Articles

Efficacy and safety of dapagliflozin in patients with heart

failure with mildly reduced or preserved ejection fraction by
baseline glycaemic status (DELIVER): a subgroup analysis
from an international, multicentre, double-blind,
randomised, placebo-controlled trial
Silvio E Inzucchi, Brian L Claggett, Muthiah Vaduganathan, Akshay S Desai, Pardeep S Jhund, Rudolf A de Boer, Adrian F Hernandez,
Mikhail N Kosiborod, Carolyn S P Lam, Felipe Martinez, Sanjiv J Shah, Subodh Verma, Yaling Han, Jose F Kerr Saraiva, Olof Bengtsson,
Magnus Petersson, Anna Maria Langkilde, John J V McMurray, Scott D Solomon

Summary
Background Type 2 diabetes and prediabetes are risk factors for heart failure and adverse heart failure outcomes. Lancet Diabetes Endocrinol
The Dapagliflozin Evaluation to Improve the Lives of Patients with Preserved Ejection Fraction Heart Failure 2022; 10: 869–81

(DELIVER) trial showed that dapagliflozin was associated with a reduction in the primary outcome of worsening heart Published Online
November 10, 2022
failure or cardiovascular mortality in patients with heart failure with mildly reduced or preserved ejection fraction.
https://doi.org/10.1016/
We aimed to assess the efficacy and safety of oral dapagliflozin in these patients by their baseline glycaemia categories. S2213-8587(22)00308-4
See Comment page 831
Methods DELIVER was an international, multicentre, double-blind, randomised, placebo-controlled trial done in Section of Endocrinology and
350 health-care centres and hospitals across 20 countries. Patients aged 40 years or older with New York Heart Metabolism, Yale School of
Association class II–IV, left ventricular ejection fraction of more than 40%, elevated natriuretic peptides (N-terminal Medicine, and Yale New Haven
pro B-type natriuretic peptide ≥300 pg/mL or ≥600 pg/mL for patients in atrial fibrillation or flutter), and evidence of Hospital, New Haven, CT, USA
(Prof S E Inzucchi MD);
structural heart disease were randomly assigned (1:1) to 10 mg dapagliflozin or placebo, administered orally, and Cardiovascular Division,
followed up for a median of 2∙3 years (IQR 1∙7–2∙8). The primary outcome, a composite of time from randomisation Brigham and Women’s
to first worsening heart failure events (defined as an unplanned hospitalisation or urgent heart failure visit requiring Hospital, Harvard Medical
intravenous therapy) or cardiovascular death, in participants with type 2 diabetes (history of or identified by School, Boston, MA, USA
(B L Claggett PhD,
HbA1c ≥6∙5% [48 mmol/mol] at baseline) or prediabetes (HbA1c 5∙7 to <6·5% [39 mmol/mol to <48 mmol/mol] at M Vaduganathan MD,
baseline) was compared with those with normoglycaemia (HbA1c <5∙7% [39 mmol/mol]). Efficacy of dapagliflozin A S Desai MD,
versus placebo was assessed according to glycaemic status and based on HbA1c as a continuous measure. The full- Prof S D Solomon MD); BHF
Glasgow Cardiovascular
analysis set comprised all patients who were randomly assigned to study treatment, with patients analysed according
Research Centre, University of
to their randomised treatment assignment, irrespective of the treatment received (ie, intention to treat). The safety Glasgow, Glasgow, UK
analysis set comprised patients who were randomly assigned to study treatment and who took at least one dose of (Prof P S Jhund PhD,
investigational product, with patients analysed according to the treatment actually received. This trial is registered Prof J J V McMurray MD);
Department of Cardiology,
with ClinicalTrials.gov, NCT03619213.
Erasmus Medical Center,
Rotterdam, Netherlands
Findings Between Sept 1, 2018, and Jan 18, 2021, 6263 patients were randomly assigned to oral dapagliflozin (n=3131) or (Prof R A de Boer MD); Duke
placebo (n=3132). Of these patients, 1175 had normoglycaemia, 1934 had prediabetes, and 3150 had type 2 diabetes and Clinical Research Institute,
Duke University Medical
were included in the glycaemia subgroup analysis (3515 [56·2%] of 6263 patients were men and 4435 [70·9%] were
Center, Durham, NC, USA
White). The incidence rate of the primary outcome was 6∙9 per 100 patient-years in the normoglycaemia subgroup (Prof A F Hernandez MD); Saint
(reference), increasing to 7∙6 per 100 patient-years in the prediabetes subgroup (hazard ratio 1∙09 [95% CI 0∙90–1∙31]) Luke’s Mid America Heart
and 10∙1 per 100 patient-years in the type 2 diabetes subgroup (1∙46 [1∙24–1∙73]; p<0·0001 for trend). Dapagliflozin Institute and University of
Missouri-Kansas City, Kansas
reduced the risk of the primary outcome versus placebo in each subgroup (hazard ratio 0∙77 [95% CI 0∙57–1∙04], log-rank
City, MO, USA
p=0·088, for patients with normoglycaemia, 0∙87 [0∙69–1∙08], log-rank p=0·21, for patients with prediabetes, and 0∙81 (Prof M N Kosiborod MD);
[0∙69–0∙95], log-rank p=0·0077, for patients with type 2 diabetes; pinteraction=0∙82) and across the continuous HbA1c range National Heart Centre
(pinteraction=0∙85). Volume-related or renal serious adverse events or adverse events leading to discontinuation of the study Singapore and Duke–National
University of Singapore,
drug, hypoglycaemia, and amputations were not differentially affected by treatment in any of the glycaemia categories. Singapore
(Prof C S P Lam MBBS);
Interpretation In patients with heart failure with mildly reduced or preserved ejection fraction, oral dapagliflozin improved Department of Cardiology,
heart failure outcomes to a similar extent in three glycaemia subgroups: normoglycaemia, prediabetes, and type 2 diabetes. University of Cordoba,
Cordoba, Argentina
Moreover, the heart failure benefits of dapagliflozin seem to be consistent across a continuous glycaemic range. (Prof F Martinez MD); Bluhm
Cardiovascular Institute,
Funding AstraZeneca. Northwestern University
Feinberg School of Medicine,
Chicago, IL, USA
Copyright © 2022 Elsevier Ltd. All rights reserved.

www.thelancet.com/diabetes-endocrinology Vol 10 December 2022 869

 Articles

(Prof S J Shah MD); Division of

Cardiac Surgery, St Michael’s Research in context
Hospital, University of Toronto,
Toronto, ON, Canada Evidence before this study Patients with Preserved Ejection Fraction Heart Failure
(Prof S Verma PhD); We searched PubMed for research articles published in English (DELIVER) trial, dapagliflozin reduced the risk of worsening
Cardiovascular Research between Jan 1, 2012, and Sept 1, 2022, using the search terms heart failure or cardiovascular death by 18% in 6263 patients
Institute, Department of
Cardiology, General Hospital of
“SGLT2 inhibitor”, “heart failure”, and “randomised controlled with heart failure and mildly reduced or preserved ejection
Northern Theater Command, clinical trial”. In large cardiovascular outcome trials involving fraction. In a follow-up meta-analysis of DAPA-HF and
Shenyang, China patients with type 2 diabetes at high cardiovascular risk, DELIVER, the effects of dapagliflozin were consistent across
(Prof Y Han PhD); Clinical SGLT2 inhibitors, originally developed as glucose-lowering the range of LVEFs.
Research Institute of Campinas,
medications, have been shown to reduce hospitalisations for
Sao Paulo, Brazil Added value of this study
(Prof J F Kerr Saraiva PhD); heart failure. The Dapagliflozin and Prevention of Adverse
Our results show that the heart failure benefits of dapagliflozin
AstraZeneca, Gothenburg, Outcomes in Heart Failure (DAPA-HF) and EMPEROR-Reduced
Sweden (O Bengtsson Ph Lic, in DELIVER were consistent across three glycaemia subgroups:
trials subsequently showed that the SGLT2 inhibitors
M Petersson PhD, normoglycaemia, prediabetes, and type 2 diabetes and, within
dapagliflozin and empagliflozin, respectively, reduced the risk
A M Langkilde PhD) these subgroups, along the trial’s entire range of LVEF values. In
of heart failure hospitalisation or cardiovascular death in
Correspondence to: the type 2 diabetes subgroup, the effects of dapagliflozin were
Prof Silvio E Inzucchi, Section of individuals with heart failure with a left ventricular ejection
also similar in patients using versus those not using metformin
Endocrinology and Metabolism, fraction (LVEF) of 40% or less. The EMPEROR-Preserved trial
Yale School of Medicine, New or insulin at baseline, those with longer versus those with
(using empagliflozin) extended these findings to heart failure
Haven, CT 06520-8020, USA shorter duration of diabetes, and those with higher versus
with a LVEF higher than 40%, a category of heart failure
[email protected] those with lower HBA1c levels at baseline.
highly prevalent in type 2 diabetes populations and for which
there were no previous clear, evidence-based therapies. Implications of all the available evidence
Notably, the benefits of SGLT2 inhibitors were consistent in In patients with heart failure with mildly reduced or preserved
patients with and those without type 2 diabetes in all three ejection, dapagliflozin imparts substantial cardiovascular
trials. In the Dapagliflozin Evaluation to Improve the Lives of benefits across a broad range of glycaemia.

Introduction showed clear efficacy in reducing adverse heart failure

The prevalence of heart failure, particularly heart failure outcomes, including in patients with reduced and those
with preserved ejection fraction, is higher in people with with preserved ejection fraction.5–8 As a result, SGLT2
type 2 diabetes than in the general population.1 Heart inhibitors are now considered foundational therapy in
failure with preserved ejection fraction and diabetes in heart failure, endorsed by several guideline statements
fact share several risk factors, including obesity, insulin from both international cardiology9,10 and diabetology11
resistance, hypertension, and coronary artery disease. professional societies and national regulatory agencies.
Patients with heart failure with preserved ejection fraction Dapagliflozin was shown to be highly efficacious and safe
and diabetes are also at an increased risk of adverse in both patients with heart failure with reduced ejection
outcomes compared with those without diabetes.1 fraction in the Dapagliflozin and Prevention of Adverse
Although more stringent control of blood glucose Outcomes in Heart Failure (DAPA-HF) trial5 and, more
alone in these individuals does not seem to improve recently, patients with heart failure with preserved
heart failure outcomes,2 epidemiological studies have ejection fraction in the Dapagliflozin Evaluation to
shown a clear relationship between HbA1c levels and the Improve the Lives of Patients with Preserved Ejection
incidence of heart failure as well as hospitalisations and Fraction Heart Failure (DELIVER) trial, which additionally
mortality in those with prevalent heart failure.3 These included patients with mildly reduced ejection fraction.8
trends seem to extend to more mildly elevated ranges of How well the observed benefits of dapagliflozin might
HbA1c that are not yet at the diagnostic threshold for apply to patients who have heart failure with preserved
diabetes. ejection fraction or those who have heart failure with
SGLT2 inhibitors, originally developed as oral glucose- mildly reduced ejection fraction across glycaemia
lowering medications for type 2 diabetes, were sub­ categories has not been previously reported. The benefits
sequently found to have unexpected benefits in reducing of dapagliflozin could be important for people with
heart failure hospitalisations. These data emerged from diabetes and those with prediabetes, given their higher
large cardiovascular outcome trials, which had been risk of adverse cardiac outcomes, but also in individuals
assembled to meet regulatory requirements for the with normoglycaemia who, although at relatively lower
demonstration of cardiovascular safety of newer risk, would not be expected to have any glucose-lowering
antihyperglycaemic therapies, involving patients with effect from an SGLT2 inhibitor.
type 2 diabetes who were at high cardiovascular risk.4 In the DELIVER trial, oral dapagliflozin was compared
Such findings quickly led to dedicated large, randomised with placebo in 6263 patients with heart failure and
heart failure trials, including both participants with and mildly reduced ejection fraction or heart failure with
those without diabetes, in which SGLT2 inhibitors preserved ejection fraction, defined as having a clinical

870 www.thelancet.com/diabetes-endocrinology Vol 10 December 2022

 Articles

diagnosis of heart failure with a left ventricular ejection from criteria of the American Diabetes Association:14
fraction (LVEF) of more than 40%.8 The main findings patients with normoglycaemia (no history of diabetes and
from DELIVER have been previously reported, with baseline HbA1c <5∙7% [39 mmol/mol]); those with
dapagliflozin reducing the risk of the primary outcome prediabetes (no history of diabetes and baseline HbA1c 5∙7
(worsening heart failure or cardiovascular death) to <6∙5% [39 mmol/mol to <48 mmol/mol]; and those
by 18% (hazard ratio [HR] 0∙82 [95% CI 0∙73 to 0∙92], with type 2 diabetes (history of or prevalent use of
p<0∙001). Reflecting the known increased prevalence of a glucose-lowering agent [unless specifically prescribed
diabetes in patients with heart failure—especially heart for an indication other than diabetes]) or baseline HbA1c
failure with preserved ejection fraction12—more than half ≥6∙5% [48 mmol/mol]). The protocol (appendix 1
of the DELIVER study population had either a history pp 93–199) was approved by institutional review boards or
of type 2 diabetes or had been newly diagnosed with ethics committees at each study site, and every patient
type 2 diabetes on the basis of an elevated HbA1c at provided written informed consent.
enrolment. Among the no diabetes group, nearly
two-thirds had evidence of prediabetes based on Procedures
HbA1c levels (comprising nearly a third of the entire trial After informed consent and a 21-day screening period,
cohort). With diabetes status being a prespecified patients were randomly assigned (1:1) to 10 mg
subgroup in DELIVER, we took this opportunity to dapagliflozin or matching placebo, taken orally once
conduct an in-depth analysis of overall event rates and the daily. Randomisation was stratified by type 2 diabetes
efficacy and safety of dapagliflozin in the population of status at baseline. Concomitant medical treatment
trial participants with various stages of dysglycaemia, and for diabetes, hypertension, dyslipidaemia, and other
compared them with those in the normoglycaemia comorbidities, including heart failure management,
subgroup. Our primary objective was to determine was recommended according to the local standard of
whether there was any differential effect on the primary care. Following randomisation, study visits took place at
outcome between baseline glycaemia categories. 30 days, 120 days, 240 days, 360 days, and 480 days after
randomisation, and then every 120 days thereafter until
Methods the final study visit. Because of the intervening
Study design and patients COVID-19 pandemic, several amendments needed to be
The design and baseline characteristics of the study made during conduct of the trial to ensure data quality
population in the DELIVER trial have been previously and integrity. These adaptations included the
described.13 Briefly, DELIVER, an international, ran­ conversion, when necessary, of in-person to virtual
domised, double-blind, placebo-controlled trial, was study site visits, remote data collection for patient-
done in 350 health-care centres and hospitals across reported outcomes, and the collection and reporting
20 countries. Eligibility criteria included either ambu­ of all COVID-19-related adverse events and adjudication
latory or hospitalised patients aged 40 years or older who of COVID-19-related hospitalisations and deaths.
had an LVEF of more than 40%, a New York Heart A sensitivity analysis was performed with patients being
Association functional class II–IV, evidence of structural censored at the time of a COVID-19 diagnosis; we then
heart disease (such as left atrial enlargement or left assessed for competing risk for non-cardiovascular
ventricular hypertrophy), and elevation in natriuretic death. This did not change the primary results of the
peptides (N-terminal pro B-type natriuretic peptide DELIVER trial. 8
≥300 pg/mL or ≥600 pg/mL for patients in atrial
fibrillation or flutter). The trial recruited patients Outcomes
irrespective of their glycaemic status—ie, both those with All prespecified study outcomes were adjudicated by
and those without type 2 diabetes were eligible. Key a Clinical Events Committee, which was masked to
exclusion criteria included probable alternative diagnoses treatment assignment. The primary outcome was the
potentially accounting for the patients’ symptoms; composite of time from randomisation to first worsening
uncontrolled hypertension (systolic blood pressure heart failure event (defined as an unplanned hospitalisation
≥160 mm Hg if not on three or more antihypertensive or urgent heart failure visit requiring intravenous therapy)
medications, or ≥180 mm Hg regardless of number of or cardiovascular death. Key secondary outcomes included
medications); an estimated glomerular filtration rate total number of worsening heart failure events (ie, first
(eGFR) of less than 25 mL/min per 1∙73 m²; type 1 diabetes and subsequent hospitalisations for heart failure or urgent
(owing to the known risk of diabetic ketoacidosis); and heart failure visits or cardiovascular death), cardiovascular
treatment with any SGLT2 inhibitor within 4 weeks of deaths, and all-cause deaths. Other prespecified outcomes
randomisation or previous intolerance to an SGLT2 included time from randomisation to first worsening
inhibitor. For purposes of this analysis, the trial population heart failure event, time from randomisation to first
was divided as prespecified in the trial’s statistical analysis hospitalisation for heart failure, and time from ran­
plan (appendix 1 pp 225–293)13 into the following domisation to all-cause mortality. In light of the previous See Online for appendix 1
categories based on glycaemic status at baseline, derived extensive safety data regarding dapagliflozin, only

www.thelancet.com/diabetes-endocrinology Vol 10 December 2022 871

 Articles

Additionally, events related to major hypoglycaemic events,

Normoglycaemia Prediabetes Type 2 diabetes
subgroup (n=1175) subgroup (n=1934) subgroup (n=3150) diabetic ketoacidosis, and amputation were collected.
Accordingly, we did not collect and cannot report other
Age, years 71·3 (10·5) 73·0 (9·5) 71·0 (9·1)
non-serious adverse events, such as genital mycotic
Diabetes duration, years NA NA 9·2 (3·8–17·0)
infections, which are already known to be increased by
Sex
SGLT2 inhibition.
Men 660 (56·2%) 1037 (53·6%) 1818 (57·7%)
Women 515 (43·8%) 897 (46·4%) 1332 (42·3%)
Statistical analysis
Race
Details of the DELIVER statistical analysis plan have
White 793 (67·5%) 1364 (70·5%) 2278 (72·3%)
been previously published,8,13 and the main analysis
Asian 265 (22·6%) 422 (21·8%) 587 (18·6%)
reported in this Article was prespecified (see the
Black or African American 29 (2·5%) 37 (1·9%) 93 (3·0%) academic statistical analysis plan, version 1.3, in
American Indian or Alaska Native 33 (2·8%) 57 (2·9%) 99 (3·1%) appendix 1 pp 294−303). Baseline characteristics were
Other 55 (4·7%) 54 (2·8%) 93 (3·0%) compared between the three glycaemia subgroups
Ethnicity using ANOVA for continuous variables and χ² or Fisher
Hispanic 237 (20·2%) 329 (17·0%) 662 (21·0%) exact tests for categorical variables. For the overall trial,
Geographical region a sample size of 6100 participants followed up for
Asia 262 (22·3%) 412 (21·3%) 552 (17·5%) a minimum of 13∙5 months and a maximum of
Europe and Saudi Arabia 528 (44·9%) 933 (48·2%) 1543 (49·0%) 39 months was estimated to provide 1117 events. This
Latin America 228 (19·4%) 313 (16·2%) 638 (20·3%) resulted in 93% power to detect a 20% relative risk
North America 157 (13·4%) 276 (14·3%) 417 (13·2%) reduction for the primary endpoint with a two-sided α
Medical history value of 0∙024; the HR of 0·80 was prespecified and
Type 2 diabetes 0 0 2806 (89·1%) considered to be a relevant target treatment effect.
Myocardial infarction 234 (19·9%) 456 (23·6%) 949 (30·1%) However, the current subgroup analyses had no formal
Any coronary artery disease 462 (39·3%) 882 (45·6%) 1819 (57·7%) statistical consid­erations. The primary and secondary
Any atherosclerotic cardiovascular 540 (46·0%) 997 (51·6%) 2014 (63·9%) event-based objectives were evaluated under the
disease treatment policy estimand, including differences in
Previous heart failure hospitalisation 446 (38·0%) 764 (39·5%) 1327 (42·1%) outcomes over the entire study period until the primary
Atrial fibrillation or flutter 688 (58·6%) 1198 (61·9%) 1663 (52·8%) analysis censoring date to reflect the effect of the initial
Stroke 95 (8·1%) 158 (8·2%) 344 (10·9%) treatment assignment, irrespective of exposure to
Dyslipidaemia 648 (55·1%) 1107 (57·2%) 2233 (70·9%) concomitant treatment with or subsequent treatment
Smoking status discontinuation of the investigational product. The
Current 97 (8·3%) 153 (7·9%) 234 (7·4%) analysis was done in the full-analysis set and comprised
Former 389 (33·1%) 697 (36·0%) 1172 (37·2%) all events that occurred on or before the primary analysis
Never 689 (58·6%) 1084 (56·0%) 1744 (55·4%) censoring date, including those following premature
BMI, kg/m² 28·4 (5·7) 29·0 (6·0) 30·9 (6·2) discon­tinuation of the investigational product. The full-
BMI group, kg/m² analysis set comprised all patients who were randomly
<18·5 (underweight) 15 (1·3%) 23 (1·2%) 16 (0·5%) assigned to study treatment, irrespective of their
18·5–24·9 (normal weight) 351 (29·9%) 492 (25·5%) 500 (15·9%) protocol adherence and continued participation in the
25·0–29·9 (overweight) 399 (34·0%) 680 (35·2%) 993 (31·5%) study. Patients were analysed according to their
30·0–34·9 (class I obesity) 264 (22·5%) 446 (23·1%) 863 (27·4%) randomised treatment assignment, irrespective of the
35·0–39·9 (class II obesity) 101 (8·6%) 187 (9·7%) 508 (16·1%) treatment received (ie, intention to treat). The full-
≥40 (class III obesity) 45 (3·8%) 102 (5·3%) 268 (8·5%) analysis set was therefore the primary analysis set for
Time since heart failure diagnosis the intention-to-treat analysis of primary and secondary
0–3 months 108 (9·2%) 187 (9·7%) 273 (8·7%) variables. The safety analysis set comprised patients
>3 to 6 months 123 (10·5%%) 168 (8·7%) 301 (9·6%)
who were randomly assigned to study treatment and
>6 to 12 months 175 (14·9%) 281 (14·5%) 386 (12·3%)
who took at least one dose of investigational product,
>1 to 2 years 178 (15·2%) 310 (16·0%) 505 (16·0%)
with patients analysed according to the treatment
>2 to 5 years 259 (22·1%) 511 (26·4%) 798 (25·3%)
actually received. Safety events included those on or
before the last dose of study drug plus 30 days. The
>5 years 330 (28·1%) 475 (24·6%) 886 (28·1%)
safety analysis set was the primary analysis set for all
(Table 1 continues on next page)
safety outcomes. For these analyses, time-to-event data
were evaluated using Kaplan-Meier estimates and Cox
proportional hazards models, with the treatment group
collection of serious adverse events and adverse events assignment as a fixed-effect factor. Event rates per
leading to discontinuation of study drug were required; 100 patient-years, adjusted for age, sex, and region, were
these included volume depletion and renal events. further assessed across the HbA1c range (as a continuous

872 www.thelancet.com/diabetes-endocrinology Vol 10 December 2022

 Articles

variable) by Poisson regression using restricted cubic

Normoglycaemia Prediabetes Type 2 diabetes
splines with knot placement at the 10th, 50th, and subgroup (n=1175) subgroup (n=1934) subgroup (n=3150)
90th percentiles. A similar technique was used to assess
(Continued from previous page)
event rates across the LVEF range in each of the
Baseline New York Heart Association class
glycaemia subgroups. As in the primary DELIVER
I 1 (0·1%) 0 0
analysis, treatment effects were examined by Cox
II 911 (77·5%) 1491 (77·1%) 2308 (73·3%)
proportional hazards models stratified by type 2 diabetes
at baseline and including interaction terms for effect III 261 (22·2%) 440 (22·8%) 829 (26·3%)

modification by glycaemia category. IV 2 (0·2%) 3 (0·2%) 13 (0·4%)

Heart failure outcomes might be in part affected by the Left ventricular ejection fraction, % 54·3% (8·8) 54·3% (8·9) 54·0% (8·7)
duration and severity of diabetes, and conceivably by Left ventricular ejection fraction group, %
concurrent anti-hyperglycaemic therapy. We therefore ≤40% 2 (0·2%) 1 (0·1%) 1 (<0·1%)
analysed (post hoc) the primary outcome by several other 41–49% 383 (32·6%) 645 (33·4%) 1083 (34·4%)
subgroups and reported treatment effect HRs in patients 50–59% 431 (36·7%) 684 (35·4%) 1140 (36·2%)
with previously diagnosed type 2 diabetes by duration ≥60% 359 (30·6%) 604 (31·2%) 926 (29·4%)
of type 2 diabetes of less than 10 years versus 10 years or ECG atrial fibrillation or flutter 491 (41·8%) 939 (48·6%) 1211 (38·5%)
longer; baseline median HbA1c 7∙1% [54 mmol/mol] Systolic blood pressure, mm Hg 127·2 (15·5) 126·8 (15·0) 129·5 (15·4)
or lower versus higher than 7∙1% [54 mmol/mol];) Diastolic blood pressure, mm Hg 74·3 (10·4) 74·0 (10·2) 73·7 (10·4)
baseline metformin use versus no metformin HbA1c, % 5·4% (0·2) 6·0% (0·2) 7·4% (1·6)
use; baseline sulfonylurea use versus no sulfonylurea use; HbA1c, mmol/mol 36 (2) 42 (2) 57 (17)
base­line DPP-4 inhibitor use versus no DPP-4 inhibitor eGFR, mL/min per 1·73 m² 65·4 (19·1) 60·9 (18·2) 59·5 (19·5)
use; and baseline insulin use versus no insulin use. We eGFR ≥60 mL/min per 1·73 m² 707 (60·2%) 996 (51·5%) 1488 (47·2%)
also assessed (post hoc) efficacy of dapagliflozin within N-terminal pro-B-type natriuretic 1012 (608–1790) 1104 (656–1904) 951 (606–1673)
each glycaemia subgroup by LVEF. peptide, ng/L
p values of less than 0∙05 were considered to be Heart failure therapies
significant. The p values for the subgroup analyses were Loop diuretics 846 (72·0%) 1480 (76·6%) 2481 (78·8%)
not adjusted for multiple comparisons because the Angiotensin-converting enzyme 431 (36·7%) 709 (36·7%) 1152 (36·6%)
tests were exploratory and were interpreted descriptively. inhibitor

Statistical analyses were conducted using STATA Angiotensin receptor blocker 393 (33·5%) 636 (32·9%) 1243 (39·5%)
(version 16.1). Angiotensin receptor−neprilysin 68 (5·8%) 94 (4·9%) 139 (4·4%)
inhibitor
The trial is registered with ClinicalTrials.gov,
β blocker 952 (81·1%) 1587 (82·1%) 2634 (83·6%)
NCT03619213.
Mineralocorticoid receptor antagonist 508 (43·3%) 883 (45·7%) 1275 (40·5%)
Pacemaker 135 (11·5%) 210 (10·9%) 316 (10·0%)
Results
Implantable cardioverter defibrillator 18 (1·5%) 38 (2·0%) 57 (1·8%)
Between Sept 1, 2018, and Jan 18, 2021, 10 418 patients
were screened, with 6263 randomly assigned to oral Data are mean (SD), median (IQR), or n (%). NA=not applicable.
dapagliflozin (n=3131) or placebo (n=3132; appendix 2 p 11) Table 1: Baseline characteristics by glycaemic status (full-analysis set)
and were included in the intention-to-treat population for
the primary analysis, which has been previously
published. At baseline, clinical characteristics were At baseline, 3150 (50∙3%) participants had type 2 diabetes See Online for appendix 2
balanced between groups. For the intention-to-treat (2806 [44∙8%] based on history, 26 [0∙4%] based on
population, mean age of was 71∙7 years (SD 9∙6), mean medication use, and 318 [5∙1%] newly identified by
BMI was 29∙8 kg/m² (6∙1), and 2787 (44∙5%) of baseline HbA1c level), 1934 (30·9%) had prediabetes, and
participants had obesity, 2747 (43∙9%) were women, and 1175 (18∙8%) had normoglycaemia. Using the more
4439 (70∙9%) were White. The mean of the most recent restrictive criteria of the International Expert Committee14
LVEF was 54∙2% (SD 8∙8), and the mean baseline eGFR that defines prediabetes as HbA1c 6·0% to less than 6·5%
was 61∙0 mL/min per 1∙73 m² (SD 19∙1; 3069 [49∙0%] [42–48 mmol/mol], 3150 (50∙3%) had type 2 diabetes,
had an eGFR <60 mL/min per 1∙73 m²). Baseline HbA1c 1033 (16∙5%) had prediabetes, and 2076 (33∙2%) had
was available in 6247 participants and an additional normoglycaemia. Baseline characteristics by glycaemia
12 patients were identified as having type 2 diabetes at subgroups are shown in table 1. Baseline characteristics
baseline (ie, did not need an HbA1c for diagnosis); by treatment assignment in each glycaemia subgroup are
therefore, a total of 6259 patients were included in the in appendix 2 (pp 1–3). Patients with type 2 diabetes had
full-analysis set (3515 [56·2%] were men and 4435 [70·9%] higher BMI, somewhat worse kidney function, and had
were White). In all patients with HbA1c available, the mean a greater prevalence of atherosclerotic disease, including
value was 6∙6% (SD 1∙4; 49 mmol/mol [15]), with a range coronary artery disease, than the normoglycaemia
of 4∙2–17∙2% (22–165 mmol/mol), and the median value subgroup. They had a higher New York Heart Association
was 6∙2% (IQR 5∙7–7∙0%; 44 mmol/mol [39–53]). class and had also been more frequently hospitalised for

www.thelancet.com/diabetes-endocrinology Vol 10 December 2022 873

 Articles

A B
100 Normoglycaemia 100 Prediabetes vs normoglycaemia HR 1·19 (95% CI 0·95–1·49)
Prediabetes Type 2 diabetes vs normoglycaemia HR 1·58 (95% CI 1·30–1·94)
30 Type 2 diabetes 30

Worsening heart failure event (%)

Prediabetes vs normoglycaemia HR 1·09 (95% CI 0·90–1·31)
25 Type 2 diabetes vs normoglycaemia HR 1·46 (95% CI 1·24–1·73) 25
Primary outcome (%)

20 20

15 15

10 10

5 5

0 0

C D
100 Prediabetes vs normoglycaemia HR 0·85 (95% CI 0·65–1·11) 100 Prediabetes vs normoglycaemia HR 0·93 (95% CI 0·77–1·12)
Type 2 diabetes vs normoglycaemia HR 1·07 (95% CI 0·85–1·36) Type 2 diabetes vs normoglycaemia HR 1·17 (95% CI 0·99–1·38)
30 30
Cardiovascular death (%)

25 25

All-cause death (%)

20 20

15 15

10 10

5 5

0 0
0 0·5 1·0 1·5 2·0 2·5 3·0 0 0·5 1·0 1·5 2·0 2·5 3·0
Time from randomisation (years) Time from randomisation (years)

Figure 1: Kaplan-Meier curves for key outcomes by baseline glycaemic status

Primary outcome (A; the composite of time to first worsening heart failure events, defined as an unplanned hospitalisation or urgent heart failure visit requiring
intravenous therapy, or cardiovascular death), worsening heart failure events (B), cardiovascular death (C), and all-cause death (D) in the major glycaemia subgroups
(normoglycaemia, prediabetes, and type 2 diabetes).

heart failure during the year before study enrolment, but and 10∙1 per 100 patient-years in the type 2 diabetes
had a lower prevalence of atrial fibrillation or flutter than subgroup (1∙46 [1∙24–1∙73]; p<0·0001 for trend; figure 1;
patients with normoglycaemia. In general, the prediabetes appendix 2 p 4). The same patterns were observed for heart
subgroup had features intermediate between the other failure-specific components of the primary outcomes,
two, with the exception of atrial fibrillation, which was such as worsening heart failure events (appendix 2 p 4).
most prevalent in this group. Use of angiotensin receptor Cardiovascular death and all-cause death occurred at
blocker and less loop diuretic and mineralocorticoid similar frequency across the subgroups with a trend
receptor antagonist was more frequent in patients with towards higher all-cause death in the type 2 diabetes
type 2 diabetes than in patients in the other subgroups. subgroup only (appendix 2 p 4). For cardiovascular and all-
Treatment with devices such as pacemakers and cause mortality, there was no evidence of any intermediate
implantable cardioverter defibrillators was similar across outcome rates in patients with prediabetes (appendix 2 p 4).
the subgroups. The median duration of follow-up was Finally, using HbA1c as a continuous variable, all event
2∙3 years (IQR 1∙7–2∙8). The trial was stopped when the rates generally increased at higher levels (figure 2).
required number of primary outcome events was reached. In the normoglycemia subgroup, the primary composite
Crude event rates for the primary composite outcome of outcome rate was 7∙8 per 100 patient-years in the placebo
first worsening heart failure event or cardiovascular death group and 6∙0 per 100 patient-years in the dapagliflozin
and most of the secondary outcomes varied across the group, with an absolute risk reduction of 1∙8 per
three glycaemia subgroups, with primary outcome and 100 patient-years (numbers needed to treat 56; HR 0∙77
worsening heart failure events in the type 2 diabetes [95% CI 0∙57–1∙04], log-rank p=0·088; table 2; figure 3A;
subgroup being significantly higher than in the appendix 2 p 5). The corresponding values (placebo vs
normoglycaemia subgroup (figure 1; appendix 2 p 4). The dapagliflozin) in the dysglycaemia categories were 8∙2 per
primary composite outcome incidence rate was 6∙9 per 100 patient-years versus 7·1 per 100 patient-years (absolute
100 patient-years in the normoglycaemia subgroup risk reduction 1∙1 per 100 patient-years; numbers needed
(reference), increasing to 7∙6 per 100 patient-years in the to treat 91; HR 0∙87 [95% CI 0∙69–1∙08]; log-rank p=0·21;
prediabetes subgroup (HR 1∙09 [95% CI 0∙90–1∙31]) table 2; figure 3B; appendix 2 p 6) for patients with

874 www.thelancet.com/diabetes-endocrinology Vol 10 December 2022

 Articles

A B
14 10

Incidence rate of worsening heart failure

12
Incidence rate of primary outcome

10

6
8

6 4

C D
5·0
10
Incidence rate of cardiovascular death

Incidence rate of all-cause death

4·5
9

4·0
8

3·5
7

3·0
6
5 6 7 8 9 10 5 6 7 8 9 10
Baseline HbA1c (%) Baseline HbA1c (%)

Figure 2: HbA1c distribution, incidence rates, and dapagliflozin treatment effect by baseline HbA1c
Incidence rates (indicated by the solid red line) with 95% CIs (indicated by shaded areas) for the primary composite outcome (the composite of time to first
worsening heart failure events, defined as an unplanned hospitalisation or urgent heart failure visit requiring intravenous therapy, or cardiovascular death; (A),
worsening heart failure events (B), cardiovascular death (C), and all-cause death (D) by baseline HbA1c as a continuous variable. Data have been adjusted by sex,
age, and region.

prediabetes and 11·2 versus 9·0 per 100 patient-years baseline median HbA1c of 7∙1% [54 mmol/mol] or lower
(absolute risk reduction 2·2 per 100 patient-years; (1471 [52·6%]) versus higher than higher than 7∙1%
numbers needed to treat 46; HR 0∙81 [95% CI 0∙69–0∙95]; (1323 [47·4%]); and baseline use of the most prevalent
log-rank p=0·0077; table 2; figure 3C; appendix 2 p 7) glucose-lowering agents in the DELIVER population
for patients with type 2 diabetes (pinteraction=0∙82; (1641 [58∙5%] on metformin, 843 [30∙0%] on insulin,
table 2; figure 3D). Similarly, dapagliflozin’s treatment 471 [16∙8%] on DPP-4 inhibitors, 603 [21∙5%] on
effect did not differ significantly across the glycaemia sulfonylureas, and 61 [2∙2%] on GLP-1 receptor agonists).
subgroups for the following components of the primary There proved to be generally no statistical heterogeneity
outcome and key secondary outcomes: cardiovascular across the subgroups on the primary outcome, indicating
death; worsening heart failure events; hospitalisations for similar benefits from dapagliflozin (appendix 2 p 16). The
heart failure; the composite of cardiovascular death and all exception was for sulfonylureas: patients taking these
heart failure events (including recurrent); and all-cause medications at baseline had significantly more benefit
death (all pinteraction values >0·10; table 2; appendix 2 pp 13–15). from dapagliflozin (HR 0∙50 [95% CI 0∙35–0∙72]) than
Consistent with the categorical analysis, results were those not taking these medications (HR 0∙94 [95% CI
similar with HbA1c modelled as a continuous variable 0∙79–1∙13]; pinteraction=0∙0026). Finally, we did a post-hoc
(pinteraction>0∙20 for all; figure 4) analysis of the effect of dapagliflozin on the primary
Our post-hoc analyses of several type 2 diabetes outcome rates within each glycaemia subgroup based on
subgroups examined dapagliflozin treatment effect by LVEF as a continuous variable, and found no signifi­cant
duration, severity, and baseline treatment of established heterogeneity (pinteraction=0∙92 for all subgroups;
diabetes as follows: diabetes duration less than 10 years appendix 2 p 17).
(1505 [53·7%] of 2801 participants with history of Serious adverse events as well as amputations,
diabetes]) versus 10 years or longer (1296 [46·3%]); hypoglycaemia, and diabetic ketoacidosis (the latter

www.thelancet.com/diabetes-endocrinology Vol 10 December 2022 875

 Articles

hypoglycaemia; and two [0·1%] vs none vs none had

Normoglycaemia Prediabetes Type 2 diabetes Pinteraction value
subgroup subgroup subgroup diabetic ketoacidosis; appendix 2 p 8). However, there
(n=1175) (n=1934) (n=3150) was no evidence of any influence by treatment
Primary composite outcome assignment (appendix 2 pp 9–10). Specifically, patients in
Events 172 305 644 ·· the dapagliflozin group had no differences in volume-
Rate in dapafliglozin group (n=3129), 6·0 7·1 9·0 ·· related or renal serious adverse events or adverse events
per 100 patient-years leading to discontinuation of the study drug,
Rate in placebo group (n=3130), per 7·8 8·2 11·2 ·· hypoglycaemia, or amputations versus placebo.
100 patient-years
HR (95% CI) 0·77 0·87 0·81 0·82 Discussion
(0·57–1·04) (0·69–1·08) (0·69–0·95)
In patients with heart failure and mildly reduced
Cardiovascular death ejection fraction or preserved ejection fraction
Events 93 132 266 ·· receiving contemporary therapy, there was a stepwise
Rate in dapafliglozin group (n=3129), 3·2 3·1 3·5 ·· increase in heart failure outcomes across the three
per 100 patient-years
glycaemia categories (normoglycaemia, prediabetes,
Rate in placebo group (n=3130), per 3·9 3·1 4·1 ··
100 patient-years
and type 2 diabetes). The SGLT2 inhibitor dapagliflozin,
HR (95% CI) 0·82 1·02 0·85 0·63
administered orally, reduced the composite primary
(0·54–1·23) (0·72–1·43) (0·67–1·08) outcome of time from randomisation to first worsening
Worsening heart failure event heart failure events or cardiovascular mortality in the
Events 118 229 475 ·· three glycaemia subgroups, without heterogeneity, as
Rate in dapafliglozin group (n=3129), 4·1 4·8 6·7 ·· well as across the entire range of baseline HbA1c values.
per 100 patient-years Given the higher absolute risk of the primary outcome
Rate in placebo group (n=3130), per 5·4 6·6 8·2 ·· observed in patients with diabetes, the absolute risk
100 patient-years reduction with dapagliflozin tended to be most
HR (95% CI) 0·76 0·73 0·83 0·74 favourable for this subgroup (2∙2 per 100 patient-years;
(0·53–1·10) (0·56–0·95) (0·69–0·99)
numbers needed to treat 46) than in the other two
Hospitalisation for heart failure
(1∙8 per 100 patient-years [numbers needed to treat 56]
Events 113 204 429 ··
for the normoglycaemia subgroup; 1∙1 per 100 patient-
Rate in dapafliglozin group (n=3129), 3·7 4·7 5·9 ··
years [numbers needed to treat 91] for the prediabetes
per 100 patient-years
subgroup). These results extend the efficacy of
Rate in placebo group (n=3130), per 5·3 5·8 7·4 ··
100 patient-years dapagliflozin in patients with heart failure with reduced
HR (95% CI) 0·69 0·74 0·81 0·72 ejection fraction from DAPA-HF to those with heart
(0·47–1·00) (0·56–0·98) (0·67–0·98) failure and mildly reduced ejection fraction or preserved
Urgent heart failure visit ejection fraction. DAPA-HF5 demonstrated an overall
Events 11 41 86 ·· 26% relative risk reduction (95% CI 15–35; p<0∙001) for
Rate in dapafliglozin group (n=3129), 0·5 0·7 1·1 ·· the same primary outcome, with an HR of 0∙73 (95% CI
per 100 patient-years 0∙60–0∙88) in patients without type 2 diabetes
Rate in placebo group (n=3130), per 0·3 1·2 1·4 ·· (including prediabetes) versus 0∙75 (0∙63–0∙90) in the
100 patient-years
type 2 diabetes subgroup (pinteraction=0∙80).15 Taken
HR (95% CI) 1·73 0·59 0·78 0·38 together, these data support the notion that the clinical
(0·51–5·91) (0·31–1·11) (0·51–1·19)
benefits of dapagliflozin in patients with heart failure
Composite of cardiovascular death and all heart failure events (including recurrent)
not only extend over the entire range of ejection
Events 284 502 1084 ··
fractions but also across the glycaemic range, including
Rate in dapafliglozin group (n=3129), 9·0 9·7 14·1 ··
per 100 patient-years
patients considered to be normoglycaemic. The results
Rate in placebo group (n=3130), per 12·7 13·8 17·2 ··
are also generalisable, as the DELIVER cohort seems to
100 patient-years be representative of the broader population of patients
Risk ratio (95% CI) 0·71 0·70 0·82 0·58 with heart failure and mildly reduced ejection fraction
(0·50–1·01) (0·54–0·92) (0·68–0·99) or preserved ejection fraction seen in clinical practice.
(Table 2 continues on next page) Additional post-hoc analyses within DELIVER’s
type 2 diabetes subgroup also found generally no
interaction with dapagliflozin’s treatment effect based on
two being rare in the trial) were more frequent in the duration, severity, diabetes treatment, or ejection
type 2 diabetes subgroup than in the prediabetes and fraction. The interaction regarding baseline sulfonylurea
normoglycaemia subgroups (1492 [47·4%] of 3150 vs therapy has not been reported before in other SGLT2
806 [41·7%] of 1934 vs 484 [41·2%] of 1175 had serious inhibitor trials and is likely to be a chance finding.
adverse events; 36 [1·1%] vs five [0·3%] vs three [0·3%] For example, in DAPA-HF, the HRs for the primary
had amputations; 13 [0·4 %] vs none vs none had endpoint in favour of dapagliflozin was 0∙77 and 0∙75 in

876 www.thelancet.com/diabetes-endocrinology Vol 10 December 2022

 Articles

patients taking versus not taking sulfonylureas

Normoglycaemia Prediabetes Type 2 diabetes Pinteraction value
(pinteraction=0·93).16 Similar data from the EMPEROR trials subgroup subgroup subgroup
are not available. If this finding is confirmed in (n=1175) (n=1934) (n=3150)
EMPEROR-Preserved, further investigation would be (Continued from previous page)
warranted. An obvious mechanism is not apparent. All-cause death
Trials testing other SGLT2 inhibitors in heart failure
Events 181 284 556 ··
populations have found similar effects. In EMPEROR-
Rate in dapafliglozin group (n=3129), 6·1 7·1 7·6 ··
Reduced,6 involving 3730 patients with heart failure with per 100 patient-years
reduced ejection fraction, empagliflozin lowered the risk Rate in placebo group (n=3130), per 7·7 6·2 8·4 ··
of the primary outcome of time-to-first heart failure 100 patient-years
hospitalisation or cardiovascular death by 25% (p<0∙001), HR (95% CI) 0·80 1·14 0·91 0·14
with an HR of 0∙72 (95% CI 0∙60–0∙87) in participants (0·60–1·07) (0·90–1·44) (0·77–1·07)
without diabetes and 0∙78 (0∙64–0∙97) in those with HR=hazard ratio.
diabetes (pinteraction=0∙57).17 In the EMPEROR-Preserved
Table 2: Treatment effect estimates (dapagliflozin versus placebo) on primary and secondary outcomes
trial,7 involving 5988 patients with heart failure with by baseline glycaemic status (full-analysis set)
preserved ejection fraction or mildly reduced ejection
fraction, empagliflozin reduced the risk of the same
primary outcome by 21% (95% CI 10–31; p<0∙001), with Reflecting the known epidemiology of heart failure with
nearly identical HRs in patients without (0∙78 [95% CI preserved ejection fraction,12 most of the patients enrolled
0·64–0·95]) versus with (0∙79 [0∙67–0∙94] diabetes in DELIVER had dysglycaemia, with more than
(pinteraction=0∙92).18 For direct comparison, risk of first heart 50% having type 2 diabetes and more than 30% having
failure hospitalisation or cardiovascular death in prediabetes. Indeed, only two of every ten DELIVER
DELIVER was reduced by 20% with dapagliflozin in the participants were normoglycaemic. These proportions
overall population (HR 0∙80 [95% CI 0∙71–0∙91]). Finally, are consistent with other recent heart failure with
in SOLOIST, which involved 1222 patients with preserved ejection fraction trials, including EMPEROR-
type 2 diabetes and recent hospitalisation for heart Preserved7 and PARAGON-HF.24 These findings reinforce
failure, treatment with sotagliflozin, an inhibitor of both the importance of the awareness among clinicians about
SGLT2 and SGLT1, resulted in a 29% risk reduction the frequency of heart failure with preserved ejection
(HR 0∙71 [95% CI 0∙56–0∙89]) in the same composite fraction in dysglycaemic states (and vice versa).
outcome.19 The role of SGLT2 inhibitors as heart failure therapies
Regarding safety concerns, we observed more frequent is now solidified. In contrast to the original cardiovascular
serious adverse events in the type 2 diabetes subgroup outcome trials with these agents, which involved only
but no increased risk based on randomised treatment. patients with type 2 diabetes, the heart failure trials have
This finding is in keeping with previous heart failure included both patients with and without diabetes, clearly
studies5–8 and is notable given the frequently frail nature demonstrating that these medications extend to those
of patients with heart failure who are commonly treated with prediabetes and normoglycaemia.16–18 In heart failure
with multiple vasoactive therapies and potent diuretics guidelines, SGLT2 inhibitors are now endorsed for both
that can substantially alter plasma volume and blood patients with heart failure with reduced ejection fraction
pressure. and those with heart failure with preserved ejection
Our findings are particularly important for medical fraction.9,10 Initial trials documenting their substantive
professionals involved in diabetes care. Heart failure, benefits involved patients with LVEFs of 40% or less.5,6
particularly heart failure with preserved ejection fraction, EMPEROR-Preserved7 and now DELIVER8 have extended
is extremely common in people with diabetes. In the these to patients with heart failure who have higher
USA, it has been one of the leading causes for hospital ejection fractions. Previously, there remained
admission in the Medicare population (age >65 years),20 considerable uncertainty as to whether this class of
with similar reports from other parts of the world.21 Once medication would be effective in patients with heart
heart failure is established, adverse clinical outcomes, failure with preserved ejection fraction, given the
including hospitalisations and mortality, are worse when extensive history of clinical trial failures in this space,
diabetes coexists.21,22 Reducing hyperglycaemia by itself involving evidence-based therapies known to be effective
does not seem to improve heart failure outcomes.2 in heart failure with reduced ejection fraction.25 In fact, in
Indeed, several glucose-lowering agents have been found a meta-analysis of the EMPEROR trials, there appeared
to increase heart failure hospitalisations.23 Accordingly, to be attenuation of treatment benefit from empagliflozin
understanding the optimal management of heart failure, at the highest ejection fractions.26 We found no such
especially heart failure and mildly reduced ejection trends in DELIVER, neither in the overall cohort nor in
fraction or with preserved ejection fraction, which is the individual glycaemia subgroups, when analysed
highly prevalent in patients with obesity and hypertension with LVEF as a continuous variable. Indeed, in a pooled
who also have type 2 diabetes, is critically important. analysis of data from both EMPEROR-Preserved and

www.thelancet.com/diabetes-endocrinology Vol 10 December 2022 877

 Articles

A Normoglycaemia B Prediabetes
100 Placebo group Log-rank p=0·21
Dapagliflozin group
Cumulative incidence (%) 30 Log-rank p=0·088

25

20

15

10

0
0 0·5 1·0 1·5 2·0 2·5 3·0 0 0·5 1·0 1·5 2·0 2·5 3·0
Time from randomisation (years) Time from randomisation (years)
Number at risk
Placebo group 584 547 520 445 361 222 64 974 901 848 737 603 336 98
Dapagliflozin group 591 567 545 463 372 229 66 960 909 870 735 593 337 98

C Type 2 diabetes D
100 Log-rank p=0·0077 HR (95% CI) pinteraction

30 Baseline glycaemic 0·82

status
Cumulative incidence (%)

25

20 Normoglycaemia 0·77 (0·57–1·04)

15
Prediabetes 0·87 (0·69–1·08)
10

5
Type 2 diabetes 0·81 (0·69–0·95)
0
0 0·5 1·0 1·5 2·0 2·5 3·0 0·50 0·75 1·00 1·50
Time from randomisation (years)
Number at risk
Placebo group 1572 1442 1334 1120 906 520 152
Dapagliflozin group 1578 1470 1384 1173 969 560 168

Figure 3: Kaplan-Meier curves for the primary outcome, dapagliflozin versus placebo, in the normoglycaemia (A), prediabetes (B), and type 2 diabetes (C)
subgroups, and forest plot for the primary outcome (D)

DELIVER, the HRs for the primary outcomes were outcome trial, we did not obtain fasting plasma glucose
nearly identical in all three ejection fraction subgroups concentrations and we did not perform oral glucose
(HR 0·78 [95% CI 0·67–0·90] for <50%; 0·79 [0·68–0·93] tolerance testing. More precise classification of newly
for 50–59%, and 0·81 [0·69–0·96] for >60%).27 In fact, identified type 2 diabetes, prediabetes, and even nor­
the SGLT2 inhibitors are now the only category of moglycaemia would have necessitated these tests for
medication with clear heart failure benefits throughout a more rigorous assessment. The number of patients
the full ejection fraction range in both heart failure with with prediabetes by HbA1c who might actually have had
preserved ejection fraction and in heart failure with type 2 diabetes based on fasting plasma glucose or oral
reduced ejection fraction. glucose tolerance testing is considerable.29,30 Conversely,
There remains uncertainty as to the manner through some patients with prediabetes by HbA1c might not have
which SGLT2 inhibitors exert their cardiac effects; the met criteria for prediabetes by fasting plasma glucose or
effects certainly do not appear to be mediated through by oral glucose tolerance testing.30,31 Accordingly, the
glucose reduction. Theories include reductions in plasma numbers and individuals in our subgroups might have
volume and interstitial fluid with concurrent changed if these additional tests had been used. Second,
improvements in ventricular afterload and preload; HbA1c was determined from only a single blood sample.
improved myocardial energetics through altered fuel Although this is adequate for identifying prediabetes,
supply involving ketone bodies; direct cardiac effects via official policies are for the HbA1c to be repeated
the sodium–hydrogen ion exchanger; reductions in and confirmed before establishing a diagnosis of
sympathetic or vascular tone, or both; and improvements type 2 diabetes.14 Accordingly, it is possible that some
in cardiac remodelling.28 participants in the small group of newly identified
Our analysis has several limitations. First, given the diabetes might have had lower HbA1c on a second
nature of DELIVER as a large global heart failure determination, leading to re-categorisation. Third, less

878 www.thelancet.com/diabetes-endocrinology Vol 10 December 2022

 Articles

A B
1·4 1·4

1·2 1·2

Rate ratio for worsening heart failure

Rate ratio for primary outcome

1·0 1·0

0·8 0·8

0·6 0·6
pinteraction=0·85 pinteraction=0·79

C D
1·4 1·4
Rate ratio for cardiovascular death

1·2 1·2
Rate ratio for all-cause death

1·0 1·0

0·8 0·8

0·6 0·6
pinteraction=0·72
pinteraction=0·26
5 6 7 8 9 10 5 6 7 8 9 10
Baseline HbA1c (%) Baseline HbA1c (%)

Figure 4: Treatment effect of dapagliflozin versus placebo by baseline HbA1c

Restricted cubic spline analyses of the effect of dapagliflozin on the primary composite outcome (the composite of time to first worsening heart failure events,
defined as an unplanned hospitalisation or urgent heart failure visit requiring intravenous therapy, or cardiovascular death; A), worsening heart failure events (B),
cardiovascular death (C), and all-cause death (D) across the range of baseline HbA1c. The solid red line indicates a continuous hazard ratio, and the shaded areas show
the 95% CIs.

than 5% of our study population self-identified as being which is, to our knowledge, the largest SGLT2 inhibitor
of Black race. This constitutes an under-represented heart failure trial so far. Moreover, the lack of any
group in DELIVER and is a concern in many heart failure effect modification by baseline glycaemia adds to the
trials. Given the higher prevalence of type 2 diabetes in grow­ing evidence that the heart failure benefits from
Black people, it is important to study this population dapagliflozin are independent of glycaemic status of
further, perhaps through meta-analyses, to ensure no the patient. These results are consistent with those
heterogeneity in effect, although one has not emerged demonstrated in patients with heart failure with reduced
across SGLT2 inhibitor trials so far. In this regard, we ejection fraction in the DAPA-HF trial. The benefits of
would also point out that several investigators have found dapagliflozin on adverse heart failure outcomes are
a higher mean HbA1c in Black people versus White robust across the spectra of both ejection fraction and
people for a given degree of hyperglycaemia.14 So, any glycaemia.
such future study would ideally entail complementary Contributors
measures of glycaemia to confirm proper assignment of SEI, BLC, JJVM, and SDS conceived of and designed the study. SEI drafted
baseline glycaemic status. Finally, we did not measure the manuscript. SEI and SDS had final responsibility for the decision to
submit for publication. BLC and OB had access to the raw data, conducted
HbA1c in follow-up, so could not determine the impact of the analyses, and verified the data. All authors contributed to data
dapagliflozin on this glycaemic measure or analyse interpretation and writing of the final version of the manuscript and take
outcomes based on the degree of HbA1c lowering. responsibility for the accuracy and integrity of the data.
Oral dapagliflozin reduced the composite of worsening Declaration of interests
heart failure and cardiovascular death regardless of SEI has served on clinical trial committees or as a consultant to
glycaemia category or baseline HbA1c in patients with AstraZeneca, Boehringer Ingelheim, Novo Nordisk, Merck, Pfizer,
Bayer, Abbott, Lexicon Pharmaceuticals, vTv Therapeutics, and
heart failure and mildly reduced ejection fraction or Esperion, and has delivered lectures sponsored by AstraZeneca and
with preserved ejection fraction enrolled in DELIVER, Boehringer Ingelheim. BLC has received consulting fees from

www.thelancet.com/diabetes-endocrinology Vol 10 December 2022 879

 Articles

Boehringer Ingelheim. MV has received research grant support or Akros, Alnylam, Amgen, Arena, AstraZeneca, Bayer, Boehringer
served on advisory boards for American Regent, Amgen, AstraZeneca, Ingelheim, Bristol Myers Squibb, Cardior, Cardurion, Corvia,
Bayer, Baxter Healthcare, Boehringer Ingelheim, Cytokinetics, Lexicon Cytokinetics, Daiichi-Sankyo, GlaxoSmithKline, Lilly, Merck, Myokardia,
Pharmaceuticals, Novartis, Pharmacosmos, Relypsa, Roche Diagnostics, Novartis, Roche, Theracos, Quantum Genomics, Cardurion, Janssen,
and Sanofi; received speaker fees from AstraZeneca, Novartis, and Cardiac Dimensions, Tenaya, Sanofi-Pasteur, Dinaqor, Tremeau,
Roche Diagnostics; and participates on clinical trial committees for CellProThera, Moderna, American Regent, and Sarepta.
studies sponsored by Galmed, Novartis, Bayer, Occlutech, and Impulse
Data sharing
Dynamics. ASD has received research grants (to his institution) from
The sponsor of this trial is committed to sharing access to patient-level
Abbott, AstraZeneca, Alnylam, Bayer, and Novartis, and consulting fees
data and supporting clinical documents from eligible studies by
or honoraria from Abbott, Alynylam, AstraZeneca, Axon Therapeutics,
qualified external researchers. Such requests are reviewed by an
Biofourmis, Boston Scientific, Bayer, Cytokinetics, GlaxoSmithKline,
independent review panel and approved based on scientific merit.
Lupin Pharma, Merck, Medpace, Novartis, Parexel, Regeneron, Roche,
For link to request data see Requests to access data can be made to AstraZeneca. All data eventually
and Verily. PSJ’s employer has been remunerated by AstraZeneca,
https://vivli.org/ourmember/ shared will be anonymised to respect the privacy of patients who have
Bayer, and Novo Nordisk for clinical trial work. PSJ also reports
astrazeneca/ participated in the trial, in keeping with applicable laws and regulations.
consulting and speaker fees from Novartis, AstraZeneca, and
For trial data availability see The trial data availability is according to previously published criteria
Boehringer Ingelheim; and research funding from Boehringer
https://astrazenecagrouptrials. and processes, which can be found online.
Ingelheim. RAdB has received research grants from AstraZeneca,
pharmacm.com/ST/Submission/ Abbott, Boehringer Ingelheim, Cardior Pharmaceuticals, Ionis Acknowledgments
Disclosure Pharmaceuticals, Novo Nordisk, and Roche, and has received speaker The DELIVER trial was funded by AstraZeneca. We thank all investigators,
fees from Abbott, AstraZeneca, Bayer, Bristol Myers Squibb, Novartis, coordinators, other members of the trial teams, Clinical Events
and Roche. AFH has received research grants from American Regent, Committee, and Data Safety Monitoring Committee (appendix 1 p 311),
Amgen, AstraZeneca, Bayer, Boehringer Ingelheim, Merck, Novartis, as well as the patients, for their participation in the trial.
Somologic, and Verily, and has served as a consultant or on the advisory
References
board for Amgen, AstraZeneca, Bayer, Boehringer Ingelheim, Boston 1 Kenny HC, Abel ED. Heart failure in type 2 diabetes mellitus.
Scientific, Bristol Myers Squibb, Cytokinetics, Eidos, Intercept, Merck, Circ Res 2019; 124: 121–41.
and Novartis. MNK reports research grant support from AstraZeneca 2 Gerstein HC, Swedberg K, Carlsson J, et al. The hemoglobin
and Boehringer Ingelheim, and consulting fees from Alnylam, A1c level as a progressive risk factor for cardiovascular death,
AstraZeneca, Amgen, Bayer, Boehringer Ingelheim, Cytokinetics, hospitalization for heart failure, or death in patients with chronic
Esperion, Eli Lilly, Janssen, Lexicon, Merck (Diabetes and heart failure: an analysis of the Candesartan in Heart failure:
Cardiovascular), Pharmacosmos, Novo Nordisk, Sanofi, and Vifor. CSPL Assessment of Reduction in Mortality and Morbidity (CHARM)
is supported by a Clinician Scientist Award from the National Medical program. Arch Intern Med 2008; 168: 1699–704.
Research Council of Singapore; has received research support from 3 Castagno D, Baird-Gunning J, Jhund PS, et al. Intensive glycemic
Bayer and Roche Diagnostics; has served as consultant or on the control has no impact on the risk of heart failure in type 2 diabetic
advisory board, steering committee, or executive committee for patients: evidence from a 37,229 patient meta-analysis. Am Heart J
Actelion, Alleviant Medical, Allysta Pharma, Amgen, AnaCardio AB, 2011; 162: 938–48.e2.
Applied Therapeutics, AstraZeneca, Bayer, Boehringer Ingelheim, 4 McGuire DK, Shih WJ, Cosentino F, et al. Association of SGLT2
Boston Scientific, Cytokinetics, Darma, EchoNous, Eli Lilly, Impulse inhibitors with cardiovascular and kidney outcomes in patients
Dynamics, Ionis Pharmaceutical, Janssen Research & Development, with type 2 diabetes: a meta-analysis. JAMA Cardiol 2021;
6: 148–58.
Medscape/WebMD Global, Merck, Novartis, Novo Nordisk, Prosciento,
Radcliffe Group, Roche Diagnostics, Sanofi, Siemens Healthcare 5 McMurray JJV, Solomon SD, Inzucchi SE, et al. Dapagliflozin in
patients with heart failure and reduced ejection fraction.
Diagnostics, and Us2.ai; and is a co-founder of and serves as and non-
N Engl J Med 2019; 381: 1995–2008.
executive director of Us2.ai. FM has received personal fees from
6 Packer M, Anker SD, Butler J, et al. Cardiovascular and renal
AstraZeneca. SJS reports research grants from the US National
outcomes with empagliflozin in heart failure. N Engl J Med 2020;
Institutes of Health, Actelion, AstraZeneca, Corvia, Novartis, and Pfizer, 383: 1413–24.
and consulting fees from Abbott, Actelion, AstraZeneca, Amgen, Aria
7 Anker SD, Butler J, Filippatos G, et al. Empagliflozin in heart
CV, Axon Therapies, Bayer, Boehringer Ingelheim, Boston Scientific, failure with a preserved ejection fraction. N Engl J Med 2021;
Bristol Myers Squibb, Cardiora, Coridea, CVRx, Cyclerion, Cytokinetics, 385: 1451–61.
Edwards Lifesciences, Eidos, Eisai, Imara, Impulse Dynamics, Intellia, 8 Solomon SD, McMurray JJV Claggett BL, et al. Dapagliflozin in
Ionis, Ironwood, Lilly, Merck, MyoKardia, Novartis, Novo Nordisk, heart failure with mildly reduced or preserved ejection fraction.
Pfizer, Prothena, Regeneron, Rivus, Sanofi, Shifamed, Tenax, Tenaya, N Engl J Med 2022; 387: 1089–98.
and United Therapeutics. SV reports receiving research grants or 9 McDonagh TA, Metra M, Adamo M, et al. 2021 ESC Guidelines for
speaking honoraria from Amarin, Amgen, AstraZeneca, Bayer, the diagnosis and treatment of acute and chronic heart failure.
Boehringer Ingelheim, Eli Lilly, EOCI Pharmacomm, HLS Eur Heart J 2021; 42: 3599–726.
Therapeutics, Janssen, Novartis, Novo Nordisk, Pfizer, PhaseBio, Sanofi, 10 Heidenreich PA, Bozkurt B, Aguilar D, et al. 2022 AHA/ACC/
Sun Pharmaceuticals, and the Toronto Knowledge Translation Working HFSA guideline for the management of heart failure: a report of
Group. JFKS has received grants or fees for working on clinical trials, the American College of Cardiology/American Heart Association
consultancy, advisory board or steering committee membership, and Joint Committee on Clinical Practice Guidelines. J Am Coll Cardiol
other activities from AstraZeneca, Novo Nordisk, Bayer, Boehringer 2022; 79: 1757–80.
Ingelheim, Novartis, Merck Sharp & Dohme, and Janssen. OB, MP, and 11 Buse JB, Wexler DJ, Tsapas A, et al. 2019 Update to: Management of
AML are employees of AstraZeneca. JJVM has received funding to his hyperglycemia in type 2 diabetes, 2018. A consensus report by the
institution for his work on clinical trials, consulting, and other activities American Diabetes Association (ADA) and the European
from Alnylam, Amgen, AstraZeneca, Bayer, Bristol Myers Squibb, Association for the Study of Diabetes (EASD). Diabetes Care 2020;
43: 487–93.
Cardurion, Cytokinetics, GlaxoSmithKline, Novartis, Pfizer, and
Theracos, and has received personal lecture fees from the Corpus, 12 McHugh K, DeVore AD, Wu J, et al. Heart failure with preserved
ejection fraction and diabetes: JACC State-of-the-Art Review.
Abbott, Hickma, Sun Pharmaceuticals, and Medsca. SDS has received
J Am Coll Cardiol 2019; 73: 602–11.
research grants from Actelion, Alnylam, Amgen, AstraZeneca,
13 Solomon SD, de Boer RA, DeMets D, et al. Dapagliflozin in heart
Bellerophon, Bayer, Bristol Myers Squibb, Celladon, Cytokinetics, Eidos,
failure with preserved and mildly reduced ejection fraction: rationale
Gilead, GlaxoSmithKline, Ionis, Lilly, Mesoblast, MyoKardia, US and design of the DELIVER trial. Eur J Heart Fail 2021; 23: 1217–25.
National Heart, Lung, and Blood Institute (National Institutes of
14 American Diabetes Association Professional Practice Committee. 2.
Health), Neurotronik, Novartis, NovoNordisk, Respicardia, Sanofi Classification and diagnosis of diabetes: Standards of Medical Care
Pasteur, Theracos, and Us2.ai; and has consulted for Abbott, Action, in Diabetes—2022. Diabetes Care 2022; 45 (suppl 1): S17–38.

880 www.thelancet.com/diabetes-endocrinology Vol 10 December 2022

 Articles

15 Docherty KF, Jhund PS, Bengtsson O, et al. Effect of dapagliflozin 24 Solomon SD, McMurray JJV, Anand IS, et al. Angiotensin–
in DAPA-HF according to background glucose-lowering therapy. neprilysin inhibition in heart failure with preserved ejection
Diabetes Care 2020; 43: 2878–81. fraction. N Engl J Med 2019; 381: 1609–20.
16 Petrie MC, Verma S, Docherty KF, et al. Effect of dapagliflozin on 25 Wintrich J, Kindermann I, Ukena C, et al. Therapeutic approaches
worsening heart failure and cardiovascular death in patients with in heart failure with preserved ejection fraction: past, present, and
heart failure with and without diabetes. JAMA 2020; 323: 1353–68. future. Clin Res Cardiol 2020; 109: 1079–98.
17 Anker SD, Butler J, Filippatos G, et al. Effect of empagliflozin on 26 Butler J, Packer M, Filippatos G, et al. Effect of empagliflozin in
cardiovascular and renal outcomes in patients with heart failure by patients with heart failure across the spectrum of left ventricular
baseline diabetes status: results from the EMPEROR-Reduced trial. ejection fraction. Eur Heart J 2022; 43: 416–26.
Circulation 2021; 143: 337–49. 27 Vaduganathan M, Docherty KF, Claggett BL, et al. SGLT-2 inhibitors
18 Filippatos G, Butler J, Farmakis D, et al. Empagliflozin for heart in patients with heart failure: a comprehensive meta-analysis of five
failure with preserved left ventricular ejection fraction with and randomised controlled trials. Lancet 2022; 400: 757–67.
without diabetes. Circulation 2022: 146: 676–86. 28 Zelniker TA, Braunwald E. Mechanisms of cardiorenal effects of
19 Bhatt DL, Szarek M, Steg PG, et al. Sotagliflozin in patients with sodium-glucose cotransporter 2 inhibitors: JACC State-of-the-Art
diabetes and recent worsening heart failure. N Engl J Med 2021; Review. J Am Coll Cardiol 2020; 75: 422–34.
384: 117–28. 29 Gyberg V, De Bacquer D, Kotseva K, et al. Screening for
20 Kilgore M, Patel HK, Kielhorn A, Maya JF, Sharma P. Economic dysglycaemia in patients with coronary artery disease as reflected by
burden of hospitalizations of Medicare beneficiaries with heart fasting glucose, oral glucose tolerance test, and HbA1c: a report
failure. Risk Manag Healthc Policy 2017; 10: 63–70. from EUROASPIRE IV—a survey from the European Society of
21 Seferović PM, Petrie MC, Filippatos GS, et al. Type 2 diabetes Cardiology. Eur Heart J 2015; 36: 1171–77.
mellitus and heart failure: a position statement from the Heart 30 Ferrannini G, De Bacquer D, De Backer G, et al. Screening for
Failure Association of the European Society of Cardiology. glucose perturbations and risk factor management in dysglycemic
Eur J Heart Fail 2018; 20: 853–72. patients with coronary artery disease—a persistent challenge in
22 Scheen AJ. Counteracting heart failure with diabetes drugs: need of substantial improvement: a report from ESC EORP
a review into the pharmacokinetic and pharmacodynamic EUROASPIRE V. Diabetes Care 2020; 43: 726–33.
properties. Expert Opin Drug Metab Toxicol 2022; 18: 381–93. 31 Saukkonen T, Cederberg H, Jokelainen J, et al. Limited overlap
23 Nassif ME, Kosiborod M. A review of cardiovascular outcomes trials between intermediate hyperglycemia as defined by A1C 5.7–6.4%,
of glucose-lowering therapies and their effects on heart failure impaired fasting glucose, and impaired glucose tolerance.
outcomes. Am J Cardiol 2019; 124 (suppl 1): S12–19. Diabetes Care 2011; 34: 2314–16.

www.thelancet.com/diabetes-endocrinology Vol 10 December 2022 881