Author + information
- Received September 15, 2014
- Revision received January 6, 2015
- Accepted January 7, 2015
- Published online May 1, 2015.
- Beth A. Davison, PhD∗∗ (, )
- Marco Metra, MD†,
- Gad Cotter, MD∗,
- Barry M. Massie, MD‡,
- John G.F. Cleland, MD§,‖,
- Howard C. Dittrich, MD¶,
- Christopher Edwards, BS∗,
- Gerasimos Filippatos, MD#,
- Michael M. Givertz, MD∗∗,
- Barry Greenberg, MD††,
- Piotr Ponikowski, MD‡‡,
- Adriaan A. Voors, MD, PhD§§,
- Christopher M. O’Connor, MD‖‖,
- John R. Teerlink, MD‡,¶¶,
- on behalf of the PROTECT and RELAX-AHF Executive Committees
- ∗Momentum Research Inc., Durham, North Carolina
- †Cardiology, Department of Medical and Surgical Specialties, Radiological Sciences, and Public Health, University of Brescia, Brescia, Italy
- ‡Division of Cardiology, School of Medicine, University of California-San Francisco, San Francisco, California
- §Department of Cardiology, University of Hull, Kingston upon Hull, United Kingdom
- ‖National Heart and Lung Institute, Royal Brompton and Harefield Hospitals National Health Service Trust, Imperial College, London, United Kingdom
- ¶NovaCardia, San Diego, California
- #Department of Cardiology, Athens University Hospital, Attikon, Athens, Greece
- ∗∗Cardiovascular Division, Brigham and Women’s Hospital, Boston, Massachusetts
- ††Division of Cardiology, University of California at San Diego, San Diego, California
- ‡‡Department of Cardiology, Medical University, Clinical Military Hospital, Wroclaw, Poland
- §§Department of Cardiology, University of Groningen, Groningen, the Netherlands
- ‖‖Division of Cardiology, Duke University Medical Center, Durham, North Carolina
- ¶¶Section of Cardiology, San Francisco Veterans Affairs Medical Center, San Francisco, California
- ↵∗Reprint requests and correspondence:
Dr. Beth A. Davison, Momentum Research Inc., 3100 Tower Boulevard, Suite 802, Durham, North Carolina 27707.
Objectives These studies conducted analyses to examine patient characteristics and outcomes associated with worsening heart failure (WHF).
Background WHF during an admission for acute heart failure (AHF) represents treatment failure and is a potential therapeutic target for clinical trials of AHF.
Methods Individual patient data from the PROTECT (Placebo-Controlled Randomized Study of the Selective A1 Adenosine Receptor Antagonist Rolofylline for Patients Hospitalized with Acute Decompensated Heart Failure and Volume Overload to Assess Treatment Effect on Congestion and Renal Function) and RELAX-AHF (Relaxin in Acute Heart Failure) phase II and III studies were pooled for analysis.
Results Of 3,691 patients, death or WHF through day 5 occurred in 12.4%, ranging from 9.5% to 14.5% among studies. A multivariable model provided modest discrimination between patients who did or did not develop WHF (C-index = 0.68). After multivariable adjustment, WHF was associated with a mean increase in length of stay of 5.2 days (95% confidence interval [CI]: 4.6 to 5.8 days) and increased risks of 60-day HF or renal failure readmission or cardiovascular death (hazard ratio [HR]: 1.64, 95% CI: 1.34 to 2.01) and 180-day mortality (HR: 1.93, 95% CI: 1.55 to 2.41) (all p < 0.001). The risk of mortality was higher in patients whose WHF required intravenous inotropes or mechanical therapy (HR: 3.03, 95% CI: 2.11 to 4.36) compared with patients whose WHF was treated with intravenous loop diuretic alone (HR: 1.80, 95% CI: 1.36 to 2.36) (both p < 0.001). WHF was associated with larger increases in markers of renal and hepatic dysfunction during the first days of admission, but remained significantly associated with adverse outcomes after adjustment for these changes.
Conclusions WHF during the first 5 days of admission for AHF occurred in approximately 10% to 15% of patients and was associated with longer length of stay and higher risk for readmission and death.
During a hospital admission for acute heart failure (AHF), the majority of patients’ symptoms are relieved. However, not all patients recover well, and limited information is available on the factors contributing to lengthy hospital stays, repeat admissions, and high mortality in these patients. Guidance regarding therapies in current use is based on limited data (1).
Worsening heart failure (WHF) can be defined as worsening of symptoms and signs of heart failure requiring intensification of intravenous (IV) therapy or initiation of mechanical therapy during an admission for AHF. WHF has been associated with adverse outcomes including longer hospital stay (2,3), rehospitalization (2,4), and death through 6 months (2,5–7) in several studies. WHF, with slightly varying definitions, has been incorporated into the primary efficacy endpoint in several clinical studies of investigational drugs in patients with AHF, including VERITAS (Value of Endothelin Receptor Inhibition With Tezosentan in Acute Heart Failure) (8,9), REVIVE (Randomized Evaluation of Intravenous Levosimendan Efficacy) (10), PROTECT (Placebo-Controlled Randomized Study of the Selective A1 Adenosine Receptor Antagonist Rolofylline for Patients Hospitalized with Acute Decompensated Heart Failure and Volume Overload to Assess Treatment Effect on Congestion and Renal Function) (11,12), and RELAX-AHF (Relaxin in Acute Heart Failure) (13,14). In addition, WHF is a component of the primary endpoint in the ongoing TRUE-AHF (Trial to Evaluate the Efficacy and Safety of Ularitide [Urodilatin] Intravenous Infusion in Patients Suffering From Acute Decompensated Heart Failure) trial (NCT01661634).
We designed a pooled analysis of individual patient data from several recent studies of AHF to examine patient characteristics and outcomes associated with WHF.
Studies and data included
A pooled analysis was conducted using patient-level data from the PROTECT (12,15) and RELAX-AHF phase II/III (14,16) studies (Table 1). In these studies, physician-assessed HF signs and symptoms were collected daily using standard scales and WHF was defined as worsening signs and/or symptoms of HF requiring intensification of IV therapy or mechanical support. Ethics committee approval was obtained at each site and informed consent obtained from each patient who participated in these studies.
Potential predictors were chosen based on important baseline measures identified and considered in models for similar outcomes developed for the PROTECT (17,18) and RELAX-AHF (7) studies. Analyses were restricted to data available across all the studies and excluded variables with a high proportion of missing values. Most covariates considered were available for >99% of patients; laboratory parameters had the highest proportion of missing values, with 11% the highest proportion missing (Online Table 1). Missing values for continuous variables were imputed with the median, and categorical variables with the mode, within study. Highly right-skewed variables were log-transformed for analysis. Extreme outliers not physiologically plausible were either set to missing and then imputed or truncated prior to any transformation.
When considered as an outcome, 29 patients who died without a preceding WHF event by 5 days were considered as having a WHF event. In both the PROTECT and RELAX-AHF programs, the treatment for the WHF event was collected on a standardized worksheet. The treatment required for the first WHF event, if a patient experienced more than 1, was categorized as IV loop diuretic alone, IV inotrope (e.g., dobutamine, norepinephrine, levosimendan, phenylephrine) or mechanical therapy (e.g., mechanical ventilation, circulatory support, ultrafiltration), or other treatment (e.g., IV nitrates, nesiritide, nonloop diuretic). Other outcomes examined included the length of stay (LOS) of the initial hospitalization, cardiovascular (CV) death or HF/renal failure (RF) rehospitalization through day 60, and all-cause death through day 180. LOS was truncated at 30 days for 79 patients and was imputed as 31 days for 126 in-hospital deaths.
A single variable was chosen among collinear (e.g., body mass index, height, and weight) and/or highly correlated (e.g., hemoglobin and hematocrit) candidate predictors. Linear regression models were used for LOS, and Cox regression models for 60-day CV death or HF/RF rehospitalization and 180-day mortality. The linearity of the associations of each continuous variable and each outcome were assessed by evaluating the statistical significance of a test of the nonlinear terms for restricted cubic spline transformations with 4 knots. Where associations were nonlinear (p < 0.05), transformations were chosen through examination of plots, changes in Akaike information criterion, and clinical judgment. Plots of the associations of each ordinal predictor with each outcome were examined; these covariates were treated as linear terms. Univariable associations were assessed following imputation and transformation. Multivariable models were developed using backward selection with p < 0.10 as the criterion for inclusion in the model for all outcomes except LOS, for which p < 0.05 was the criterion. All models were stratified by study. Discrimination of the final models was assessed using a C-statistic for Cox regression and adjusted R-square for linear regression. Bias-corrected C-statistics, differences in C-statistics between models, and associated confidence intervals (CIs) were obtained through bootstrapping with 1,000 bootstrap samples.
Estimates of the effect of WHF by day 5 were adjusted for the covariates included in the final multivariable model for each of the other outcomes. Thirty-nine patients who died by day 5 were excluded from the analysis of LOS, and censored or event times by day 5 were excluded from the time-to-event analyses. The proportional hazards assumption was tested by including a WHF-by-time interaction in the Cox regression model for all-cause mortality, and hazard ratios (HRs) for WHF estimated assuming the ln(HR) decreases linearly over time. Alternate models for the time-to-event outcomes included all patients and treated WHF as a time-dependent covariate.
Changes in laboratory parameters were compared between patients with and without WHF using linear regression models including as covariates the laboratory parameter’s baseline level and study. WHF associations were then further adjusted for the change in laboratory values (blood urea nitrogen [BUN], creatinine, alanine aminotransferase [ALT] or aspartate aminotransferase [AST], albumin, and cholesterol) from baseline to day 2 and the baseline value. Either ALT or AST was included, depending on which value had the lower Akaike information criterion for the unadjusted association with the outcome. Laboratory parameters were adjusted for individually first and then all simultaneously.
Data from 3,734 patients were included (Table 1). Slight differences in patient characteristics (Online Table 1) reflect differences in eligibility criteria between the 2 phase II/III programs. For example, RELAX-AHF excluded patients with systolic blood pressure <125 mm Hg compared with <95 mm Hg in PROTECT, and PROTECT excluded patients with de novo AHF.
Predictors of WHF
A total of 461 of 3,691 patients (12.4%) with a nonmissing outcome died or experienced WHF by day 5 (Table 2). The strongest multivariable predictors of WHF were higher BUN, respiratory rate, hematocrit, and systolic blood pressure, which had a V-shaped association with WHF (Table 3). The C-statistic for the model was 0.67 (95% CI: 0.65 to 0.70).
Association of WHF with other clinical outcomes
The occurrence of WHF was associated with a mean increase in LOS of 5.7 days (95% CI: 5.1 to 6.4 days) ranging from 4.2 days in the PROTECT pilot study to 6.2 days in the PROTECT main study (Figure 1A). Neither unadjusted nor adjusted estimates varied significantly by study (p = 0.29 and 0.08, respectively). Online Table 2 provides the multivariable linear regression model developed for LOS. After multivariable adjustment, WHF was associated with a mean increase in LOS of 5.2 days (95% CI: 4.6 to 5.8 days, p < 0.0001) (Figure 1A). The R-squared for the multivariable model for LOS increased from 0.18 to 0.24 with the inclusion of WHF as a predictor, with a bias-corrected difference of 0.05 (95% CI: 0.03 to 0.07).
Unadjusted estimates of the association of WHF with CV death or HF/RF rehospitalization through day 60 ranged from a HR of 2.08 in PROTECT to 3.86 in Pre-RELAX-AHF, with a combined unadjusted estimate of 2.19 (95% CI: 1.80 to 2.67) (Figure 1B). Neither unadjusted nor adjusted estimates varied significantly by study (p = 0.58 and p = 0.43, respectively). Online Table 3 provides the multivariable Cox regression model developed for CV death or HF/RF rehospitalization through day 60. After multivariable adjustment, the HR for the association of WHF with 60-day CV death or HF/RF readmission was 1.64 (95% CI: 1.34 to 2.01; p < 0.0001) (Figure 1B). The C-statistic for the multivariable model increased from 0.70 to 0.71 with the inclusion of WHF as a predictor, with a bias-corrected difference of 0.01 (95% CI: 0.003 to 0.02). An alternate model including all patients and treating WHF as a time-dependent covariate resulted in nearly the same estimated association (HR: 1.68, 95% CI: 1.38 to 2.04; p < 0.0001).
HR for 180-day all-cause mortality for patients with WHF compared with those without ranged from 1.91 in RELAX-AHF to 4.15 in Pre-RELAX-AHF, with a combined unadjusted estimate of 2.61 (95% CI: 2.20 to 3.10) (Figure 1C). Neither unadjusted nor adjusted estimates varied significantly by study (p = 0.45 and p = 0.72, respectively). After adjustment for covariates associated with all-cause mortality in the multivariable model (Online Table 4), the HR for the association of WHF with 180-day all-cause mortality was 1.93 (95% CI: 1.55 to 2.41; p < 0.0001) (Figure 1C). The C-statistic for the multivariable model increased from 0.75 to 0.76 with the inclusion of WHF as a predictor, with a bias-corrected difference of 0.0050 (95% CI: −0.0033 to 0.0132). An alternate model including all patients and treating WHF as a time-dependent covariate resulted in an estimated HR of 1.98 (95% CI: 1.60 to 2.44; p < 0.0001).
The association of WHF with all-cause mortality appeared to decrease over time (WHF-by-time interaction p = 0.0264). Estimated HR were 2.34 (95% CI: 1.79 to 3.07) at 30 days, 2.00 (95% CI: 1.60 to 2.50) at 60 days, 1.70 (95% CI: 1.32 to 2.20) at 90 days, 1.45 (95% CI: 1.03 to 2.06) at 120 days, and 1.06 (95% CI: 0.59 to 1.91) at 180 days. The increased risk of dying was very high immediately following the WHF event, but as the patient survived the associated risk declined—to almost no increased risk after 6 months.
Subgroups of WHF
Compared with patients who did not experience WHF by day 5, the combined unadjusted estimated HR for death by day 180 was 2.29 (95% CI: 1.77 to 2.98) for WHF treated with IV loop diuretic alone, 4.86 (95% CI: 3.51 to 6.74) for WHF requiring an IV inotrope or mechanical therapy, and 1.62 (95% CI: 0.98 to 2.66) for WHF that was treated with other therapy (Table 4). After multivariable adjustment, these HR were 1.80 (95% CI: 1.36 to 2.36), 3.03 (95% CI: 2.11 to 4.36), and 1.23 (95% CI: 0.71 to 2.16), respectively.
Association between WHF and biomarkers and signs and symptoms of congestion
Patients who had WHF had larger increases in creatinine (p < 0.001) and smaller decreases in ALT and AST (p < 0.001 for both) than did patients who did not experience WHF (Figure 2). Adjustment for changes to day 2 in markers of renal and hepatic dysfunction and metabolic abnormalities (cholesterol and albumin) explained little of the associations between WHF and other outcomes (Table 5). Simultaneous adjustment for all laboratory markers had the largest impact on the association of WHF with 180-day mortality, with an 11% reduction in the estimated HR (from 1.93 to 1.71).
Simultaneous adjustment for changes in signs and symptoms of congestion (peripheral edema, rales, and dyspnea on exertion) at day 2 had little effect on the association of WHF with CV death or HF/RF rehospitalization to day 60, reduced the HR for association with 180-day mortality by 10%, and reduced the mean difference in length of stay by 15% (Table 5). Adjustment for changes in all laboratory markers and HF signs and symptoms reduced the mean difference in LOS from 5.20 to 4.31 days (17%), reduced the HR for CV death or HF/RF rehospitalization through day 60 from 1.64 to 1.57 (4%), and reduced the HR for death through day 180 from 1.93 to 1.58 (18%). The associations of WHF with these outcomes remained highly significant with all p < 0.0001.
WHF occurred in 10% to 15% of patients within the first 5 days of admission. A model incorporating baseline demographics, medical history, and clinical assessments provided modest discrimination between patients who did and did not develop WHF (C-statistic 0.68). The strongest predictor of WHF in the pooled database was BUN. BUN is not only a marker of renal dysfunction and hypoperfusion but may also reflect neurohormonal activation in HF (19). The occurrence of WHF by day 5 was associated with a prolonged LOS, an increased risk of 60-day CV death or HF/RF rehospitalization, and an approximate doubling of the risk of 180-day mortality. The limited reduction in the magnitude of these associations after covariate adjustment suggests that only a small part of the risk of more adverse outcomes in patients with WHF is explained by more severe HF and comorbidities at baseline. Patients with WHF had laboratory markers of more severe end-organ damage at baseline that showed greater subsequent deterioration. However, changes in markers of end-organ dysfunction, worsening metabolic abnormalities, or symptoms and signs of HF during the first 2 days of admission only partially explained the increase in risks associated with WHF. WHF treated with IV loop diuretic alone appears to confer a somewhat smaller risk of 180-day mortality than does WHF requiring IV inotropes or mechanical therapy, but WHF confers increased risk regardless of the therapy required. This finding suggests that WHF is an important adverse event in the course of an AHF admission regardless of its therapy. As is the case with other adverse CV events, those who require more intensive therapy carry a higher risk of adverse outcome; however, the event itself—namely WHF—is a significant occurrence regardless of its severity.
WHF is gaining recognition as an important short-term outcome and a potential therapeutic target in AHF (20,21), and hence it has been incorporated as an endpoint in many clinical studies exploring new therapies in AHF. The current analysis suggests that WHF is an important event that cannot simply be explained by other measured clinical characteristics. This may be because the cause of WHF is heterogeneous or because many potential precipitants were not measured, suggesting that other unidentified mechanisms contribute to the occurrence of such events. That the association of WHF with outcomes remains robust after multivariable adjustment for baseline characteristics and further adjustment for markers of worsening organ dysfunction, metabolic abnormalities, and degree of congestion during the first 2 days of admission strengthens this conclusion.
The question remains whether WHF is preventable and whether its prevention can modify longer-term outcomes. Studies of diuretics have failed to demonstrate a benefit on either WHF or other outcomes (9,12,22) consistent with the current finding that changes in symptoms and signs of congestion did not significantly modify the associations of WHF with outcomes. However, in the REVIVE II study, compared with placebo, levosimendan treatment was associated with a reduction in WHF events (15% vs. 26%, p value not provided) (10), but with a trend toward increased early mortality. In the RELAX-AHF study, serelaxin administration was associated with 30% or greater reductions in WHF to day 14 and 180-day CV and all-cause mortality, but not with 60-day HF/RF rehospitalization or CV death (14). Thus, administration of some IV therapies during the first days of admission for AHF has led to a reduction in the rate of WHF, but this has not translated consistently into improved longer-term outcomes.
The unavailability of data regarding previous hospitalization for HF and biomarkers such as lymphocyte percentage, troponin, and natriuretic peptides may have limited our ability to predict the occurrence of WHF and somewhat exaggerated the associations of WHF with other outcomes. Additionally, the linear regression model for LOS demonstrated some lack of fit, and alternate models such as survival analysis might be explored. Because WHF occurred in many instances during the nighttime or weekend, HF symptoms and signs were not captured at the time of WHF and hence are not available for us to analyze. We assumed that data were missing at random; single imputation of missing values may have overestimated the precision of covariate effects. External validation of the model for WHF, particularly in a broader group of patients such as in an AHF clinical registry in which the occurrence of WHF is collected prospectively, would provide greater confidence in the model’s performance.
WHF occurs more commonly in sicker patients, particularly in those with high BUN values, but its occurrence can be predicted with only modest accuracy from baseline variables and early changes in laboratory measurements. WHF is associated with a longer LOS and increased risks of 60-day CV death or HF/RF rehospitalization and 180-day mortality. These associations cannot be explained by a different risk profile of patients with WHF, suggesting that the occurrence of WHF conveys a specific independent risk for short- and medium-term adverse outcomes in AHF.
COMPETENCY MEDICAL KNOWLEDGE: WHF during an admission for AHF, that is, worsening of symptoms and signs of HF requiring intensification of IV or mechanical therapy occurs in 10% to 15% of patients. This event is only partially predicted by more severe HF or other baseline characteristics at admission and is associated with a substantial increase in adverse short-term outcome, suggesting that these events should become a focus of attention in patients admitted for AHF.
TRANSLATIONAL OUTLOOK: Further larger studies are needed to confirm the independent association of WHF with outcomes. Further research is need to examine whether prevention of WHF should become a therapeutic priority in patients admitted for AHF and whether specific therapies can be developed to reduce adverse outcomes after a WHF event has occurred.
For supplemental tables, please see the online version of this article.
The PROTECT studies were sponsored by NovaCardia, a subsidiary of Merck. The RELAX-AHF studies were sponsored by Corthera, a Novartis affiliate company. No external support was provided for these analyses. Drs. Davison, Cotter, and Edwards are employees of Momentum Research Inc., which has received research grants from Novartis, Corthera, Novacardia, Trevena, Merck, Amgen, ChanRx Corp., Cardio3 Biosciences, Sorbent Therapeutics, and Singulex. Dr. Metra has received consulting fees from Bayer, Novartis, Servier, and Trevena. Dr. Massie has received consulting fees and research funding from Merck/Novacardia and Corthera/Novartis. Dr. Cleland has received honoraria and research funding from MSD, Novartis, and Amgen. Dr. Dittrich is a former employee of NovaCardia and Sorbent; and is a consultant to Corthera. Dr. Filippatos is a member of the steering committees of trials sponsored by Novartis, Cardiorentis, and Bayer. Dr. Givertz is an employee of Brigham and Women’s Hospital. Dr. Greenberg has received consulting fees from Novartis. Dr. Ponikowski has received consulting and lecture fees from Novartis, Johnson & Johnson, Bayer, Coridea, and Cardiorentis. Dr. Voors serves on the steering committee of RELAX-AHF, and has received speaker, consultancy, or research fees from Amgen, Bayer Health Care, Cardio3Biosciences, Celladon, GlaxoSmithKline, Merck, Nephera Ltd., Novartis, Servuier, Singulex, Trevena, and Vifor. Dr. O’Connor has received consulting fees from Cardiorentis, Resmed, and Stealth Peptides; and research funding from Otsuka and Resmed. Dr. Teerlink has received consulting fees and research funding from Merck/NovaCardia, Corthera, Novartis, and Amgen/Cytokinetics. Anthony DeMaria, MD, served as Guest Editor for this paper.
- Abbreviations and Acronyms
- acute heart failure
- alanine aminotransferase
- aspartate aminotransferase
- blood urea nitrogen
- confidence interval
- heart failure
- hazard ratio
- length of stay
- renal failure
- worsening heart failure
- Received September 15, 2014.
- Revision received January 6, 2015.
- Accepted January 7, 2015.
- 2015 American College of Cardiology Foundation
- Cotter G.,
- Metra M.,
- Weatherley B.D.,
- et al.
- Metra M.,
- Cotter G.,
- Davison B.A.,
- et al.,
- for the RELAX-AHF Investigators
- Packer M.,
- Colucci W.,
- Fisher L.,
- et al.,
- for the REVIVE Heart Failure Study Group
- Weatherley B.D.,
- Cotter G.,
- Dittrich H.C.,
- et al.
- Cleland J.G.,
- Chiswell K.,
- Teerlink J.R.,
- et al.