucose has been suggested to lead to both systolic and diastolic dysfunction [4, 13]. The high incidence of HF in patients with diabetes, even when coronary artery disease is absent, suggests a substantial direct relation between diabetes and development of HF [13]. The higher mortality associated with prevalent diabetes in this study correlates with previous studies describing diabetes as an independent risk factor for death and adverse cardiovascular outcomes in patients with HF [1, 12, 14‐17]. Obesity is closely related to diabetes and its presence in patients with diabetes is associated with an increased risk of developing HF, but surprisingly reduced mortality [18]. In established HF, this so-called obesity paradox of reduced mortality exists for patients without diabetes but was not found in patients with diabetes [19]. Our finding of a poorer prognosis in HF patients with incident diabetes is in line with a prior study which showed a graded relation between blood glucose levels at the time of hospitalization for HF and long-term outcomes in patients without known diabetes [20]. An older study using Danish registry data showed that mortality risk in HF patients was lower in women, but we found no interaction between diabetes status, sex and mortality risk in the present analysis [21].

The stepwise increase in the incidence of diabetes as HF persists in our study is consistent with the hypothesis that HF over time can lead to diabetes. Apart from including real-world patients, our study highlights important information on how the annual incidence changes over time and similar to that reported from clinical trials [22‐25]. Several studies have shown that insulin sensitivity decreases as HF progresses. In the Bezafibrate Infarction Prevention study development of diabetes among HF patients occurred in a stepwise manner from 13% in New York Heart Association (NYHA) Class I to 20% in NYHA Class III during a mean follow-up of 7.7 years. Being in NYHA Class III was an independent risk factor for the development of diabetes [26]. Furthermore in patients with advanced HF, left ventricular assist device implantation have been shown to improve diabetes control [27]. In this cohort of real-life patients with HF, around 10% of patients without diabetes at baseline, developed new-onset diabetes during follow-up, which is similar to findings from the Carvedilol Or Metoprolol European Trial (COMET) where around 10–12% developed diabetes during 5 years of follow-up [22]. Prior observational studies have demonstrated that increasing HF severity (as defined by dosages of loop diuretics) is associated with new-onset diabetes in a severity-dependent manner [7, 28]. The decreased cardiac output as seen in progressive HF may lead to diminished oxygen, glucose and insulin distribution to peripheral muscular tissue as well as a loss of muscle mass which in turn may increase insulin resistance [29]. Impaired blood flow can increase systemic levels of adrenaline (epinephrine) and noradrenaline (norepinephrine) which is suggested to increase insulin resistance and hepatic gluconeogenesis as well as decrease the insulin release from pancreatic beta cells. Sympathetic overdrive has also been shown to reduce insulin sensitivity [5, 30]. This relation between HF and incident diabetes has been supported by observational studies [7, 26]. The bidirectional relationship between diabetes and HF is supported by several clinical and epidemiological studies, but the causality remains unclear. The idea of a common disease origin, in a recent review, proposes inflammation and oxidative stress could be the common ground for the development of both diabetes and cardiovascular disease [31]. Targeted anti-inflammatory therapy decreased the risk of recurrent cardiovascular events in patients with known cardiovascular disease and several studies showed novel genetic variants linked to increased risk of both diabetes and cardiovascular disease development [32‐34].

As expected, patients with new-onset diabetes were not primarily prescribed insulins, thiazolidinediones or DPP-4 inhibitors as these drugs are usually used as add-ons with the progression of diabetes. Only rarely insulin is used as first-line treatment in patients diagnosed with severely dysregulated diabetes. Likewise, among patients with prevalent diabetes, we found a higher prevalence of all types of antidiabetic drugs, as these patients have had their diabetes for a longer period and therefore often require second-line antidiabetics. In our crude analysis, the risk of death among patients with new-onset diabetes was significantly different and attenuated compared to patients with prevalent diabetes and no diabetes. After adjustment for age, sex, duration of HF, baseline comorbidities and antidiabetic medication, the risk was higher among patients with new-onset diabetes. The cardiovascular safety profile for antidiabetic drugs in patients has been questioned. Overly aggressive antidiabetic therapy may lead to hypoglycemia, and both insulin, as well as thiazolidinediones, can lead to fluid retention and worsening of HF [13, 35]. Compared to SGLT2 inhibitors DPP-4 inhibitors were associated with increased risk of HF hospitalization [36]. The small number of patients receiving SGLT2 inhibitors and GLP-1 inhibitors within our cohort limits us in terms of assessing the impact of these drugs. But, the increased risk in patients with new-onset diabetes after adjustment for antidiabetic medication and comorbidity status could be suggestive of an association that is not fully explained by the uncertain safety profile of antidiabetic medication nor the increased comorbidities among patients with new-onset diabetes. Our study underlines the need to further explore risk factors associated with the development of new-onset diabetes in HF patients.

The main limitations of the present study is the lack of information on clinical variables reflecting HF and diabetes severity including ejection fraction, NYHA class and smoking status, vital parameters e.g. heart rate and blood pressure, body mass index, biochemical parameters such as natriuretic peptides (e.g. NT-pro-BNP), glucose levels, hemoglobin A1c and type of diabetes. As we were not able to fully adjust for these possible confounders, residual bias cannot be ruled out. Secondary we lack information on diet treated diabetes patients, which confers a selections bias towards a more ill cohort than in the general population.

We have shown the impact on prognosis development of new-onset diabetes in patients with HF compared to patients without diabetes. Around 10% of the HF cohort developed new-onset diabetes and the incidence of new-onset diabetes was rising during follow-up aiding important information from a real-world cohort. We hope our findings will aid clinicians in assessing important subgroups among patients with HF in need of close monitoring and supervision of co-existing diabetes illness.