Author + information
- Received July 22, 2013
- Revision received September 16, 2013
- Accepted October 8, 2013
- Published online February 25, 2014.
- Tonje Thorvaldsen, MD∗,†,
- Lina Benson, MSc‡,
- Marcus Ståhlberg, MD, PhD∗,†,
- Ulf Dahlström, MD, PhD§,
- Magnus Edner, MD, PhD∗ and
- Lars H. Lund, MD, PhD∗,†∗ ()
- ∗Department of Medicine, Karolinska Institutet, Stockholm, Sweden
- †Department of Cardiology, Karolinska University Hospital, Stockholm, Sweden
- ‡Department of Clinical Science and Education, Karolinska Institutet, SöS, Stockholm, Sweden
- §Division of Cardiovascular Medicine, Department of Medicine and Health Sciences, Faculty of Health Sciences, Linköping University, and Department of Cardiology UHL, County Council of Östergötland, Linköping, Sweden
- ↵∗Reprint requests and correspondence:
Dr. Lars H. Lund, Unit of Cardiology, Karolinska Institutet, Karolinska University Hospital, N305, Stockholm 17176, Sweden.
Objectives The purpose of this study was to evaluate simple criteria for referral of patients from the general practitioner to a heart failure (HF) center.
Background In advanced HF, the criteria for heart transplantation, left ventricular assist device, and palliative care are well known among HF specialists, but criteria for referral to an advanced HF center have not been developed for generalists.
Methods We assessed observed and expected all-cause mortality in 10,062 patients with New York Heart Association (NYHA) functional class III to IV HF and ejection fraction <40% registered in the Swedish Heart Failure Registry between 2000 and 2013. Next, 5 pre-specified universally available risk factors were assessed as potential triggers for referral, using multivariable Cox regression: systolic blood pressure ≤90 mm Hg; creatinine ≥160 μmol/l; hemoglobin ≤120 g/l; no renin-angiotensin system antagonist; and no beta-blocker.
Results In NYHA functional class III to IV and age groups ≤65 years, 66 to 80 years, and >80 years, there were 2,247, 4,632, and 3,183 patients, with 1-year observed versus expected survivals of 90% versus 99%, 79% versus 97%, and 61% versus 89%, respectively. In the age ≤80 years group, the presence of 1, 2, or 3 to 5 of these risk factors conferred an independent hazard ratio for all-cause mortality of 1.40, 2.30, and 4.07, and a 1-year survival of 79%, 60%, and 39%, respectively (p < 0.001).
Conclusions In patients ≤80 years of age with NYHA functional class III to IV HF and ejection fraction <40%, mortality is predominantly related to HF or its comorbidities. Potential heart transplantation/left ventricular assist device candidacy is suggested by ≥1 risk factor and potential palliative care by multiple universally available risk factors. These patients may benefit from referral to an advanced HF center.
Heart failure (HF) affects 2% of the Western population (1–4) and is associated with poor quality of life and high mortality. Pharmacologic therapy, cardiac resynchronization therapy (CRT), and implantable cardioverter-defibrillators (ICDs) have improved prognosis in HF with reduced ejection fraction (EF) but have also increased the number of patients living with advanced HF. Improved survival after acute coronary syndromes and aging of the population are further contributing to an increased prevalence of HF. Thus, there is an unmet and growing need for advanced HF therapy (1–3).
Heart transplantation (HTx) and left ventricular assist devices (LVADs) improve quality of life and survival in advanced HF (4–6). LVADs are used mainly as bridge to transplantation (BTT), but with donor organ shortage and improving outcomes with continuous flow devices, they are increasingly being used as permanent therapy, or destination therapy (DT) (7). Palliative care improves quality of life and is indicated in refractory HF, especially if HTx and LVAD are ruled out (2,8).
However, patients with advanced HF are believed to be underserved by these treatments (1,3,5,6,8,9). Reasons may include HF care performed by generalists with lack of awareness of prognosis and indications for advanced treatment, and inadequate or delayed referral to advanced HF centers, which are best suited to perform triage to different interventions. Indeed, although there are well-established criteria for HF specialists performing in particular HTx and LVAD selection, there are no tools or criteria for generalists to determine who to refer for evaluation.
Therefore, the aims were: first, to assess the contemporary observed and expected survival in an unselected population with New York Heart Association (NYHA) functional class III to IV HF and reduced EF; and second, to test the hypothesis that simple universally available variables can independently predict prognosis and can be used as triggers for referral to advanced HF centers. Because NYHA functional class II versus III is a subjective distinction and many clinicians base referrals on symptoms alone, we also assessed the triggers separately in NYHA functional class II.
The Swedish Heart Failure Registry has been previously described (10). The inclusion criterium is clinician-judged HF. Eighty variables are recorded at discharge from hospital or outpatient visit and entered into a Web-based database managed by the Uppsala Clinical Research Center (Uppsala, Sweden). The database is run against the Swedish death registry monthly. (The protocol, registration form, and annual report are available at www.rikssvikt.se.) Establishment of the registry and analysis of data was approved by a multisite ethics committee and conforms to the Declaration of Helsinki. Individual patient consent is not required, but patients are informed of entry into national registries and are allowed to opt out.
Between May 11, 2000, and June 5, 2013, there were 85,291 registrations from 68 of approximately 75 hospitals and 102 of approximately 1,000 primary care outpatient clinics in Sweden. Of these, there were 10,062 first registrations with NYHA functional class III to IV and EF <40% and 9,463 first registrations with NYHA functional class II and EF <40%. The main analysis was NYHA functional class III to IV and included baseline characteristics, observed and expected all-cause mortality, and risk factors for all-cause mortality. The separate analysis was NYHA functional class II and included baseline characteristics and risk factors for all-cause mortality (Fig. 1).
Statistical analysis was performed in R version 2.15.3 (R Foundation for Statistical Computing, Vienna, Austria). The level of significance was set to 5%, and all reported p values and confidence intervals (CIs) are 2-sided.
Forty-six clinically-relevant baseline variables were included for analysis (Table 1) and were compared among 3 age groups: ≤65 years, 66 to 80 years, and >80 years. The age cut-offs were based on general European practice: HTx considered mainly for age ≤65 years, DT-LVAD for patients in their 70s, and for carefully selected patients, up to age 80 years; and palliation for age >80 years or for younger patients with contraindications to HTx or LVAD.
Observed and expected all-cause mortality
Observed mortality for the overall study population and the 3 age groups was charted with the Kaplan-Meier method together with the expected mortality (Fig. 2). The expected mortality is for the study population if it had the same mortality probability as the general Swedish population matched to the sex, age, and year of observation of the study population (11). The mortality probabilities for the Swedish population were obtained from the Human Mortality database (http://www.mortality.org). The difference between observed and expected yields the “excess” mortality, which can be interpreted as the mortality related to HF itself and/or to associated comorbidities.
Risk factors for all-cause mortality
Because patients with advanced HF and age >80 years are generally not candidates for HTx or LVAD and are generally agreed to be suitable for palliation, they were excluded from the following risk factor analysis (Fig. 1).
Five simple and universally available variables, and cut-offs for continuous variables, were prospectively selected as potential independent risk factors for all-cause mortality based on previous studies (12–15) and as potential triggers for referral to an HF center (Table 2): systolic blood pressure ≤90 mm Hg (a criterion for cardiogenic shock); creatinine ≥160 μmol/l (which represents considerable end-organ impairment but generally not yet a contraindication to HTx or LVAD ); hemoglobin ≤120 g/l (a marker of the cardiorenal syndrome and progressive HF ); and absent renin-angiotensin system antagonist or beta-blocker treatment (12,15). We cannot show that absent drug therapy is due to intolerance, and certainly every effort should be made to utilize these medications. However, whether the lack of drug therapy is because of true intolerance or is a reflection of the generalist’s perception of patient frailty or unease about follow-up and monitoring, both of these reasons warrant referral to an advanced HF center for optimization.
Twelve additional secondary variables were selected for assessment of prognostic utility (Fig. 3). Continuous variables contain more information when analyzed continuously, but for ease of interpretation and, importantly, for ease of use as referral criteria (5 variables) and prognostic markers (12 secondary variables), they were categorized at cut-offs that were previously shown or proposed to be prognostically useful (12–15).
A multivariable Cox regression, using the Efron method for tie handling, was performed with all 46 baseline variables, and the hazard ratios (HRs) for mortality associated with each of the 5 risk factors (Table 2, parts A and B) and secondary 12 pre-specified risk factors were displayed in a Forest plot (Fig. 3). To assess the HR associated with cumulative number of risk factors, another Cox regression was performed in which a new variable “number of risk factors” was included, and the 5 risk factors themselves were excluded, yielding 42 variables (Table 2, parts C and D).
The proportional hazards assumption was investigated for the scaled Schoenfeld residuals from the model containing 46 variables. In the main NYHA functional class III to IV analysis, NYHA functional class exhibited nonproportional hazards; therefore, a consistency Cox analysis was performed in which the time axis for NYHA functional class was partitioned into ≤6 months and >6 months. The dfbetas (a measure of how much the HR changes due to the deletion of a single observation) from the models were inspected for outliers (16), but no problems were detected. The pre-specified continuous risk factors were categorized, but for remaining continuous variables, the assumption of linearity was investigated by smoothed Martingale residuals plots and found to be acceptably linear in relation to mortality.
To avoid bias and confounding due to baseline variables missing not at random, multiple imputation (n = 10) (aregImpute [Hmisc] in R) was performed using predictive mean matching. The imputation model was for all patients, NYHA functional class II to IV, and included the 46 numbered variables in Table 1, and the outcome, all-cause mortality, was included as the Nelson-Aalen estimator but was not imputed itself because it contained no missing values. Mortality was assessed with the Kaplan-Meier method for each of the 5 risk factors separately and according to the cumulative number of pre-specified risk factors, 0 to 5 (Table 2, Fig. 4).
To assess concordance between model predictions and observed outcomes, c-indexes (17) were calculated: 1) for the multivariable Cox regression model including all 46 baseline variables; 2) for a model containing only the 5 risk factors; and 3) for a model containing only 1 variable defined as “number of risk factors” (n = 0 to 5).
In NYHA functional class III to IV, age groups ≤65 years, 66 to 80 years, and >80 years, there were 2,247, 4,632, and 3,183 patients, respectively. Younger patients were more commonly male with lower EF but otherwise generally healthier and with more evidence-based treatment (Table 1). The NYHA functional class II patients were younger and generally healthier than the NYHA functional class III to IV patients, but the differences between age groups exhibited similar patterns (Online Table 1).
Observed and expected all-cause mortality
For NYHA functional class III to IV, overall observed 1- and 5-year survival was 76% and 39%, respectively, and expected survival was 95% and 76%, respectively (Fig. 2A). The corresponding figures by age group were as follows: age ≤65 years, 90% and 68% observed versus 99% and 96% expected; age 66 to 80 years, 79% and 40% observed versus 97% and 83% expected; and age >80 years, 61% and 17% observed versus 89% and 52% expected (Figs. 2B and 2D). The absolute difference between observed and expected mortality (“excess mortality,” distance between curves) was similar for all age groups, but the relative difference (distance between curves in proportion to actual mortality) was dramatically higher in the ≤65 years and 66 to 80 years age groups compared with the >80 years group.
Risk factors for all-cause mortality
Table 2, part A, and Figure 3 depict the 5 risk factors for all-cause mortality for NYHA functional class III to IV. Hemoglobin ≤120 g/l was the most common risk factor. All 5 risk factors and most of the 12 pre-specified risk factors were independent predictors of mortality. Creatinine ≥160 μmol/l was associated with the highest HR. Table 2, part B, and Figure 4 depict mortality by number of risk factors for NYHA functional class III to IV. Number of risk factor distribution was as follows: 0, 59%; 1, 28%; 2, 10%; 3 to 5, 3%. In other words, 41% of patients had at least 1 risk factor. The presence of 0 to 5 risk factors was associated with progressively worse survival; 0 risk factors (90% 1-year survival) was similar to a post-HTx prognosis (18) and 1 risk factor (79% 1-year survival) was similar to a post-LVAD prognosis (19). Two risk factors and 3 to 5 risk factors were associated with 60% and 39% 1-year survival, respectively. The HR increased progressively by number of risk factors: 1.40, 2.30, and 4.07 for 1, 2, and 3 to 5 risk factors, respectively, versus 0 risk factors (Table 2, part B). The HR for NYHA functional class IV versus III was 1.39 (95% CI: 1.25 to 1.55, p < 0.001). With partitioned functional class time axis for NYHA functional class, the HR for NYHA functional class IV versus III was 3.02 (95% CI: 2.58 to 3.54, p < 0.001) for the first 6 months of follow-up and 0.83 (95% CI: 0.71 to 0.97, p = 0.020) thereafter. Partitioned NYHA functional class time axis in the Cox regression did not alter the results for the other variables.
Table 2, parts C and D, and Figure 4B show data for the separate NYHA functional class II analysis. The 5 risk factors were less common in NYHA functional class II. As expected, the absolute risk was lower; namely, overall survival for any given risk factor and for any given number of risk factors was higher in NYHA functional class II. Systolic blood pressure ≤90 mm Hg and absent beta-blocker were rare and were not significant alone. The “number of risk factors” was highly significant, but the HR associated with any given risk factor or with any given number of risk factors was slightly lower in NYHA functional class II than NYHA functional class III to IV.
In NYHA functional class III to IV, the c-indexes for the model containing the 5 risk factors and for the model with only 1 variable defined as “number of risk factors” were 0.71 and 0.73, respectively. The c-index for the multivariable Cox regression model including all 46 baseline variables was 0.75. In NYHA functional class II, the corresponding c-indexes were 0.69, 0.69, and 0.75.
In a large, unselected population age ≤80 years with NYHA functional class III to IV HF and EF <40%: 1) mortality was mainly directly or indirectly related to HF and/or its comorbidities; and 2) the presence of any 1 or more of 5 pre-specified universally available risk factors conferred a 1-year survival of 79% or worse, and could thus identify patients who may benefit from referral to an advanced HF center. Also for patients with NYHA functional class II HF, the risk factor model was predictive of relative risk, but calibration and suitability as a referral tool was worse.
Under-referral for advanced HF therapy
Before advanced therapy consideration, basic HF therapy should be optimized. The 5% to 22% untreated with renin-angiotensin system antagonists or beta-blockers in our study was better than in other observational HF studies (20). However, fewer than one-half of patients received mineralocorticoid receptor antagonists, and only 1% to 11% received CRT and/or ICD, which certainly should be improved. However, advanced HF therapy likely remains underutilized (1,3,5,6,8,9). One-year survival after HTx approaches 90% (18), and 1-year survival with LVAD now approaches 80% (7,19). In the United States, an estimated 100,000 people would benefit from HTx (5,6), and 25,000 to 250,000 may benefit from LVAD, primarily as DT (1,3). Yet, there are only 2,200 HTx (18) and 1,800 LVADs performed annually (7). Utilization of palliative care varies considerably and is substantially less for HF than for cancer (21).
The main reason for underutilization of HTx is donor organ shortage (5). Although organs are increasingly going to high urgency or LVAD patients, even the listing of ambulatory patients has decreased dramatically (6). Earlier referral may not increase the total number of HTx, but would identify potential candidates early and allow careful monitoring and optimal timing for HTx and/or BTT-LVAD if needed. The reasons for under-referral for LVAD are unknown. BTT-LVAD is generally accepted but referrals are often too late, with prohibitive right ventricular or end-organ failure (4,7). DT-LVAD has not been widely adopted, possibly because of poor awareness and slow acceptance among referring physicians, remaining debates among HF specialists (1,22,23), and limited patient acceptance (24).
However, the major reason for underutilization is likely unawareness and difficulty for generalists in assessing the need. In Sweden, more than one-half of patients with HF are cared for exclusively by primary care physicians (25), which may in part explain our observed low penetrance of CRT and ICD. In our study, nearly one-half of patients were seen by internal medicine/geriatrics, and follow-up was in primary care in 7% to 50%, despite being in NYHA functional class III to IV and EF <40%. In Sweden, a majority of hospitals have an HF nurse clinic, and referral occurred in 33% to 63% of patients, but few of these clinics have expertise in HTx or LVAD selection, and most lack formal palliative care programs.
Prognostic tools such as the peak VO2, the Heart Failure Survival Score (HFSS), the Seattle Heart Failure Model (SHFM), and the recent MAGGIC (Meta-Analysis Global Group in Chronic Heart Failure) risk score (4,6,13,26,27) are well known among HF specialists but may be too complex for busy generalists. Even with these tools, prognosis remains notoriously difficult to predict (28). NYHA functional class discriminates well (HR: 1.39 for NYHA functional class IV vs. III in our study) but is subjective, is variable (nonproportional hazards), and does not calibrate well (cannot be used alone as a referral criterion). Therefore, we assessed our model also in NYHA functional class II. With better overall prognosis, the calibration as a tool for referral for advanced therapy was worse. Namely, in NYHA functional class II, 1-year survival with 1 risk factor was 91% (vs. 79% in NYHA functional class III to IV) and with 2 risk factors, 84% (vs. 60% in NYHA functional class III to IV), and the simple 1 risk factor criterion was no longer applicable. The risk factor model was still predictive of relative risk, but discrimination, as measured by area under the curve, was slightly lower in NYHA functional class II compared with NYHA functional class III to IV. The risk factors are related to the severity of HF and are thus less common in NYHA functional class II, but once present, they may signify progressive decline and should certainly be considered when assessing risk, and they may justify referral even if the subjective NYHA functional class assessment is only II.
With elderly patients, clinicians may underestimate the symptomatic and prognostic role of HF relative to age itself. However, in our patients age ≤80 years, mortality was primarily HF and/or comorbidity related (and thus potentially treatable with HF therapy) rather than from age. It is likely that comorbidities (e.g., a risk factor for HF such as hypertension) cause death independently of HF (e.g., fatal stroke). Thus, although 41% of patients age ≤80 years had at least 1 of 5 risk factors and thus higher mortality than after HTx or LVAD, we certainly do not suggest that all would benefit from or be candidates for such treatment. However, in well-selected elderly, DT-LVAD is associated with favorable outcomes (19), and regardless of the relative roles of HF and age, the elderly should not be excluded from referral for expert evaluation, optimal symptom relief, and palliation if indicated.
Five simple and universally available triggers to improve referral
In the NYHA functional class III to IV analysis, the 90% and 79% 1-year survival in the age ≤65 years group and the 66 to 80 years group, respectively, was better than the overall 70% to 80% in other unselective observational studies (29). However, with only 1 of the 5 risk factors, 1-year survival in the age ≤80 years group dropped to 79%, worse than after HTx and LVAD (18,19), and with additional risk factors, prognosis worsened considerably, suggesting these risk factors may be useful triggers for referral to advanced HF centers. Peak VO2 <12 to 14 ml/kg/min and medium-high risk HFSS criteria for HTx correspond to <80% to 90% 1-year survival (13,26). Analogously, the correspondence of 1 risk factor with 79% 1-year survival, worse than after HTx (18) and similar to LVAD (19), suggests proper calibration and utility for referral to an advanced HF center. However, ≥1 risk factor certainly does not automatically confer an indication for HTx or LVAD. Indeed, the appropriateness of and criteria for DT-LVAD continue to be debated (1,22,23). Instead, it serves as a proposed trigger for referral to an advanced HF center, where detailed indications and contraindications and potential other interventions can be assessed. For NYHA functional class II, >1 risk factor was needed for a prognosis worse than post-HTx or LVAD, but the risk factors were still predictive of increased relative risk. Thus, referral may still be reasonable if numerous risk factors are present or the distinction between NYHA functional class II versus III is difficult.
Renal function is better assessed by glomerular filtration rate than by creatinine. Continuous variables (blood pressure, creatinine, hemoglobin) provide more statistical resolution when assessed continuously rather than categorically. However, importantly, our aim was not to validate predictors with optimal resolution but to provide simple, universally available, and memorable criteria that do not require calculation (e.g., glomerular filtration rate) or access to risk models (e.g., HFSS, SHFM, or MAGGIC) and that can help generalist clinicians quickly identify patients with poor prognosis who may benefit from referral to an advanced HF center. In the NYHA functional class III to IV analysis, the discriminatory performance of the simple 5 variables alone (c-index 0.71) and of the simple “number of risk factors” (n = 0 to 5) variable (c-index 0.73) was close to that of the complete model based on 46 variables (c-index 0.75), was better than that for the peak VO2, and was similar to the HFSS and SHFM (13) in HTx-referred patients. The 12 additional risk factors were also generally predictive of mortality, suggesting that, for example, longer duration of HF, ischemic etiology, progressively increasing N-terminal pro-brain natriuretic peptide, and of course NYHA functional class IV should also raise concern. However, these factors may be difficult to assess or unavailable in generalist settings. In contrast, the 5 main risk factors are completely objective and universally available and were also among the strongest predictors (Fig. 3).
Thus, we propose that referring generalists should not focus on cumbersome detailed criteria for HTx, LVAD, or palliative care, but instead should rely on simple, universally available triggers for referral to advanced HF centers. All 5 risk factor variables have previously been shown to be independent predictors of mortality, but generally with less covariate adjustment and not as specific referral criteria (12–15). Although the c-index was high, we do not suggest that our simple model provides superior prognostication compared with sophisticated selection tools (12,13). However, we do propose that our universally available criteria are simpler and more likely to be used in generalist care and emphasize that they should be used for referral, not for selection. Although our proposed strategy would likely increase identification of appropriate candidates for advanced therapy, we cannot determine the potential impact of such clinical routines in terms of the extent of increase in referrals or (potentially unnecessary) testing. But it does not seem unreasonable that 41% of patients (i.e., with ≥1 risk factor) age ≤80 years, with NYHA functional class III to IV and EF <40%, get the opportunity for at least a 1-time expert assessment at an advanced HF center.
The 5 risk factors were chosen based on their universal availability and their known independent prognostic power, but were not prospectively validated as the best predictors in our population. However, our aim was not to validate predictors with optimal resolution but to provide simple, universally available, and memorable criteria that can help generalist clinicians identify patients for referral. Furthermore, the unselective nature, large number of patients, and large number of variables for multivariable analyses lend reliability and generalizability to the findings.
Among unselected patients with NYHA functional class III to IV HF and EF <40%, the death of patients up to age 80 years is primarily death related to HF and/or its comorbidities and may therefore be served by advanced HF therapy. The presence of 1 or more universally available risk factors conferred a 1-year survival of 79% or worse. We propose that the presence of 1 or more risk factors should be a trigger for generalists to refer to an advanced HF center for optimization and potentially evaluation for HTx, LVAD, palliative care, or other potential interventions.
The authors thank the local investigators at the 68 hospitals and 102 primary care outpatient clinics reporting to the Swedish Heart Failure Registry for this study.
The Swedish Heart Failure Registry is funded by the Swedish National Board of Health and Welfare, the Swedish Association of Local Authorities and Regions, the Swedish Society of Cardiology, and the Swedish Heart Lung Foundation.
This work was supported in part by the Swedish Heart Lung Foundation (grants 20080409 and 20100419), the Stockholm County Council (grants 00556-2009 and 20110120), Thoratec Europe, and Boston Scientific (all to Dr. Lund’s institution). Dr. Lund has received research funding for his institution and consulting fees and speaker’s honoraria from the manufacturers of cardiac resynchronization therapy/implantable cardioverter-defibrillators (Boston Scientific, Medtronic, and St. Jude) and left ventricular assist devices (Thoratec and HeartWare). Dr. Ståhlberg has received research funding for his institution and consulting fees and speaker’s honoraria from the manufacturers of cardiac resynchronization therapy/implantable cardioverter-defibrillators (Boston Scientific, Medtronic, and St. Jude). All other authors have reported that they have no relationships relevant to the contents of this paper to disclose.
- Abbreviations and Acronyms
- bridge to transplant
- confidence interval
- destination therapy
- cardiac resynchronization therapy
- ejection fraction
- heart failure
- hazard ratio
- heart transplantation
- implantable cardioverter-defibrillator
- left ventricular assist device
- New York Heart Association
- renin-angiotensin system
- Received July 22, 2013.
- Revision received September 16, 2013.
- Accepted October 8, 2013.
- American College of Cardiology Foundation
- Miller L.W.
- Yancy C.W.,
- Jessup M.,
- Bozkurt B.,
- et al.
- Miller L.W.,
- Guglin M.
- Mancini D.,
- Lietz K.
- Levy W.C.,
- Mozaffarian D.,
- Linker D.T.,
- et al.
- Therneau T.,
- Grambsch P.
- Owens A.T.,
- Jessup M.
- Mehra M.R.,
- Domanski M.J.
- Pocock S.J.,
- Ariti C.A.,
- McMurray J.J.,
- et al.
- Goldberg L.R.,
- Jessup M.