The Seattle Heart Failure Model: Prediction of Survival in Heart Failure

The Seattle Heart Failure ModelPrediction of Survival in Heart Failure

Wayne C. Levy, MD; Dariush Mozaffarian, MD, MPH; David T. Linker, MD;Santosh C. Sutradhar, PhD; Stefan D. Anker, MD; Anne B. Cropp, PharmD; Inder Anand, MD;

Aldo Maggioni, MD; Paul Burton, MBBS, PhD; Mark D. Sullivan, MD, PhD; Bertram Pitt, MD;Philip A. Poole-Wilson, MD; Douglas L. Mann, MD; Milton Packer, MD

Background—Heart failure has an annual mortality rate ranging from 5% to 75%. The purpose of the study was to developand validate a multivariate risk model to predict 1-, 2-, and 3-year survival in heart failure patients with the use of easilyobtainable characteristics relating to clinical status, therapy (pharmacological as well as devices), and laboratoryparameters.

Methods and Results—The Seattle Heart Failure Model was derived in a cohort of 1125 heart failure patients with the useof a multivariate Cox model. For medications and devices not available in the derivation database, hazard ratios wereestimated from published literature. The model was prospectively validated in 5 additional cohorts totaling 9942 heartfailure patients and 17 307 person-years of follow-up. The accuracy of the model was excellent, with predicted versusactual 1-year survival rates of 73.4% versus 74.3% in the derivation cohort and 90.5% versus 88.5%, 86.5% versus86.5%, 83.8% versus 83.3%, 90.9% versus 91.0%, and 89.6% versus 86.7% in the 5 validation cohorts. For the lowestscore, the 2-year survival was 92.8% compared with 88.7%, 77.8%, 58.1%, 29.5%, and 10.8% for scores of 0, 1, 2, 3,and 4, respectively. The overall receiver operating characteristic area under the curve was 0.729 (95% CI, 0.714 to0.744). The model also allowed estimation of the benefit of adding medications or devices to an individual patient’stherapeutic regimen.

Conclusions—The Seattle Heart Failure Model provides an accurate estimate of 1-, 2-, and 3-year survival with the useof easily obtained clinical, pharmacological, device, and laboratory characteristics. (Circulation. 2006;113:1424-1433.)

Key Words: diuretics � heart failure � hemoglobin � lymphocytes � prognosis

In the United States, �5 million individuals have heartfailure. Each year, 550 000 new cases are diagnosed, and

there are �1 million hospitalizations. The direct and indirectcosts of heart failure are estimated at $29 billion per year.1

Although heart failure presents enormous healthcare burdens,outcomes in heart failure are highly variable, with annualmortality varying from 5%2 to 75%.3 Physicians need tocounsel patients about prognosis to enable informed decisionsabout medications, devices, transplantation, and end-of-lifecare.

Clinical Perspective p 1433Existing models to predict the risk of death/urgent trans-

plantation in heart failure all have features that may limit theirapplicability. These models relied on either peak oxygenconsumption or invasive measures of cardiac performance,2

were validated during a hospitalization for heart failure,4,5

have not been validated in another database,6 or may beinaccurate with �-blockers.7,8 The predictive accuracy of onesuch model, the Heart Failure Survival Score (HFSS), hasbeen suboptimal in some validation data sets (Figure 1).

The purpose of this study was to develop and validate amultivariate risk model to estimate survival of heart failurepatients that incorporates easily obtainable clinical and labo-ratory variables, heart failure medications, and devices.

MethodsParticipantsThe present study used data previously collected in 6 cohorts ofpatients with predominantly left ventricular systolic heart failure.One cohort was used to develop the model (the Prospective Ran-domized Amlodipine Survival Evaluation [PRAISE1]; n�1125), and

Received August 30, 2005; revision received December 12, 2005; accepted January 13, 2006.From the University of Washington, Seattle (W.C.L., D.T.L., M.D.S.); Channing Laboratory, Brigham and Women’s Hospital and Harvard Medical

School, Boston, Mass (D.M.); Merck Research Laboratories, Blue Bell, Pa (S.C.S.); Imperial College School of Medicine, London, UK (S.D.A.,P.A.P.-W.); Pfizer Laboratories, Groton, Conn (A.B.C.); University of Minnesota, Minneapolis (I.A.); Italian Association of Hospital CardiologistsResearch Center, Florence, Italy (A.M.); Amgen Inc, Thousand Oaks, Calif (P.B.); University of Michigan, Ann Arbor (B.P.); Baylor College of Medicineand Houston Veterans Affairs, Houston, Tex (D.L.M.); and University of Texas Southwestern Medical Center at Dallas (M.P.).

The online-only Data Supplement can be found at http://circ.ahajournals.org/cgi/content/full/CIRCULATIONAHA.105.584102/DC1.Correspondence to Wayne C. Levy, MD, Division of Cardiology, University of Washington, Box 356422, 1959 NE Pacific St, Seattle, WA 98177.

E-mail [email protected]© 2006 American Heart Association, Inc.

Circulation is available at http://www.circulationaha.org DOI: 10.1161/CIRCULATIONAHA.105.584102

1424

Heart Failure

by guest on February 21, 2016http://circ.ahajournals.org/Downloaded from

http://circ.ahajournals.org/

5 other cohorts (n�9942) were used to prospectively validate themodel. PRAISE1 was a randomized trial of amlodipine versusplacebo among 1153 patients in the United States and Canada withejection fraction (EF) �30% and New York Heart Association(NYHA) functional class IIIB to IV heart failure.9 We excluded 32patients with incomplete baseline data. Evaluation of Losartan in theElderly (ELITE2) was a randomized trial of captopril versus losartanamong 3152 patients in 46 countries with EF �40%, age �60 years,and NYHA class II to IV heart failure.10 We excluded 165 patientswith incomplete baseline data. Valsartan Heart Failure Trial (Val-HeFT) was a randomized trial of valsartan versus placebo in 5010patients in 16 countries with EF �40 and NYHA class II to IV heartfailure. Allopurinol use and implantable cardioverter/defibrillator(ICD) use were not available. We excluded 1 patient with probabledata entry error for bumetidine.11 University of Washington (UW)was a prospective cohort study of 148 consecutive outpatients at atertiary US heart failure clinic.12,13 Randomized Enbrel North Amer-ican Strategy to Study Antagonism of Cytokines (RENAISSANCE)was a randomized trial of etanercept (Enbrel, Amgen, ThousandOaks, Calif) in 925 patients with NYHA class II to IV heart failureand EF �30 in the United States and Canada.14 Italian Heart FailureRegistry (IN-CHF) is a database of consecutive heart failure patientsseen by local participating cardiologists in Italy and entered into anational database. There were no exclusion criteria for entry in theregistry, and patients with any heart failure etiology, age, EF, orcomorbidities could be enrolled.15 For the IN-CHF, percent lympho-cytes were imputed with the use of white blood cell count and othervariables. Potassium-sparing diuretic use was not available. In UW,RENAISSANCE, and Val-HeFT, any patients with missing data wereassigned the median value for the covariate in that data set, except formissing drug or device variables, in which case they were assigned nodrug/device. The studies were approved by the institutional reviewboards of each participating institution, and all participants gave in-formed written consent. The authors had full access to the data and takefull responsibility for its integrity. All authors have read and agree to themanuscript as written.

Ascertainment of EventsIn PRAISE1, ELITE2, Val-HeFT, RENAISSANCE, and In-CHF,events were classified by a centralized adjudication committees.9–11,14,15

In the UW cohort, events were classified by one of the study cardiolo-gists (W.C.L.) using review of medical records. For this analysis, theprimary outcome was survival free of left ventricular assist device(LVAD) implantation or cardiac transplantation. Death, rather than

LVAD implantation or transplantation, represented the majority ofevents (98%) in these data sets.

Statistical AnalysisA participant level database from PRAISE1 was used for derivationof the prediction model. All databases were deidentified. Clinicalvariables previously reported to be associated with mortality wereevaluated with the use of the Cox proportional hazards model. Inaddition, diuretic type and dose were evaluated as predictors.Statistically significant univariate variables were visually inspectedfor linearity with the use of a plot of the variable by deciles versusthe natural log of the hazard ratio. If the response appeared nonlinear,simple transformations or cut points were used. For each covariate,categorical variables and both the continuous and transformedunivariate variables were allowed to enter in a stepwise forwardmultivariate Cox model with the use of a probability value of �0.05for inclusion or deletion. The hazard ratio for certain heart failuremedications and devices could not be effectively estimated in thePRAISE1 because of either widespread or rare use, includingangiotensin-converting enzyme (ACE) inhibitors, �-blockers, angio-tensin receptor blockers, aldosterone blockers, ICDs, biventricularpacing, and LVADs. For these medications and devices, benefitswere estimated from large published randomized trials or meta-anal-yses to determine the �-coefficients (natural log of the hazard ratio)for adding the medication/device to a patient’s regimen.3,10,15–29 Forpatients already on the medication/device, the impact of medicationor device use on systolic blood pressure (SBP), NYHA class, and EFwere incorporated into the estimate of the hazard ratio with the useof effects seen in large published trials. Thus, these hazard ratioswere estimated from published results of clinical trials and not fromthe validation data sets.

To calculate the Seattle Heart Failure Model (SHFM) Score, eachvariable in the multivariate model was multiplied by its �-coefficient(natural log of the hazard ratio), and the products were summed.2,6

Baseline survival (survival for score�0) was estimated bySurvival�e��t, where t�time and ��the slope/year were derivedfrom the PRAISE1 data (see Results; ��0.0405). Survival at time t(between 0 and 5 years) for any score is estimated by the followingequation:

Survival(t)�e(��t)e(SHFM Score)

The SHFM Score derived in PRAISE1 was then prospectivelyapplied to each patient in ELITE2, UW, Val-HeFT, RENAIS-SANCE, and IN-CHF to provide individual estimates of survival at1, 2, 3, and 5 years. For each study, the predicted survival by deciles(quintiles for smaller studies) was plotted against the actual (Kaplan-Meier) survival, and a correlation coefficient and standard error ofthe estimate (SEE) was calculated. Accuracy of the model acrossdata sets was determined by comparing the mean 1-, 2-, and 3-yearpredicted versus actual survival. Model discriminant ability wasdetermined by the 1-year receiver operating characteristic (ROC)area under the curve for each data set and for all data sets combined.Mean survivals were estimated by integrating the area under thesurvival curve for an annual mortality �30%. Because the formermethod will overestimate years of life remaining for lower-riskpatients, mean survival in these patients (annual mortality �30%)was estimated by using the predicted 5-year mortality from themodel and then by deriving total years of life remaining by themortality life table method,30 combining the most recent mortalitytables available for the United States, England and Wales, WestGermany, and France.31 The discriminative ability of the SeattleHeart Failure Model was compared with the Toronto4 and AcuteDecompensated Heart Failure National Registry (ADHERE) heartfailure models.5 For calculation of the Toronto model, comorbiditiesof cancer, stroke, dementia, cirrhosis, and chronic obstructive pul-monary disease were assumed to be absent, and the resting respira-tory rate was �20. Statistics were analyzed with the use of Statview5 (SAS Institute, Cary, NC), with ROC determined with the use ofSPSS 11 (SPSS Inc, Chicago, Ill). Statistical significance wasdefined as P�0.05 (2 tailed).

Figure 1. Predicted vs actual 1-year survival by the Heart Fail-ure Survival Score for low-, moderate-, and high-riskgroups.2,7,8,37–40 The diagonal line is the line of identity. BB indi-cates �-blockers.

Levy et al The Seattle Heart Failure Model 1425



ResultsPatient Characteristics and EventsThe patients in PRAISE1 were younger than those in ELITE2but had more severe heart failure by most clinical measures,including NYHA class, EF, SBP, and diuretic dose (Table 1).The patients in UW were younger but had lower SBP,hemoglobin, and percent lymphocytes; higher diuretic dose;and greater �-blocker, statin, and ICD use. The patients inRENAISSANCE had lower SBP and cholesterol and thehighest use of potassium-sparing diuretics and statins. Thepatients in IN-CHF had the lowest NYHA class and highestEF, with one third having an EF �40%. The Val-HeFTpatients were the largest group and were reflective of thebroad range of systolic heart failure.

Derivation of the ModelThe univariate and multivariate predictors of survival inPRAISE1 are shown in Table 2. In univariate analyses, olderage, male gender, ischemic etiology, and lower body massindex, EF, blood pressure, serum sodium, cholesterol, hemo-globin, and percent lymphocytes were associated with in-creased mortality. Higher NYHA class, white blood cellcount, creatinine, uric acid, and allopurinol use were also

associated with increased mortality in univariate analysis.Conversion of the various weight-adjusted diuretic regimensinto a uniform daily furosemide equivalent dose was per-formed by comparing the hazard ratio for the individualdiuretics in the multivariate model on the basis of thePRAISE1 data. Thiazide diuretics only added risk in themodel when added to a loop diuretic. The conversion usedwas furosemide 80 mg�torsemide 40 mg�bumetidine 3mg�metolazone 2 mg�hydrochlorothiazide 25 mg. Thefurosemide equivalent doses of other thiazide and loopdiuretics used in ELITE2 and IN-CHF were estimated withthe use of Micromedex (Thomson Micromedex, Greenwood,Colo). The resulting calculated daily diuretic dose per kilo-gram of body weight was the most powerful univariatepredictor of mortality (�2�77), with a 30% increase in risk ofdeath for each 1 mg/kg per day. Statin use was associatedwith lower mortality, consistent with prior analyses.24 The 1/xtransformations of EF and cholesterol had similar or greater�2 values in comparison to the continuous variables. Forserum sodium, the increased risk was associated with values�138 mEq/L, whereas values �138 did not alter risk.Hemoglobin had a U-shaped relationship with mortality inunivariate analysis (P�0.001 for the hemoglobin2 term) but

TABLE 1. Patient Characteristics

PRAISE9 ELITE210 UW12 RENAISSANCE14 Val-HeFT11 IN-CHF15

No. of patients 1125 2987 148 925 5010 872

Age, y 65�11 71�7 53�10 62�12 63�11 64�12

Range 23–95 52–100 19–74 N/A 18–96 14–93

Gender (male) 76% 69% 78% 78% 80% 76%

NYHA class 3.6 2.5 2.7 2.9 2.4 2.2

Range 3B–4 2–4 1–4 2–4 2–4 1–4

EF 21�6 31�7 27�9 22�6 27�7 35�11

Range 1–32 4–40 3–53 N/A 4–40 10–75

Ischemic etiology 64% 74% 34% 62% 58% 47%

SBP, mm Hg 118�18 134�19 105�15 109�17 123�18 129�22

Medications

Diuretic dose, mg/kg 1.45�1.33 0.41�0.63 1.67�2.05 1.30�1.43 0.82�1.10 0.73�0.94

ACE inhibitor 99% 50% 80% 79% 93% 84%

�-Blocker 0% 24% 72% 61% 34% 35%

Angiotensin receptor blocker 0% 50% 26% 19% 50% 7%

K-sparing diuretic use 3% 22% 28% 35% 5% N/A

Allopurinol use 10% 4% 8% 12% N/A 18%

Statin use 8% 13% 36% 52% 32% 16%

ICD 0% 0% 22% 18% N/A N/A

Laboratory

Sodium, mEq/L 139�4 141�4 137�3 139�3 139�3 140�4

Creatinine, mg/dL 1.37�0.48 1.17�0.33 1.25�0.39 1.26�0.43 1.29�0.31 1.27�0.69

Cholesterol, mg/dL 202�50 207�46 186�55 182�44 205�71 193�46

Uric acid, mg/dL 8.9�2.6 6.3�2.0 7.9�2.8 8.4�2.5 7.5�2.1 6.9�2.2

Hemoglobin, g/dL 13.9�1.7 14.0�1.6 13.1�1.8 13.8�1.6 13.7�1.4 13.6�1.8

White blood cells 7.7�2.3 7.2�2.1 7.4�2.3 7.7�2.1 6.9�2.0 7.5�2.2

Percent lymphocytes 26�9 25�12 22�9 24�8 25�9 25�4 (estimated)

N/A indicates data not available.

1426 Circulation March 21, 2006



was best fit with the use of a simple linear model �16 and�16 g/dL.15,32 Lymphocytes (�47%), uric acid (�3.4 mg/dL), and SBP (�160 mm Hg) were truncated for valuesoutside the database and/or normal range (ie, SBP of180 mm Hg was assigned a value of 160).

Age and gender were forced into the model as importantdemographic variables. Potassium-sparing diuretics wereforced into the derivation model as a planned addition to themodel based on external data.22,23 In the multivariate stepwiseregression (Table 2, Figure 2), body mass index, blood urea

TABLE 2. Univariate and Multivariate Predictors of Survival in PRAISE1 Derivation Cohort

Univariate Hazard Ratio(95% CI) Wald �2

Multivariate Hazard Ratio(95% CI) Wald �2

Demographic

Age (decade)* 1.134 (1.034–1.245) 7.1 1.090 (0.985–1.205) 2.8

Gender (male) 1.256 (0.986–1.599) 3.4 1.089 (0.839–1.414) 0.4

Body mass index, kg/m2 0.960 (0.943–0.979) 17.9 2.7

NYHA class 2.402 (1.546–3.734) 15.2 1.600 (1.019–2.511) 4.2

Ejection fraction (0–30) 0.971 (0.955–0.987) 12.0 �0.1

100/Ejection fraction 1.023 (1.009–1.038) 10.2 1.030 (1.010–1.050) 8.4

Ischemic etiology 1.483 (1.194–1.841) 12.7 1.354 (1.074–1.707) 6.6

SBP, 10 mm Hg* (for SBP �160 mm Hg) 0.803 (0.755–0.855) 47.9 0.877 (0.823–0.935) 16.0

Medications

Diuretic dose, mg/kg per day 1.305 (1.229–1.386) 77.0 1.178 (1.097–1.266) 20.0

Allopurinol use 1.520 (1.144–2.021) 8.3 1.571 (1.170–2.109) 9.0

Statin use 0.600 (0.390–0.922) 5.4 0.63 (0.410–0.978) 4.2

Laboratory

Sodium, mEq/L 0.940 (0.916–0.965) 21.6 0.1

If sodium �138, 138�sodium 1.117 (1.074–1.162) 30.2 1.050 (1.005–1.097) 4.8

Creatinine, mg/dL 1.554 (1.319–1.831) 27.9 0.5

Cholesterol,* each 40 mg/dL 0.814 (0.744–0.890) 20.3 0.1

100/Cholesterol,*dL/mg 8.401 (4.16–16.968) 35.2 2.206 (1.045–4.656) 4.3

White blood cell 1.079 (1.036–1.123) 13.5 3.4

Hemoglobin, g/dL 0.884 (0.833–0.937) 16.9 �0.1

If hemoglobin �16, 16�hemoglobin 1.186 (1.110–1.266) 25.9 1.124 (1.053–1.200) 12.3

If hemoglobin �16, hemoglobin�16 1.363 (1.038–1.79) 5.0 1.336 (1.010–1.767) 4.1

% Lymphocytes,* each 5% (for lymphocytes �47%)* 0.823 (0.777–0.871) 44.9 0.897 (0.846–0.951) 13.2

Uric acid, mg/dL (for uric acid �3.4)* 1.12 (1.008–1.16) 36.9 1.064 (1.022–1.108) 9.1

Hazard Ratio:Add Medication/Device

Hazard Ratio:Patient on Medication/Device

Medications

ACE inhibitor 0.7716 0.7715,17

�-Blocker 0.6619 0.6615,17

Angiotensin receptor blocker 0.8710,20,21 0.85

K-sparing diuretic 0.7022 0.7423

Statin 0.7818 0.6324

Devices

Biventricular pacemaker 0.7425,26,28 1.00

Implantable cardioverter-defibrillator 0.7427 0.7329

Biventricular implantable cardioverter-defibrillator 0.6425 0.79

Left ventricular assist device 0.523 N/A

Unadjusted and adjusted hazard ratios with 95% CIs are shown. A �2 of �4 corresponds to a P value �0.05. N/A indicates datanot available.

*These characteristics were evaluated as continuous variables in the model. The Seattle Heart Failure Score is calculated with theuse of the natural log of the multivariate hazard ratios that are in italic typeface as described in the online-only Data SupplementAppendix. The �2 value is shown for adding each of the other variables individually to the Seattle Heart Failure Score. The hazard ratiosfor adding a medication or device or for a patient who is already on a medication or device were estimated from large randomizedtrials or meta-analyses (see Methods for details).




nitrogen, creatinine, white blood cell count, heart rate, dia-betes, diastolic blood pressure, albumin, and total bilirubinwere no longer independent predictors. The estimated mor-tality impact of adding versus already being on a medicationor device was generally similar except for biventricularpacing (Table 2), for which all of the predicted benefit ofbiventricular pacing was already accounted for among pa-tients already on the device by improvements in NYHA class(�0.6 U), SBP (�6.3 mm Hg), and EF (�6.9 U).28 A hazardratio of 1.13 was used for patients on etanercept inRENAISSANCE.14

Performance of the Model in PRAISE1The overall model was highly significant, with Wald �2�208.The baseline survival function was e�0.0405�t (r�0.999). The

predicted 1- and 2-year survival rates for the entire cohortwere 73.4% and 56.7% versus actual survival of 74.3% and56.0% (Table 3). The correlation between predicted andactual survival by deciles was 0.97, SEE��5% (Figure 3A).

Validation of the ModelThe Seattle Heart Failure Model was applied prospectively toELITE2, Val-HeFT, UW, RENAISSANCE, and IN-CHF(Table 3). In ELITE2, the Wald �2 of the score was 205. Thepredicted 1- and 2-year survivals for the entire cohort were90.5% and 82.4% versus actual survival of 88.5% and 80.0%.The correlation between predicted and actual survival bydeciles was 0.97, SEE��3% (Figure 3B). In Val-HeFT, theWald �2 of the score was 461. The predicted 1-, 2-, and 3-yearsurvivals for the entire cohort were 90.9%, 83.3%, and 76.8%versus actual survival of 91.0%, 81.6%, and 71.7%. Thecorrelation between predicted and actual survival by decileswas 0.98, SEE��3% (Figure 3C), although the modeloverestimated 3-year survival by �5%. In the UW study, theWald �2 was 30. The predicted 1-, 2-, and 3-year survivals forthe entire cohort were 86.5%, 76.5%, and 68.6% versus actualsurvival of 86.5%, 79.7%, and 71.8%. The correlation be-tween predicted and actual survival by quintiles was 0.99,SEE��2% (Figure 3D). In RENAISSANCE, the Wald �2

was 78. The predicted 1- and 2-year survivals for the entirecohort were 83.8% and 72.3% versus actual survival of83.3% and 65.4%. The correlation between predicted andactual survival by quintiles was 0.97, SEE��4% (Figure3E). In IN-CHF, the Wald �2 was 99. The predicted versusactual 1-year survivals for the entire cohort was 89.6% versus86.7%. The correlation between predicted and actual survivalby quintiles was 0.99, SEE��1% (Figure 3F).

Figure 2. The final variables and their contribution to the predic-tive power of the model in PRAISE1 are shown. Lymph indi-cates lymphocytes; Hgb, hemoglobin.

TABLE 3. Survival in Derivation and Validation Cohorts

DemographicPRAISE19

(n�1125)ELITE210

(n�2987)UW12

(n�148)RENAISSANCE14

(n�925)Val-HeFT11

(n�5010)IN-CHF15

(n�872)

Mean follow-up of survivors, y 1.5 1.6 3.6 1.1 2.0 1.0

Death/LVAD/transplantation, n 403 505 48 179 979 115

Death, n 395 505 19 160 979 115

Transplantation/LVAD, n 8 N/A 29 19 N/A N/A

LVAD/UNOS 1, n N/A N/A 19 N/A N/A N/A

UNOS 2, n N/A N/A 10 N/A N/A N/A

1-Year survival

Actual* 74.3�1.4% 88.5�0.6% 86.5�2.8% 83.3�1.4% 91.0�0.4% 86.7�1.2%

Predicted† 73.4�0.5% 90.5�0.1% 86.5�1.0% 83.8�0.5% 90.9�0.1% 89.6�0.4%

2-Year survival

Actual* 56.0�1.8% 80.0�1.0% 79.7�3.3% 65.4�4.6% 81.6�0.6% � � �

Predicted† 56.7�0.6% 82.4�0.2% 76.5�1.6% 72.3�0.7% 83.3�0.2% � � �

3-Year survival

Actual* � � � � � � 71.8�3.7% � � � 71.7�1.3% � � �

Predicted† � � � � � � 68.6�1.8% � � � 76.8�0.2% � � �

1-Year ROC 0.725 0.679 0.810 0.682 0.694 0.749

95% CI 0.69–0.76 0.65–0.71 0.72–0.90 0.63–0.73 0.68–0.72 0.70–0.80

UNOS indicates United Network for Organ Sharing.*Survival�Kaplan-Meier survival free from LVAD or transplantation�SEE.†Survival predicted by the Seattle Heart Failure Model�SEE.




Figure 3. Predicted vs actual survival by deciles of survival inPRAISE1 (the derivation cohort) (A) and validation cohortsELITE2 (B), Val-HeFT (C), UW (D), RENAISSANCE (E), andIN-CHF (F). G, Predicted vs actual 1-, 2-, and 3-year survival bydeciles of survival in all 6 data sets combined. The diagonal lineis the line of identity in all 7 graphs.




The 6 data sets were combined into 1 data set. Thepredicted 1-, 2-, and 3-year survivals were 88.2%, 79.2%, and71.8% versus actual survival of 87.8%, 77.6%, and 68.0%(Figure 3G). The overall fit by deciles was very good(r�0.98, SEE��3) but with a slight underestimation ofmortality, most evident at 2 and 3 years for lower-riskpatients. The model overestimated survival by �2% in theELITE2 data set, as reflected by the mean 1-year survival. InVal-HeFT, the largest data set, the model correctly predicted1-year survival and 2-year survival.

When we rounded each individual’s score to the nearestinteger between �1 and 4 (a natural log increase in risk�2.7-fold), the discriminatory power for overall mortalitywas excellent (Figure 4).

The 1-year ROC for PRAISE1, the derivation cohort, was0.725 (95% CI, 0.69 to 0.76). In the validation cohorts, the1-year ROC was 0.682 (95% CI, 0.65 to 0.73) for ELITE2,0.810 (95% CI, 0.72 to 0.90) for UW, 0.682 (95% CI, 0.63 to0.73) for RENAISSANCE, 0.694 (95% CI, 0.68 to 0.72) forVal-HeFT, and 0.749 (95% CI, 0.70 to 0.80) for IN-CHF. Ifoutcomes in UW were restricted to death/LVAD/UnitedNetwork for Organ Sharing (UNOS) 1 transplantations, withcensoring alive for UNOS 2 transplantations, the 1-year ROCwas 0.88 (95% CI, 0.80 to 0.96). The combined data set of all6 trials had a 1-year ROC of 0.729 (95% CI, 0.714 to 0.744).

Four data sets (PRAISE1, Val-HeFT, IN-CHF, and UW)had data available to allow comparison of the Seattle HeartFailure Model with the ADHERE and the Toronto heartfailure models. In these combined data sets, the 1-year ROCwas 0.75 (95% CI, 0.73 to 0.77) for the Seattle Heart FailureModel, 0.59 (95% CI, 0.57 to 0.61) for the ADHERE model,and 0.68 (95% CI, 0.66 to 0.70) for the Toronto model.

Practical Calculation of the Estimated SurvivalThe calculation of estimated survival included 14 continuousvariables and 10 categorical values, making it impractical forcomputation by hand (an example is provided in the onlineData Supplement). A Web-based calculator (http://www.SeattleHeartFailureModel.org) has been developed, allowingconvenient interactive calculation of estimated survival andalso of the predicted effects on survival of adding (orsubtracting) medications or devices to a patient’s regimen(Figures 5 and 6).

DiscussionThe Seattle Heart Failure Score was derived in the PRAISE1database and validated prospectively in 5 additional data-bases: ELITE2, Val-HeFT, UW, RENAISSANCE, and IN-CHF. It accurately predicts survival of heart failure patientswith the use of commonly obtained clinical characteristics.Importantly, the validation cohorts included patients with awide range of countries or origins (46), ages (14 to 100years), EFs (1% to 75%), and heart failure symptoms (NYHAclass I to IV), including patients at intermediate and higherrisk in whom prediction of risk can be most problematic.Additionally, 2 of the validation cohorts consisted of patientsfrom a more general population, followed either in a US heartfailure clinic (UW) or Italian general cardiology clinics(IN-CHF), representing the populations in whom validationresults may be most widely applicable. The model performedextremely well in these groups, among whom the ROC forpredicting 1-year survival was highest (0.75 to 0.81).

The score indicated that NYHA class, ischemic etiology,diuretic dose, EF, SBP, sodium, hemoglobin, percent lym-phocytes, uric acid, and cholesterol each had independentpredictive power. Renal function was not an independentpredictor, consistent with the HFSS2 but inconsistent withother models.6 In post hoc analyses, adding creatinine hadlittle effect on the ROC (�0.001). Most previous heart failurerisk models have not included medications and also requiredeither invasive hemodynamic measurements or peak V̇O2.2

For example, peak V̇O2 was available in �1% of the patientsin these 6 data sets (95 of 148 from UW) and did not add tothe score in this small subset (P�0.09). In addition, theSeattle Heart Failure Model incorporates the most currentdata on medications and devices available in clinical practice.

The Seattle Heart Failure Model also had a higher ROCthan the ADHERE or Toronto models applied to the samedata sets. ADHERE and Toronto models were derived inhospitalized patients and have not been previously validatedamong nonhospitalized patients. The accuracy of the previ-ously developed HFSS model has been suboptimal in valida-

Figure 4. The combined data set of the derivation and 5 valida-tion cohorts for a Seattle Heart Failure Score rounded to �1 to4. The score, the predicted 1-year survival for the score, andthe number of patients with that score are shown.

Figure 5. The predicted effects on survival of sequentially add-ing medications and an ICD for a heart failure patient with anannual mortality of 20% and a mean survival of 4.1 years atbaseline. Adding an ACE inhibitor (ACEI), a �-blocker, an aldo-sterone (Aldo) blocker, and an ICD decreases the annual mortal-ity by 70% (20% to 6%) and increases the mean survival by 5.6years (mean survival 4.1, 5.0, 6.6, 8.2, and 9.7 years,respectively).




tion data sets (Figure 1) and may be particularly problematicfor patients on �-blockers.7,8

Neurohormones,33 cytokines,34 and peak V̇O22 are prognos-

tically important in multivariate models, although recentstudies have suggested that peak V̇O2 may have limited utilityin patients on �-blockers.7,8 The HFSS, which includes peakV̇O2, had a 1-year ROC for death/LVAD/UNOS 1 transplan-tation of 0.69 to 0.762,8 versus 0.88 for the Seattle HeartFailure Model in a similar population (UW). Several vari-ables in the Seattle Heart Failure Model may in part incor-porate some of the prognostic power of neurohormones andcytokines, including SBP and sodium (associated with nor-epinephrine and renin35), diuretic dose (associated with re-nin35), and cholesterol, uric acid, percent lymphocytes, andhemoglobin (associated with tumor necrosis factor).13,34 In aprevious publication from Val-HeFT, B-type natriuretic pep-tide (BNP) (P�0.0001) was added to a multivariate model,whereas renin (P�0.01) and norepinephrine (P�0.02) hadmodest effects.33 The addition of BNP in Val-HeFT to theSeattle Heart Failure Model increased the ROC by �0.03.The addition of peak V̇O2, BNP, and other inflammatoryvariables may add to the prognostic power of the score, andfurther investigation is warranted.

The Seattle Heart Failure Model also incorporates the useof medications and devices and predicts the associatedchange in survival. The inclusion of medications and devicesin a prognostic heart failure model is critical because medi-cations and devices are altered by healthcare providers toimprove survival in their patients. Furthermore, such a modelmay facilitate translation of clinical trial results into clinicalpractice. Additional potential uses include educating patientson the preventive value of medications, listing patients forcardiac transplantation, comparing heart failure severity be-tween various facilities, counseling patients about end-of-lifeissues, and selecting patients for device use or entry intoclinical trials. The Seattle Heart Failure Model can also be

updated as new heart failure medications or devices areavailable in the future.

Although the hazard ratios for individual components ofthe score may vary in specific databases, the ability of theoverall score to predict survival was high in all 5 validationcohorts with minimal systematic bias. Similar results havebeen shown with the Framingham Coronary Heart DiseaseScore, in which the hazard ratios for individual componentsof the score varied between cohorts, but the overall Framing-ham Coronary Heart Disease Score performed well.36 How-ever, the Framingham score required recalibration for ethnicpopulations with a markedly lower coronary heart diseaserisk.36 Whether the Seattle Heart Failure Model will needrecalibration in different ethnic populations is unknown, butthe model has been validated in data sets with patients from46 countries.

LimitationsThe hazard ratios for a subset of medications and deviceswere estimated from prior published literature, and resultsfrom prior clinical trials may not be generalizable to a widerpopulation of heart failure patients. However, large observa-tional databases of nonclinical trial heart failure patients haveshown similar hazard ratios for ACE inhibitors, �-blockers,aldosterone blockers, and ICDs as seen in randomized clinicaltrials.3,10,15–29 Additionally, the model performed extremelywell in all 5 validation cohorts, including studies of both heartfailure clinic and community-based patients. The model doesnot provide an estimate of whether a heart failure medication/device can be safely added, which is a clinical judgment, butrather predicts benefit if it were added. The estimated benefitfor medications was based on the doses used by patients inclinical trials, and effects may be different if higher or lowerdoses are used. The Seattle Heart Failure Model was devel-oped and validated among outpatients participating in clinicaltrials, observational studies, or clinical registries, and themodel may not be generalizable to hospitalized patients or

Figure 6. The Seattle Heart FailureModel has been implemented as aninteractive program that employs theSeattle Heart Failure Score to estimatemean, 1-, 2-, and 5-year survival and thebenefit of adding medications and/ordevices for an individual patient.This model is available at www.SeattleHeartFailureModel.org. ACE-I indi-cates ACE inhibitor; ARB, angiotensinreceptor blocker; HCTZ, hydrochlorothia-zide; Hgb, hemoglobin; and BiV,biventricular.




those with major life-altering comorbidities such as cirrhosis,renal failure, dementia, or cancer4; inclusion of such comor-bidities may add prognostic power to the Seattle Heart FailureModel. The potential benefit of statin therapy was based onthe benefit on cardiac mortality in a meta-analysis.18 Twolarge clinical trials of statins in heart failure (ControlledRosuvastatin Multinational Trial in Heart Failure [CORONA]and Gruppo Italian per lo Studio della SopravvivenzaNell’Infarto Miocardioco-Heart Failure [GISSI-HF]) are on-going. The estimate of mean life-years is provided to illus-trate the potential change in long-term mortality with theaddition of medications/devices; however, the potential errorfor such long-term estimates may be large for very-low-riskpatients. For the most accurate survival estimate, the scoreshould be recalculated after changes in clinical status ormedications/device. The benefit of heart failure medications/devices in diastolic heart failure is less certain because thescore was derived and validated mainly in patients withsystolic heart failure. However, the accuracy of the modelwas similar in the IN-CHF population, in whom one third ofpatients had an EF �40%.

In conclusion, the Seattle Heart Failure Model allowsprediction of survival of heart failure patients with the use ofeasily obtained clinical characteristics. The model providesan accurate estimate of mean, 1-, 2-, and 3-year survival andallows estimation of effects of adding medications or devicesto a patient’s regimen. Use of this model by both healthcareproviders and their patients may facilitate estimation of progno-sis, enhance compliance, and increase the use of life-savingmedications and devices; further investigation of such potentialbenefits is needed.

AcknowledgmentsThis research was supported in part by an unrestricted gift fromAmgen. The UW Depression Study was funded by the AmericanHeart Association, Dallas, Tex, and the Dana Foundation, NewYork, NY. Statistical assistance and/or database access was kindlyprovided by Pfizer Laboratories for PRAISE1, Merck ResearchLaboratories for ELITE2, and Amgen Inc for RENAISSANCE.Database access was kindly provided by Novartis Pharmaceuticalsfor Val-HeFT (Jay N. Cohn, MD, Val-HeFT Primary Investigator)and by the Italian Heart Failure Registry for IN-CHF. The authorsthank all of the subjects, research coordinators, and investigatorswho participated in the trials.

DisclosuresWayne Levy received honoraria or consulted with Pfizer, Glaxo-SmithKline, Novartis, Boehringer Ingelheim, Cardiac Dimensions,and Scios. He received partial research support for this project fromAmgen. He has received a research grant from GlaxoSmithKline andPfizer. He has equity holdings in Cardiac Dimensions. SantoshSutradhar, Anne Cropp, and Paul Burton are employees of Merck,Pfizer, and Amgen (previous employer), respectively. Inder Anandhas received research grants and honoraria from Novartis Pharma-ceuticals. Bertram Pitt is a consultant to Pfizer and Novartis. BertramPitt, Philip Poole-Wilson, Douglas Mann, and Milton Packer werethe primary investigators for the clinical trials used for the analysesand for other large clinical heart failure trials. Other authors do notreport conflicts of interest.

References1. Heart disease and stroke statistics: 2004 update. Available at: http://

www.americanheart.org. Accessed April 28, 2004.

2. Aaronson KD, Schwartz JS, Chen TM, Wong KL, Goin JE, Mancini DM.Development and prospective validation of a clinical index to predictsurvival in ambulatory patients referred for cardiac transplant evaluation.Circulation. 1997;95:2660–2667.

3. Rose EA, Moskowitz AJ, Packer M, Sollano JA, Williams DL, TierneyAR, Heitjan DF, Meier P, Ascheim DD, Levitan RG, Weinberg AD,Stevenson LW, Shapiro PA, Lazar RM, Watson JT, Goldstein DJ, GelijnsAC. The REMATCH trial: rationale, design, and end points: RandomizedEvaluation of Mechanical Assistance for the Treatment of CongestiveHeart Failure. Ann Thorac Surg. 1999;67:723–730.

4. Lee DS, Austin PC, Rouleau JL, Liu PP, Naimark D, Tu JV. Predictingmortality among patients hospitalized for heart failure: derivation andvalidation of a clinical model. JAMA. 2003;290:2581–2587.

5. Fonarow GC, Adams KF Jr, Abraham WT, Yancy CW, Boscardin WJ.Risk stratification for in-hospital mortality in acutely decompensatedheart failure: classification and regression tree analysis. JAMA. 2005;293:572–580.

6. Brophy JM, Dagenais GR, McSherry F, Williford W, Yusuf S. A multi-variate model for predicting mortality in patients with heart failure andsystolic dysfunction. Am J Med. 2004;116:300–304.

7. Lund LH, Aaronson KD, Mancini DM. Predicting survival in ambulatorypatients with severe heart failure on beta-blocker therapy. Am J Cardiol.2003;92:1350–1354.

8. Koelling TM, Joseph S, Aaronson KD. Heart failure survival scorecontinues to predict clinical outcomes in patients with heart failurereceiving beta-blockers. J Heart Lung Transplant. 2004;23:1414–1422.

9. Packer M, O’Connor CM, Ghali JK, Pressler ML, Carson PE, Belkin RN,Miller AB, Neuberg GW, Frid D, Wertheimer JH, Cropp AB, DeMetsDL, for the Prospective Randomized Amlodipine Survival EvaluationStudy Group. Effect of amlodipine on morbidity and mortality in severechronic heart failure. N Engl J Med. 1996;335:1107–1114.

10. Pitt B, Poole-Wilson PA, Segal R, Martinez FA, Dickstein K, Camm AJ,Konstam MA, Riegger G, Klinger GH, Neaton J, Sharma D, ThiyagarajanB. Effect of losartan compared with captopril on mortality in patients withsymptomatic heart failure: randomised trial: the Losartan Heart FailureSurvival Study ELITE II. Lancet. 2000;355:1582–1587.

11. Cohn JN, Tognoni G. A randomized trial of the angiotensin-receptorblocker valsartan in chronic heart failure. N Engl J Med. 2001;345:1667–1675.

12. Sullivan MD, Levy WC, Crane BA, Russo JE, Spertus JA. Usefulness ofdepression to predict time to combined end point of transplant or death foroutpatients with advanced heart failure. Am J Cardiol. 2004;94:1577–1580.

13. Huehnergarth KV, Mozaffarian D, Sullivan MD, Crane BA, WilkinsonCW, Lawler RL, McDonald GB, Fishbein DP, Levy WC. Usefulness ofrelative lymphocyte count as an independent predictor of death/urgenttransplant in heart failure. Am J Cardiol. 2005;95:1492–1495.

14. Mann DL, McMurray JJ, Packer M, Swedberg K, Borer JS, Colucci WS,Djian J, Drexler H, Feldman A, Kober L, Krum H, Liu P, Nieminen M,Tavazzi L, van Veldhuisen DJ, Waldenstrom A, Warren M, Westheim A,Zannad F, Fleming T. Targeted anticytokine therapy in patients withchronic heart failure: results of the Randomized Etanercept WorldwideEvaluation (RENEWAL). Circulation. 2004;109:1594–1602.

15. Maggioni AP, Opasich C, Anand I, Barlera S, Carbonieri E, Gonzini L,Tavazzi L, Latini R, Cohn J. Anemia in patients with heart failure:prevalence and prognostic role in a controlled trial and in clinical practice.J Card Fail. 2005;11:91–98.

16. Garg R, Yusuf S, for the Collaborative Group on ACE Inhibitor Trials.Overview of randomized trials of angiotensin-converting enzyme inhib-itors on mortality and morbidity in patients with heart failure. JAMA.1995;273:1450–1456.

17. Sin DD, McAlister FA. The effects of beta-blockers on morbidity andmortality in a population-based cohort of 11,942 elderly patients withheart failure. Am J Med. 2002;113:650–656.

18. Studer M, Briel M, Leimenstoll B, Glass TR, Bucher HC. Effect ofdifferent antilipidemic agents and diets on mortality: a systematic review.Arch Intern Med. 2005;165:725–730.

19. Lee S, Spencer A. Beta-blockers to reduce mortality in patients withsystolic dysfunction: a meta-analysis. J Fam Pract. 2001;50:499–504.

20. Dickstein K, Kjekshus J. Effects of losartan and captopril on mortalityand morbidity in high-risk patients after acute myocardial infarction: theOPTIMAAL randomised trial: Optimal Trial in Myocardial Infarctionwith Angiotensin II Antagonist Losartan. Lancet. 2002;360:752–760.

21. Pfeffer MA, Swedberg K, Granger CB, Held P, McMurray JJ, MichelsonEL, Olofsson B, Ostergren J, Yusuf S, Pocock S. Effects of candesartan




on mortality and morbidity in patients with chronic heart failure: theCHARM-Overall programme. Lancet. 2003;362:759–766.

22. Pitt B, Zannad F, Remme WJ, Cody R, Castaigne A, Perez A, PalenskyJ, Wittes J, for the Randomized Aldactone Evaluation Study Investi-gators. The effect of spironolactone on morbidity and mortality in patientswith severe heart failure. N Engl J Med. 1999;341:709–717.

23. Domanski M, Norman J, Pitt B, Haigney M, Hanlon S, Peyster E.Diuretic use, progressive heart failure, and death in patients in the Studiesof Left Ventricular Dysfunction (SOLVD). J Am Coll Cardiol. 2003;42:705–708.

24. Mozaffarian D, Nye R, Levy WC. Statin therapy is associated with lowermortality among patients with severe heart failure. Am J Cardiol. 2004;93:1124–1129.

25. Bristow MR, Saxon LA, Boehmer J, Krueger S, Kass DA, De Marco T,Carson P, DiCarlo L, DeMets D, White BG, DeVries DW, Feldman AM.Cardiac-resynchronization therapy with or without an implantable defi-brillator in advanced chronic heart failure. N Engl J Med. 2004;350:2140–2150.

26. McAlister FA, Ezekowitz JA, Wiebe N, Rowe B, Spooner C, Crumley E,Hartling L, Klassen T, Abraham W. Systematic review: cardiac resyn-chronization in patients with symptomatic heart failure. Ann Intern Med.2004;141:381–390.

27. Nanthakumar K, Epstein AE, Kay GN, Plumb VJ, Lee DS. Prophylacticimplantable cardioverter-defibrillator therapy in patients with left ventric-ular systolic dysfunction: a pooled analysis of 10 primary preventiontrials. J Am Coll Cardiol. 2004;44:2166–2172.

28. Cleland JG, Daubert JC, Erdmann E, Freemantle N, Gras D,Kappenberger L, Tavazzi L. The effect of cardiac resynchronization onmorbidity and mortality in heart failure. N Engl J Med. 2005;352:1539–1549.

29. Chan PS, Hayward RA. Mortality reduction by implantable cardioverter-defibrillators in high-risk patients with heart failure, ischemic heartdisease, and new-onset ventricular arrhythmia: an effectiveness study.J Am Coll Cardiol. 2005;45:1474–1481.

30. Pickin DM, McCabe CJ, Ramsay LE, Payne N, Haq IU, Yeo WW,Jackson PR. Cost effectiveness of HMG-CoA reductase inhibitor (statin)treatment related to the risk of coronary heart disease and cost of drugtreatment. Heart. 1999;82:325–332.

31. Human Mortality Database. University of California, Berkeley (USA),and Max Planck Institute for Demographic Research (Germany).Available at: http://www.mortality.org. Accessed September 16, 2004.

32. Mozaffarian D, Nye R, Levy WC. Anemia predicts mortality in severeheart failure: the Prospective Randomized Amlodipine Survival Eval-uation (PRAISE). J Am Coll Cardiol. 2003;41:1933–1939.

33. Latini R, Masson S, Anand I, Salio M, Hester A, Judd D, Barlera S,Maggioni AP, Tognoni G, Cohn JN. The comparative prognostic value ofplasma neurohormones at baseline in patients with heart failure enrolledin Val-HeFT. Eur Heart J. 2004;25:292–299.

34. Rauchhaus M, Clark AL, Doehner W, Davos C, Bolger A, Sharma R,Coats AJ, Anker SD. The relationship between cholesterol and survival inpatients with chronic heart failure. J Am Coll Cardiol. 2003;42:1933–1940.

35. Packer M, Lee WH, Kessler PD, Gottlieb SS, Bernstein JL, Kukin ML.Role of neurohormonal mechanisms in determining survival in patientswith severe chronic heart failure. Circulation. 1987;75:IV-80–IV-92.

36. D’Agostino RB Sr, Grundy S, Sullivan LM, Wilson P. Validation of theFramingham coronary heart disease prediction scores: results of amultiple ethnic groups investigation. JAMA. 2001;286:180–187.

37. Anker SD, Doehner W, Rauchhaus M, Sharma R, Francis D, Knosalla C,Davos CH, Cicoira M, Shamim W, Kemp M, Segal R, Osterziel KJ,Leyva F, Hetzer R, Ponikowski P, Coats AJ. Uric acid and survival inchronic heart failure: validation and application in metabolic, functional,and hemodynamic staging. Circulation. 2003;107:1991–1997.

38. Lund LH, Aaronson KD, Mancini DM. Validation of peak exerciseoxygen consumption and the Heart Failure Survival Score for serial riskstratification in advanced heart failure. Am J Cardiol. 2005;95:734–741.

39. Zugck C, Kruger C, Kell R, Korber S, Schellberg D, Kubler W, Haass M.Risk stratification in middle-aged patients with congestive heart failure:prospective comparison of the Heart Failure Survival Score (HFSS) anda simplified two-variable model. Eur J Heart Fail. 2001;3:577–585.

40. Deng MC, De Meester JM, Smits JM, Heinecke J, Scheld HH, for theComparative Outcome and Clinical Profiles in Transplantation (COCPIT)Study Group. Effect of receiving a heart transplant: analysis of a nationalcohort entered on to a waiting list, stratified by heart failure severity.BMJ. 2000;321:540–545.

CLINICAL PERSPECTIVEThe number of individuals living with heart failure is steadily increasing. A well-validated method has not been availablefor healthcare providers to estimate the survival of a patient with heart failure. The Seattle Heart Failure Model, derivedand validated in �11 000 patients, makes use of commonly assessed clinical variables such as age, gender, weight, systolicblood pressure, New York Heart Association functional class, ejection fraction, basic laboratory values, and heart failuremedications and devices (eg, defibrillators and biventricular pacemakers) to provide an estimate of survival in patients withheart failure. The effect on survival of adding medications and devices to the patient’s regimen can be estimated, includingangiotensin-converting enzyme (ACE) inhibitors, angiotensin receptor blockers, �-blockers, statins, aldosterone blockers,biventricular pacemakers, defibrillators, and left ventricular assist devices. For example, a heart failure patient with anestimated 20% annual mortality would have an average survival of �4 years without treatment. The addition of an ACEinhibitor, �-blocker, and aldosterone blocker would reduce the predicted annual mortality from 20% to 8%, adding �4additional years of life. The authors anticipate that the model could be commonly used in clinical practice to estimate thesurvival of heart failure patients and illustrate the estimated benefit of adding medications/devices to a given patient’sregimen.




Philip A. Poole-Wilson, Douglas L. Mann and Milton PackerAnne B. Cropp, Inder Anand, Aldo Maggioni, Paul Burton, Mark D. Sullivan, Bertram Pitt,

Wayne C. Levy, Dariush Mozaffarian, David T. Linker, Santosh C. Sutradhar, Stefan D. Anker,The Seattle Heart Failure Model: Prediction of Survival in Heart Failure

Print ISSN: 0009-7322. Online ISSN: 1524-4539 Copyright © 2006 American Heart Association, Inc. All rights reserved.

is published by the American Heart Association, 7272 Greenville Avenue, Dallas, TX 75231Circulation doi: 10.1161/CIRCULATIONAHA.105.584102

2006;113:1424-1433; originally published online March 13, 2006;Circulation.

http://circ.ahajournals.org/content/113/11/1424World Wide Web at:

The online version of this article, along with updated information and services, is located on the

http://circ.ahajournals.org/content/suppl/2006/03/15/CIRCULATIONAHA.105.584102.DC1.htmlData Supplement (unedited) at:

http://circ.ahajournals.org//subscriptions/

is online at: Circulation Information about subscribing to Subscriptions:

http://www.lww.com/reprints Information about reprints can be found online at: Reprints:

document. Permissions and Rights Question and Answer this process is available in the

click Request Permissions in the middle column of the Web page under Services. Further information aboutOffice. Once the online version of the published article for which permission is being requested is located,

can be obtained via RightsLink, a service of the Copyright Clearance Center, not the EditorialCirculationin Requests for permissions to reproduce figures, tables, or portions of articles originally publishedPermissions:


http://circ.ahajournals.org/content/113/11/1424

http://circ.ahajournals.org/content/suppl/2006/03/15/CIRCULATIONAHA.105.584102.DC1.html

http://www.ahajournals.org/site/rights/

http://www.lww.com/reprints

http://circ.ahajournals.org//subscriptions/


The Seattle Heart Failure Model: Prediction of Survival in Heart Failure

Documents

Transcript of The Seattle Heart Failure Model: Prediction of Survival in Heart Failure