All procedures were conducted in accordance with the Declaration of Helsinki and its amendments. The study protocol was approved by the Institutional Review Board of Kumamoto University (approval number, Senshin 2,225). This study is registered in the University Hospital Medical Information Network (UMIN) Clinical Trials Registry (UMIN000036884). There is no written consent because we obtained the consent of the participants by using “the opt-out material.” Opt-out materials are available at the following address: https://kumadai-junnai.com/wp-content/uploads/houkatsu.pdf.

We prospectively investigated 948 consecutive patients with HF who were hospitalized in Kumamoto University Hospital between January 2007 and September 2013. We recorded each patient’s medical history and relevant clinical characteristics. We excluded 443 patients for the following reasons: severe valvular disease (n = 118), chronic renal failure requiring hemodialysis (n = 65), systemic inflammatory disease (n = 5), acute renal failure with dehydration (n = 1), and not meeting the diagnostic criteria for HFpEF as subsequently described (including HF with a reduced LVEF [HFrEF], n = 251). We subsequently calculated the HFA-PEFF score of each of these HFpEF patients, and the subjects were subdivided into low-score (0–1), intermediate-score (2–4), and high-score (5–6) groups according to the original concepts of HFA-PEFF scoring as described below. We excluded patients with insufficient data (n = 3), and we also excluded the low-score group (n = 3) because this group had a low probability of having HFpEF. Finally, the remaining 502 patients were enrolled in this study and followed for up to 1,500 days to track the occurrence of composite cardiovascular events and HF-related events. A flow chart of the study is shown in Fig. 1.

HFpEF was clinically defined according to the European Society of Cardiology task force, as follows: (1) symptoms or signs of HF; (2) normal or mildly reduced LVEF (LVEF > 50% and LV end-diastolic volume index < 97 mL/m²); (3) evidence of abnormal LV relaxation, filling, diastolic distensibility, and diastolic stiffness.

We excluded HFpEF patients who had shown even a transient reduction in EF. Hence, HFpEF patients whose LVEF was < 50% and was improved by optimal medical therapy were not included in the present study. In our study, we stratified patients by the ratio of early transmitral flow velocity to early diastolic mitral annular velocity (E/e′), classified as either a ≥ 15 ratio or a ratio between 8 and 15; we also stratified patients by plasma brain-type natriuretic peptide (BNP) levels, with a cutoff of 100 pg/mL. Physicians further confirmed that the patients had HF by determining their New York Heart Association (NYHA) functional class [15], which was assessed by the standard questionnaire while the patients were in stable condition after optimal therapy.

The calculation of the HFA-PEFF score has been described previously [11]. In brief, the score has functional, morphological, and biomarker domains. Within each domain, a major criterion is worth 2 points, and a minor criterion is worth 1 point (Supplementary Table 1).

A total score ≥ 5 points is considered to be diagnostic of HFpEF, while a score of ≤ 1 point is considered to make a diagnosis of HFpEF very unlikely and to mandate investigations for alternative causes. Patients with an intermediate score (2–4 points) need further evaluation [11]. We subsequently calculated the HFA-PEFF scores of these HFpEF patients, and the subjects were subdivided into low- (0–1), intermediate- (2–4), and high-score (5–6) groups.

The clinical parameters have been described previously [16–18].

The patients underwent echocardiography while they were in stable condition after admission; the examinations were performed by experienced cardiac sonographers who had no knowledge of the study data. The methods have been described in detail previously [9].

Early-morning fasting blood samples were obtained while the patients were in stable condition. The patients’ BNP levels were analyzed using a commercially available assay (Abbott Japan, Matsudo, Japan) in the hospital clinical laboratory on admission. The estimated glomerular filtration rate (eGFR) was calculated using the Japanese Society of Nephrology formula [19].

Patients were followed up prospectively at our outpatient clinics or by a primary care physician every month until July 2017 or until the occurrence of a cardiovascular event, including the following: cardiovascular death, hospitalization for HF decompensation, nonfatal myocardial infarction (MI), unstable angina pectoris, coronary revascularization for a new diagnosis of angina or in-stent restenosis after percutaneous coronary intervention, and nonfatal ischemic stroke. Cardiovascular death was defined as death within 30 days of a documented cardiovascular event, sudden death without apparent non-cardiovascular causes, MI, death from HF, or death from stroke. Hospitalization for HF decompensation was defined as having occurred if the patient was admitted for at least an overnight stay in the hospital because of HF with typical symptoms and had objective signs of worsening HF requiring intravenous drug administration. MI was diagnosed by an increase or decrease in cardiac biomarkers (plasma creatine kinase-MB or cardiac troponin) above the 99th percentile of the upper limit of the normal range together with evidence of myocardial ischemia and at least 1 of the following symptoms: electrocardiographic changes (new ST-T changes, left bundle branch block, or pathological Q wave), imaging evidence of new viable myocardial loss, or a new regional wall motion abnormality [20]. Unstable angina pectoris was diagnosed according to new or accelerating symptoms of myocardial ischemia accompanied by new ischemic ST-T-wave changes. Ischemic stroke was diagnosed according to a documented focal neurological deficit with radiological evidence of brain infarction, excluding intracranial hemorrhage. The primary and secondary endpoints were composite cardiovascular event (cardiovascular death, hospitalization for HF decompensation, and nonfatal stroke) occurrence and HF-related event (hospitalization for HF decompensation) occurrence, respectively, at 1,500 days. Cardiovascular events were ascertained by reviewing medical records and were confirmed by direct contact with the patients, their families, and physicians or by an annual telephone interview with each patient. An Events Committee comprising at least three independent physicians reviewed all events to avoid intraobserver biases.

Continuous variables are expressed as the mean ± standard deviation for normally distributed variables according to the Shapiro-Wilk test. Variables with a nonnormal distribution are expressed as the median value with the interquartile range. Categorical variables are presented as frequencies and percentages. Intergroup differences in categorical variables were determined using Fisher’s exact test. Differences in continuous variables were analyzed by the unpaired t test or the Mann-Whitney U test, as appropriate. Missing data were excluded from the analyses. A Kaplan-Meier curve was used to determine the cumulative incidence of composite cardiovascular events and HF-related events. The log-rank test was used to compare the incidence of these events between the two groups. The Cox proportional hazards model was used to estimate the hazard ratios (HRs) for HF-related events by univariate and multivariate analyses. We selected variables that were statistically significant in the univariate analyses (p < 0.05) and excluded variables that would cause internal correlations. The factors of plasma BNP level, E/e′ ratio, maximum velocity of tricuspid regurgitation (TRVmax) and LVMI were components of the HFA-PEFF score, and we expected that plasma BNP level, E/e′, tricuspid regurgitation pressure gradient, pulmonary artery pressure, and LVMI would cause internal correlations with the HFA-PEFF score variable. Furthermore, the estimates of Harrell’s C-statistics in the Cox proportional hazards regression models were compared after the addition of high HFA-PEFF score to conventional factors identified in the subanalysis of the Irbesartan in Patients with Heart Failure and Preserved Ejection Fraction (I-PRESERVE) trial [21,22]. The factors are age, previous hospitalization for HF, diabetes mellitus (DM), and BNP level; however, we excluded BNP level because plasma BNP is a component of the HFA-PEFF score. Therefore, we defined age, previous hospitalization for HF and DM as the 3 prognostic factors (PF3). We also assessed the incremental effects of adding a high HFA-PEFF score to the PF3 to predict HF-related events using the net reclassification improvement (NRI). The categorical NRI has been described previously [23,24]. Decision curve analysis (DCA) was used to quantify the clinical usefulness of the prediction models [25]. A p < 0.05 was considered statistically significant. Harrell’s C-statistic, NRI, and integrated discrimination improvement (IDI) were calculated by the R package PredictABEL (R Foundation for Statistical Computing, Vienna, Austria). The DCA analysis was performed using the R package ggplot2. The software SPSS version 26.0 (IBM Japan, Tokyo, Japan) was used for other statistical analyses.

A total of 502 patients with HFpEF were enrolled in this study. The numbers of patients (percentage) with intermediate and high scores were 311 (62.0%) and 191 (38.0%), respectively. The baseline characteristics of the HFpEF patients are shown in Table 1. Overall, the patients had a mean age of 71.6 ± 9.5 years, and 55% were male. The prevalence rates of DM and ischemic heart disease (IHD) were reduced in the high-score group. Plasma BNP levels, left atrial diameter, LVMI, and E/e′ were significantly increased in the high-score group. Hemoglobin and eGFR levels were significantly reduced in the high-score group. While the rate of diuretic prescription was significantly increased in the high-score group, the other drugs, including angiotensin-converting enzyme inhibitors, angiotensin II receptor blockers, calcium channel blockers, beta blockers and statins, were prescribed at similar rates in both groups.

Overall, 236 cardiovascular events were recorded during the follow-up period (median, 1,157 days). Table 2 shows the details of cardiovascular events during follow-up. We found a significantly higher rate of hospitalization for HF decompensation events in the high-score group than in the intermediate score group (p < 0.001). We also found a significantly higher rate of coronary revascularization in the high-score group than in the intermediate-score group (p = 0.001); however, other events did not differ in frequency between the two groups.

We performed a Kaplan-Meier analysis and observed that there were significant differences in the rates of composite cardiovascular events and HF events between the two groups (p = 0.003 and p < 0.001, respectively) (Fig. 2).

Table 3 shows the results of univariate and multivariate Cox proportional hazards analyses for HF-related events. Univariate Cox proportional hazards analysis identified previous hospitalization for HF (HR, 2.30; 95% confidence interval [CI], 1.51 to 3.50; p < 0.001), NYHA III or IV (HR, 2.83; 95% CI, 1.89 to 4.23; p < 0.001), presence of hypertension (HR, 0.57; 95% CI, 0.38 to 0.85; p = 0.007), presence of IHD (HR, 0.56; 95% CI, 0.38 to 0.82; p = 0.003), presence of atrial fibrillation (AF; HR, 1.76; 95% CI, 1.20 to 2.59; p = 0.004), hemoglobin level (HR, 0.85; 95% CI, 0.76 to 0.94; p = 0.001), high HFA-PEFF score (HR, 2.18; 95% CI, 1.49 to 3.19; p < 0.001), TRVmax > 2.8 m/sec, and high BNP level as significant factors associated with HF-related events. In a multivariate Cox proportional hazard analysis that included PF3 by forced-entry logistic regression (Model 1), previous hospitalization for HF (HR, 2.02; 95% CI, 1.32 to 3.10; p = 0.001) and a high HFA-PEFF score (HR, 1.98; 95% CI, 1.35 to 2.92; p = 0.001) were independently and significantly associated with HF-related events. Multivariate Cox proportional hazards analysis using the abovementioned significant factors from the univariate analysis (Model 2) identified NYHA III or IV (HR, 1.70; 95% CI, 1.07 to 2.70; p = 0.026), hemoglobin levels (HR, 0.87; 95% CI, 0.79 to 0.97; p = 0.011) and a high HFA-PEFF score (HR, 1.66; 95% CI, 1.11 to 2.50; p = 0.014) as independent predictors of HF-related events in patients with HFpEF.

Receiver operating characteristic (ROC) curves were constructed to assess the ability of the HFA-PEFF score to predict HF-related events (Fig. 3). The area under the curve of the HFA-PEFF score for the detection of HF-related events was 0.633 (95% CI, 0.574 to 0.692; p < 0.001). When a cutoff value of 4.5 was applied to HFA-PEFF scores, the sensitivity and specificity were 57.8% and 67.4%, respectively, for the detection of HF-related events.

The value of Harrell’s C-statistic for PF3 was 0.598 (95% CI, 0.539 to 0.657); after an HFA-PEFF score ≥ 5 was added as a factor, the value was 0.630 (95% CI, 0.571 to 0.689; p = 0.021). We reclassified the risk of HF-related events after adding an HFA-PEFF score ≥ 5 to PF3; the continuous NRI was 27.3% (p = 0.007), and the IDI was 4.1% (p < 0.001) (Table 4).

We analyzed the DCA of PF3 and the combined model to predict 1,500-day HF-related events in patients. Fig. 4 shows the DCA for PF3 and the combined model. Both models were useful between 0% and 40%. However, the combined model was superior to PF3 for a threshold probability of 40% to 60% (Fig. 4).

The HFA-PEFF score differs depending on whether the patient has sinus rhythm (SR) or AF. Thus, we divided the Kaplan-Meier analysis by separately assessing the prognostic relevance of this score in patients with AF and in patients with SR (Supplementary Fig. 1). The Kaplan-Meier analyses showed that, among patients with SR, the high-score group had a higher risk of composite cardiovascular events than the AF group (p = 0.015) (Supplementary Fig. 1A). Additionally, these analyses showed that there was no significant difference in composite cardiovascular events between the AF patients in the intermediate and high groups (p = 0.064) (Supplementary Fig. 1B). Additionally, they showed that the high-score group had a higher risk of HF-related events than the intermediate-score group whether the patients had SR or AF (SR, p = 0.001, Supplementary Fig. 1C; AF, p = 0.008, Supplementary Fig. 1D).

Supplementary Table 2 shows the results of univariate and multivariate Cox proportional hazards analyses for HF-related events at 365, 730, and 1,095 days. Multivariate Cox proportional hazard analyses revealed that a high HFA-PEFF score was an independent and significant predictor of HF-related events at 1,095 days but not at 365 or 730 days.