Association of left ventricular ejection fraction with worsening renal function in patients with acute heart failure: insights from the RELAX-AHF-2 study

Study design and participants

As described previously, RELAX-AHF-2 was a multicentre, randomized, double-blind, placebo-controlled phase 3 study of serelaxin, a recombinant form of human relaxin-2 as a treatment for AHF.^{20, 21} The trial was approved at each study site and written informed consent was obtained from all participants. A total of 6600 who were hospitalized with AHF were recruited between 2 October 2013 and 1 February 2017 at 546 centres in 35 countries. Patients were required to be ≥18 years old with all of the following at entry: dyspnoea; congestion on chest radiograph; brain natriuretic peptide (BNP) ≥500 pg/mL or N-terminal pro BNP (NT-proBNP) ≥2000 pg/mL; systolic blood pressure ≥125 mmHg; mostly mild to moderate renal impairment [estimated glomerular filtration rate (eGFR) 25–75 mL/min/1.73 m²] and persistent HF symptoms after initial intravenous loop diuretic treatment equivalent to ≥40 mg furosemide. Patients received a 48 h infusion of either serelaxin (30 µg/kg/day) or placebo, in addition to standard care. The main study outcome was neutral for either of the co-primary endpoints of time to first event of cardiovascular death at 180 days or occurrence of worsening HF through day 5.²⁰ Serelaxin reduced the incidence of WRF, but this effect was consistent across the spectrum of LVEF (online supplementary Figure S1) and there was no significant interaction between Serelaxin and LVEF on WRF (P-interaction = 0.553). Thus, the treatment groups were pooled for this analysis.

Left ventricular ejection fraction measurement

In RELAX-AHF-2 patients, LVEF was measured by echocardiography during screening or early post-randomization. The method of LVEF quantification was determined at each study site. Patients without LVEF measurement during index hospitalization were excluded (online supplementary Figure S2). Patients were categorized into quartiles according to LVEF: Q1, 7–29%; Q2, 30–38%; Q3, 39–50% and Q4, >50%. These quartiles were similar to guideline classification of HFrEF, Q1 and Q2, HF with mid-range ejection fraction, Q3 and HFpEF, Q4.

Outcome and clinical assessment

According to protocol, serum creatinine was measured during screening (considered the baseline value) and then daily through day 5 or discharge.²⁰ WRF was defined as any rise in creatinine of ≥0.3 mg/dL from baseline through day 5. Patients with missing baseline creatinine were excluded (online supplementary Figure S2). Correlations between LVEF and WRF were assessed analysing LVEF as a categorical variable in quartiles and as a continuous variable.

Physical examination, laboratory tests, and treatment at baseline and daily through day 5 were recorded. Clinical assessment included determination of orthopnoea, oedema, jugular venous pulse, rales and body weight. A composite congestion score ranging from 0 to 8 points was calculated by summing scores for orthopnoea (0–3), peripheral oedema (0–3) and jugular venous pulse (0–2) as described.²² Diuretic response was defined as weight change from baseline through day 5 per 40 mg of intravenous furosemide (or equivalent doses of bumetanide or torsemide) administered during the corresponding period.²³ eGFR was calculated using the simplified Modification of Diet in Renal Disease formula.²⁴ Creatinine and other laboratory tests were performed at each study site using validated assays.

Patients enrolled in RELAX-AHF-2 were assessed at pre-defined clinic visits post-discharge. The main outcome for this analysis was time to first event of the composite of cardiovascular death or HF/renal failure rehospitalization through day 180. Other outcomes of interest were the two components of the composite and all-cause mortality. An independent Clinical Event Committee adjudicated all deaths and rehospitalization events according to pre-defined criteria.^{20, 21}

Statistical analysis

Categorical variables are presented as absolute and relative frequencies for baseline characteristics across the LVEF quartiles. Normally and non-normally distributed variables, objectively determined via Shapiro–Wilk testing for normality, are depicted as mean ± standard deviation and as median (25th, 75th), respectively. Means for continuous variables were compared by ANOVA or Kruskal–Wallis test, while categorical variables were compared by Chi-squared test. A two-tailed P-value of ≤0.05 was considered statistically significant.

Kaplan–Meier curves for WRF incidence in the LVEF quartiles through day 5 were compared using log-rank test. For multivariable analysis, Cox regression models were used to adjust for effect of covariates and to calculate hazard ratios (HR). HR between LVEF quartiles were computed using the lowest LVEF (Q1) group as reference. The HRs in multivariable analysis were adjusted using the following variables: study treatment (serelaxin); geographical region; previous hypertension; peripheral arterial disease; atrial fibrillation/flutter; diabetes mellitus; history of HF; systolic blood pressure; pulse rate; respiratory rate; oedema; haemoglobin; baseline eGFR; bilirubin; received intravenous loop diuretic dose in furosemide equivalent at the time of screening; angiotensin-converting enzyme inhibitor (ACEi) or angiotensin receptor blocker (ARB); beta-blockers; mineralocorticoid receptor antagonist (MRA). Selection of variables was based on P-value <0.2 in univariable analysis and from their availability. In multivariable analysis (online supplementary Table S1), the total number was reduced to 4815 patients as participants with missing variables were excluded (online supplementary Figure S3). Smoothing spline functions taking a reference LVEF value of 50% were generated to explore the non-linear relationship between WRF and LVEF as a continuous variable. To determine if the interaction between LVEF and WRF was influenced by in-hospital treatment, changes in creatinine in Q1 and Q4 were compared using repeated mixed effects measures analysis with adjustment for intravenous diuretic, ACEi, ARB, beta-blocker and MRA dose from baseline through day 5 in addition to adjustment for the baseline variables described above.

All analyses were performed using R statistical software version 3.4.3 (R Foundation for Statistical Computing, Vienna, Austria).

Baseline characteristics

Of the 6600 patients randomized in RELAX-AHF-2, 6545 were eligible for this analysis (online supplementary Figure S2). Of these, 417 patients without LVEF measurement at index hospitalization and 16 without baseline creatinine measurement were excluded, leaving a study population of 6112 patients.

Table 1 summarizes baseline characteristics of the patients in each LVEF quartile. Age, likelihood of female sex and history of hypertension and other comorbidities (with the exception of coronary artery disease) increased as the LVEF quartile advanced from lowest to highest. Systolic blood pressure and body mass index also increased progressively as LVEF increased. The baseline congestion composite score dropped as LVEF increased. Likelihood of chronic kidney disease increased and eGFR dropped progressively from lowest to highest LVEF quartile. Loop diuretic dose and likelihood of receiving an MRA were less in patients with higher LVEF whereas ACEi/ARB or beta-blocker usage were similar across the quartiles.

In-hospital treatment and outcomes

Table 2 summarizes in-hospital treatments and outcomes through hospital day 5 according to LVEF. The total dose of both intravenous and oral loop diuretic decreased progressively as LVEF increased. Neither use of intravenous vasoactive therapy nor mechanical circulatory support differed significantly between quartiles. Weight loss and diuretic responsiveness decreased as the LVEF increased while residual congestion was slightly more likely in Q4 patients than in the other quartiles. Systolic blood pressure reduction was greatest in Q1 patients. Creatinine and blood urea nitrogen increased (and eGFR decreased) in stepwise manner from Q1 to Q4.

While neither worsening HF nor death during the first 5 days of hospitalization differed significantly between the LVEF quartiles, incidence of WRF increased progressively going from 23.1% in Q1 to 34.4% in Q4 (P < 0.001).

Association between worsening renal function and left ventricular ejection fraction

To further assess the association between LVEF and WRF, Kaplan–Meier curves for patients in the four quartiles over the first 5 days of hospitalization were compared (Figure 1). The risk of WRF between quartiles differed significantly (P < 0.0001 by log-rank test) with Q4 patients experiencing the highest risk. To account for baseline differences between the patients in the quartiles that might have influenced this outcome, the HR for WRF was determined without adjustment (Model 1), after adjustment for age and sex (Model 2) and after adjustment for multiple baseline differences identified during univariate analysis (Model 3). As summarized in online supplementary Table S2, in comparison to Q1, the risk of developing WRF was significantly higher in Q4 in Model 1 [HR 1.57, 95% confidence interval (CI) 1.34–1.84; P < 0.001], Model 2 (HR 1.39, 95% CI 1.17–1.64, P < 0.001) and Model 3 (HR 1.2, 95% CI 1–1.43, P = 0.050). Smooth spline modelling with log-transformed HR for the correlation between LVEF and WRF (depicted in online supplementary Figure S4) demonstrates increased risk for WRF as LVEF increases, although the slope of the association flattens after baseline adjustment. As shown online supplementary Figure S5, change in serum creatine over the 5-day period seen in Q4 patients remained significantly greater than in Q1 patients after adjustment for baseline characteristics and for intravenous diuretic and neurohormonal agent dose.

Factors associated with worsening renal function

Baseline haemodynamic abnormalities, in-hospital diuretic treatment and post-treatment residual congestion have been identified as potential contributors to the development of WRF. Consequently, the association between selected variables reflecting these factors and WRF in patients in the highest and lowest LVEF quartiles were compared to determine whether their impact on WRF differed according to LVEF. As noted in Table 3 and online supplementary Table S1, treatment with serelaxin and greater diuretic responsiveness were associated with reduced risk of WRF but the effects of each were seen across the spectrum of LVEF and there was no interaction of either variable with LVEF. In patients with the lowest LVEF (Q1), rales >1/3 way up the lung fields on admission were associated with reduced risk of developing WRF while residual 3+ peripheral oedema and higher clinical congestion score at day 5 were associated with increased risk of WRF. In patients with the highest LVEF (Q4), only higher respiratory rate on admission was associated with increased risk of WRF. When the impact of selected variables on the risk WRF was compared between LVEF quartiles, the only significant interaction was for baseline rales, which was associated with reduced risk of WRF in Q1 and nominally higher risk in Q4. Modest interactions towards higher baseline respiratory rate (P = 0.051) associated with higher risk of WRF in Q4 than in Q1 and greater risk of WRF for residual congestion in Q1 than Q4 were noted. Neither baseline natriuretic peptide levels nor haemoconcentration (as evidenced by change in haemoglobin) was associated with WRF risk in either quartile and there was no significant interaction of the impact of these variables on WRF and LVEF. As shown in online supplementary Figure S5, increases in creatinine were greater in patients in Q4 than in Q4 at all time points. A significant interaction between WRF and LVEF persisted after repeated mixed effects measures analysis which incorporated dose of intravenous loop diuretics and neurohormonal agents as well as the other covariates described previously.

Association between worsening renal function and post-discharge events

As shown in online supplementary Table S3, patients who experienced WRF during index hospitalization were at increased risk for the composite of cardiovascular mortality or HF/renal failure rehospitalization through day 180 (HR 1.18, 95% CI 1.05–1.31; P = 0.004). This was driven predominantly by increased risk for cardiovascular death (HR 1.25, 95% CI 1.04–1.50; P = 0.018). WRF during the first 5 days of hospitalization was also associated with increased risk for all-cause mortality at 180 days (HR 1.29, CI 1.10–1.51; P = 0.002). Occurrence of the 180 composite endpoint according to incident WRF during the index hospitalization is shown for each quartile in Figure 2. While occurrence of WRF did not significantly affect this outcome in patients in the lowest LVEF quartile, patients in Q2–4 who developed WRF in hospital were at higher risk for events that patients without WRF. As shown in Table 4, there was a significant interaction between WRF and LVEF for the composite outcome, with similar directional trends for both components.

Study limitations

In this post-hoc analysis of RELAX-AHF-2 data, confounding variables that were not identified or considered may have affected the results. Entry criteria which excluded patients with systolic blood pressure <125 mmHg or receiving inotrope treatment may have reduced the number of patients with WRF, particularly in the lower LVEF quartiles. Since patients were required to have an eGFR 25–75 mL/min/1.73 m², our findings are most applicable to patients who have evidence of mild to moderate chronic kidney disease during AHF hospitalization. LVEF measurements and serum creatinine values were accessed at local study sites rather than at a central core laboratory. The evaluation of some clinical signs is subjective and considerable interobserver variation has been noted. A change in the absolute creatinine value was used to define WRF and this approach might have failed to detect WRF in elderly cachectic patients with lower baseline creatinine level. Absence of variables from multivariable analysis resulted in a substantial reduction in patient numbers and some important in-hospital variables including inotrope use^{19, 36} and changes in neurohormonal blocking medications were not included due to incompleteness of data.