Dapagliflozin across the range of ejection fraction in patients with heart failure: a patient-level, pooled meta-analysis of DAPA-HF and DELIVER

Figures

• 

  Fig. 1. Effect of dapagliflozin on key clinical outcomes in pooled DAPA-HF and DELIVER dataset. a–f, Incidence of: death from CV causes (a); death from all causes (b); the total number of hospital admissions for HF (c); time to first hospital admission for HF (d); death from CV causes, MI or stroke (e); and death from CV causes or hospital admission for HF (f), according to randomized therapy. Participants randomized to dapagliflozin are shown in blue and those randomized to placebo in red. All figures are Kaplan–Meier curves with an HR and 95% CI estimated from Cox’s model with two-sided P values except for the total number of hospital admissions for HF, which was plotted using the Gosh and Lin method accounting for death from CV causes (the RR is estimated from the joint frailty model with a two-sided P value). No adjustment for multiple comparisons was made. NNT indicates the number of patients who need to be treated over the median duration of follow-up to prevent one event (of the type in each panel). An NNT could not be calculated for the total number of hospital admissions for HF because this was an episode-based rather than a patient-based analysis (that is, patients may have had…

• 

  Fig. 2. Effect of dapagliflozin on clinical outcomes across the range of ejection fraction. a–f, Effect of dapagliflozin on: death from CV causes (a); death from all causes (b); the total number of hospital admissions for HF (c); time to first hospital admission for HF (d); death from CV causes, MI or stroke (e); and death from CV causes or hospital admission for HF (f), according to baseline LVEF. The horizontal blue line shows the continuous HR across the range of LVEF and the shaded area around this line represents the 95% CI from Cox’s model. The overall effect of treatment in the pooled population is shown in each panel as an HR (95% CI) with the two-sided P value from Cox’s model for Wald’s test of interaction between treatment and LVEF. No adjustment for multiple comparisons was made. aRestricted cubic spline and interaction P value derived from LWYY model for total HF hospitalization.

• 

  Fig. 3. Effect of randomized treatment on CV death according to the pre-specified subgroups. Estimates are HRs with error bars representing 95% CIs from Cox’s model and a two-sided P value for interaction from Wald’s test of Cox’s model. No adjustment for multiple comparisons was made. aNot a pre-specified subgroup.

• 

  Fig. 4. Effect of dapagliflozin on clinical outcomes across the range of NT-proBNP. a–f, Effect of dapagliflozin on: death from CV causes (a); death from all causes (b); the total number of hospital admissions for HF (c); time to first hospital admission for HF (d); death from CV causes, MI or stroke (e); and death from CV causes or hospital admission for HF (f), according to baseline NT-proBNP level. The horizontal blue line shows the continuous HR across the range of NT-proBNP levels at baseline and the shaded area around this line represents the 95% CI from Cox’s model. The overall effect of treatment in the pooled population is shown in each panel as an HR (95% CI) with the two-sided P value for Wald’s test of interaction between treatment and NT-proBNP level from Cox’s model. No adjustment for multiple comparisons was made. aRestricted cubic spline and interaction P value derived from LWYY model for total HF hospitalization.

• 

  Extended Data Fig. 1. Distribution of LVEF in pooled DAPA-HF and DELIVER dataset complete. Distribution of left ventricular ejection fraction (LVEF) in the total population in DAPA-HF and DELIVER.

• 

  Extended Data Fig. 2. Effect of dapagliflozin on clinical outcomes across the range of NT-proBNP. Effect of dapagliflozin on death from cardiovascular causes (CV death) and CV death or hospitalisation for heart failure (HF hospitalisation) where the definition of CV death used excluded undetermined deaths from the definition of CV death (top two panels) and according to the original trial definitions (that is, including undetermined deaths in DAPA-HF and excluding undetermined deaths in DELIVER) (bottom two panels). The horizontal blue line shows the continuous hazard ratio (HR) across the range of left ventricular ejection fraction (LVEF) and the shaded area around this line represents the 95% confidence interval (95%CI) estimated from a Cox model. The overall effect of treatment in the pooled population is shown in each panel as a HR (95%CI) with the two-sided p-value estimated from a Cox model for the Wald test of interaction between treatment and LVEF. No adjustment for multiple comparisons was made.