Rates and predictors of cardiovascular and non‐cardiovascular outcomes in heart failure with preserved ejection fraction

Abstract

Aims

The detailed sub‐categories of death and hospitalization, and the impact of comorbidities on cause‐specific outcomes, remain poorly understood in heart failure (HF) with preserved ejection fraction (HFpEF). We sought to evaluate rates and predictors of cardiovascular (CV) and non‐CV outcomes in HFpEF.

Methods

The Karolinska–Rennes study was a bi‐national prospective observational study designed to characterize HFpEF (ejection fraction ≥45%). Patients were followed for cause‐specific death and hospitalization. Baseline characteristics were pre‐selected based on clinical relevance and potential eligibility criteria for HFpEF trials. The associations between characteristics and cause‐specific outcomes were assessed with univariable and multivariable Cox regressions.

Results

Five hundred thirty‐nine patients [56% females; median (inter‐quartile range) age 79 (72–84) years; NT‐proBNP/BNP 2448 (1290–4790)/429 (229–805) ng/L] were included. Over 1196 patient‐years follow‐up [median (min, max) 744 days (13–1959)], there were 159 (29%) deaths (13 per 100 patient‐years: CV 5.1 per 100, dominated by HF 3.9 per 100; and non‐CV 5.8 per 100, dominated by cancer, 2.3 per 100). There were 723 hospitalizations in 338 patients (63%; 60 per 100 patient‐years: CV 33 per 100, dominated by HF 17 per 100; and non‐CV 27 per 100, dominated by lung disease 5 per 100).

Higher age and natriuretic peptides, lower serum natraemia and NYHA class III–IV were independent predictors of CV death; lower serum natraemia, anaemia and stroke of non‐CV death; and anaemia and lower serum natraemia of non‐CV death or hospitalizations. There were no apparent predictors of CV death or hospitalization.

Conclusions

In a clinical cohort hospitalized and diagnosed with HFpEF, death and hospitalization rates were roughly similar for CV and non‐CV causes. CV deaths were predicted primarily by severity of HF; non‐CV deaths primarily by anaemia and prior stroke. Lower serum sodium predicted both. Hospitalizations were difficult to predict.

Introduction

Heart failure (HF) with preserved ejection fraction (HFpEF) is a heterogeneous syndrome characterized by extensive comorbidities. ¹ The detailed causes and sub‐categories of death and hospitalization and the impact of comorbidities on cause‐specific outcomes, remain poorly understood. This presents important challenges in trial design and may explain why multiple HFpEF drug trials have failed.

The greater prevalence and prognostic role of non‐cardiovascular (non‐CV) comorbidities in HFpEF versus in HF with reduced ejection fraction (HFrEF) have been highlighted. ² , ³ , ⁴ Causes of death appear to be more non‐CV in HFpEF ⁵ , ⁶ , ⁷ ; however, the role of detailed sub‐categories of cardiovascular (CV) versus non‐CV hospitalizations as well as predictors of CV versus non‐CV events in HFpEF remain unclear. To address these issues, we assessed the rates and predictors of cause specific death and hospitalization in the multicentre Karolinska–Rennes (KaRen) study of HFpEF.

Methods

Study design and setting

The KaRen rationale and design has been previously described. ⁸ KaRen is a prospective, bi‐national, multicentre (3 centres in Sweden and 11 centres in France), observational cohort study aiming to characterize clinical characteristics and outcomes in HFpEF. The minimum required sample size for the overall KaRen study was 400. The power calculation has been described in detail and was based on detecting differences in the overall primary endpoint based on electrical dyssynchrony. ⁸

Patients

Inclusion criteria were all three of acute presentation with HF (defined by the Framingham criteria), ⁹ and left ventricular ejection fraction (EF) ≥ 45%, and N‐terminal pro‐brain natriuretic peptide (NT‐proBNP) > 300 ng/L or brain natriuretic peptide (BNP) > 100 ng/L within 72 h of presentation. Exclusion criteria have been described ⁸ and included primary restrictive or obstructive cardiomyopathy or pericardial constriction, any cardiovascular disorder with an indication for surgical or percutaneous intervention (e.g., valve disease considered suitable for intervention), existing cardiac resynchronization therapy, or kidney disease requiring dialysis or pulmonary disease requiring chronic supplemental oxygen.

The study conformed to the Declaration of Helsinki, was approved by local ethics committees, and all patients provided written informed consent. There were no investigational interventions and all patients received standard care.

Data

Extensive baseline (at time of presentation with acute HFpEF) history, symptoms, physical exam, and laboratory data were determined by local investigators. Other than the Framingham, echocardiographic and natriuretic peptide inclusion criteria, there were no other HF validation criteria at entry, allowing inclusion of a relatively heterogeneous and representative population. All data were entered into a web‐based electronic case report form (ClinSource, Brussels, Belgium). Selected baseline variables are shown in Table 1.

Table 1.

Baseline characteristics of all 539 patients.

Variable	Missing value n (%)	Median [IQR, (min–max)] or n (%), unless otherwise stated
Duration of follow‐up
Follow‐up time to death or censor, days	0 (0)	744 [13 (505–1175) 1959]
Demographics
Age a , years	0 (0)	79 [42 (72–84) 100]
Sex a , female	0 (0)	303 (56)
Clinical
NYHA in stable state prior to presentation a	76 (14)
I		87 (19)
II		284 (61)
III		88 (19)
IV		4 (0.9)
NYHA at presentation	12 (2.2)
I		4 (0.9)
II		49 (9.3)
III		211 (40)
IV		263 (50)
Ejection fraction a , %	8 (1.5)	55 [45 (50–60) 80]
Systolic blood pressure a , mmHg	7 (1.3)	148 [20 (130–169) 246]
Diastolic blood pressure, mmHg	8 (1.5)	75 [11 (64–90) 168]
Supine heart rate, bpm	7 (1.3)	80 [39 (68–100) 180]
Body mass index a , kg/m2	31 (5.8)	28 [16 (24–32) 64]
Laboratory
GFR, MDRD, mL/min/1.73 m2	6 (1.1)	61 [4 (43–76) 185]
NT‐proBNP, ng/L a	101 (19)	2448 [305 (1290–4790) 32 200]
BNP, ng/L a	438 (81)	429 [107 (229–805) 3585]
Haemoglobin, g/L	11 (2.0)	120 [5 (109–135) 206]
White blood cell count (×109)	14 (2.6)	8.3 [0.9 (6.7–10.5) 860.0]
Platelet count (×109)	16 (3.0)	224 [28 (179–283) 642]
Sodium a , mEq/L	7 (1.3)	139 [112 (137–141) 147]
Potassium, mEq/L	11 (2.0)	4.2 [2.4 (3.8–4.6) 7.5]
Comorbidities
Known heart failure prior to presentation a	4 (0.7)	216 (40)
Hypertension	1 (0.2)	419 (78)
Atrial fibrillation/flutter a	7 (1.3)	350 (66)
Coronary disease a	2 (0.4)	158 (29)
History of myocardial infarction	10 (1.9)	77 (15)
Cardiovascular syncope	13 (2.4)	17 (3.2)
Non‐cardiovascular syncope a	12 (2.2)	16 (3)
Stroke a	0 (0)	56 (10)
Peripheral arterial disease	1 (0.2)	79 (15)
Diabetes mellitus a	0 (0)	161 (30)
Smoking	23 (4.3)
Previous		179 (35)
Current		28 (5.4)
Any lung disease a	0 (0)	135 (25)
Any cancer a	0 (0)	86 (16)
Lung cancer		2 (0.4)
Colorectal cancer		7 (13)
Breast cancer		16 (3)
Prostate cancer		25 (4.6)
Cirrhosis	0 (0)	5 (0.9)
Non‐cirrhotic liver disease	0 (0)	7 (1.3)
Chronic kidney disease a (GFR < 60 mL/min/1.73 m2)	6 (1.1)	257 (48)
Anaemia a (<12 g/dL women, <13 g/dL men)	11 (2.0)	271 (51)
Medications at discharge
ACEi and/or ARB	12 (2.2)	376 (71)
β‐blocker	12 (2.2)	387 (73)
Digoxin	12 (2.2)	48 (9.1)
Potassium sparing diuretic	12 (2.2)	81 (15)
Calcium channel blocker	12 (2.2)	160 (30)
Thiazide diuretic	12 (2.2)	47 (8.9)
Loop diuretic standing and/or as needed	12 (2.2)	424 (80)
Loop diuretic standing	12 (2.2)	420 (80)
Loop diuretic as needed	12 (2.2)	12 (2.3)
Nitrate	12 (2.2)	35 (6.6)
Anti‐arrhythmic	12 (2.2)	84 (16)
Aspirin and/or other anti‐platelet	12 (2.2)	211 (40)
Anticoagulant	12 (2.2)	276 (52)
Among patients with atrial fibrillation/flutter, how many did not have anticoagulant		83 (24)
Statin	12 (2.2)	222 (42)
Insulin	12 (2.2)	74 (14)
Oral diabetes medication	12 (2.2)	82 (16)
History of interventions
Conventional pacemaker	2 (0.4)	60 (11)
CABG or PCI	3 (0.6)	102 (19)
Valve intervention	4 (0.7)	38 (7.1)

Note: Values are n (%) for categorical variables and median [min (1st and 3rd interquartile range) max] for continuous variables.

Abbreviations: ACEi, angiotensin‐converting enzyme inhibitor; ARB, angiotensin receptor blocker; BNP, brain natriuretic peptide; CABG, coronary artery bypass graft; GFR, glomerular filtration rate; IQR, inter‐quartile range; MDRD, Modification of Diet in Renal Disease equation; NT‐proBNP, N‐terminal pro‐B‐type natriuretic peptide; NYHA, New York Heart Association class; PCI, percutaneous coronary intervention.

^a Eighteen pre‐specified variables for inclusion in multivariable Cox models.

Outcomes

Patients were included between May 2007 and December 2011 and followed prospectively from the initial hospitalization to November 2012. Vital status was assessed by clinical visit, telephone contact with patient and/or family in France, or in Sweden by the Swedish National Patient Register and Population Register. Detailed CV and non‐CV cause‐specific outcomes were adjudicated by local investigators (with reporting of underlying cause rather than immediate mode of death). CV outcomes were defined as death or hospitalization due to myocardial infarction, HF, stroke, sudden cardiac death, or other CV.

Statistical analysis

Baseline data of the overall population were expressed as n (%) for categorical variables and median [inter‐quartile range (IQR)] for continuous variables.

The total numbers of each cause‐specific event were assessed and expressed as events per 100 patient‐years. Death and hospitalizations and their categories and specific causes were displayed in pie charts. Time to CV death and the composite of CV death or first CV hospitalization, with censoring at non‐CV, unknown cause of death or last follow‐up, and conversely, time to non‐CV death and the composite of non‐CV death or first non‐CV hospitalization, with censoring at CV death, unknown cause of death or last follow‐up, was assessed by Kaplan–Meier analysis.

Predictors of outcomes were assessed with univariable and multivariable Cox regression using a set of pre‐specified clinically relevant independent baseline variables (the same n = 18 variables, annotated with ^a in Table 1, for all dependent outcomes) and displayed in forest plots. Either NT‐proBNP or BNP was allowed for enrolment. NT‐proBNP was missing in 101 (19%) and BNP in 438 (81%) patients. To include both measurements, values were categorized into tertiles based on the entire cohort and composite to one variable in the Cox regressions.

Of 539 patients, one or more of these 18 variables were missing in 122 patients, leaving 417 patients in the multivariable models. CV death reduces the time exposed to risk of non‐CV death and vice versa. Therefore, consistency analyses were performed using the Fine and Gray model for competing events. ¹⁰

Multiple hospitalizations can be experienced by the same individual, inducing time dependence. Hence, the main analysis for predictors of recurrent cause‐specific hospitalizations were fitted using adjusted proportional intensity models for CV and non‐CV hospitalizations, and a consistency analysis was performed using proportional means models.

To assess the functional form of continuous covariates and checking the proportional hazards assumption for each covariate in the Cox models, we used the graphical and numerical methods from cumulative sums of Martingale residuals over follow‐up times or covariate values. ¹¹

All P values were two‐sided, and statistical significance was set at 0.05. All analyses were performed in SAS (SAS Institute, Cary, NC, USA) and RStudio version 4.0.4.

Results

Patients and baseline characteristics

Of 584 patients included in KaRen, 29 did not meet inclusion criteria, and 16 withdrew consent, leaving 539 patients for analysis. Table 1 shows selected baseline characteristics. Overall, median (IQR) age was 79 (72–84) years, and 56% were females. Most patients were in New York Heart Association (NYHA) class II in stable state prior to admission (61%) and in NYHA III (40%) and IV (50%) on admission; median (IQR) EF was 55% (50–60), with NT‐proBNP 2448 (1290–4790) /BNP 429 (229–805) ng/L, glomerular filtration rate 61 (43–76) mL/min/1.73 m² and body mass index 28 (24–32) kg/m². Most patients had hypertension (78%) and atrial fibrillation or flutter (66%); half of the patients had chronic kidney disease and anaemia; about one third had coronary disease, diabetes, history of cancer and were former smokers (Table 1).

Causes and rates of events

Among the 539 patients with 1196 patient‐years follow‐up [median (min, max) 744 days (13–1959)], there were 159 (29%) deaths [13 (95% confidence interval, CI 11–16)] per 100 patient‐years. There were 61 (38%) CV deaths 5.1 (95% CI 4.0–6.5) per 100, dominated by HF 3.9 (95% CI 3.0–5.2) per 100; and 70 (44%) non‐CV deaths 5.8 (95% CI 4.6–7.4) per 100, dominated by cancer, 2.3 (95% CI 1.5–3.3) per 100. Additional CV causes were acute myocardial infarction, sudden death and stroke, each around 0.2–0.3 deaths per 100 patient‐years. Additional non‐CV causes were pneumonia and lung disease, around 0.5–0.6 deaths per 100 patient‐years, and sepsis, osteoporotic fracture, gastrointestinal disease, kidney disease, liver disease and trauma, each around 0.1–0.4 deaths per 100 patient‐years (Figure 1A; Table S1).

Figure 1.

(A) Causes of death; (B) causes of hospitalization. Event rates (95% confidence interval) are presented per 100 patient‐years.

There were 723 hospitalizations in 338 (63%) patients (60 [95% CI 56–65] per 100 patient‐years: CV 33 (95% CI 30–37) per 100, dominated by HF 17 (95% CI 15–19) per 100; non‐CV 27 (95% CI 24–30) per 100, dominated by lung disease 4.9 (95% CI 3.8–6.4) per 100. Additional CV causes were acute myocardial infarction at 2.1 (95% CI 1.4–3.1) and stroke and cardiac syncope, each at 0.8 (95% CI 0.4–1.6) hospitalization per 100 patient‐years. Additional non‐CV causes were infection, cancer, and kidney disease, at 1.8 (95% CI 1.2–2.8), 1.4 (95% CI 0.9–2.3) and 1.0 (95% CI 0.6–1.87) hospitalization per 100‐patient‐years, respectively (Figure 1B; Table S2). There were no major differences in causes of death and hospitalization when stratified according to EF 45–49% versus ≥50%.

Time to events

In Kaplan–Meier analyses, the survival curves to CV death and to non‐CV death were similar. The survival curve free from CV death or first CV hospitalization fell steeply and then flattened out whereas the survival curve free from non‐CV death or first non‐CV hospitalization had a more uniform decline (Figure 2A–D). Overall, the long‐term rates of CV and non‐CV deaths were similar, and CV and non‐CV hospitalizations were also similar (Figure 1A–B).

Figure 2.

Kaplan–Meier survival curves fitted for time to first event (A) time to CV death; (B) time to non‐CV death; (C) time to CV death or first CV hospitalization and; (D) time to non‐CV death or first non‐CV hospitalization. CV, cardiovascular; non‐CV, non‐cardiovascular.

Predictors of cause‐specific events

In multivariable analysis, higher NYHA class (III–IV vs. I), higher natriuretic peptides, higher age and lower serum sodium were independently associated with greater risk of CV death and history of stroke, anaemia and lower serum sodium of non‐CV death. There were no apparent predictors associated with the combined outcome CV death or first CV hospitalization; however, anaemia and lower serum sodium were independently associated with higher risk of non‐CV death or first non‐CV hospitalization (Figure 3A–D).

Figure 3.

Forest plots for 18 pre‐selected variables and their adjusted association with (A) CV death; (B) non‐CV death; (C) CV death or first CV hospitalization; and (D) non‐CV death or first non‐CV hospitalization. AF, atrial fibrillation; BMI, body mass index; BNP, brain natriuretic peptide; CAD, coronary artery disease; CI, confidence interval; CKD, chronic kidney disease; HF, heart failure; HR, hazard ratio; NT‐pro‐BNP, N‐terminal pro‐B‐type natriuretic peptide; NYHA, New York Heart Association; SBP, systolic blood pressure.

In competing risk regression analysis, elevated systolic blood pressure was associated with CV death. On the other hand, natriuretic peptides and NYHA class were no longer associated with CV death. Competing risk models and variables associated with outcomes are reported in Table S4.

In proportional intensity models, chronic kidney disease, diabetes, anaemia, lung disease and coronary artery disease, but not NYHA class or natriuretic peptides, were independently associated with greater risk of CV hospitalization. History of cancer appeared to be associated with lower risk of CV hospitalization. Finally, anaemia, lung disease and stroke were independently associated with higher risk of non‐CV hospitalization. EF in the higher range appeared to be inversely associated with non‐CV hospitalization, although not statistically significant (Figure 4A–B).

Figure 4.

Adjusted fitted intensity models for 18 pre‐selected variables and their association with (A) CV hospitalization and (B) non‐CV hospitalization. AF, atrial fibrillation; BMI, body mass index; BNP, brain natriuretic peptide; CAD, coronary artery disease; CI, confidence interval; CKD, chronic kidney disease; CV, cardiovascular; HF, heart failure; HR, hazard ratio; non‐CV, non‐cardiovascular; NT‐pro‐BNP, N‐terminal pro‐B‐type natriuretic peptide; NYHA, New York Heart Association; SBP, systolic blood pressure.

Using proportional means models, only anaemia and chronic kidney disease remained independently associated with CV hospitalization. There were no differences in variables associated with non‐CV hospitalization (Table S5).

Discussion

In the present prospective observational multicentre study, we assessed the incidence and predictors of novel detailed cause‐specific death and hospitalization in HFpEF (EF ≥ 45%) among 539 patients over a median follow‐up time of 744 (13–1959) days. The major findings were: (i) the mortality rate was 13.2 per 100 patient‐years, with 5.1 for CV, 5.8 for non‐CV, and 2.3 per 100‐patient‐years for unknown causes, respectively; (ii) the total hospitalization rate was 60 per 100‐patient years, with 33 for CV and 27 hospitalizations per 100‐patient‐years for non‐CV; (iii) HF was the leading cause of CV death, followed by acute myocardial infarction, sudden death and stroke, which were much less common, and cancer the leading cause of non‐CV death, followed by pneumonia and lung disease and then multiple different and more rare causes; (iv) HF was the most common cause of CV hospitalization, followed by acute myocardial infarction, cardiac syncope and stroke, which were again much less common, and lung disease the most common cause of non‐CV hospitalizations, followed by infection, cancer and kidney disease; (v) higher age and a greater severity of HF predicted CV death, stroke and anaemia predicted non‐CV death and lower serum sodium predicted both; (vi) there were no apparent predictors associated with CV causes of death and hospitalization; however, anaemia and lower serum sodium predicted non‐CV death and hospitalization; (vii) chronic kidney disease, anaemia, diabetes, lung disease, coronary artery disease and the absence of cancer predicted recurrent CV hospitalization, whereas anaemia, a history of lung disease and stroke predicted recurrent non‐CV hospitalization.

Cause specific death in heart failure with preserved ejection fraction

The overall mortality rate was 13.2 per 100‐patient‐years, consistent with previous findings in HFpEF. ¹² , ¹³ Prognostic studies in HF have shown variable rates of cause‐specific deaths, although the proportion of non‐CV deaths appears to be higher in HFpEF compared with HFrEF. ¹⁴ We observed a CV‐death proportion of 38%, lower than the 50–60% in observational studies ⁵ , ¹² , ¹³ , ¹⁵ , ¹⁶ , ¹⁷ and around 70% in randomized controlled trials, ⁶ , ⁷ , ¹⁸ , ¹⁹ although the opposite pattern has also been observed. ²⁰

In accordance with previous reports, ⁵ , ¹⁶ our findings showed that cancer was the most common non‐CV cause of death, with infection/sepsis and lung diseases as other major causes. In some HFpEF trials, ⁶ , ⁷ , ¹⁸ the majority of cause‐specific CV deaths were due to sudden death and to HF and stroke, findings confirmed in the Minnesota Heart Survey. ²⁰ Similar to most population‐based studies, ¹⁶ , ²¹ the majority of CV deaths in our study was due to HF (77% of CV deaths). Interestingly, sudden death caused between 24% and 38% of all deaths in two previous studies ⁶ , ¹² ; however, only 7% in our cohort died from sudden death. The discrepancy might be explained by differences and uncertainty of classification of deaths especially sudden cardiac death. It may also reflect differences in the understanding of cause versus mode of death. In HFrEF, the majority of sudden death are due to ventricular arrhythmias and progressive pump failure, features not thought to be predominant in HFpEF. ²² But some are caused by bradyarrhythmia, electromechanical dissociation or non‐CV causes such as haemorrhagic stroke. Compared with HFrEF, patients with HFpEF have fewer deaths related to coronary heart disease and a higher proportion of non‐CV deaths, which might at least partially explain why improvements in prognosis are smaller in most HFpEF drug trials with CV‐specific endpoints. ¹⁸ Patients enrolled in KaRen were slightly older than patients in other HFpEF studies, ⁵ , ⁶ , ⁷ , ¹⁶ , ¹⁸ and therefore, some non‐CV comorbidities might be age‐related rather than HF‐related. Further, the age‐related non‐CV comorbidities might contribute to the high number of non‐CV deaths found in our study.

Misdiagnosis of HFpEF and lack of precision of inclusion criteria in studies may further bias the estimates of causes of death in general. In addition to clinical signs of HFpEF and elevated natriuretic peptides, multiple echocardiographic parameters are helpful to ascertain a diagnosis of HFpEF. A previous paper from the KaRen cohort showed that a majority of our patients met objective diagnostic echocardiographic criteria for HFpEF, and the presence of a high number of abnormal diastolic parameters was associated with worse prognosis. ²³ Worsening HF in HFrEF might be explained by low output, congestion, cardiogenic shock or a combination of all three. However, HF as a cause of death in HFpEF could be due to an exacerbation of other comorbidities as patients with HFpEF are more likely to have progressive ventricular dysfunction, pulmonary hypertension, and or kidney dysfunction, culminating in multi‐organ failure. ²²

Predictors of cardiovascular and non‐cardiovascular death

Higher NYHA class and natriuretic peptides were, as previously established, ¹⁴ , ²⁴ predictive of CV death, but interestingly, this association was not seen when competing events were taken into consideration.

The findings from the current analysis corroborate the known burden of non‐CV comorbidities in HFpEF ⁴ , ²⁵ and by demonstrating the important role of age and comorbidities in driving cause‐specific mortality events. Reflecting neurohormonal activation and volume overload, hyponatraemia is a well‐known predictor of adverse outcome, both in the inpatient and outpatient clinical setting. ²⁶ Patel et al. found that lower sodium, measured both as a continuous variable and with cut‐off values, were independently associated with greater risk of all‐cause mortality in patients with de novo HF diagnosis. ²⁷ In our cohort where only 40% had a HF diagnosis prior to index, we were able to capture outcomes more accurately by showing that lower sodium was associated with CV death and also non‐CV events. Additionally, these findings were confirmed in the presence of competing events.

The prevalence of anaemia varies across HF studies, but appears more common with higher EF and age. ²⁸ In the present study, patients with anaemia had more than a two‐fold risk of non‐CV events, even when censored for the presence of competing events. The prognostic role of anaemia on mortality and hospitalization has previously been investigated in observational and nationwide cohorts, ²⁹ suggesting a complex interrelation with higher age, obesity, kidney dysfunction and diabetes. Compared with HFrEF, and to some extent HF with mildly reduced EF (HFmrEF), these features are generally more common in HFpEF, and highly prevalent in the KaRen cohort. More importantly, these features are also consistent with the comorbidity–inflammation paradigm observed in HFpEF. ³⁰ Iron deficiency is the main cause of anaemia and has also been shown to impact prognosis. ³¹ However, we did not have information on the iron status of the patients included.

Prevalence and predictors of cardiovascular and non‐cardiovascular hospitalization

The risk of readmissions in HF remains high and irrespective of EF, hospitalizations have generally been reported as predominately CV with HF as main cause. ¹⁹ , ³² In our HFpEF cohort, we found that hospitalization rates were roughly similar between CV and non‐CV causes, but that CV hospitalizations occurred earlier. This is consistent with results from the I‐PRESERVE trial (54% for CV causes, 44% non‐CV causes) ³² whereas most studies have suggested significantly more non‐CV hospitalizations. ³³ In prior studies the majority of CV events are due to HF hospitalization, and non‐CV hospitalization due to lung disease, ³⁴ which was also observed in the present study. However, clinical diagnosis of HFpEF and lung‐related conditions might be challenging to separate during acute admissions, particularly in elderly patients.

A novel aspect of our analysis was the determination of prognostic predictors for recurrent cause‐specific hospitalization. History of coronary artery disease and stroke were the only CV comorbidities with a predictive power of CV and non‐CV hospitalizations, respectively. Nonetheless, the majority of comorbidities associated with recurrent hospitalizations where non‐CV. Iorio et al. addressed the impact of chronic kidney disease, diabetes, lung disease and anaemia on first all‐cause, non‐CV and HF readmissions in HFpEF, ²⁵ we extend these findings by showing their independent association with recurrent CV hospitalization over time. Additionally, lung disease and anaemia raise the risk for both hospitalization causes. Finally, treatment strategies in HF are mostly targeted towards reducing CV burden and HF exacerbation, our results raise important questions in considering non‐CV comorbidities as entry criterion and outcomes in HFpEF trials. More recently, therapies such as sodium‐glucose cotransporter 2 (SGLT2) inhibitors and angiotensin receptor neprilysin inhibitors (ARNI) have been introduced, ³⁵ , ³⁶ which have reduced the risk of adverse outcomes. If the patients in the present study had been treated with these therapies, the rates and associations with CV and non‐CV outcomes might have been different.

Study limitations

In this observational study, EF inclusion criteria was set at threshold ≥45% in accordance with previous definitions and clinical trials ⁸ ; hence, some of the patients [EF 45–49%; n = 78 (14.5%)] included in present analysis had HFmrEF according to current guidelines. ¹

The KaRen population were enrolled during acute HF presentation with a relatively broad inclusion and less restrictive exclusion criteria; therefore, findings from our cohort might not entirely reflect the overall HFpEF population. On the same note, only 40% of the KaRen population had known HF diagnosis prior to admission and given the broad inclusion criteria in the present study, patients enrolled might be more heterogenous compared with other HFpEF cohorts.

Conclusions

In this detailed assessment of adjudicated cause‐specific outcomes in HFpEF, death and hospitalization rates were roughly similar between CV and non‐CV causes. In addition, we report rates of an extensive number of specific sub‐categories of death and hospitalization. This confirms that patients with HFpEF not only have complex non‐CV comorbidities but that outcomes are also commonly non‐CV. However, even in this elderly and comorbid population, there was substantial CV and especially HF mortality and hospitalization, suggesting that therapy targeting HF remains promising also in HFpEF.

Funding

The Prospective KaRen study was supported in part by grants from Fédération Française de Cardiologie/Société Française de Cardiologie, France, and Medtronic Bakken Research Center, Maastricht, the Netherlands. L. H. L. was supported by the Swedish Research Council (grant 2013‐23897‐104604‐23), Swedish Heart Lung Foundation (grants 20080409 and 20100419) and the Stockholm County Council (grants 00556‐2009 and 20110120). C. L. was supported by the Swedish Heart Lung Foundation (grants 20080498 and 20110406) and the Stockholm County Council (grants 20090376 and 20110610). No funding agency had any role in the design and conduct of the study; in the collection, management, analysis or interpretation of the data; or in the preparation, review or approval of the manuscript.

Conflict of interest statement

There are no commercial products involved in this study. However, to the extent that findings in KaRen may affect the use of heart failure drugs or devices, we disclose the following: L. H. L receives research grants from the Swedish Research Council, the Swedish Heart Lung Foundation and the Stockholm County Council, research grants from AstraZeneca, Novartis, Boehringer Ingelheim, Vifor Pharma, Boston Scientific; consulting or speaker's honoraria from AstraZeneca, Novartis, Boehringer Ingelheim, Vifor Pharma, Bayer, Sanofi, Fresenius, Merck, Myokardia, MedScape, Radcliffe Cardiology and Lexicon; C. L. is the principal investigator of REVERSE, a CRT study sponsored by Medtronic research grants, receives research grants from Swedish Heart Lung Foundation and Stockholm County Council; speaker honoraria from Medtronic, Abbot, Microport, Boston Scientific, Novartis, Vifor, Impulse Dynamics and Bayer; E. D. (Donal) receives research facilities from General Electric Healthcare; grant from Novartis; teaching facilities provided by Bristol Myers Squibb; J. C. D. receives research grants, speaker honoraria and consulting fees from Medtronic and St Jude Medical; H. P. receives research grants from AstraZeneca; C. H. receives consulting fees from Novartis, Roche Diagnostics and AnaCardio; research grants from Bayer and speaker and honoraria from MSD and Novartis; supported by the Swedish Research Council (grant 20180899); G. S. reports grants and personal fees from Vifor, grants from Boehringer Ingelheim, grants and personal fees from AstraZeneca, personal fees from Servier, grants from Novartis, personal fees from Cytokinetics, personal fees from Medtronic, personal fees from Teva, grants from Boston Scientific, grants from PHARMACOSMOS, grants from Merck, grants from Bayer; and received funding through the Horizon Europe Programme, outside the submitted work; I. H. L receives speaker fees from AstraZeneca, Novartis and Vifor Pharma; Other authors have no conflict of interest to declare.

Supporting information

Table S1. Causes of death, proportions in percent, and rates per 100 patient‐years, of all 159 deaths.

Table S2. Causes of hospitalization, proportions in percent, and rates per 100 patient‐years, of all 723 hospitalizations.

Table S3. Univariate Cox regressions analysis for 18 pre‐selected variables and their association with CV death, non‐CV death, CV death or first CV hospitalization and non‐CV death or first non‐CV hospitalization.

Table S4. Adjusted competing risk model for 18 pre‐selected variables and their association with CV death, non‐CV death, CV death or first CV hospitalization and non‐CV death or first non‐CV hospitalization.

Table S5. Multivariable Proportional Means model for recurrent CV and non‐CV hospitalization.

Shahim, A. , Donal, E. , Hage, C. , Oger, E. , Savarese, G. , Persson, H. , Haugen‐Löfman, I. , Ennezat, P.‐V. , Sportouch‐Dukhan, C. , Drouet, E. , Daubert, J.‐C. , Linde, C. , Lund, L. H. , and the KaRen investigators (2024) Rates and predictors of cardiovascular and non‐cardiovascular outcomes in heart failure with preserved ejection fraction. ESC Heart Failure, 11: 3572–3583. 10.1002/ehf2.14928.