Heart failure (HF) with preserved ejection fraction (HFpEF) is a clinical syndrome resulting from an increase in left ventricular filling resistance during diastole that results in congestive symptoms.1 Although our understanding of the pathophysiology of HFpEF has increased in recent years, many of its aspects remain unclear, such as the utility of consistent markers, its pathogenesis in specific populations and the treatments that show benefits in these patients.

HFpEF is considered a heterogeneous process with numerous phenotypes,2,3 some of which are related to the underlying comorbidities. Among these comorbidities, type 2 diabetes mellitus (DM2) is one of the most important and is present in approximately 45% of these patients, and this prevalence increases in newly diagnosed cases of HFpEF.4 The presence of DM2 is also associated with an increased morbidity and mortality risk for patients with HFpEF, although to a lesser degree than for patients with HF and reduced ejection fraction (HFrEF).5 An added problem is that patients with DM2 and HFpEF often have comorbidities such as arterial hypertension, dyslipidemia, obesity, chronic obstructive pulmonary disease (COPD) and atrial fibrillation, which also affect the prognosis.6 Studies are attempting to analyze the extent to which these comorbidities affect HFpEF, determining how these comorbidities group in various phenotypes and how they can be treated to achieve clinical and prognostic improvement for these patients.2,6,7 However, there are few existing patient cohorts with HFpEF, and DM2 is usually included as another comorbidity in the analyses.

Our objective in the present study is to describe the various phenotypes of associated comorbidities, both cardiovascular and noncardiovascular, which can present in patients with HFpEF and DM2. Our hypothesis is that HFpEF with DM2 is a heterogeneous entity based on the accompanying comorbidities, for which various clusters of comorbidities can be established that serve as a starting point for more accurately investigating and treating these patients.

The patients were recruited from the Spanish National Heart Failure Registry (RICA) belonging to the Heart Failure and Atrial Fibrillation Workgroup of the Spanish Society of Internal Medicine (SEMI), a multicenter, prospective registry that has been active since 2008. The registry includes, uniquely and consecutively, patients older than 50 years diagnosed with HF at hospital discharge due to decompensated or new onset HF, according to the current European cardiology guidelines at all times.8 Once included, these patients are followed for a year. The present analysis employs the data for the patients included from March 2008 to May 2017. During this period, various analyses arising from the registry have been published.9,10

The registry protocol was approved by the ethics committee of University Hospital Reina Sofia of Cordoba, Spain, and all patients had to sign an informed consent prior to their inclusion in the registry. The data were collected from the password-protected website (www.registrorica.org) that contains the database.

The present analysis exclusively included those patients with a left ventricular ejection fraction (LVEF) ≥50%, excluding patients with HF secondary to valvular heart disease (Fig. 1). We employed data from the patients’ medical history, physical examination (systolic blood pressure, weight and height) and laboratory tests (hemoglobin, glomerular filtration rate [eGFR, estimated by the modification of diet in renal disease equation formula], glycosylated hemoglobin [HbA1c] and natriuretic peptides). LVEF was assessed using 2-D echocardiography.

Figure 1.

Flow diagram.

(0.28MB).

From the entire patient sample with HFpEF, we selected those patients diagnosed with DM2, based on the patient’s history or whether they were undergoing therapy with antidiabetic agents. We excluded from the analysis the patients who did not have complete information for the study variables, those who did not complete the follow-up and those who died during the hospitalization.

Once the final sample was obtained, we divided the population according to the number of comorbidity clusters obtained from the statistical analysis.

The study’s primary objective was to describe the characteristics of each comorbidity cluster and the differences between these clusters.

Statistical analysis

To obtain the clusters, we employed the agglomerative hierarchical analysis with Ward’s method, using the hclust function of the statistical program R (version 3.5.1).11 This study considers the analysis of patients who share similar characteristics based on the reported variables, who are grouped in a timely manner (initially, there were as many groups as there were individuals), progressively constructing a group hierarchy (ascending in our case, i.e., each observation began in its own group and the pairs of groups are mixed as one rises in the hierarchy). This process allowed us to construct a classification tree (dendrogram). The following prespecified variables for performing the analysis were chosen based on their availability in the registry, on previous studies and on clinical experience: a) dichotomous qualitative variables (COPD, dyslipidemia, liver disease and dementia), which were assigned the value of 1 when present and 0 when absent; b) nondichotomous qualitative variables, such as stroke (absent 0, transient ischemic attack [TIA] 1, hemorrhagic stroke 2, cardioembolic stroke 3, atherothrombotic stroke 4) and arrhythmias (sinus rhythm 0, flutter/atrial fibrillation 1, atrioventricular block 2, other 3); and c) quantitative variables (systolic blood pressure, body mass index [BMI], eGFR and hemoglobin), which were analyzed according to their numerical value. We initially constructed a (di)similarity matrix between the observations employing the Kendall method using the factoextra function of R, given that the variables are not parametric. The color level was proportional to the similarity value between the observations: if red, the distance between the variables was 0 (high similarity); if green, the distance was high (low similarity) (Fig. 2). Once the dendrogram had been constructed, we employed resampling techniques using bootstrap (n = 1000) with the pvclust function of R to select the most appropriate clusters.12 For each potential cluster, we calculated the bootstrap probability value (BP), which corresponds to the rate at which the same cluster is detected in the various resamplings. Likewise, we calculated the approximately unbiased probability (AU) for each cluster. The clusters with an AU 95% were considered strongly supported by the data. The 4 groups in the present study were chosen based on the highest possible BP and AU values (Fig. 3A). Once we obtained the groups and classified the patients according to the groups, the qualitative variables were expressed using absolute numbers and percentages and compared using the chi-squared test. The quantitative variables are expressed using their median and interquartile range and compared using Kruskal-Wallis given their nonparametric nature (analyzed with the Shapiro-Wilk test). We considered a p-value<.05 as significant. The analysis was conducted with the R program (version 3.5.1).

Figure 2.

Cluster similarity matrix performed using the Kendall method. The color level is proportional to the similarity value between the observations; if red, the distance between the variables is 0 (high similarity); if green, the distance is high (low similarity).

(0.54MB).

Figure 3.

Dendrogram resulting from the agglomerative hierarchical analysis. A) Resulting schematic dendrogram, with the degrees of significance by comorbidity. The bootstrap probability (BP) is in green, and the approximately unbiased probability (AU) is in red. B) Dendrogram resulting from the analysis. The chosen clusters are marked in color.

Abbreviations: BMI, body mass index; eGFR, estimated glomerular filtration rate; Hb, hemoglobin; SBP, systolic blood pressure.

(0.13MB).

A total of 1934 patients presented HFpEF, 907 of whom (46.9%) had DM2 (Fig. 1). The mean age was 78.5 (7.7) years, with a predominance of the female sex (552, 60.9%). The mean LVEF was 61.1% (7.9%), with a mean eGFR of 55.6 (26.6) mL/min/1.73 m2. The most prevalent comorbidities were dyslipidemia (601, 66.3%), atrial fibrillation/flutter (522, 57.5%) and COPD (217, 23.9%).

The dendrogram resulting from the cluster analysis is shown in Fig. 3B. We constructed 4 clusters whose characteristics are shown in Table 1 and Fig. 4. As it can be observed, Cluster 1 (201, 22.2%) included patients with high cardiovascular risk but no arrhythmia (only 1% presented atrial fibrillation/flutter), with a predominance of women with high systolic blood pressure (SBP), dyslipidemia and a high rate of atherothrombotic stroke.

Figure 4.

Descriptive schematic of the resulting clusters.

(0.12MB).

Cluster 2 (303, 33.4%) included older patients, also with a predominance of the female sex, with metabolically healthier patients (no dyslipidemia and lower SBP) and with an atrial fibrillation/flutter rate >50%.

Cluster 3 (140, 15.4%) included mostly male patients with COPD, dyslipidemia and liver disease. Although the presence of atrial fibrillation/flutter was 58.6%, this cluster recorded the lowest rate of cerebrovascular disease.

Lastly, Cluster 4 (263, 29.0%) was similar to Cluster 1 but had a greater presence of atrial fibrillation/flutter (93.9%), TIA (7.2%) and cardioembolic stroke (3.4%).

Although there were no significant differences related to BMI, Clusters 1, 3 and 4 included patients with obesity (BMI >30 kg/m2), while Cluster 2 included patients with excess weight (BMI between 25 and 30 kg/m2).

Based on our results, we defined 4 clusters of patients with HFpEF and DM2, according to the associated comorbidities. Two patient groups (Clusters 1 and 4) were predominantly female and had mainly cardiovascular risk factors, which differed from each other by the presence or absence of atrial fibrillation/flutter. A third group (Cluster 3) was predominantly male and had a predominance of respiratory disease and hepatic impairment. The last group (Cluster 2) encompassed older patients, predominantly women, with a high proportion of atrial fibrillation/flutter. A common finding for all the groups was the presence of excess weight or obesity.

The identification of HFpEF and DM2 phenotypes with various comorbidity clusters helps assess the relevance of differentiated approaches for these specific patient groups, in order to achieve a greater therapeutic benefit.

For patients with DM2 and HF, the first treatment option that emerges is sodium-dependent glucose cotransporter 2 inhibitors (SGLT2i). This family of drugs has been shown to reduce cardiovascular morbidity and mortality, with a dramatic reduction in hospitalizations for HF and renal protection for patients with DM2 of high cardiovascular risk.13 Specific clinical trials with SGLT2i are currently underway with patients with HFpEF. Although the hypotheses regarding the mechanisms by which SGLT2i exert their cardiorenal benefit remain open, it has been proposed that the increased availability of ketone bodies used by the heart14 and their action on the sodium-hydrogen exchanger, which reduces cytoplasmic sodium and entails coronary vasodilation,15 could be the mechanisms involved. SGLT2i have also shown beneficial effects on diastolic function and left ventricular mass for patients with HFpEF.16,17

Regarding Clusters 1, 3 and 4 in which obesity is a common denominator, there is the possibility of including glucagon-like peptide-1 receptor analogs (GLP-1ra) in the diabetic therapy. These drugs, in addition to reducing body weight, have shown benefits at the cardiovascular level (which make these advisable for Clusters 1 and 4) and could exert a positive effect on the pulmonary disease of patients in Cluster 3, as it has been demonstrated in animal models.18 In clinical trials on cardiovascular safety, GLP-1ra have shown a neutral effect on HF,19,20 but their effects on ventricular remodeling, arterial stiffness and oxidative stress21 make them potentially useful in treating HFpEF. The LIVE study conducted with patients with HFrEF, with and without diabetes, found that treatment with liraglutide (compared with placebo) showed no clinical benefits but did show an increased tendency toward adverse cardiovascular effects.22 However, the study showed a significant reduction in the E/e’ ratio, which confirmed the possible usefulness of liraglutide for HFpEF.

Dipeptidyl-peptidase-4 inhibitors (DPP4i) have shown differing effects on HF. In their respective cardiovascular safety clinical trials, sitagliptin and linagliptin were shown to be neutral,23,24 with saxagliptin significantly increasing the risk of hospitalizations for HF,25 while alogliptin showed no significant increase.26 In the Vildagliptin in Ventricular Dysfunction Diabetes (VIVIDD) study,27 vildagliptin showed no changes in LVEF, although significant increases were observed in left ventricular end-diastolic and systolic volumes. Moreover, sitagliptin has been shown to attenuate the worsening of diastolic dysfunction (E/e’ ratio) in patients with DM2.28 DPP4i offer good safety in more elderly populations,29 which favors their use in Cluster 2, which contains older patients.

In addition to the DM2 approach, these clusters offer other opportunities for individualizing the treatment of patients with HFpEF.

Although statins are the drugs of choice for treating dyslipidemia in patients with DM2,30 their role in HF is controversial. Co-enzyme Q10 inhibition, selenium deficiency and increased arterial calcification are some of the potential mechanisms that can worsen HF.31 In the patients of Clusters 1 and 4, in whom dyslipidemia is highly prevalent, we can consider the use of other hypolipidemic drugs (ezetimibe or proprotein convertase subtilisin/kexin type 9 inhibitors) without these potential adverse effects.

Moreover, a high heart rate is known to be harmful for patients with HFpEF in sinus rhythm.32 There is controversy, however, on the efficacy of beta blockers in HFpEF, regardless of cardiac rhythm (sinus or atrial fibrillation).33 A clinical trial in which 75% of the patients were also taking beta blockers showed that ivabradine improved heart rate control in patients with HFpEF, without changing the diastolic pattern (E/e’) or the 6-min walking test at 8 months of treatment.34 A previous study with a smaller percentage of patients taking beta blockers and a shorter follow-up showed that ivabradine improved the patients’ exercise capacity and decreased left ventricular filling pressures induced by exercise in the patients with HFpEF.35 Therefore, ivabradine could be considered a preferential treatment for Cluster 3, where beta blockers can jeopardize the progression of COPD. In this same cluster and in patients with atrial fibrillation, the usefulness of diltiazem may also be considered, given that it has been shown to increase end-diastolic volumes in patients with hypertrophic cardiomyopathy.36

The present study has a number of limitations. The included patients might not fit the current definition of HFpEF, given that a number of them were included prior to the publication of the current diagnostic criteria. Not all patients therefore had specific measurements in the echocardiogram or of natriuretic peptides. We did not include in the analysis a number of conflicting comorbidities of DM2 (e.g., depression), which could have a significant clinical impact.37 Given that this was a cross-sectional study, we did not analyze the progression and prognostic characteristics of each cluster. Lastly, this was a cohort study, which can be subject to various selection and randomization biases.

In conclusion, the present analysis describes various patient groups with HFpEF according to their comorbidities, which can be a starting point for future research that can help more accurately define the various comorbidity clusters and lead to a more inclusive and personalized therapeutic approach that improves these patients’ prognosis.

The authors declare that they have no conflicts of interest.