Identifying Patients Hospitalized with Heart Failure at Risk for Unfavorable Future Quality of Life

Abstract

Background

Communicating prognosis to enable shared decision-making is strongly endorsed by heart failure (HF) guidelines. Patients are concerned with both their quantity and quality of life (QoL). To facilitate the recognition of patients at high risk for unfavorable future QoL or death, we created a simple prognostic tool to estimate this combined outcome.

Methods and Results

We identified factors associated with 6-month mortality or persistently unfavorable QoL, defined by Kansas City Cardiomyopathy Questionnaire (KCCQ) scores <45 at 1 and 24 weeks after hospital discharge, among 1458 patients from the Efficacy of Vasopressin Antagonism in HF Outcome Study with Tolvaptan (EVEREST). Within 24 weeks of discharge, 478 (32.8%) patients had died and 192 (13.2%) patients had serial KCCQ scores < 45. After adjusting for 23 pre-discharge covariates, independent predictors of the combined end point included low admission KCCQ score, high B-type natriuretic peptide, hyponatremia, tachycardia, hypotension, absence of β-blocker therapy, and history of diabetes mellitus and arrhythmia. A simplified pre-discharge HF score for subsequent death or unfavorable QoL had moderate discrimination (c-statistic 0.72). Pre-discharge clinical covariates were substantially different in predicting the QoL end point as compared with traditional death or rehospitalization end points.

Conclusions

At the time of hospital discharge, readily available clinical characteristics are associated with HF patients at high risk for suffering persistently unfavorable QoL or death over the next 6 months. Such information can target patients for whom aggressive treatment options (e.g., devices or transplantation) and/or end-of-life discussions should be strongly considered prior to discharge.

In the care of patients with heart failure (HF), estimating and communicating prognosis is endorsed by clinical guidelines^1-3 and is considered to be an important component of high-quality healthcare.⁴ Without explicit education regarding future expectations regarding quantity and quality of life (QoL), patients and families are inadequately equipped to make important decisions about the optimal direction of their treatment. Yet despite the importance of estimating and discussing prognosis, this is seldom done in routine clinical practice^5-7 and the majority of patients with HF underestimate their risk for adverse outcomes.^8,9

Hospitalization is a critical event in the clinical course of HF. For some it represents a transient clinical deterioration, while for others it heralds a progressive phase of HF marked by recurrent hospitalizations, lower functional status, severe symptoms and death. This latter course, however, is common given the high rates of mortality (∼25%),^{10, 11} rehospitalization (∼50%),^{12, 13} and severe symptom burden¹⁴ in the 6 months following hospitalization for an episode of worsening HF. Thus, HF hospitalization represents an important opportunity to estimate patients' prognosis in an effort to identify those in whom health care providers should take the time to formally discuss treatment options and patients' end-of-life wishes.¹⁵

While there are numerous prognostic models to estimate survival and rehospitalization in the acute care setting,^{10, 11, 16-18} these models are fundamentally limited in that they fail to include measures of QoL, despite the availability of validated HF instruments and their frequent use in randomized trials.^{19, 20,21-25} By focusing singularly upon mortality, current prognostic models incorrectly presume that all survival represents a favorable outcome, even in light of evidence that many patients would make healthcare decisions that improve their QoL at the expense of its quantity.^25-27

One needed component of improved communication and shared decision-making between health care providers and their patients is a means for recognizing patients at high risk for either death or persistently unfavorable QoL. To address this need, we analyzed the Efficacy of Vasopressin Antagonism in Heart Failure Outcome Study with Tolvaptan (EVEREST).^{28, 29} EVEREST enrolled patients admitted with decompensated HF and collected serial health status measures at the time of hospitalization and after discharge. Our aim was to describe the frequency with which patients do not survive with favorable QoL following HF hospitalization, and then to create a simple clinical tool to enable health care providers to identify such high-risk patients approaching the time of hospital discharge. We also assessed the degree of discordance between predictors of this novel QoL end point with the more traditional outcomes of death and rehospitalization. Our goal was to create a simple tool that could potentially be used at the time of hospital discharge to facilitate communication between health care providers and their patients about prognosis—including expected QoL after discharge—to support timely discussions regarding treatment options and end-of-life care.

Methods

Study Population

The design of the EVEREST trial has been previously described.³⁰ EVEREST examined both short-²⁸ and long-term²⁹ outcomes in patients hospitalized with a primary diagnosis of HF and randomized within 48 hours of admission to tolvaptan or placebo. From October 2003 to February 2006, patients were enrolled from 359 sites in North America, South America, and Europe. The date of the last follow-up was July 2006. Important eligibility criteria included history of chronic HF for at least 30 days prior to hospitalization, evidence of volume overload, and left ventricular ejection fraction (LVEF) ≤ 40%. Key exclusion criteria were expected survival < 6 months, cardiac mechanical support implantation, cerebrovascular accident in the last 30 days, dialysis, morbid obesity, substance abuse, systolic blood pressure < 90 mm Hg, serum creatinine >3.5 mg/dL, and hemoglobin <9 g/dL. We restricted the cohort to those patients who underwent formal assessments of health status. Because these assessments were introduced mid-way through the trial, health status measures were obtained consecutively on the last 2033 of the 4133 EVEREST participants. Among these eligible patients, 1458 (72%) survived to hospital discharge and had adequate follow up. All patients continued to receive standard therapy for HF while enrolled in the trial. The trial protocol was approved by the appropriate Institutional Review Board at each study site, and all enrollees provided written informed consent.

Study Variables

Baseline data were collected within 48 hours of hospital admission,³⁰ including formal health status measures that were also obtained 1 week after discharge and 24 weeks later using the Kansas City Cardiomyopathy Questionnaire (KCCQ).¹⁹ The KCCQ is a validated, disease-specific, 23-item, self-administered questionnaire that quantifies health status in patients with HF. The KCCQ overall summary score, which includes information from the physical limitation, symptom, social limitation, and QoL scales, was used for all analyses. The KCCQ summary score ranges from 0 to 100, with higher scores reflecting better health status. Culturally and linguistically validated versions were used.³¹ Among other study covariates, glomerular filtration rate (GFR) was calculated using the Modification of Diet in Renal Disease equation.³² B-type natriuretic peptide (BNP) was measured at a centralized laboratory using the Triage Assay (Biosite, CA). History of arrhythmia was defined as any sustained atrial or ventricular arrhythmia.

Study Outcomes

The primary end point for this analysis was the composite of persistently unfavorable QoL (as defined by KCCQ <45 at weeks 1 and 24 following discharge) or all-cause mortality. The combined end point was chosen because both components were felt to warrant consideration of advanced therapies and/or end-of-life care; in a step-wise fashion, patients are first classified by level of risk for adverse future outcomes (i.e., failure to achieve favorable future QoL), and then, for those at sufficiently high risk, the type of advanced therapy (e.g., transplantation, mechanical circulatory support, hospice) is chosen based on certain patient characteristics and patient preferences. The KCCQ cut point of 45 was chosen a priori based upon an association of KCCQ <45 with advanced HF.^19,33,34 We secondarily assessed unfavorable future QoL, all-cause mortality, and rehospitalization end points individually.

Statistical Analysis

Baseline characteristics were compared using Student's t-test or the Wilcoxon rank-sum test for continuous variables and Chi-square or Fisher's exact test for categorical variables.

The association of baseline clinical variables and the primary end point was assessed using multivariable regression models. Typical analyses utilize logistic regression to estimate adjusted odds ratios, which are then generally interpreted as relative risks. However, in this study the event being modeled was not rare, in which case odds ratios are poor estimated of relative risks. To address this issue, we estimated adjusted relative risks directly using hierarchical modified Poisson regression models,³⁵ adjusted for site (given the reported association of site with outcome).³⁶ For variables with multiple measurements during the index hospitalization, the most recent data point at the time of discharge was included. Model predictors were chosen a priori based on previous inpatient HF prognostic models^{10, 11, 16-18, 37, 38} and clinical factors related to QoL.¹⁹ These covariates included demographics (age, sex, race), clinical variables (current smoking; history of hypertension, diabetes mellitus, stroke, severe chronic obstructive pulmonary disease [COPD], arrhythmia, coronary artery disease as collected on the case report form at admission; LVEF; prescription of a β-blocker at discharge; heart rate, systolic blood pressure, and the degree of edema on the day of discharge), laboratory values (pre-discharge values for GFR, blood urea nitrogen, hemoglobin, sodium, b-type natriuretic peptide [BNP], and QRS width), and baseline KCCQ. NYHA functional class was not included because of the homogeneity in NYHA assessment at the time of hospital admission and known limitations in inter-rater reproducibility.^39-42 Associations of continuous variables with outcomes were assessed for linearity using restricted cubic spine terms.⁴³ Variables found to have non-linear relationships with the outcome were categorized using cut points chosen based on prior prognostic models and clinically meaningful values. Sensitivity analyses altering the cut point for KCCQ to < 40 or < 50 in defining the primary end point were performed to evaluate the robustness of our findings. An additional sensitivity analysis was performed for the full model using continuous variables whenever possible (transformed for non-linear relationships) to evaluate whether a significant loss of predictive ability had occurred by categorizing the continuous variables.

We then created a simplified risk score which could be more readily applied in clinical practice. First, the contributions of model predictors were assessed by F-statistics. Predictors were sequentially eliminated from the full model by removing the predictor with the smallest contribution until further variable elimination led to a greater than 5% loss in model prediction (i.e., the R² for the predictions from the reduced model was >95% of the explained outcome from the full model).⁴³ Second, the β regression estimates for each variable in the reduced model were divided by the smallest β weight to create standardized β weights, which were then rounded to the nearest integer. Finally, bootstrap model validation for the risk score was conducted by replicating our risk score model on 1000 datasets generated using random sampling with replacement. A risk score was developed for each of the 1000 datasets and these were used to predict outcomes in both the bootstrap and original datasets to describe potential model optimism. Risk score calibration was assessed by graphing the observed versus predicted rates of the combined end point of death or persistently unfavorable health status within deciles of predicted risk. C-statistics were calculated to characterize discriminatory accuracy. The Hosmer-Lemeshow statistic was calculated to determine goodness of fit, and the slope of the linear predictor was calculated to assess model calibration.⁴³

In order to determine whether the predictors of future QoL differed from the more traditional clinical HF end points of death or rehospitalization (particularly since the rehospitalization is presumed to be driven by symptoms), we constructed separate risk models for the individual 24-week end points of (A) all-cause mortality, (B) persistently unfavorable QoL, and (C) rehospitalization. Mortality and rehospitalization were modeled using time to event analyses to produce hazard ratios, and QoL was modeled using modified Poisson regression models to produce relative risks. The relative contribution of each variable to each individual outcome was assessed using F-statistics.

The mean rate of missing data per patient was < 5%. Rates of missing data for individual variables was 10.5% for BNP levels, <3% for 5 variables, and <0.5% for all remaining variables. Missing data were imputed using IVEWARE.⁴⁴ Imputation was predicated upon variables collected at surrounding time intervals and variables known to correlate with model covariates.

All analyses were performed using SAS software, release 9.1.3 (SAS Institute, Cary, NC), IVEWARE⁴⁴, and R version 2.6.0.⁴⁵ A p-value of <0.05 was used to define statistical significance.

Results

For the patients included in this analysis, the mean age was 66.5 ± 11.7 years, 75% were male, and 85% were Caucasian. The etiology of HF was primarily ischemic. Mean LVEF was 27.2 ± 8.0%. The average duration of HF diagnosis prior to study enrollment was 5.9 ± 4.9 years, and 82% reported a previous hospitalization for HF. Patients frequently had significant comorbidities, including 39% with diabetes, 19% with cerebrovascular disease, and 10% with severe COPD. Median length of stay for the index hospitalization was 7 days (IQR 3-12). Additional baseline characteristics, stratified by outcome, are described in Table 1. When comparing those eligible for the analysis to the overall EVEREST cohort, no significant differences in the Table 1 variables were observed.

Table 1. In-hospital characteristics of patients, stratified by the primary outcome.

Characteristic (means ± SD or %, unless stated as median [IQR])	Unfavorable QoL (Death or KCCQ <45 at both 1 and 24 weeks post discharge)	Favorable QoL (Survival to 24 weeks with KCCQ ≥45 at either 1 or 24 weeks post discharge)	P-value
	N=670	N=788
KCCQ OS score at admission (0-100)	24.0 ± 15.4	35.8 ± 19.5	< 0.001
Demographics
Age (years)	68.1 ± 11.6	65.1 ± 11.6	< 0.001
Female	25.7%	24.9%	0.726
Race			0.226
Caucasian	86.3%	83.4%
Black	7.5%	8.1%
Other	6.3%	8.5%
BMI at admission	28.1 ± 5.7	28.5 ± 5.7	0.178
Smoking, current	9.7%	12.9%	0.054
Hypertension, history of	71.5%	72.2%	0.762
Diabetes, history of	43.1%	36.9%	0.016
Chronic kidney disease, history of	39.2%	23.2%	< 0.001
Stroke, history of	14.8%	10.1%	0.006
Peripheral vascular disease, history of	24.6%	22.2%	0.285
COPD, severe	12.5%	7.2%	< 0.001
HF characteristics
Duration of HF (years)	6.2 ± 5.2	5.6 ± 4.7	0.009
Prior hospitalization for HF	85.6%	78.1%	< 0.001
LVEF (percent)	26.4 ± 8.1	27.9 ± 7.9	< 0.001
Coronary artery disease, history of	74.7%	70.4%	0.067
Arrhythmia, history of	72.9%	59.3%	< 0.001
Atrial	58.9%	46.6%	< 0.001
Ventricular	32.2%	26.6%	0.019
Therapies
ICD	18.2%	11.2%	< 0.001
B-blocker at discharge	67.8%	79.9%	< 0.001
ACEI/ARB at discharge	78.7%	88.5%	< 0.001
Diuretic at discharge	93.6%	93.0%	0.669
Symptoms at discharge
NYHA functional class at discharge			< 0.001
I	1.9%	5.6%
II	32.8%	51.6%
III	54.1%	41.5%
IV	11.2%	1.4%
Dyspnea at discharge, frequent or constant	24.3%	12.3%	< 0.001
Fatigue at discharge, frequent or continuous	39.7%	18.0%	< 0.001
Physical signs at discharge
Systolic BP at discharge (mmHg)	116.1 ± 18.8	122.5 ± 19.0	< 0.001
Heart rate at discharge (bpm)	75.6 ± 12.0	73.2 ± 11.7	< 0.001
Pedal edema, described as moderate or marked, at discharge	9.5%	3.9%	< 0.001
Rales at discharge, any	26.2%	12.9%	< 0.001
JVP ≥ 10 cm at discharge	32.8%	27.5%	0.026
Laboratory measures nearest to discharge
Sodium (mMol, median [IQR])	139.0 (136.0, 143.0)	140.0 (138.0, 143.0)	< 0.001
<135	18.3%	8.6%	< 0.001
135-145	70.4%	80.8%
>145	11.4%	10.7%
BUN ≥43 (mMol))	31.9%	15.1%	< 0.001
GFR (ml/min, median [IQR])	49.8 (36.3, 65.9)	58.2 (44.5, 72.6)	< 0.001
≥59	33.2%	47.8%	< 0.001
30-59	53.1%	45.8%
15-29	13.3%	6.2%
<15	0.3%	0.1%
Hemoglobin (g/dL)	13.3 (11.8, 14.8)	14.1 (12.6, 15.3)	< 0.001
BNP (pg/mL, median [IQR])	794.0 (359.0, 1482.9)	404.0 (169.0, 842.7)	< 0.001
≤500	35.6%	58.7%	< 0.001
500-999	25.5%	22.2%
1000+	38.9%	19.1%
QRS duration (msec, median [IQR]	131 (106-161)	124 (99-152)	< 0.001
Index length of stay (days, median [IQR])	7.0 (4.0, 13.0)	7.0 (3.0, 12.0)	0.001

Continuous variables compared using Student's T-test, except for skewed variables expressed at medians which were compared using Wilcoxon rank-sum test, and categorical variables compared using Fisher exact test. QoL = quality of life; SD = standard deviation; IQR = interquartile range; KCCQ = Kansas City Cardiomyopathy Questionnaire; OS = Overall Summary; COPD = chronic obstructive pulmonary disease; HF = heart failure; LVEF = left ventricular ejection fraction; ICD = implantable cardioverter defibrillator; ACEI = angiotensin converting enzyme inhibitor; ARB = angiotensin II receptor blocker; NYHA = New York Heart Association; BP = blood pressure; JVP = jugular venous pressure estimation; BUN = blood urea nitrogen; GFR = estimated glomerular filtration rate; BNP = b-type natiuretic peptide

Outcomes

Mean KCCQ scores among survivors were 31.6 ± 19.0 at study enrollment, 52.9 ± 22.1 at 1 week post discharge, and 58.3 ± 23.9 at 24 weeks, indicating that significant gains in health status were typically made early with smaller improvements between 1 and 24 weeks. There were 478 deaths (32.8%) and an additional 192 (13.2%) patients who had persistently unfavorable QoL throughout follow-up (KCCQ <45 at week 1 and 24) (Figure 1). KCCQ measures of health status correlated well with NYHA assessments at 1 week after discharge: NYHA I, median KCCQ = 76 (IQR 58-87); NYHA II, median KCCQ = 63 (IQR 49-76); NYHA III, median KCCQ = 45 (IQR 33-60); NYHA IV, median KCCQ = 25 (IQR 17-36).

Figure 1.

Distribution of patients with KCCQ <45 versus >=45 at 1 week post discharge and 24 weeks post discharge.

Predictors of Persistently Unfavorable Future QoL or Death

Unadjusted associations between patient characteristics at discharge and the combined end point of persistently unfavorable QoL or death in the 24 weeks following hospital discharge are shown in Table 1. After adjusting for 23 covariates in the full model, independent predictors of the combined end point included low baseline KCCQ score (per 10 unit increase, or improvement in baseline QoL: risk ratio [RR] = 0.82, 95% confidence interval [CI] 0.78-0.87), high BNP (500-999 pg/ml: RR 1.27, CI 1.05-1.53; 1000+ pg/ml: RR = 1.41, CI 1.14-1.73; compared to <500 pg/ml), hyponatremia (sodium <135 mEq/L: RR = 1.30, CI 1.04-1.62; compared to sodium 135-145 mEq/L), increased heart rate at discharge (per 10 bpm increase: RR 1.08, CI 1.01-1.15), decreased systolic blood pressure at discharge (per 10 mmHg increase: RR = 0.92, CI 0.88-0.97), absence of β-blocker therapy at discharge (β-blocker prescribed: RR = 0.80, CI 0.64-0.99), history of diabetes (HR = 1.18, CI 1.01-1.39), and history of arrhythmia (RR = 1.32, CI 1.08-1.60) (see Appendix 1 for details). The full model had moderate discriminatory capacity (c-statistic = 0.73). Sensitivity analysis changing the cut point for the primary outcome from KCCQ < 45 to < 40 and then to < 50 did not substantially change the model predictors. Additional sensitivity analysis using continuous values for covariates, rather than dichotomous or ordinal transformations, resulted in the same covariates retaining significance with a c-statistic of 0.74.

Simplified Refractory Heart Failure Risk Score

After variable elimination, the full model was reduced to 9 independent variables presented as 12 discrete risk predictors, each assigned a score between 1 and 5 points (Figure 2). The c-statistic for the reduced model was 0.72 and for the simplified risk score was 0.72 in this cohort. Similar to the full model, the individual predictors of baseline KCCQ and BNP carried the greatest discriminatory capacity for the combined end point, and thus were assigned the highest points. Bootstrap validation resulted in c-statistics of 0.74 for the bootstrap samples and 0.73 for the original cohort based on the risk scored developed from the bootstrap samples.

Figure 2.

The refractory heart failure score: a reduced model with a simplified clinical risk score predicting unfavorable future health status, defined as death or KCCQ persistently < 45 in the 24 weeks following heart failure hospitalization, determined using clinical information at the time of hospital discharge

Risk score distribution and corresponding predicted risk can be found in Figure 3. Risk score calibration stratified by deciles of predicted risk demonstrated a graded increase in the observed rates of death or persistently unfavorable health status. The Hosmer-Lemeshow statistic was 0.051, indicating no evidence for lack of fit. The model calibration slope of the linear predictor was 1.0 (CI 0.9-1.1) on 1000 bootstrap validation samples predicting onto the original dataset, indicating good calibration (Appendix 2). Importantly, the risk score stratified risk across a broad spectrum of risk, ranging from 7.6% in the lowest decile of predicted risk to 64.9% in the highest.

Figure 3.

Distribution of refractory heart failure risk scores and their probabilities

Comparing Predictors of the Future QoL to Traditional HF End Points

To clarify the unique information provided by a novel end point that includes persistently unfavorable QoL, separate multivariable-adjusted outcomes models were created for the individual end points of (A) all-cause mortality, (B) persistently unfavorable QoL, and (C) rehospitalization with 24 weeks of hospital discharge (Figure 4). Comparison of the hazards ratios and relative risks show that the predictors of these 3 end points are different. As assessed by the F-statistics (Figure 5), low baseline KCCQ is the dominant predictor of persistently unfavorable QoL and yet does not show a significant association with either death or rehospitalization. Conversely, elevated BNP is the dominant predictor of death and rehospitalization, but was not significantly associated with persistently unfavorable QoL. Other patient characteristics, including age, diabetes mellitus, inability to prescribe a β-blocker at discharge and sodium levels were associated to varying degrees with the different outcomes (e.g., age was strongly associated with poor QoL but not with death or rehospitalization).

Figure 4.

Separate adjusted multivariable risk models for the association of in-hospital clinical covariates with selected 24-week outcomes.

Figure 5.

Relative prognostic contribution of various in-hospital clinical covariates with selected 24-week outcomes as assessed by F-testing.

Discussion

Heart Failure patients and their health care providers are concerned not only with survival but also the quality of that survival.²⁵ To our knowledge, this is the first study to provide formal estimates of future QoL in patients with HF and reduced LVEF. Modeling the combined end point of persistently low health status measures and mortality, rather than the more typical HF outcomes of rehospitalization and mortality, provides prognostic information that most directly relates to patients' concerns and experiences. Thus, these findings fundamentally extend the large body of existing literature regarding prognostication in HF by explicitly including patient-centered measures of QoL over time as part of the predicted clinical outcome. Given the growing importance of using objective evidence to engage patients in guiding their subsequent care so that decisions can be based on patients' individual goals and values,^{4, 46} we believe that these findings, once validated, may be used at the time of hospital discharge to improve the quality of HF care.

In the EVEREST cohort, we found that that nearly half of the patients discharged from the hospital did not survive to enjoy favorable QoL throughout the subsequent 24 weeks, despite eligibility criteria designed to exclude patients with end-stage HF or an expected survival < 6 months. Nearly a third of these patients survived with persistently unfavorable QoL (i.e., lived but with severe symptoms, poor function, and markedly impaired QoL). This subgroup is of particular concern because they may choose different therapies if made aware of their prognosis. Moreover, we were able to demonstrate that reasonable predictive discrimination could be accomplished using only 9 readily accessible clinical characteristics at the time of hospital discharge that could stratify patients' risk from a 7% to a 65% probability of either dying or having persistently unfavorable QoL. Importantly, the predictors of this clinically important, combined outcome differ from traditional risk models. It is noteworthy that while the KCCQ has been repeatedly demonstrated to be associated with survival, hospitalization, and costs in chronic heart failure,^{33, 47} it was not significantly associated with mortality or rehospitalization in the acute setting. Similarly, while age is one of the strongest predictors of mortality in the chronic setting, it was only associated with reduced QoL and not with mortality or rehospitalization in the setting of acute decompensated heart failure. Prior comparisons between chronic ambulatory and acute hospitalized phases of HF have shown similar differences in predictors of outcome.⁴⁸

A recent article by Quill and colleagues underscores the importance of estimating prognosis as the foundation for identifying and following the treatment preferences of very ill patients.⁴⁹ In the setting of severe HF despite optimal therapy, patients with a poor prognosis can be eligible for either more aggressive care (e.g., mechanical circulatory support and cardiac transplantation) or supportive measures to optimize their quality of life.^{15, 50} Only by transparently identifying which patients with HF are at risk for both unfavorable QoL and death can patients' preferences for the intensity of care be explicitly elicited and followed.⁴ Although the model presented here is not intended to dictate care, it can be potentially useful in identifying patients in whom an in-depth discussion of advanced treatment options or end-of-life care is most relevant.

Despite the importance of prognosis to patients with HF, studies show that patients and health care providers are relatively uninformed and overly optimistic about expectations for the future.^5,8 Our findings suggest that the formal quantification of patients' QoL at the time of hospitalization, measurement of BNP, and the integration of these values with other clinically-available data can be useful in better identifying patients with an adverse prognosis at the time of discharge. Thus, this model is complementary to traditional mortality models by explicitly incorporating QoL with mortality.

The importance of this combined outcome is supported by prior studies that have examined how patients value QoL and survival. Individuals tend to fall into 3 groups: approximately half wish to maximize length of life and thus would not trade significant quantity of life in current health for better QoL; a quarter would trade the large majority of their remaining quantity of life for a higher QoL; and a quarter fall in the middle.^26,51,52 This suggests that for approximately half of patients with symptomatic HF, patient-centered decisions about medical care will be significantly influenced by expectations and preferences for both quantity and quality of future survival. While patients' individual preferences may only be elicited by a detailed conversation between patients and their health care providers, the proposed risk-stratification model can identify those in whom such a conversation is most needed and provide a benchmark estimate of outcome to facilitate that conversation.³ Fundamentally, the data show that in order to provide expectations for future quality of life, risk models should include a baseline measure of health status.

Study Limitations

Several potential limitations should be considered when interpreting our results. The study findings are from a retrospective, post-hoc analysis of patients enrolled in a clinical trial. EVEREST excluded patients with end-stage HF, expected survival of < 6 months, significant hypotension, and severe renal dysfunction, thus eliminating some of the patients at the very highest risk for adverse outcomes. The cohort was also restricted to HF patients with reduced LVEF and who were younger, predominantly Caucasian and more likely to be male as compared with community HF populations.⁵³ However, the trial included patients with a high degree of comorbidity, and thus the clinical profile of patients enrolled in EVEREST was more similar to patients captured in recent large HF registries than has generally been the case with earlier randomized trials.^54-58 Additionally, the enrollment of patients from 359 sites on several continents, and from both academic and non-academic centers increases the generalizability of the findings. Serial KCCQ measures were obtained at 3 times, limiting the ability to assess changes in KCCQ prior to discharge and short-term fluctuations in QoL in follow up. We repeated our analysis with 1-week KCCQ scores as a model covariate in place of baseline enrollment KCCQ scores to address whether early improvements in KCCQ might reduce the predictive power of the model, and we found no major changes in model performance (data not shown). Due to the unique nature of the EVEREST data (i.e., high-quality clinical data including serial KCCQ measures in a large hospitalized HF cohort), model validation in a separate dataset is not possible at this time. External validation is a critical next step in the process of confirming the utility of this proposed risk score. In the meantime, clinical practice guidelines continue to advocate for communicating risk to patients hospitalized with heart failure. The general findings presented here help inform those discussions, suggesting a role for health status measurements in future quality of life considerations.

Conclusions

Nearly half of patients discharged after hospitalization for acute decompensated HF did not survive to enjoy favorable QoL during the subsequent 24 weeks. A risk score using commonly measured clinical data available at the time of hospital discharge, supplemented with a QoL measure, was able to identify patients at high risk for this outcome. Further work is needed to determine if this risk score can improve decisional quality and subsequent patient-centered outcomes for hospitalized patients with decompensated HF.