Highlights
-
•
Target trial emulation evaluated SGLT2 inhibitors in HFsnEF.
-
•
SGLT2i reduced cardiovascular death or heart failure hospitalization.
-
•
Findings support SGLT2i as a potential therapy for HFsnEF.
The efficacy of sodium–glucose cotransporter 2 inhibitors (SGLT2i) in heart failure (HF) with preserved ejection fraction has been established. However, their benefits in patients with HF with supra-normal ejection fraction (HFsnEF) remain unclear. We aimed to investigate the efficacy of SGLT2i therapy in patients with HFsnEF (left ventricular ejection fraction ≥65%). We conducted a single-center target trial emulation using observational data within an intention-to-treat framework. Eligible participants were adults aged ≥40 years with chronic HF who were not receiving SGLT2i at baseline and had evidence of structural heart disease and elevated natriuretic peptide levels. Patients who received SGLT2i were classified into the SGLT2i (+) group, whereas those who did not receive SGLT2i were classified into the SGLT2i (−) group. The primary outcome was a composite of cardiovascular death and HF hospitalization. Inverse probability of treatment weighting was used to adjust for confounders. During a median follow-up of 374 days (interquartile range: 156 to 747 days), the primary composite outcome occurred in 55 and 263 patients in the SGLT2i (+) and SGLT2i (−) groups, respectively. The weighted Cox hazard ratio for the composite outcome was 0.63 (95% confidence interval [CI]: 0.45 to 0.88, p = 0.007). For individual outcomes, the hazard ratio was 0.76 for cardiovascular death (95% CI: 0.33 to 1.72, p = 0.508), and 0.68 (95% CI: 0.48 to 0.96, p = 0.026) for HF hospitalization. SGLT2i use was associated with a significant reduction in the composite risk of cardiovascular death or HF hospitalization in patients with HFsnEF.
Heart failure (HF) affects an estimated 56 million individuals worldwide, and its prevalence continues to increase with population aging. Traditionally, HF has been classified by left ventricular ejection fraction (LVEF), with distinct etiologies and treatment strategies established for each category. Evidence of HF with reduced ejection fraction (HFrEF) is considerable, leading to the development of effective therapies. Recently, therapeutic research on HF with preserved ejection fraction (HFpEF) has advanced, particularly with the use of sodium–glucose cotransporter 2 inhibitors (SGLT2i). Based on the results of two large randomized controlled trials, DELIVER and EMPEROR-Preserved, , SGLT2i are now strongly recommended in major guidelines for the treatment of HFpEF. ,,
Emerging evidence suggests a distinct HFpEF phenotype: HF with supra-normal ejection fraction (HFsnEF). Proposed LVEF cutoffs for HFsnEF include ≥65% or ≥70%. ,, HFsnEF has been associated with poorer outcomes. , Despite recognition of this phenotype, therapeutic strategies specifically targeting HFsnEF remain unclear. In the DELIVER and EMPEROR-Preserved trials, patients with LVEF >60% were evaluated only in subgroup analyses. , Furthermore, the mavacamten trial in patients with LVEF >60% focused primarily on biomarker endpoints rather than clinical outcomes. To our knowledge, the prognostic effects of pharmacological therapies, including SGLT2i, in patients with HFsnEF have not been specifically assessed. To address this gap, we aimed to conduct a target trial emulation (TTE) study using observational data to evaluate the efficacy of SGLT2i in patients with HFsnEF (LVEF ≥65%). TTE provides a framework to emulate randomized control trials (RCTs) in real-world settings and may reduce several biases inherent in observational research.
If SGLT2i reduce adverse outcomes in HFsnEF, our findings would support extending treatment strategies similar to those used in HFpEF to this unique phenotype. Conversely, if no benefit is observed, clinicians may need to reconsider routine SGLT2i initiation in HFsnEF, given potential adverse effects and economic burden.
Methods
In this single-center observational cohort study, we applied a TTE framework. , We explicitly defined the key components of a hypothetical RCT—population, intervention, comparator, outcomes, follow-up, and analysis plan—and emulated them using real-world clinical data. We adopted a new-user design and performed the primary analysis using an intention-to-treat framework ( Table 1 ).
Table 1
Emulated target trial framework for cardiovascular death and heart failure hospitalization among patients with HFsnEF
| Protocol component | Target trial | Emulated trial |
|---|---|---|
| Aim | To evaluate the effect of initiating SGLT2i on the primary composite outcome (cardiovascular death or heart failure hospitalization) among patients with HFsnEF. | Same as for the target trial. |
| Eligibility criteria |
|
Same as for the target trial. |
| Treatment strategies |
|
|
| Treatment assignment | Random allocation to SGLT2i initiation or noninitiation, with open-label awareness of assignment. | Eligible individuals were classified at baseline (time zero) according to whether their observed treatment initiation was consistent with the prespecified initiation strategy. To emulate randomization, inverse probability of treatment weighting (IPTW) based on the propensity score was applied to balance baseline characteristics between strategies. |
| Time zero and follow-up |
From randomization until the earliest of:
|
Time zero was defined as the date of echocardiographic assessment.Follow-up started on the same day and continued until the earliest of:
|
| Outcomes |
Primary outcome: composite of cardiovascular death or heart failure hospitalization.
|
Same as for the target trial. |
| Causal contrasts | Intention-to-treat effect. | Observational analogue of an intention-to-treat–like effect, comparing treatment initiation strategies. |
| Statistical analysis |
|
The primary analysis estimated hazard ratios using an IPTW–weighted Cox proportional hazards model.
As a sensitivity analysis, inverse probability of censoring weights (IPCW) for loss to follow-up were applied, with administrative censoring at 1,500 days. |
HFsnEF was defined as LVEF ≥65% with both structural heart disease (left atrial enlargement or left ventricular hypertrophy) and elevated natriuretic peptide levels.
AF = atrial fibrillation; BNP = B-type natriuretic peptide; eGFR = estimated glomerular filtration rate; HFsnEF = heart failure with supra-normal ejection fraction; LVEF = left ventricular ejection fraction; MRA = mineralocorticoid receptor antagonist; NT-proBNP = N-terminal pro–B-type natriuretic peptide; RAASI = renin–angiotensin–aldosterone system inhibitor; SGLT2i = sodium–glucose cotransporter 2 inhibitor.
Eligible participants were adults aged ≥40 years with chronic HF treated between May 2013 and July 2024, for whom electronic health records, hospitalization records, prescription data, and laboratory results were available. The main study population comprised patients with HFsnEF, defined as LVEF ≥65%, with evidence of structural heart disease (left atrial enlargement or left ventricular hypertrophy) and elevated natriuretic peptide levels (B-type natriuretic peptide [BNP] ≥35 pg/ml in sinus rhythm or ≥105 pg/ml in atrial fibrillation/flutter; N-terminal pro–BNP ≥300 pg/ml in sinus rhythm or ≥600 pg/ml in atrial fibrillation/flutter). Both inpatients and outpatients were eligible for enrolment. ,, Patients who had received SGLT2i at baseline were excluded.
We excluded patients with end-stage renal disease requiring dialysis, pregnancy, contraindications to SGLT2i, or a history of cardiac surgery within 30 days before the index date.
The index date (time zero) was defined as the date of echocardiographic assessment, and follow-up began on the same day. The intervention strategy was initiation of an SGLT2i (dapagliflozin or empagliflozin) within a prespecified 7-day exposure window after the index date, whereas the comparator strategy was no initiation of an SGLT2i during the same 7-day window. We selected this exposure window a priori to reflect routine clinical practice and to operationalize treatment initiation within the TTE framework.
Follow-up ended at the earliest of the following: (1) occurrence of the primary composite outcome (cardiovascular death or first hospitalization for HF), (2) all-cause death, (3) end of the observation period (July 31, 2025), (4) loss to follow-up (e.g., missing data or transfer to another hospital), or (5) reaching the maximum follow-up duration of 1,500 days. We performed time-to-first-event analysis as the primary analysis.
Missing baseline covariates were handled using multiple imputation. To minimize confounding, propensity scores (PS) were estimated using logistic regression, including age, sex, body mass index, comorbidities (atrial fibrillation, ischemic heart disease, chronic kidney disease, and diabetes mellitus), HF severity markers (BNP and LVEF), renal function, and baseline medications (renin-angiotensin-aldosterone system inhibitors, angiotensin receptor–neprilysin inhibitors, beta-blockers, mineralocorticoid receptor antagonists, and diuretics). We applied inverse probability of treatment weighting (IPTW) to estimate the intention-to-treat–like effect of the initiation strategy. Covariate balance between groups was assessed using standardized mean differences, with values <0.1 considered well balanced and values between 0.1 and 0.2 considered acceptable. Effect estimates were obtained within each imputed dataset and pooled using Rubin’s rules.
This study was conducted following the Declaration of Helsinki and current ethical guidelines. The Institutional Review Board of Chutoen General Medical Center approved this study (approval number: 1313250829). Informed consent was obtained through an opt-out procedure, whereby study information was posted on the institutional website in compliance with the Act on the Protection of Personal Information of Japan. Upon request from patients or their families, the relevant cases were excluded from the study cohort.
The primary endpoint was a composite of cardiovascular death or HF hospitalization, as defined in prior studies. , HF hospitalization was diagnosed when patients presented with worsening HF symptoms accompanied by evidence of congestion on chest radiography, elevated natriuretic peptide levels, or echocardiographic findings of increased filling pressures, requiring initiation of intravenous diuretic therapy. HF hospitalization was defined as the first admission for worsening HF after the index date; any HF admissions before the index date were treated as medical history and were not counted as outcome events.
Cardiovascular death was defined as sudden cardiac death (including cardiac arrest); death due to fatal arrhythmia, myocardial infarction, or ischemia; HF-related death; death due to stroke; or any other death directly attributable to cardiovascular causes.
Outcome ascertainment was based on an integrated review of electronic health records, hospitalization summaries, prescription records, laboratory data, and echocardiographic findings rather than on diagnostic codes alone. Cardiologists reviewed medical records to exclude admissions clearly attributable to non–HF causes.
Categorical variables are expressed as counts and percentages and were compared using the chi-square test or Fisher’s exact test, as appropriate. Continuous variables were assessed for normality and analyzed using appropriate parametric tests or the Kruskal–Wallis test. Missing data were addressed with multiple imputation by chained equations: predictive mean matching for continuous variables, logistic regression for binary variables, and multinomial logistic regression for categorical variables, generating 20 imputed datasets. For each dataset, PS were estimated based on baseline covariates, and IPTW was applied. Extreme weights were trimmed at the 1st and 99th percentiles, and weighted Cox proportional hazards models were fitted. We pooled hazard ratios (HRs) with 95% confidence intervals (CIs) across imputations using Rubin’s rules. Weighted cumulative incidence curves were generated using IPTW within a representative imputed dataset, with pooled effect estimates provided in the annotations. Between-group comparisons in the figures are presented descriptively. Furthermore, an unweighted log-rank test was performed for descriptive purposes. The primary composite outcome and its components were analyzed using IPTW-adjusted weighted Cox models. We further assessed HF hospitalization using the Fine–Gray model to account for cardiovascular death as a competing risk.
Several sensitivity analyses were conducted to assess the robustness of the primary findings. First, to address potential bias due to loss to follow-up, we incorporated inverse probability of censoring weighting (IPCW) in addition to IPTW. The maximum follow-up duration was set at 1,500 days, and patients were classified as experiencing an event, loss to follow-up, or administrative censoring. Stabilized IPCW was estimated using a Cox proportional hazards model that treated loss to follow-up as the outcome and multiplied it by IPTW to generate the final analytical weights. We then re-estimated cumulative incidence and weighted Cox models using these combined weights. Second, to evaluate short-term treatment effects, follow-up was restricted to 365 days, and analyses were repeated using the same weighted Cox models. Third, to examine whether the observed associations were influenced by valvular heart disease, an additional sensitivity analysis was performed after excluding patients with aortic stenosis or aortic regurgitation at baseline. Fourth, to account for potential confounding related to temporal changes in HF management and evolving standards of care, an additional sensitivity analysis incorporating calendar time at cohort entry (index year) was conducted. Given that SGLT2i became reimbursed and widely adopted for HF treatment in Japan around 2020 to 2021, index year was categorized as pre-2021 versus 2021 or later and included in the PS model. In addition, a doubly robust analysis was performed by further adjusting for index year as a covariate in the IPTW-weighted Cox proportional hazards model. Across all sensitivity analyses, the primary estimand of interest remained the intention-to-treat–like effect of SGLT2i initiation.
A two-sided p value <0.05 was considered statistically significant. All analyses were performed using R software version 4.5.1 (R Foundation for Statistical Computing, Vienna, Austria). Statistical analyses were primarily conducted by a coauthor with biostatistics training according to a predefined analytic plan.
Results
In total, 875 patients with HFsnEF were screened, and 562 were included in the final analysis ( Figure 1 ). Among them, 129 were classified into the SGLT2i (+) group and 433 into the SGLT2i (−) group ( Table 2 ).
Flow chart. The diagram shows the selection of study participants. AF = atrial fibrillation; AFL = atrial flutter; NT-proBNP = N-terminal pro–B-type natriuretic peptide.
Table 2
Comparison of baseline characteristics of HFsnEF patients by SGLT2i initiation status before and after IPTW
| Before weighting | After weighting | |||||
|---|---|---|---|---|---|---|
| SGLT2i (+) group ( n = 129) | SGLT2i (−) group ( n = 433) | Standardized difference | SGLT2i (+) group ( n = 125) | SGLT2i (−) group ( n = 437) | Standardized difference | |
| Age, year | 79.6 ± 10.2 | 79.6 ± 11.3 | 0.001 | 79.8 ± 9.35 | 79.6 ± 11.3 | 0.013 |
| Male sex | 47 (36.4) | 197 (45.5) | 0.185 | 53 (42.2) | 197 (43.6) | 0.028 |
| Body-mass index, kg/m 2 | 21.8 ± 4.53 | 21.6 ± 4.26 | 0.174 | 21.5 ± 4.37 | 21.3 ± 4.44 | 0.050 |
| Laboratory data | ||||||
| eGFR, ml/min/1.73 m 2 | 55.5 ± 26.2 | 53.2 ± 18.3 | 0.101 | 52.0 ± 21.9 | 53.2 ± 18.6 | 0.053 |
| HbA1c, % | 6.12 ± 0.95 | 5.99 ± 0.69 | 0.150 | 6.04 ± 0.74 | 6.03 ± 0.74 | 0.006 |
| BNP, pg/ml | 305.1 ± 484.6 | 304.9 ± 352.4 | 0.001 | 297.8 ± 435.9 | 308.0 ± 363.8 | 0.024 |
| Medical history | ||||||
| Hypertension | 85 (65.9) | 287 (66.3) | 0.008 | 83 (66.5) | 291 (66.6) | 0.002 |
| Diabetes mellitus | 26 (20.2) | 75 (17.3) | 0.073 | 21 (16.5) | 80 (18.4) | 0.048 |
| Dyslipidemia | 26 (20.2) | 115 (26.6) | 0.152 | 32 (25.4) | 109 (25.1) | 0.009 |
| Atrial fibrillation | 39 (30.2) | 123 (28.4) | 0.040 | 36 (28.6) | 127 (29.0) | 0.010 |
| Prior CAD | 14 (10.9) | 66 (15.2) | 0.131 | 12 (9.1) | 61 (13.9) | 0.142 |
| Prior stroke | 11 (8.5) | 57 (13.2) | 0.150 | 15 (12.2) | 52 (12.0) | 0.007 |
| Myocardial infarction | 5 (3.9) | 29 (6.7) | 0.126 | 7 (5.3) | 26 (6.0) | 0.032 |
| Treatment at baseline | ||||||
| SGLT2i category | ||||||
| Empagliflozin | 86 (66.7) | – | – | – | – | – |
| Dapagliflozin | 43 (33.3) | – | – | – | – | – |
| DPP-4 inhibitors | 13 (10.1) | 31 (7.2) | 0.104 | 12 (9.7) | 37 (8.5) | 0.043 |
| Beta-blockers | 48 (37.2) | 148 (34.2) | 0.063 | 38 (30.6) | 151 (34.6) | 0.084 |
| CCB | 70 (54.3) | 225 (52.0) | 0.046 | 62 (49.8) | 229 (52.5) | 0.055 |
| RAASI | 52 (40.3) | 205 (47.3) | 0.142 | 52 (41.2) | 201 (46.0) | 0.096 |
| ARNI | 8 (6.2) | 14 (3.2) | 0.140 | 5 (4.3) | 16 (3.7) | 0.025 |
| Diuretics | 80 (62.0) | 253 (58.4) | 0.073 | 69 (55.5) | 258 (59.1) | 0.075 |
| MRA | 74 (57.4) | 115 (26.6) | 0.657 | 43 (34.7) | 147 (33.6) | 0.023 |
| Echocardiographic data | ||||||
| LVEF, % | 70.3 ± 3.88 | 70.1 ± 3.77 | 0.027 | 70.3 ± 4.00 | 70.2 ± 3.78 | 0.040 |
| AR > mild | 64 (51.2) | 260 (60.0) | 0.180 | 79 (63.3) | 258 (59.0) | 0.087 |
| AS > mild | 20 (15.5) | 59 (13.6) | 0.053 | 17 (13.5) | 61 (14.0) | 0.016 |
| MR > mild | 110 (85.3) | 328 (75.8) | 0.242 | 102 (81.9) | 339 (77.7) | 0.107 |
Stay updated, free articles. Join our Telegram channel
Full access? Get Clinical Tree
