Prognostic impact of the addition of peak oxygenconsumption to the Seattle Heart Failure Model in atransplant referral population
Wayne C. Levy, MD,a Keith D. Aaronson, MD,b Todd F. Dardas, MD,a,b
Paula Williams, BS,c Jennifer Haythe, MD,c and Donna Mancini, MDc
From the aDivision of Cardiology, University of Washington, Seattle, Washington; bDivision of Cardiology, University of Michigan,
Ann Arbor, Michigan; and cDivision of Cardiology, Columbia Presbyterian Medical Center, New York, New York.
BACKGROUND: In this study we investigated whether the addition of peak oxygen consumption (VO2)improves the predictive accuracy of the Seattle Heart Failure Model (SHFM). The SHFM is a validatedmultivariate risk model that uses NYHA classification to assess functional capacity rather than peakoxygen consumption (VO2).METHODS: Outpatients (n � 1,240) evaluated for transplant at three centers had their SHFM scorecalculated and peak VO2 measured. The outcomes assessed were death/LVAD/urgent transplant withcensoring at the time of elective transplant.RESULTS: Over the course of 4.0 (mean) years of observation, there were 571 events. Both the SHFMscore (�2 � 227) and peak VO2 (�2 � 88, both p � 0.0001) were highly predictive of outcomes. TheSHFM and peak VO2 were modestly correlated (r � 0.39, p � 0.0001). In a multivariate Cox model,peak VO2 added to the SHFM with a hazard ratio of 0.949 (p � 0.0001) for each 1-ml/kg/min increase.Peak VO2 improved both the net reclassification improvement and integrated discrimination index(both p � 0.0002). Peak VO2 provided additive prognostic information within each SHFM score (p �0.05). The 1-year areas under the receiver-operating characteristic curve were obtained for peak VO2
(0.645, 95% CI 0.606 to 0.684), SHFM (0.758, 95% CI 0.721 to 0.795) and SHFM with peak VO2
Heart failure carries a risk of death varying from �5% to75% per year.1 Peak oxygen consumption (VO2) is a powerfulprognostic predictor that is widely used for selection of patientsfor cardiac transplantation.2,3 However, the use of heart failuremedications such as �-blockers has been shown to impact theprognostic significance of peak VO2 with a much lower risk ofdeath for a given level of peak VO2.4–8
Reprint requests: Wayne C. Levy, MD, Division of Cardiology, Uni-versity of Washington, Box 356422, 1959 NE Pacific Street, Seattle, WA98195. Telephone: 206-221-4507. Fax: 206-221-6835.
The Seattle Heart Failure Model (SHFM) is a 20-variablemodel that uses New York Heart Association (NYHA)classification as a surrogate for peak VO2.1 It provides a 1-to 5-year estimate of survival and average life expectancyand has been widely validated.9–12 However, the SHFM hasnot been widely validated in a transplant referral population,as 98% of the events in the derivation and validation data-bases were death.9,10
This analysis addresses three questions: Is the SHFMuseful in predicting outcomes in a transplant referral popu-lation? Does the addition of peak VO2 add to the prognostic
significance of the SHFM? Can the SFHM aid in selection
818 The Journal of Heart and Lung Transplantation, Vol 31, No 8, August 2012
of ambulatory high-risk patients for destination left ventric-ular assist devices (LVADs)?
Methods
A total of 1,240 ambulatory outpatients with heart failure had peakVO2 measured during maximal cardiopulmonary exercise testing per-formed for clinical reasons, most for evaluation for cardiac transplan-tation, and had variables available for calculation of the SHFM score.Patients were evaluated at three different transplant centers: ColumbiaPresbyterian Medical Center (n � 647, era 1990 to 2006); Universityof Michigan (n � 498, era 1993 to 2001); and University of Wash-ington (n � 91, era 1999 to 2001). The SHFM score and estimatedsurvival were calculated as previously described.1 Survival was eval-uated at each institution using chart review, telephone and socialsecurity death index. Research use of the clinically obtained data wasapproved by the review board at each institution.
The primary outcome was survival free from death, LVAD orurgent transplant (United Network for Organ Sharing [UNOS]
OutcomesDeath 340 (27%)Left ventricular assist device 59 (5%)UNOS Status 1A or 1B transplant 170 (14%)UNOS Status 2 transplant 93 (8%)Alive 578 (47%)
Seattle Heart Failure Model Score 0.84 � 1.06Estimated 1-year SHFM survival 86 � 15%Observed 1-year survival free of
death/LVAD/urgent transplant (mean �SEM)
83 � 1%
CRT, cardiac resynchronization therapy.
aData presented as mean � SE, percent, mean or mean (percent).
Status 1A or 1B). Survival time for patients who were transplantedas UNOS Status 2 was censored on the date of the transplant. First,a Cox proportional hazards model was used to estimate the hazardratio for the primary outcome for SHFM and peak VO2 alone andin combination. The peak VO2 values were stratified for presen-tation purposes using �10 ml/kg/min, 10 to �14 ml/kg/min and�14 ml/kg/min. The SHFM score was rounded to the nearestinteger between 0 and 3. The hazard ratio for peak VO2 was alsocalculated within a given SHFM score.
To determine the predictive power for adding peak VO2 to theSHFM, we used four methods: (1) change in Cox proportional model�2 with the addition of peak VO2 to the SHFM; (2) change in 1-yearreceiver operating characteristic area under the curve (ROC AUC)was obtained; (3) incremental value of the SHFM and peak VO2 inpredicting outcomes was determined using net reclassification im-provement (NRI); and (4) integrated discrimination improvement(IDI) was calculated.13 NRI is the difference in the number of patientsmoving up or down clinical risk groups, stratified according towhether they develop the outcomes during follow-up. Here, clinicallymeaningful risk annual event cut-points of �5%, �5% to 10%, �10to 20%, �20% to 40% and �40% risk were defined. IDI quantifiesthe improvement in predictive models by comparing the averagedifferences in predicted probabilities at 1 year between those whoexperienced the outcome versus those who did not have an event.Because a binary outcome is required for risk reclassification analy-ses, subjects censored before 1 year were excluded from NRI and IDIanalyses (n � 1,075). A nomogram was developed for the annualevent rates over 2 years with peak VO2 alone and in conjunction withthe SHFM stratified by SHFM scores of 0 to 3. Statistical analyseswere performed using STATVIEW (version 5; SAS, Inc., Cary, NC),SPSS (version 16; SPSS, Inc., Chicago, IL) and STATA (version11.2; Statcorp, College Station, TX).
All investigators had full access to the data and take fullresponsibility for the integrity of the findings.
Results
Study population
The population reflected a transplant referral populationwith a preponderance of middle-aged men, low ejection
0%
10%
20%
30%
40%
50%
60%
70%
80%
90%
100%
0 1 2 3 4 5 6 7 8 9 10
Sur
viva
l
Years
Death
Death/LVAD/UTx
Figure 1 The 10-year event-free survival for the whole cohort,with events defined as either death alone or the combination ofdeath, left ventricular assist device and urgent transplantation.
fraction and high prevalence of standard heart failure med-
819Levy et al. SHFM and PVO2
ication use (see Table 1). Many of these patients wereevaluated prior to the known benefits of �-blockers, aldo-sterone blockers and device use in heart failure. The 1,240patients were followed for an average of 4.0 (interquartilerange 1.0 to 6.7) years. During this time there were 569outcome events (46% of patients). There were 340 deaths,59 LVAD implantations and 170 urgent transplants. Cen-soring occurred when patients were alive at the end offollow-up (n � 578) and for elective transplants (n � 93).The 1-, 2- and 5-year survival (censoring for LVAD ortransplant) was 91%, 84% and 69%, respectively. The 1-, 2-and 5-year survival free from LVAD/urgent transplant forthis cohort was 83%, 73% and 53%, respectively (Figure 1).
Peak VO2 and freedom from death, LVAD andurgent transplant
Peak oxygen consumption (VO2) was 14.9 � 5.1 ml/kg/min. Peak VO2 was a significant univariate risk marker with
Figure 2 Kaplan–Meier estimate of event-free survival fromdeath/LVAD/urgent transplant when: (a) stratified by peak VO2
�10 ml/kg/min, �10 to 14 ml/kg/min and �14 ml/kg/min and thenumber of patients in each peak VO2 group (b) stratified by SHFMscore of 0 to 3. The number of patients in each SHFM group isshown above.
a hazard ratio 0.915 ml/kg/min (95% CI 0.899 to 0.933,
�2 � 88, p � 0.0001) for the principal outcome. The 1-yearsurvival free of LVAD/urgent transplant was 72% for apeak VO2 �10 ml/kg/min, 80% for a peak VO2 10 to 14ml/kg/min and 88% for peak VO2 �14 ml/kg/min. Thecorresponding 5-year survival free from LVAD/urgenttransplant was 39%, 45% and 63%. Kaplan–Meier curvesfor groupings of peak VO2 �10 ml/kg/min, �10 to 14ml/kg/min and �14 ml/kg/min are shown in Figure 2a. The1-year ROC AUC for peak VO2 was 0.645 (95% CI 0.606to 0.684).
SHFM and freedom from death, LVAD and urgenttransplants
The mean SHFM score was 0.84 � 1.06. SHFM-estimated1-year survival was 86 � 15% for all patients. SHFM scorewas a powerful predictor of survival free from LVAD/urgent transplant with a hazard ratio was 1.80 for each unitchange in the score (95% CI 1.66 to 1.94, �2 � 227, p �0.0001). The 1-year survival free of LVAD/urgent trans-plant was 95% for a value of 0% and 49% for a score of 3or a 10-fold increase in risk. Figure 2b shows the Kaplan–
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
Sen
sitiv
ity
1-Specificity
Peak VO2
SHFM
SHFM & Peak VO2
A
B
Figure 3 (a) SHFM score plotted vs peak VO2. (b) One-yearROC as follows: peak VO2 (0.645, 95% CI 0.606 to 0.684); SHFM(0.758, 95% CI 0.721 to 0.795); and SHFM with peak VO (0.766,
2
95% CI 0.731 to 0.802).
2
820 The Journal of Heart and Lung Transplantation, Vol 31, No 8, August 2012
Meier survival curve with rounding values of the SHFMscore to the nearest integer between 0 and 3. The 1-yearROC AUC for SHFM was 0.758 (95% CI 0.721 to 0.795).
Association of peak VO2 and SHFM
The peak VO2 and the SHFM score had a significant inverseassociation (r � 0.39, p � 0.0001; Figure 3a). For SHFMscore 0, the mean peak VO2 was 17.1 � 5.7 ml/kg/min. For
Figure 4 (A-D) The additional prognostic information adding peml/kg/min is shown for SHFM scores of 0, 1, 2 and 3, respectiveclinical impact of peak VO2 was in lower risk patients (i.e., SHFrisk, even if peak VO2 was �14 ml/kg/min. (E) Patients with peashown in.
each unit increase in SHFM score, the peak VO2 decreased
by �2 ml/kg/min. The average peak VO2 for an SHFMscore of 3 was 12.4 � 3.9 ml/kg/min.
Joint effect of SHFM and peak VO2 on outcomes
The addition of peak VO2 to the SHFM increased the Coxmodel �2 from 227 to 242 (p � 0.0001; Figure 4a–d).Addition of peak VO2 to the SHFM increased the 1-yearROC AUC from 0.758 (95% CI 0.721 to 0.795) to 0.766
2 by categories of �10 ml/kg/min, �10 to 14 ml/kg/min and �14g with SHFM-estimated mortality for SHFM score. The greatestes 0 and 1). The patients with SHFM scores of 2 or 3 were highof �10 ml/kg/min can be further risk stratified by the SHFM as
10
Sur
viva
l Fre
e of
LV
AD
/Urg
ent T
rans
plan
t
10
87
10
D
B
ak VOly, alonM scork VO
(95% CI 0.731 to 0.802; Figure 3b), although the change in
821Levy et al. SHFM and PVO2
the AUC was not significant (p � 0.16). The hazard ratio forpeak VO2 when added to the SHFM was 0.949 ml/kg/minand similar within all SHFM scores (Table 2). The interac-tion between SHFM score and peak VO2 was non-signifi-cant, suggesting that peak VO2 had similar prognostic sig-nificance across the range of SHFM scores. A nomogramfor a peak VO2 of 5 to 25 ml/kg/min alone and in combi-nation with the SHFM is shown in Figure 5.
Given the relative insensitivity in detecting clinicallysignificant risk differences using AUC methods, we alsoinvestigated reclassification using NRI and IDI. We as-sessed whether adding peak VO2 to the SHFM score im-proved classification of patients with and without an event(Table 4). We used clinically meaningful annual event cut-points of �5%, �5% to 10%, �10% to 20%, �20% to 40%and �40% risk. After classifying patients into risk groupsusing the SHFM, the addition of peak VO2 reclassified13.6% of patients into more appropriate risk groups (p �
Table 3
SHFM with peak VO2
�5% �10% �20% �40% �40% Total
Subjects who experience death, LVAD, or urgent transplant�5% 26 9 0 0 0 35�10% 7 118 35 0 0 160�20% 0 20 204 23 0 247�40% 0 0 14 91 7 112�40% 0 0 0 3 12 15Total 33 147 253 117 19 569Subjects who do not experience death, LVAD, or urgent
NRI: 0.136 (95% CI 0.077 to 0.195), p � 0.0001; NRI (events):0.053 (95% CI 0.154 to 0.090), p � 0.006; NRI (non-events):�0.083 (95% CI �0.130, �0.037), p � 0.0005. IDI: 0.59%, (95% CI0.0028, 0.0090), p � 0.0002; IDI (events): 0.33% (95% CI 0.0009 to0.0056), p � 0.006; IDI (non-events): �0.26% (95% CI �0.0046 to�0.0007), p � 0.009.
Table 2 Effect in a Cox Proportional Hazards Model of Adding2 or 3
0.0001). The IDI, or difference in the average predictedprobabilities between the two models (SHFM alone vsSHFM and peak VO2) in differentiating patient risk, wasalso significant at 0.59% (p � 0.0002). The change in theCox proportional model �2, NRI and IDI all suggest asignificant, yet modest, improvement in risk classificationwith the addition of peak VO2 to the SHFM.
VO2 to SHFM Overall and Stratified by SHFM Scores of 0, 1 and
Hazard Ratio Peak VO2 (ml/kg/min) 0.915Chi square 88, p<0.0001
0%
10%
20%
30%
40%
50%
60%
5 10 15 20 25
Ann
ual R
ate
of D
eath
/LV
AD
/Urg
ent T
rans
plan
t
Peak VO2 (ml/kg/min)
SHFM 0
SHFM 1
SHFM2
SHFM 3
Hazard Ratio Peak VO20.949, p<0.0001
A
B
Figure 5 A nomogram derived from the Cox model for theannual rate of death/LVAD/urgent transplant over 2 years by peakVO2 (a) as a continuous variable with the observed annual eventrate by peak VO2 quartiles (dots) and (b) in combination with theSHFM score along with observed event rates by the SHFM scoreand peak VO categories of �10 ml/kg/min, �10 to 14 ml/kg/min
Peak
2
and �14 ml/kg/min (dots).
822 The Journal of Heart and Lung Transplantation, Vol 31, No 8, August 2012
Calibration of SHFM and peak VO2
The observed survival free of LVAD/urgent transplant wasslightly lower than predicted at 1 year (83% vs 86%), butwas the same at 4 years (63% vs 63%; Figure 6). Although�-blocker and angiotensin-receptor blocker (ARB) use washigher in patients evaluated in the 2000s vs 1990s (75% vs49% and 18% vs 12%, both p � 0.005), the era of patientevaluation was not predictive of outcomes when adjustedfor the SHFM (1990s vs 2000s, p � 0.32), likely becausethe SHFM includes current heart failure medications anddevices.
The calibration of the SHFM was evaluated by compar-ing the predicted vs the observed 1- and 4-year event ratesfor SHFM score values of 0 to 3. Most observed values werewithin �5% of the estimated rate. Table 4 shows the ob-served 1-year Kaplan–Meier survival free of LVAD/urgenttransplant by peak VO2 groups and SHFM score. Patientswith an SHFM score of 0 were at relatively low risk even ifthe peak VO2 was �10 ml/kg/min (Table 4). Patients withan SHFM score of 2 or 3 are high risk, even if peak VO2 isrelatively preserved (i.e., �14 ml/kg/min). The addition ofpeak VO2 would most effectively aid in risk stratificationfor LVAD/transplant in patients with an intermediate riskwith a SHFM score of 1 (�90% estimated survival). Within
Table 4 Observed Kaplan–Meier 1-Year Survival Free of LVAD/Score and Peak VO2 (ml/kg/min)
Overall
SHF95(n
Peak VO2 �14 ml/kg/min 88 � 1% 96(n � 619) (n
Peak VO2 �10–14 ml/kg/min 80 � 2% 95(n � 425) (n
Peak VO2 �10 ml/kg/min 72 � 3% 94(n � 196) (n
Survival data expressed as mean � SE.
0%
10%
20%
30%
40%
50%
60%
70%
80%
90%
100%
0% 10% 20% 30% 40% 50% 60% 70% 80% 90% 100%
SH
FM
Pre
dict
ed S
urvi
val
Observed Survival free of LVAD/Urgent Transplant
1 Year
4 Years
Predicted Observed1 Yr 86% 83%4 Yr 63% 63%
Figure 6 The predicted vs observed survival free of LVAD/urgent transplant for SHFM scores of 0 to 3 at 1 and 4 years. Theactual event rate at 1 year was �3% higher than that predicted by
the model, although there was no difference at 4 years.
this group of patients, patients with peak VO2 �10 ml/kg/min, 10 to 14 ml/kg/min and �14 ml/kg/min had an ob-served 1-year survival free of LVAD/urgent transplant of75%, 80% and 90%, respectively.
As all patients were ambulatory outpatients, they wouldhave been listed as UNOS Status 2, although many even-tually progressed to UNOS Status 1 at the time of transplant.Using the SHFM, the 1-year SHFM-estimated survival ofpatients who ultimately proceeded to death/LVAD/urgenttransplant was 80 � 18% vs 92 � 8% (p � 0.0001) forthose alive at the end of follow-up. The average time to anevent was 2.7 years. The 1-year SHFM survival estimateswere similar for patients who died (82 � 17%) or receivedan LVAD (80 � 20%) or an urgent transplant (77 � 19%).The patients who received an UNOS Status 2 transplantwere lower risk at baseline (87 � 9%), but higher risk thanthose who remained alive at the end of follow-up (92 � 8%,p � 0.0001 for both comparisons). The type of event overa 4-year period varied accorording to the baseline SHFMscore. For the low-risk group (SHFM score 0), the majorityof events were death, with equal numbers of UNOS Status1 and 2 transplants. In the highest risk group (SHFM score3), death and urgent transplant were predominant, followedby LVADs and rare elective transplants. Intermediate scoreshad roughly equal numbers of urgent transplants, LVADsand deaths (Figure 7).
Discussion
In this analysis of a transplant referral population, both peakVO2 and the SHFM were strong and complementary pre-dictors of survival free from LVAD/urgent transplant, asconfirmed by reclassification using the NRI and the IDI.However, as demonstrated previously in a comparison ofpeak VO2 with another predictive model (i.e., the HeartFailure Survival Score [HFSS]), the use of a predictivemodel such as the SHFM provides better discrimination ofhigh-risk candidates than a single variable. One recent anal-ysis failed to show that peak VO2 added to a model with theindividual components of the SHFM and HFSS, althoughthe HFSS score did add prognostic value to the SHFM.14
Transplant and Number of Patients for Categories of SHFM
Utilization of the SHFM may facilitate earlier identification
823Levy et al. SHFM and PVO2
of patients at increased risk and appropriate referral foradvanced heart failure therapy including LVADs and trans-plant. It is also possible that a model like the SHFM couldmore effectively allocate transplants in an ambulatoryUNOS Status 2 population by allowing high-risk ambula-tory patients to have a higher priority for an available organ.
The SHFM was derived and validated in populations inwhich 98% of the events were death. Thus, in reality, theSHFM estimate is the risk of death rather than the combineddeath/LVAD/urgent transplant rate. As the number of pa-tients who receive an LVAD or urgent transplant increases,the combined event rate will increase in a given population.As described in other samples, the combined LVAD/urgenttransplant event rate in transplant-eligible patients is �15%to 20% higher than the death rate (i.e., a 10% mortalitywould have a �12% combined event rate).9,10
The decision to list ambulatory patients for cardiac trans-plantation remains a difficult clinical decision. With ad-vances in medical therapy, survival for ambulatory patientswith heart failure, LVADs, and post transplant all continueto improve. This requires continual reassessment of thecriteria needed to advance to the next stage of therapy. Thecurrent 1- and 5-year survival of UNOS Status 2 transplantrecipients is 92% and 76%, respectively.15
Most patients listed as UNOS Status 2 candidates prog-ress to a UNOS Status 1/LVAD listing prior to transplant,given the scarcity of donor organs. This is reflected in thefinding that only 29% of patients were listed as UNOSStatus 2 at the time of transplant. Patients who underwentUNOS Status 2 transplant had a 1-year SHFM survivalestimate of 87% and 5-year survival of 54%. Because allUNOS Status 2 patients are treated equally, being priori-tized only by the waiting time, it is unlikely that high-riskambulatory outpatients (SHFM score 2 or 3) will survive toreceive an elective transplant without progression to UNOSStatus 1 or LVAD (Figure 7).
As LVADs become more common in ambulatory non–inotrope-dependent Stage D heart failure patients, it be-comes very important to use risk stratification models like
0%
5%
10%
15%
20%
25%
30%
35%
40%
0%
5%
10%
15%
20%
25%
30%
35%
40%A
nnua
l Eve
nt R
ate
over
4 y
ears
Death
LVAD
Urgent Transplant
Elective Transplant
SHFM 0 SHFM 1 SHFM 2 SHFM 3
Each EventLogrank p<0.0001
Figure 7 Annual event rate over 4 years for a baseline SHFMscore rounded to the nearest integer of 0 to 3.
the SHFM to assist in appropriate patient selection. The
National Heart, Lung, and Blood Institute (NHLBI)-spon-sored REVIVE-IT study for early vs usual timing forLVADs in ambulatory NYHA Status 3 or 4 patients willrequire a SHFM-estimated 1-year mortality �16.5% fortrial entry.16 In patients with a peak VO2 of �10 ml/kg/min,the SHFM identified a wide range of observed survival freeof LVAD and urgent transplant with an observed survival at1 year varying from 94% to 44% and at 5 years from 70%to 7% for SHFM scores of 0 and 3 (Figure 4e). The 5-yearobserved survival in patients with a peak VO2 of �10ml/kg/min for the SHFM 0 group (�70%) was similar to the5-year survival for a UNOS Status 2 transplant (�76%),suggesting that, considering mortality alone, aggressiveend-stage therapies like transplant and LVAD may be safelydeferred in patients with a low-risk SFHM score, regardlessof their peak VO2.15 We acknowledge, however, that thisanalysis does not address the important quality of life con-siderations that may appropriately influence treatment deci-sions in patients with severely reduced exercise capacity.
Given that the 1-year survival with LVAD as a bridge toa transplant (�68% to �91%)15,17 exceeds the 1-year sur-vival with medical therapy in the SHFM score 3 strata(�50%), it would seem reasonable to proceed to a LVAD asa bridge to transplantation or destination therapy in thisambulatory outpatient population unless a transplant can beobtained quickly. There are strengths and limitations to thisanalysis. The strengths include the utilization of three dif-ferent centers that routinely measure peak VO2 with long-term follow-up. The SHFM was not used for risk stratifi-cation during the observation period. It is uncertain hownewer therapies that were not as widely utilized in the 1990smight influence these results (�-blockers, aldosteroneblockers, implantable cardioverter-defibrillators [ICDs] andbiventricular pacing). A recent analysis using a variant ofthe SHFM applied prospectively to the Sudden CardiacDeath Heart Failure Trial suggested that high-risk ambula-tory patients with an annual mortality of �25% had nomortality benefit from use of an ICD for primary preven-tion.11 We did not measure exercise ventilatory efficiency,which has been shown to be a powerful predictor.18 It isquite likely that the addition of B-type natriuretic peptide(BNP) level would have modestly improved prediction.1,12
It is very possible that patients with a preserved peak VO2
(�14 ml/kg/min) and SHFM estimated 1-year mortality of�25% may be better served with optimization of heartfailure medications and devices, such as an ICD with/with-out a biventricular pacemaker, to prevent sudden death, andto defer transplant until peak VO2 is decreased. We do nothave information on modes of death or when patients werelisted for transplant.
It is also very possible that variables within the SHFMwould have different hazard ratios if peak VO2 were in-cluded with derivation of the model. To accurately estimatethe alteration in hazard ratios, it is likely that the data setwould need to be substantially larger than the current co-horts to ensure that the newly derived � coefficients would
remain similar in a validation cohort.
824 The Journal of Heart and Lung Transplantation, Vol 31, No 8, August 2012
In conclusion, the Seattle Heart Failure Model is awidely validated model that worked very well in risk-strat-ifying patients referred for consideration of transplant in thisvalidation cohort. The SHFM may facilitate appropriate useof guideline-based heart failure therapy and appropriatereferral of patients for advanced heart failure therapy, suchas destination LVADs and cardiac transplantation. The ad-dition of peak VO2 improved the risk stratification, espe-cially for those patients at moderate risk (i.e., �10% annualmortality). The model may also have utility in the allocationof hearts for transplantation, especially with regard to am-bulatory outpatients. Table 3.
Disclosure statementThe authors have no conflicts of interest to disclose. The Univer-sity of Washington Center for Commercialization holds the copy-right to the Seattle Heart Failure Model. W.C.L. has receivedresearch funding for the SHFM from HeartWare, General Electric,CardioMems and Thoratec and licensing revenue from Epocrates.D.M. is a consultant to the Celladon Corporation. T.F.D. receivededucational support from Thoratec. K.D.A. and W.C.L. receivedfunding from the NHLBI for REVIVE-IT.
References
1. Levy WC, Mozaffarian D, Linker DT, et al. The Seattle Heart FailureModel: prediction of survival in heart failure. Circulation 2006;113:1424-33.
2. Mancini DM, Eisen H, Kussmaul W, et al. Value of peak exerciseoxygen consumption for optimal timing of cardiac transplantation inambulatory patients with heart failure. Circulation 1991;83:778-86.
3. Aaronson KD, Schwartz JS, Chen TM, et al. Development and prospec-tive validation of a clinical index to predict survival in ambulatory patientsreferred for cardiac transplant evaluation. Circulation 1997;95:2660-7.
4. Lund LH, Aaronson KD, Mancini DM. Predicting survival in ambu-latory patients with severe heart failure on beta-blocker therapy. Am J
Cardiol 2003;92:1350-4.
5. 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;95:734-41.
6. Peterson LR, Schechtman KB, Ewald GA, et al. The effect of beta-adrenergic blockers on the prognostic value of peak exercise oxygenuptake in patients with heart failure. J Heart Lung Transplant 2003;22:70-7.
7. Zugck C, Haunstetter A, Kruger C, et al. Impact of beta-blockertreatment on the prognostic value of currently used risk predictorsin congestive heart failure. J Am Coll Cardiol 2002;39:1615-22.
8. O’Neill JO, Young JB, Pothier CE, et al. Peak oxygen consumption asa predictor of death in patients with heart failure receiving beta-blockers. Circulation 2005;111:2313-8.
9. Allen LA, Yager JE, Funk MJ, et al. Discordance between patient-predicted and model-predicted life expectancy among ambulatory pa-tients with heart failure. JAMA 2008;299:2533-42.
10. Kalogeropoulos AP, Georgiopoulou VV, Giamouzis G, et al. Utility ofthe Seattle Heart Failure Model in patients with advanced heart failure.J Am Coll Cardiol 2009;53:334-42.
11. Levy WC, Lee KL, Hellkamp AS, et al. Maximizing survival benefitwith primary prevention implantable cardioverter-defibrillator therapyin a heart failure population. Circulation 2009;120:835-42.
12. May HT, Horne BD, Levy WC, et al. Validation of the Seattle HeartFailure Model in a community-based heart failure population andenhancement by adding B-type natriuretic peptide. Am J Cardiol2007;100:697-700.
13. Pencina MJ, D’Agostino RB, D’Agostino RB Jr, et al. Evaluating theadded predictive ability of a new marker: from area under the ROCcurve to reclassification and beyond. Stat Med 2008;27:157-72.
14. Goda A, Lund LH, Mancini D. The Heart Failure Survival Scoreoutperforms the peak oxygen consumption for heart transplantationselection in the era of device therapy. J Heart Lung Transplant 2011;30:315-25.
15. Lietz K, Miller LW. Improved survival of patients with end-stage heartfailure listed for heart transplantation: analysis of organ procurementand transplantation network/U.S. United Network of Organ Sharingdata, 1990 to 2005. J Am Coll Cardiol 2007;50:1282-90.
16. Baldwin JT, Mann DL. NHLBI’s program for VAD therapy for mod-erately advanced heart failure: the REVIVE-IT pilot trial. J Card Fail2010;16:855-8.
17. Slaughter MS, Rogers JG, Milano CA, et al. Advanced heart failuretreated with continuous-flow left ventricular assist device. N EnglJ Med 2009;361:2241-51.
18. Arena R, Myers J, Guazzi M. The clinical and research applications ofaerobic capacity and ventilatory efficiency in heart failure: an evi-
