Long‐term benefit of SGLT2 inhibitors to prevent heart failure hospitalization in patients with diabetes, with potential time‐varying benefit

Abstract

SGLT2 inhibitors show promise in reducing hospitalization for heart failure in diabetics, but their long‐term effects and time‐dependency remain unclear. We conducted a retrospective nested case–control study within a large type 2 diabetic cohort (n = 11,209) using electronic health records. Cases (heart failure hospitalization, n = 352) were matched to controls (n = 1372) based on age, sex, cohort entry date, and diabetes duration. Matched‐set conditional logistic regression was used to estimate hazard ratios (HRs) for antidiabetic drug class and heart failure hospitalization risk. SGLT2 inhibitors were associated with a significant reduction in heart failure hospitalization risk (adjusted HR 0.56, 95% CI 0.38–0.82, p = 0.028). This protective effect appeared more pronounced with a longer duration of treatment, suggesting a potential cumulative benefit. Time‐varying analysis within propensity score‐matched cohorts revealed a progressive decrease in hospitalization risk with continued SGLT2 inhibitor use, indicating a strengthening effect over time (greedy nearest neighbor: HR 0.52, CI 0.31–0.87, p = 0.015; optimal matching: HR 0.54, CI 0.34–0.85, p = 0.008). While promising, further investigation with larger datasets is warranted to definitively confirm these findings.

Study Highlights.

• WHAT IS THE CURRENT KNOWLEDGE ON THE TOPIC?

Prior to this study, the long‐term effects of SGLT2 inhibitors on heart failure hospitalization and the possibility of a time‐dependent benefit (increasing effectiveness with longer use) were unclear.

• WHAT QUESTION DID THIS STUDY ADDRESS?

This study addressed whether SGLT2 inhibitors, compared to other diabetes medications, reduce the risk of heart failure hospitalization in routine clinical practice, and particularly whether this effect is influenced by the duration of SGLT2 inhibitor use.

• WHAT DOES THIS STUDY ADD TO OUR KNOWLEDGE?

This study provides strong evidence for a long‐term protective effect of SGLT2 inhibitors against heart failure hospitalization, with the benefit potentially increasing with longer use.

• HOW MIGHT THIS CHANGE CLINICAL PHARMACOLOGY OR TRANSLATIONAL SCIENCE?

The progressive efficacy of SGLT2 inhibitors might necessitate modifications to clinical trial designs in clinical pharmacology to assess their long‐term benefits and potential time‐dependent effects.

INTRODUCTION

Sodium‐glucose cotransporter 2 (SGLT2) inhibitors are a class of antidiabetic medications that primarily lower blood glucose by inhibiting SGLT2 proteins in the kidneys, leading to increased urinary glucose excretion. ¹ Beyond their hypoglycemic effects, SGLT2 inhibitors offer multifaceted benefits, including cardiovascular protection, renal protection, diuresis, weight loss, and potential anti‐inflammatory properties. ¹ , ² , ³ , ⁴ , ⁵ By reducing blood volume and blood pressure, they decrease the cardiac workload, thereby reducing the risk of heart failure hospitalization and improving cardiovascular outcomes in patients with type 2 diabetes mellitus. ¹ Recent randomized controlled trials (RCTs) have shown that SGLT2 inhibitors improve heart failure with reduced ejection fraction (HFrEF), heart failure with preserved ejection fraction (HFpEF), and chronic kidney disease (CKD), as well as reduce mortality and adverse outcomes in patients with chronic heart failure (HF), regardless of diabetes status. ⁶ , ⁷ , ⁸ , ⁹ This growing body of evidence has led to the incorporation of SGLT2 inhibitors in treatment guidelines for specific HFrEF populations. ¹⁰ Multiple RCTs have consistently demonstrated that SGLT2 inhibitors significantly reduce heart failure hospitalization risk. However, translating these findings to real‐world clinical practice requires consideration of factors such as effectiveness in routine care, long‐term benefit durability, patient‐specific characteristics (e.g., comorbidities), and potential drug interactions with existing medications (polypharmacy). Additionally, uncertainties remain regarding the long‐term effects and the potential time‐dependent relationship between SGLT2 inhibitor use and heart failure outcomes. This study evaluated whether SGLT2 inhibitors, compared to other antidiabetic medications, are associated with a reduced risk of heart failure hospitalization in a real‐world setting over long‐term use. Specifically, we aimed to determine if the effectiveness of SGLT2 inhibitors in this regard is influenced by the duration of treatment.

MATERIALS AND METHODS

Data source

We obtained health care data on patients with diabetes from the Nihon University School of Medicine's Clinical Data Warehouse (NUSM's CDW), which is a centralized data repository that integrates detailed clinical information including patient demographic, diagnosis, prescription and laboratory data from the hospital information systems at three hospitals affiliated with NUSM. ¹¹ Several epidemiological studies have investigated the effects of various drugs on laboratory parameters using the CDW of the NUSM. ¹² , ¹³ , ¹⁴ , ¹⁵ The experimental protocol was approved by the Ethical Committee of NUSM, and the study was conducted in compliance with the Ethical Guidelines for Medical and Health Research Involving Human Subjects of the Ministry of Education, Culture, Sports, Science and Technology and the Ministry of Health, Labor and Welfare, Japan.

Study population

We assembled a foundational cohort consisting of 26,060 patients who had started their first‐ever prescription for a noninsulin antidiabetic drug including biguanides, sulfonylureas, thiazolidinediones, dipeptidyl peptidase 4 (DPP‐4) inhibitors, glucagon‐like peptide 1 (GLP‐1) analogues, alpha‐glucosidase inhibitors, meglitinides, and sodium‐glucose cotransporter 2 (SGLT2) inhibitors, or combinations thereof, as delineated in Table S1. This cohort was identified from January 1, 2005, to November 30, 2022, and included the entire patient population from their first to the most recent health care information available in the NUSM's CDW. The date of the initial prescription for a noninsulin antidiabetic drug served as the entry date for cohort inclusion. From this foundational cohort, we systematically excluded individuals who met the following predefined criteria: (i) <18 years of age, (ii) <2 weeks of usage for any antidiabetic drug, (iii) <365 days of continuous coverage, (iv) <30 days of follow‐up after the cohort‐entry date, (v) use of insulin prior to the cohort‐entry date, (vi) history of polycystic ovary syndrome before the cohort‐entry date, (vii) diagnosis of gestational diabetes in the year preceding cohort entry, (viii) use of anthracycline drugs before the cohort‐entry date, and (ix) use of antiretroviral medication before the cohort‐entry date. To avoid potential biases associated with known effects on heart failure, we excluded those diseases and medications with reference to the previous study by Filion et al. ¹⁶ Consequently, we established a focused study cohort consisting of 11,209 patients with diabetes who were receiving treatment with a noninsulin antidiabetic drug. A detailed flowchart outlining the identification process for the study cohort is presented in Figure 1.

FIGURE 1.

Identification of study population. The figure shows numbers of patients in the study cohort after a screening procedure (i.e., some patients were excluded for the reasons shown in the figure).

Data elements

For each participant, we gathered baseline information, including patient demographics such as age and sex, medical history, prescribed medication, and pertinent laboratory results. The comprehensive design for data collection is presented in Figure S1.

Assessment of medical history revealed discerning the presence or absence of various conditions, including heart failure (identified through the International Classification of Diseases, 10th Revision (ICD‐10) codes, specifically I50), ¹⁷ alcohol‐related disorders (F10), atrial fibrillation (I48), chronic obstructive pulmonary disease (COPD) (J44.9), ischemic heart disease (I20–I25), hyperlipidemia (E78), hypertension (I10), peripheral artery disease (I73.9), coronary revascularization (I25.10, I25.5), myocardial infarction (I21, I22), cerebrovascular disease (I60‐I69), and diabetic complications (encompassing diabetic neuropathy (E10.40), diabetic nephropathy (E11.21), diabetic retinopathy (E11.3), and diabetic arteriopathy (E11.51)). These diagnoses were established at any point preceding the cohort‐entry date. We recorded current users of medications including antihypertensive agents (angiotensin type II receptor blockers (ARBs), angiotensin‐converting enzyme (ACE) inhibitors, β‐blockers, calcium channel blockers (CCBs), thiazides, and other antidiabetic drugs), diuretics, lipid‐lowering drugs, aspirin, nonsteroidal anti‐inflammatory drugs (NSAIDs), antiarrhythmic drugs, and coronary vasodilators, defined as patients who had received these agents within the 90 days preceding the cohort‐entry date. Furthermore, blood test data, including serum total protein, glycated hemoglobin (HbA1c), low‐density lipoprotein (LDL) cholesterol, triglyceride, alanine aminotransferase (ALT), aspartate aminotransferase (AST), creatinine, blood urea nitrogen (BUN), potassium, hemoglobin level, and red blood cell (RBC) count, were collected for each individual during the 90 days leading up to the cohort‐entry date. The levels of brain natriuretic peptide (BNP) and N‐terminal pro B‐type natriuretic peptide (NT‐proBNP) were excluded from the list of collected parameters because of the limited number of patients with these data.

We employed a categorization scheme for the variables detailed below. The age groups were as follows: <60 years, 60–69 years, 70–79 years, and ≥80 years. The HbA1c level was categorized using the following thresholds: <5.6, 5.6–6.4, ≥6.5, and missing. The Cohortentry years were delineated as follows: before 2010, 2010–2014, 2015–2019, and from 2020. The estimated glomerular filtration rate (eGFR) was calculated according to the formula for Japanese individuals specified by the Japanese Society of Nephrology: eGFR (mL/min/1.73 m²) = 194*SCr^−1.094*Age^−0.287 (*0.739 if female). ¹⁸ The GFR stages were assigned as follows: G1 (eGFR ≥90 mL/min/1.73 m²), G2 (eGFR 60–89 mL/min/1.73 m²), G3a (eGFR 45–59 mL/min/1.73 m²), G3b (eGFR 30–44 mL/min/1.73 m²), G4 (eGFR 15–29 mL/min/1.73 m²), and G5 (eGFR <15 mL/min/1.73 m²).

Outcome assessment

In this study, we followed participants for the earliest event: hospitalization for heart failure (as defined below) or censoring. The follow‐up period began 30 days after cohort entry and continued until the earliest event. A 30‐day lag time was implemented to minimize reverse causality bias. Censoring events included death, withdrawal from the database, or the last access to the database (August 31, 2023). In our comprehensive analysis, which incorporated both a nested case–control methodology and a traditional case–control approach, we systematically identified patients through hospitalization episodes related to heart failure. This encompassed a wide spectrum of events, including both fatal and nonfatal cases, all of which were classified by the ICD‐10 code I50.x. For patients lacking a prior history of heart failure, cases were identified by the presence of a heart failure diagnosis (either primary or secondary) in conjunction with corresponding admission records. It is essential to note that the event definition for patients with a documented history of heart failure excluded instances where heart failure was a secondary diagnosis.

Exposure assessment

We implemented a nested case–control design to evaluate the association between the outcomes and exposure to eight classes of antidiabetic drugs, namely biguanides, sulfonylureas, thiazolidinediones, DPP‐4 inhibitors, GLP‐1 analogs, alpha‐glucosidase inhibitors, meglitinides, and SGLT2 inhibitors. Current exposure to an antidiabetic drug was defined as any prescription with a duration, along with a 30‐day grace period, encompassing the date of the outcome for both case patients and their time‐matched controls. The classification of current exposure for both case patients and controls was systematically conducted, creating two mutually exclusive categories based on the utilization of antidiabetic drugs (both alone and in combination) that either included or did not include a specific target class.

A nested case–control study

Selection of controls: To investigate the association between a specific medication and heart failure hospitalization, we employed risk‐set sampling to select up to five control patients per case in a nested case – control study. This method matches cases with control patients from the same source population. The source population included individuals free of heart failure at the time of each hospitalization (index date). Matching was based on key characteristics such as sex, age group, date of study cohort entry (±180 days), and duration of treated diabetes at the index date (±90 days), ensuring comparable groups for analysis. This approach strengthens the analysis by controlling for potential confounding variables that might influence both medication exposure and heart failure risk.

Variable selection

We initially considered 45 potential explanatory variables, spanning demographics, medical history, medications, and laboratory results. After evaluating their distributions, we excluded variables with excessive missing data (>20%) or extreme prevalence (<5% or >95%). To address multicollinearity, we removed variables identified through correlation analysis (visualized in Figure S2) and clinical judgment, resulting in a pool of 38 candidate variables. We assumed missing data were missing at random and imputed missing values 20 times using an approximate Bayesian bootstrap (ABB), obviating the necessity of assuming multivariate normality. We then employed conditional logistic regression, using stepwise, backward, and full model selection techniques on the imputed datasets: To compare model fit, we averaged the Bayesian Information Criterion (BIC) across the 20 imputation pools (Table S2). To mitigate the impact of confounding variables, our analysis primarily employed two adjustment models:

• (1)Stepwise selection model: This focused model included key clinical factors such as GFR stage, beta‐blocker and diuretic use, heart failure and atrial fibrillation history, and relevant laboratory values including total protein and hemoglobin. This was used for the primary analysis unless otherwise specified.
• (2)Full model: This comprehensive model incorporated all available and relevant baseline data, including medications, medical history, and laboratory values. While this approach maximized information, it also carried the risk of overfitting and was used for sensitivity analysis.

Statistical analysis

A nested case–control study was conducted to investigate the association between a specific drug and heart failure. Conditional logistic regression analysis with risk‐set sampling was employed. The primary analysis utilized conditional logistic regression, incorporating covariates selected through a stepwise selection process. This analysis was conducted on 100 datasets generated by ABB imputation. Odds ratios (ORs) with 95% confidence intervals (CIs) were estimated. Importantly, the estimated ORs are directly interpretable as HRs, providing a measure of the relative risk associated with drug exposure compared to non‐exposure. ¹⁹

To investigate potential interactions between drug exposure and baseline characteristics on the risk of heart failure, we conducted secondary analyses using conditional logistic regression models with interaction terms. These interaction terms assess whether the effect of drug exposure on heart failure risk varies depending on the level of each relevant baseline variable (e.g., prior heart failure history, prior myocardial infarction). The models also included the main effects of drug exposure and the baseline variables. To further explore potential effect modifications identified through interaction terms, we conducted subgroup analyses. Individuals were stratified by pre‐existing conditions potentially influencing the association, such as prior heart failure history and prior myocardial infarction history. We additionally examined whether the risk of heart failure associated with drug exposure varied by the duration of pre‐existing diabetes at the index date. Individuals were categorized into groups based on the duration of treated diabetes (<5 years and ≥5 years). The 5‐year cutoff point was chosen based on the distribution of diabetes duration in our study population.

Propensity score‐matched cohort analysis

To validate the association between SGLT2 inhibitor use and hospitalization for heart failure, we conducted a propensity score‐matched cohort analysis employing two separate cohorts described below. ²⁰ We utilized the PSMATCH procedure in SAS 9.4 to estimate the propensity score for SGLT2 inhibitor use compared to that for other antidiabetic medications. The propensity score model was a logistic regression with treatment assignment (SGLT2 inhibitor vs. other) as the dependent variable. The independent variables included baseline demographics (age, sex), cohort‐entry year, medical history, presence of diabetic microvascular complications (nephropathy, retinopathy), medications prescribed, and stage of GFR, as shown in Table S3. To create matched cohorts, we employed 1:1 ratio propensity score matching with either greedy nearest neighbor matching (caliper = 0.25) or optimal matching techniques. To account for the matched design, Cox proportional hazards models were employed to estimate the association between SGLT2 inhibitor use and heart failure hospitalization. Additionally, a time‐varying Cox model was used within the matched cohorts to assess the time‐dependent effect of SGLT2 inhibitor use on hospitalization risk. ²¹

We performed a comprehensive set of sensitivity analyses, including analyses of missing data, parameter sensitivity, and unmeasured confounders, and employed distinct datasets, to thoroughly evaluate the robustness of our findings in the field of clinical research. Regarding missing laboratory parameter data, the primary analysis employed multiple imputations based on ABB. To gauge the robustness of the primary analysis in relation to missing data, sensitivity analyses were conducted, incorporating various imputation approaches such as maximum likelihood (ML) imputation or multiple imputation with a Markov chain Monte Carlo (MCMC) method. To rigorously assess the degree of unmeasured confounding variables necessary to elucidate the observed associations, we calculated the E‐value for any exposure that exhibited a significant association. The E‐value serves as the minimum threshold for the strength of associations between a potential unmeasured confounder and both the exposure and outcome, surpassing the influence of the measured covariates. ²² , ²³ Additionally, to evaluate the potential impact of unmeasured covariates on our findings, we conducted a sensitivity analysis using Greenland's approach. ²⁴ , ²⁵ This involved specifying a range of ORs for the relationship between the outcome and an unmeasured confounder. We assumed ORs ranging from 1.5 to 7 for the outcome‐confounder relationship and from 1.5 to 4 for the confounder‐variant relationship. Utilizing these specified ranges, we computed the bias‐adjusted OR for the relationship between the outcome and the exposure. Moreover, bootstrap resampling was employed to evaluate the uncertainty or variability in HR estimates. We iteratively conducted the estimation process using 100 and 1000 bootstrap datasets, yielding confidence intervals for each estimation. Additionally, we repeated the primary analysis using the full model to assess the potential influence of overfitting. Furthermore, to explore the potential influence of GFR on our risk estimates, we conducted a sensitivity analysis using an alternative matching strategy for risk‐set sampling. This strategy included sex, age, cohort‐entry date, duration of treated diabetes, and GFR stage as matching criteria. We then repeated our primary analysis with this alternative dataset to assess the robustness of our findings.

To compare differences in baseline characteristics between groups, we employed Student's t‐tests for continuous variables and chi‐square tests for categorical variables. Statistical significance was determined at a p‐value threshold of 0.05. The primary statistical analyses were performed using SAS software (version 9.4). For E‐value calculations and HR visualizations, we utilized R (version 4.2.1) with the E‐value and casebase packages, ²² , ²⁶ respectively. A correlation heatmap was generated using Python (version 3.10.10), and Stata (version 18.0) was employed to construct the time‐varying Cox proportional hazards model. Please refer to the supplementary information for examples of code.

RESULTS

Study population

This study included 11,209 individuals with diabetes identified from NUSM's CDW. The cohort was followed for a total of 54,190 person‐years. During follow‐up, 352 patients (3.2%) were hospitalized for heart failure, corresponding to an unadjusted incidence rate of 6.5 events per 1000 person‐years. A nested case–control design was employed. Each patient with heart failure hospitalization was matched to a maximum of five controls from the cohort using risk‐set sampling, based on sex, age group, date of cohort entry, and duration of treated diabetes. This resulted in a final study population of 352 cases and 1372 controls. The detailed demographic and clinical characteristics are presented in Table 1 and Table S4. Compared to controls, patients exhibited a significantly greater prevalence of coexisting cardiovascular conditions (e.g., atrial fibrillation, ischemic heart disease, myocardial infarction, hypertension, and previous heart failure; all p < 0.05). Additionally, they were more likely to be taking medications such as ARBs, beta‐blockers, diuretics, and coronary vasodilators (all p < 0.05). Several laboratory parameters (e.g., total serum protein, creatinine, BUN, hemoglobin level, RBC count) also differed significantly between patients and controls at baseline.

TABLE 1.

Baseline characteristics of patients with diabetes who were hospitalized for heart failure (case patients) and matched controls.

Characteristics	Case patients	Controls	p‐value
	(N = 352)	(N = 1372)
Female sex – no. (%)	92 (26.1)	282 (26.1)	1.0000
Age – year
Mean ± SD	68.3 ± 0.7	67.9 ± 0.3	0.4648
Age group – no. (%)			0.6455
<60 years	71 (20.2)	304 (22.2)
60–69	111 (31.5)	448 (32.7)
70–79	105 (29.8)	400 (29.2)
≥80 years	65 (18.5)	220 (16.0)
Cohort entry year – no (%)			0.9135
<2010	172 (48.9)	390 (48.0)
2010–2014	109 (31.0)	215 (31.9)
2015–2019	48 (13.6))	96 (13.7)
≥2020	23 (6.5)	72 (6.4)
Mean duration of treated diabetes – year ± SD	5.6 ± 0.2	5.1 ± 0.1	0.9929
HbA1c (%) – no. (%)			0.8658
<5.6	9 (2.6)	39 (2.8)
5.6–6.4	52 (14.8)	204 (14.8)
≥6.5	264 (75.0)	1015 (74.4)
Missing	27 (7.7)	114 (8.1)
GFR stage (mL/min/1.73 m2) – no. (%)			0.1617
G1	37 (10.5)	212 (14.6)
G2	99 (28.1)	540 (38.5)
G3a	90 (25.6)	273 (20.7)
G3b	56 (15.9)	135 (10.2)
G4	37 (10.5)	77 (6.5)
G5	14 (4.0)	27 (2.0)
Missing	19 (5.4)	108 (7.6)
Medical history – no. (%)
Heart failure	212 (60.2)	512 (38.7)	<0.001
Alcohol‐related disorder	6 (1.7)	21 (1.4)	0.7089
Atrial fibrillation	74 (21.0)	168 (12.2)	0.0017
COPD	14 (4.0)	57 (4.6)	0.7068
Ischemic heart disease	217 (61.6)	591 (45.2)	<0.001
Hyperlipidemia	161 (45.7)	652 (48.1)	0.5291
Hypertension	301 (85.5)	995 (73.8)	0.0001
Peripheral artery disease	84 (23.9)	261 (19.7)	0.1820
Coronary revascularization	11 (3.1)	35 (2.6)	0.6701
Myocardial infarction	145 (41.2)	370 (27.4)	0.001
Cerebrovascular disease	74 (21.0)	316 (22.7)	0.5978
Diabetic complications – no. (%)
Diabetic neuropathy	5 (1.4)	27 (1.8)	0.6590
Diabetic nephropathy	71 (20.2)	230 (17.0)	0.2845
Diabetic retinopathy	67 (19.0)	218 (15.1)	0.1685
Diabetic arteriopathy	4 (1.1)	12 (0.8)	0.6689
Prescription drug use – no. (%)
Antihypertensive drug	278 (79.0)	869 (64.6)	<0.001
Alpha blocker	17 (4.8)	49 (4.2)	0.6874
ACE inhibitor	56 (15.9)	178 (11.9)	0.1241
ARB	177 (50.3)	510 (38.8)	0.0022
Beta‐blocker	135 (38.4)	315 (24.1)	<0.001
CCB	177 (50.3)	585 (44.0)	0.0970
Thiazide	16 (4.5)	56 (4.2)	0.8080
Other antihypertensive drug	35 (9.9)	76 (5.1)	0.0157
ARB with CCB combination	4 (1.1)	11 (0.7)	0.5349
ARB with thiazide combination	5 (1.4)	22 (1.5)	0.9046
Diuretic	162 (46.0)	360 (27.6)	<0.001
Lipid‐lowering drug	126 (35.8)	425 (32.3)	0.3346
Aspirin	40 (11.4)	125 (9.8)	0.5139
NSAID	137 (38.9)	506 (37.8)	0.7496
Antiarrhythmic drug	35 (9.9)	92 (7.3)	0.2188
Coronary vasodilator	120 (34.1)	293 (23.3)	0.0016

Note: Cases and controls were matched for sex, age group, year of study cohort entry, and duration of treated diabetes. Comparisons of differences in patient characteristics between groups were performed using Student's t‐test for continuous variables and the chi‐squared test for categorical data. Means and percentages in controls were first weighted by the number of controls per patient.

Abbreviations: ACE inhibitor, angiotensin‐converting enzyme inhibitor; ARB, angiotensin type II receptor blocker; CCB, calcium channel blocker; COPD, chronic obstructive pulmonary disease; GFR, glomerular filtration rate; NSAID, nonsteroidal anti‐inflammatory drug.

SGLT2 inhibitors reduced risk of hospitalization of heart failure patients

Our analysis demonstrated a substantial 43% reduction in heart failure hospitalization risk among patients treated with SGLT2 inhibitors compared to patients in the control group receiving other antidiabetic medications (adjusted HR 0.57, 95% confidence interval (CI) 0.39–0.83, p = 0.003; Figure 2). Importantly, no significant interaction was observed between baseline characteristics and SGLT2 inhibitor use (Figure 3), suggesting a broad treatment effect.

FIGURE 2.

Risk of hospitalization for heart failure and treatment with antidiabetic drugs. This figure indicates a decreased risk of heart failure hospitalization among users of SGLT2 inhibitors compared to users of other antidiabetic medications, adjusting for potential confounders in the stepwise selection model. Patients and controls were matched for sex, age group, year of cohort entry, and duration of treated diabetes. Stepwise selection model: HRs were adjusted for clinically relevant factors, including GFR stage, beta‐blocker and diuretic use, history of heart failure and atrial fibrillation, and key laboratory values (total protein and hemoglobin). CI, confidence interval; DPP‐4, dipeptidyl peptidase 4; GFR, glomerular filtration rate; GLP‐1, glucagon‐like peptide 1; HR, hazard ratio; SGLT2, sodium‐glucose cotransporter‐2.

FIGURE 3.

Effect modification of SGLT2 inhibitor use on heart failure hospitalization stratified by baseline variables. This figure examines how baseline variables (listed in the left column) influence the risk of heart failure (HF) hospitalization and whether different patient characteristics (baseline variables) influence how SGLT2 inhibitors affect the risk of heart failure hospitalization (i.e., effect modification). We employed conditional logistic regression models with interaction terms for each baseline variable and SGLT2 inhibitor use. HRs and 95% CIs are presented for the baseline variables. p‐values for interaction terms indicate whether the association between a baseline variable and HF hospitalization risk differs according to SGLT2 inhibitor use status. The models were adjusted for clinically relevant confounders, including GFR stage, beta‐blocker and diuretic use, history of heart failure and atrial fibrillation, and key laboratory values (total protein and hemoglobin). ACE inhibitor, angiotensin‐converting enzyme inhibitor; ARB, angiotensin type II receptor blocker; ALT, alanine aminotransferase; AST, aspartate aminotransferase; BUN, blood urea nitrogen; CCB, calcium channel blocker; COPD, chronic obstructive pulmonary disease; CI, confidence interval; CPK, creatine phosphokinase; GFR, glomerular filtration rate; HbA1c, hemoglobin A1c; HR, hazard ratio; LDLC, low‐density lipoprotein cholesterol; NSAID, nonsteroidal anti‐inflammatory drug; RBC, red blood cell count; SGLT, sodium‐glucose cotransporter 2.

Our nested case–control analysis revealed a statistically significant reduction in heart failure hospitalization risk associated with SGLT2 inhibitors, regardless of prior heart failure diagnosis (Table 2). Myocardial infarction (MI) history did not significantly modify the protective effect, with HR suggesting benefit in both analyses, regardless of prior MI diagnosis, although the p‐value for history of MI in the nested case–control analysis showed borderline significance (0.0604). Interestingly, the protective effect appeared more pronounced with a longer duration of diabetes. Patients treated with SGLT2 inhibitors for at least 5 years had a significant risk reduction (adjusted HR 0.56, 95% CI 0.38–0.82, p = 0.028), suggesting a potential cumulative benefit. These findings, particularly the time‐dependent effect, were corroborated in our propensity score‐matched cohort analysis described below, which strengthened the observed associations.

TABLE 2.

Subgroup analysis of heart failure hospitalization risk according to SGLT2 inhibitor use.

Category of subgroup	Case patients	Controls	HR (95% CI)	p‐value
	Number
History of heart failure (HF)
Yes	212	512	0.60 (0.38, 0.95)	0.0296
No	140	860	0.43 (0.20, 0.92)	0.0299
History of myocardial infarction (MI)
Yes	145	370	0.56 (0.30, 1.03)	0.0604
No	207	1002	0.59 (0.36, 0.99)	0.0440
Duration of treated diabetes
<5 years	232	771	0.60 (0.28, 1.29)	0.1879
≥5 years	120	601	0.56 (0.38, 0.82)	0.0028

Note: This table presents hazard ratios (HRs) for the association between SGLT2 inhibitors and the risk of hospitalization for heart failure stratified by prior heart failure history, prior myocardial infarction history, and diabetes duration. HR is a measure of how likely an individual taking an SGLT2 inhibitor is to be hospitalized for heart failure compared to those taking other antidiabetic medications. Nested case–control analyses employed conditional logistic regression with adjustment for the stepwise selection model (including key clinical factors selected through a stepwise approach; full details in Table S2).

Abbreviations: CI, confidence interval; HR, hazard ratio; HF, heart failure; MI, myocardial infarction; SGLT2, sodium‐glucose cotransporter 2.

Time‐varying effect of SGLT2 inhibitors on heart failure‐related hospitalization

We investigated whether the protective effect of SGLT2 inhibitors against heart failure‐related hospitalization changes over time. Using a Cox proportional hazards model with time‐varying covariates, we found a significant association between SGLT2 inhibitor use and reduced risk in both propensity score‐matched cohorts (greedy nearest neighbor: HR 0.43, CI 0.22–0.85, p = 0.015; optimal: HR 0.45, CI 0.26–0.77, p = 0.004; Table 3). However, a nonincreasing pattern in Schoenfeld residuals (a way to assess the proportional hazards assumption) over time suggested a potential violation (Figure S3). This indicates that the effect of SGLT2 inhibitors might change over the course of the study. The time‐varying effect analysis revealed a sustained protective effect. HR for SGLT2 inhibitor use with time‐varying covariates indicated a progressive risk reduction (greedy nearest neighbor: HR 0.52, CI 0.31–0.87, p = 0.015; optimal: HR 0.54, CI 0.34–0.85, p = 0.008). Figure 4 visually shows that HRs continually decrease over time, suggesting a strengthening protective effect with longer use of SGLT2 inhibitors. This finding warrants further investigation with larger datasets over a longer‐term period.

TABLE 3.

Time‐varying effect of SGLT2 inhibitors on heart failure‐related hospitalization.

Variable	Greedy nearest neighbor matching		Optimal matching
	HR (95% CI)	p‐value	HR (95% CI)	p‐value
Main
SGLT2I use	0.43 (0.22, 0.85)	0.015	0.45 (0.26, 0.77)	0.004
Time‐varying covariate
SGLT2I use	0.52 (0.31, 0.87)	0.015	0.54 (0.34, 0.85)	0.008

Note: Top row: HRs for SGLT2I use (average effect across study period). Bottom row: HRs for SGLT2I use interacting with natural logarithm of time (time‐varying effect).

Abbreviations: HR, Hazard ratio; CI, confidence interval; SGLT2I, sodium‐glucose cotransporter‐2 inhibitor.

FIGURE 4.

Time‐varying hazard ratio of SGLT2 inhibitors for heart failure hospitalization. This figure shows HRs with 95% CIs for the association between SGLT2 inhibitor use and the risk of heart failure hospitalization over time. HRs were estimated using a Cox proportional hazards model in two separate cohorts matched by propensity score (greedy nearest neighbor and optimal matching). CI, confidence interval; HR, hazard ratio; SGLT2 inhibitor, sodium‐glucose cotransporter‐2 inhibitor.

Sensitivity analyses

Sensitivity analyses confirmed the robustness of our main findings within the limitations of observational research (Tables S5–S11). Both the E‐value assessment and Greenland's analyses indicated that the observed associations were unlikely to be significantly influenced by unmeasured confounding factors (Tables S12 and S13).

DISCUSSION

Our study was designed to examine the effect of eight antidiabetic drug classes on the risk of hospitalization for heart failure among patients with diabetes seen in routine clinical practice. SGLT2 inhibitor use was associated with a significant 43% reduction in heart failure hospitalization risk compared to other medications. Similar results were consistent across several sensitivity analyses. Time‐varying effect analysis revealed a potentially strengthening protective effect over time.

Multiple large, well‐designed RCTs have consistently shown that SGLT2 inhibitors significantly reduce the risk of hospitalization for heart failure. Our real‐world data analysis supports these findings. RCTs such as DAPA‐HF and EMPEROR‐Reduced demonstrated the efficacy of dapagliflozin and empagliflozin, respectively, in reducing hospitalization for heart failure regardless of diabetes status in patients with heart failure and a reduced ejection fraction. ⁶ , ²⁷ Similarly, the DELIVER and EMPEROR‐Preserved trials showed that dapagliflozin and empagliflozin, respectively, reduced hospitalization for heart failure in patients with preserved ejection fraction, again with or without diabetes. ⁸ , ²⁸ VERTIS (focusing on type 2 diabetes with cardiovascular risk), DECLARE‐TIMI58 (focusing on type 2 diabetes with cardiovascular risk), EMPA‐REG (focusing on high‐risk type 2 diabetes), and CANVAS (focusing on elevated cardiovascular risk in type 2 diabetes) all supported the use of various SGLT2 inhibitors (ertugliflozin, dapagliflozin, empagliflozin, and canagliflozin) for lowering heart failure hospitalization rates. ²⁹ , ³⁰ , ³¹ , ³² These findings strongly suggest that SGLT2 inhibitors are a valuable treatment option for a broad range of patients with heart failure, including those with and without diabetes, and with varying degrees of ejection fraction. Furthermore, our subgroup analysis adds to this evidence by suggesting that SGLT2 inhibitors benefit a broad range of patients, including those with and without a history of heart failure or MI.

Our study adds to the growing body of evidence supporting the use of SGLT2 inhibitors in reducing heart failure hospitalization risk among patients with type 2 diabetes. The observed 43% reduction in risk (adjusted HR 0.57, 95% CI 0.39–0.83) aligns with these findings and suggests a substantial benefit in real‐world settings. Interestingly, our analysis using Schoenfeld residuals suggested a potential violation of the proportional hazards assumption, indicating that the effect of SGLT2 inhibitors on heart failure hospitalization risk might change over time. Time‐varying covariate analysis revealed a gradually decreasing HR with longer SGLT2 inhibitor use, suggesting a potential time‐varying effect. This finding suggests that patients with type 2 diabetes might experience a progressively greater reduction in heart failure hospitalization risk with longer duration of SGLT2 inhibitor therapy. This novel observation warrants further investigation but holds promise for understanding the long‐term benefits of this treatment strategy. This could translate to a dominant market position, particularly in chronic disease areas such as diabetes requiring long‐term treatment. These findings will contribute significantly to the growing body of evidence supporting the use of SGLT2 inhibitors for managing heart failure risk in diabetic patients, particularly in a real‐world setting. Future well‐designed RCTs with longer follow‐up periods are necessary to definitively confirm this time‐varying effect and determine the cause‐and‐effect relationship.

SGLT2 inhibitors offer a wide range of benefits beyond glycemic control, including cardiovascular, renal, and metabolic effects. ¹ , ² , ³ , ³³ SGLT2 inhibitors have demonstrated significant reductions in major cardiovascular events, heart failure hospitalizations, and the progression of chronic kidney disease. Despite their established cardiorenal benefits, SGLT2 inhibitors remain underutilized, particularly in patients with chronic kidney disease. ³⁴ Our study directly addresses this gap by providing evidence of the long‐term benefits of SGLT2 inhibitors in preventing heart failure hospitalizations. Our findings demonstrated a significant reduction in heart failure hospitalizations among patients with type 2 diabetes treated with SGLT2 inhibitors, further reinforcing their status as a valuable therapeutic option and potentially contributing to their wider adoption. Moreover, SGLT2 inhibitors have been shown to improve other outcomes, such as reducing mortality and enhancing quality of life, ³¹ , ³⁵ further highlighting their crucial role in the management of type 2 diabetes. Additionally, emerging evidence suggests that SGLT2 inhibitors may have anti‐fibrotic properties, ³⁶ expanding their potential applications to fibrotic diseases. Our real‐world data analysis provides a solid foundation for exploring the potential anti‐fibrotic effects of SGLT2 inhibitors. We will continue to investigate this in future research.

This study has several strengths that solidify the credibility and significance of our findings on the beneficial effect of SGLT2 inhibitors on heart failure hospitalization risk. First, this study included over 11,000 individuals with diabetes, providing strong statistical power. The study achieved over 54,000 person‐years of follow‐up, allowing robust assessment of long‐term effects. A nested case–control design with risk‐set sampling mitigated potential confounding factors by matching patients with controls based on relevant characteristics. We further addressed confounding factors by employing propensity score matching and performing a comprehensive set of sensitivity analyses. Propensity score matching balanced the groups for factors that might influence the outcome, while the sensitivity analyses assessed the robustness of the findings in relation to unmeasured confounding variables. Additionally, the report presents detailed information on demographics, clinical characteristics (medical history and prescribed medication), and laboratory values, allowing a thorough exploration of potential explanations for the findings.

Our study has several limitations. Despite the application of propensity score matching and adjustments, our observational study remains subject to the inherent limitations of unmeasured confounding and potential selection bias. Residual confounding by unmeasured factors, such as socioeconomic status or genetic variations, cannot be entirely ruled out. Existing literature highlights disparities in SGLT2 inhibitor use among patients with type 2 diabetes across diverse populations, including racial, ethnic, and socioeconomic groups. ³⁴ , ³⁷ , ³⁸ , ³⁹ These disparities may contribute to variations in health outcomes. Although stratified analysis is often employed to address confounding, the homogeneity of our Japanese study population and the universal coverage provided by the National Health Insurance system may limit the influence of unmeasured factors like socioeconomic status and genetic variations. Although sensitivity analyses were conducted to assess the robustness of our findings, the generalizability of our results may be limited due to the homogeneity of our study population and reliance on a single‐healthcare database. Future research including data from diverse populations and multiple healthcare systems is recommended to enhance the generalizability of these findings. Overall, our study highlights the potential benefits of SGLT2 inhibitors in reducing heart failure hospitalization risk for patients with type 2 diabetes, with a possible time‐varying effect requiring further investigation. Future well‐designed RCTs with extended follow‐up periods are essential to confirm the time‐varying effect of SGLT2 inhibitors. These studies should collect detailed data on the type and dosage of SGLT2 inhibitors used, as well as individual patient characteristics, to inform optimal treatment strategies for high‐risk populations. Additionally, conducting more specific mechanistic studies with appropriate biomarkers could help us better predict which patients are most likely to benefit from these drugs.

CONCLUSION

This nested case–control study within a diabetic cohort provides strong evidence that SGLT2 inhibitors are associated with a significant reduction in heart failure hospitalization risk. This finding was consistent across a propensity score‐matched cohort analysis and sensitivity analyses. However, the study also suggested a potential time‐dependent effect, with the protective effect of SGLT2 inhibitors on heart failure hospitalization risk potentially strengthening over time. This finding, which has significant implications for treatment strategies and patient outcomes, requires further investigation in future studies with longer follow‐up periods.

AUTHOR CONTRIBUTIONS

Y.T. and S.A. designed the research; K.M. and H.A. performed the research; Y.T., T.H., and T.N. analyzed the data; Y.T. wrote the manuscript.

FUNDING INFORMATION

No funding was received for this work.

CONFLICT OF INTEREST STATEMENT

The authors declared no competing interests for this work.

Supporting information

Data S1:

Takahashi Y, Minagawa K, Nagashima T, Hayakawa T, Akimoto H, Asai S. Long‐term benefit of SGLT2 inhibitors to prevent heart failure hospitalization in patients with diabetes, with potential time‐varying benefit. Clin Transl Sci. 2024;17:e70088. doi: 10.1111/cts.70088