predictors of clinical outcomes in elderly patients with heart failure

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

Predictors of clinical outcomes in elderly patientswith heart failureLuis Manzano1,2,3*, Daphne Babalis3,4, Michael Roughton3, Marcelo Shibata5,6,Stefan D. Anker7,8, Stefano Ghio9, Dirk J. van Veldhuisen 10, Alain Cohen-Solal11,Andrew J. Coats12, Philip P.A. Poole-Wilson4†, and Marcus D. Flather3,4, on behalf ofthe SENIORS Investigators1University of Alcala, Madrid, Spain; 2Internal Medicine Department, Hospital Universitario Ramon y Cajal, Madrid, Spain; 3Clinical Trials and Evaluation Unit, Royal Brompton andHarefield NHS Trust, London, UK; 4National Heart and Lung Institute, Imperial College London, London, UK; 5Division of Cardiology, University of Alberta, Alberta, Canada;6Covenant Health Misericordia Hospital, Alberta, Canada; 7Applied Cachexia Research, Department of Cardiology, Charite Campus Virchow-Klinikum, Berlin, Germany; 8Centre forClinical and Basic Research, IRCCS San Raffaele, Rome, Italy; 9Fondazione IRCCS Policlinico S. Matteo, piazza Golgi 1, 27100 Pavia, Italy; 10University Medical Center Groningen,University of Groningen, Groningen, The Netherlands; 11INSERM U942, University Paris 7 Denis Diderot, Hospital Lariboisiere, Paris, France; and 12Faculty of Medicine, University ofSydney, Sydney, Australia

Received 11 September 2010; revised 8 January 2011; accepted 22 January 2011; online publish-ahead-of-print 30 March 2011

See page 467 for the editorial comment on this article (doi:10.1093/eurjhf/hfr036)

Aims Heart failure (HF) in the elderly carries a poor prognosis. We used the SENIORS dataset of elderly HF patients aged≥70 years in order to develop a risk model for this population.

Methodsand results

The SENIORS trial evaluated the effects of nebivolol and enrolled 2128 patients ≥70 years with HF (ejection fraction≤35%, or recent HF admission). We randomly selected 1400 patients from the full dataset to produce a derivationcohort and the remaining 728 patients were used as a validation cohort. Baseline variables were entered into a boot-strap model with 200 iterations to determine their association with two outcomes, the composite of all-cause mor-tality or cardiovascular hospitalization, or all-cause mortality alone. Variables retaining a significant association withthese outcomes in a multivariate model were used to develop a risk prediction score tested in the validationcohort. Five factors were associated with increased risk of both outcomes in the multivariate model: higherNew York Heart Association class, higher uric acid level, lower body mass index, prior myocardial infarction, andlarger left atrial (LA) dimension. For the composite outcome, peripheral arterial disease, years with heart failure,right bundle branch block, diabetes mellitus, and orthopnoea were also retained. For all-cause mortality, creatinine,6 min walk test distance, coronary artery disease, and age were retained.

Conclusion In addition to conventional prognostic markers, uric acid and LA dimension appear to be important novel risk pre-diction markers in elderly patients with heart failure, and could be useful in guiding management.

- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -Keywords Heart failure † Risk model † Elderly † Prognosis

IntroductionHeart failure (HF) is a major public health problem affecting about 2%of populations in developed economies.1 Its prevalence and inci-dence is strongly associated with age and most patients with HFare 70 years or older. In patients .80 years of age the prevalenceof HF may be as high as 12%.1,2 Heart failure is associated withhigh rates of mortality and morbidity. Annual mortality rates in

patients with moderate stable HF are about 10–15% per year withhigher rates of hospital admissions for worsening HF.1 A major chal-lenge in the management of HF is the availability of reliable prognos-tic models that enable patients and physicians to have a realisticexpectation of prognosis, and to guide treatment options includingmedical treatment, use of devices, more intense monitoring andend-of-life care. In addition, insights into which factors relate topoor outcome may help generate hypotheses for new treatments.

* Corresponding author. Tel: +34 913368921, Fax: +34 913368922, Email: [email protected]† In memoriam.

Published on behalf of the European Society of Cardiology. All rights reserved. & The Author 2011. For permissions please email: [email protected].

European Journal of Heart Failure (2011) 13, 528–536doi:10.1093/eurjhf/hfr030

at Aston U

niversity on January 21, 2014http://eurjhf.oxfordjournals.org/

Dow

nloaded from

http://eurjhf.oxfordjournals.org/


Several risk models for HF have been developed using data from clini-cal trials or observational studies but these have mostly includedpatients younger than 70 years, and have focussed on patients withimpaired systolic function.3 –7 Information about prognosis inelderly HF outpatients, including those with preserved ejection frac-tion (EF), is lacking.8–13 In view of this knowledge gap, we usedthe SENIORS dataset of elderly HF patients aged ≥70 years,with a wide range of EFs, to explore the influence of baselinecharacteristics on key outcomes in order to develop a riskmodel for this more representative HF population.

MethodsThe SENIORS trial which is described in detail elsewhere14–16 ran-domized elderly, stable, HF patients to the b-1 selective betablocker nebivolol or placebo and showed a favourable effect on thecomposite outcome of all-cause mortality or cardiovascular (CV) hos-pital admission. To be eligible, patients had to be 70 years or older,provide written informed consent, and have a clinical history ofchronic HF with at least one of the following features: documentedhospital admission within the previous 12 months with a dischargediagnosis of congestive HF or documented left ventricular EF ≤35%within the previous 6 months. Echocardiographic variables includingEF and left atrial (LA) antero-posterior dimension were measured by2D guided M-mode echocardiography. The main exclusion criteriawere: HF primarily due to uncorrected valvular heart disease, contra-indication or previous intolerance to beta-blockers, advanced hepaticor renal dysfunction, cerebrovascular accident within the previous 3months, and other major medical conditions that may have reducedsurvival during the period of the study. Patients were enrolledbetween December 2000 and December 2002, and were followedfor a mean of 21 months. The primary outcome was the compositeof all-cause mortality or CV hospital admission (time to first event)and the main secondary outcome was all-cause mortality. Ethicalapproval was obtained from the relevant Committees or InstitutionalReview Boards and all patients provided written informed consentto be included in the SENIORS trial.

Declaration of HelsinkiThe authors confirm that the study complies with the Declaration ofHelsinki (2008), the locally appointed ethics committees haveapproved the research protocol and that informed consent has beenobtained from the participants.

Statistical methodsAn extensive list of baseline variables that might be related to patient prog-nosis was identified from the baseline case record forms used for allpatients in SENIORS. Medications at baseline or after randomizationwere not included as it is almost impossible to separate prognostic associ-ations from selection biases. Missing data were imputed using the meanvalue for all continuous variables, except for years with HF for whichvalues of zero were used. All missing categorical variables were imputedwith the value corresponding to the lowest risk category. All the variablesanalysed had ,9% missing values. The two outcomes studied here weretime to the composite of first CV hospitalization or all cause mortality, andtime to all-cause mortality alone. A derivation cohort of 1400 patients wasrandomly selected from the 2128 patients in SENIORS, and the other 728patients were used as a validation cohort to test the model. A 2:1 ratio forthe size of the training set compared with the validation set was chosen to

allow the training data to have sufficient power to detect prognostic vari-ables, while still having a suitably large dataset in which to validate themodels. From those 1400, a bootstrap resample procedure was run tocreate 200 datasets, each with N ¼ 1400. Each dataset was made up ofa random selection from the 1400 patients, with replacement (bootstrap-ping). Two hundred iterations were used for pragmatic reasons, as theprocedure is computationally intensive and can take many hours toreturn results. A forward selection process was used to select variablesto go into the model for each bootstrap dataset. All 30 variables wereavailable to be selected at the start of every iteration. Any variablewhich improved the model by an amount corresponding to a P-value of,0.05 was entered. The selection process was terminated for each iter-ation once no more variables could be added to improve the model.The model building procedure was performed for each bootstrapsample, with the final coefficients of each variable recorded for the 200separate models. Any variable that was selected at least 100 times outof the 200 separate models was included in the final model. For each ofthese variables, the mean value of all observed coefficients was taken tobe the final model coefficient (and hazard ratio). A Cox model was runon the original 1400 patients using the selected variables to obtain thebaseline survival function at 2 years, S0(2). The risk score for eachpatient was calculated by multiplying the observed value for the selectedvariables by its corresponding model coefficient. The estimated prob-ability of observing an event at 2 years was then calculated by using theformula P ¼ 1 2 S0(2)×exp(risk score).

The discriminatory properties of the model were tested by estimat-ing the Kaplan–Meier survival curves for tertiles of risk score, for boththe derivation and validation datasets. The log-rank test was then usedto see if the three curves were different from one another. This was totest whether the risk model was able to differentiate between low,medium, and high risk groups of patients. Kaplan–Meier curves werealso used to estimate the observed 2 year event rates for the 2 out-comes, split by tertiles of risk score. These estimates were then com-pared with the average predicted survival probabilities in each quintileto assess the model’s predictive accuracy.

When estimating the risk score, continuous variables were centredaround an approximate mean value (zero for years with HF). Thisvalue is given in square brackets in the model result tables. Forexample, age was centred around 75 years, meaning that the hazardratio for age corresponds to a 1 year increase or decrease in agefrom 75 years. Additionally, some continuous variables were scaledso that the hazard ratio corresponded to a meaningful clinicalchange in the measurement being taken. All analysis was performedusing Stata 10.1 (StataCorp, TX, USA), and a P-value of 0.05 was con-sidered statistically significant. The Supplementary material online,Appendix gives an example of using the model to predict the risk ofdeath at 2 years. We externally validated the risk model using the vali-dation cohort of 728 patients. A number of statistical approaches usedare based on methods described previously.6 In order to providefurther information about the prognostic discrimination of the riskmodel, we calculated the C statistic (range between 0 and 1).

We also tested the risk model developed from the CHARM dataset6

in the SENIORS cohort and compared performance with the newmodel proposed in this manuscript. The version of the CHARMmodel we used was slightly different to the one in the CHARMpaper as we did not have data for every variable in their model.

ResultsThe SENIORS study included 2128 patients. During a meanfollow-up period of 21 months, 707 died or had a CV

Predictors of clinical outcomes in elderly patients with heart failure 529

at Aston U


Dow

nloaded from

http://eurjhf.oxfordjournals.org/cgi/content/full/hfr030/DC1




. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

Table 1 Demographics and clinical characteristics of validation and derivation cohorts

Characteristic Derivation set Validation set P-value

Number 1400 728

On Nebivolol 681 (48.6%) 380 (52.2%) 0.12

Demographics

Age (year+ SD) 76.1 (+4.7) 76.1 (+4.6) 0.736

Women, n (%) 519 (37.1%) 266 (36.5%) 0.809

Clinical

NYHA Class I 39 (2.8%) 22 (3.0%) 0.934

NYHA Class II 791 (56.5%) 409 (56.2%)

NYHA Class III 540 (38.6%) 284 (39.0%)

NYHA Class IV 30 (2.1%) 13 (1.8%)

Ejection fraction (%) 35.7 (+12.0) 36.7 (+12.8) 0.072

Years with HF 4.0 (+4.8) 3.8 (+4.7) 0.489

Body mass index (Kg/m2) 26.7 (+4.0) 26.9 (+4.1) 0.234

6 min walk test distance (m) 279.4 (111.8) 274.8 (113.2) 0.366

Left atrial dimension (cm) 4.4 (+0.8) 4.4 (+0.8) 0.855

Haemodynamics

Heart rate (beats/min) 79.4 (+13.9) 78.4 (+13.2) 0.119

Sitting systolic blood pressure (mmHg) 139.5 (+20.3) 138.0 (+21.1) 0.107

Sitting diastolic blood pressure (mmHg) 80.7 (+11.0) 80.2 (+11.2) 0.326

Laboratory

MDRD (mL/min/1.73 m2) 65.0 (+19.6) 64.9 (+21.7) 0.895

Creatinine (mmol/L) 102.1 (+33.6) 104.1 (+37.4) 0.223

Haemoglobin (g/dL) 13.8 (+1.5) 13.8 (+1.5) 0.839

Sodium (mmol/L) 141.7 (+3.8) 141.6 (+3.7) 0.72

Uric acid (mmol/L) 395.7 (+120.2) 404.2 (+137.2) 0.141

Total cholesterol (mmol/L) 5.3 (+1.2) 5.4 (+1.2) 0.359

Medical history

Smoker 76 (5.4%) 33 (4.5%) 0.378

Prior history of coronary artery disease 957 (68.4%) 495 (68.0%) 0.865

Prior myocardial infarction 614 (43.9%) 316 (43.4%) 0.842

Chronic obstructive pulmonary disease 99 (7.1%) 48 (6.6%) 0.680

Hypertension 878 (62.7%) 434 (59.6%) 0.163

Hyperlipidaemia 640 (45.7%) 334 (45.9%) 0.942

Idiopathic dilated cardiomyopathy 226 (16.1%) 107 (14.7%) 0.384

Peripheral arterial disease 71 (5.1%) 31 (4.3%) 0.405

Cancer 50 (3.6%) 30 (4.1%) 0.527

Orthopnoea 197 (14.1%) 104 (14.3%) 0.893

Left bundle branch block 277 (19.8%) 144 (19.8%) 0.998

Right bundle branch block 130 (9.3%) 48 (6.6%) 0.033

Left ventricular hypertrophy 169 (12.1%) 80 (11.0%) 0.461

Atrial fibrillation 488 (34.9%) 250 (34.3%) 0.812

Diabetes 376 (26.9%) 179 (24.6%) 0.258

Medications

Diuretics 1205 (86.1%) 629 (86.4%) 0.834

ACE-inhibitors 1168 (83.4%) 588 (80.8%) 0.125

Angiotension receptor blockers 111 (7.9%) 67 (9.2%) 0.314

Aldosterone antagonists 386 (27.6%) 194 (26.7%) 0.65

Cardiac glycosides 580 (41.4%) 315 (43.3%) 0.414

Anti-arrythmics 240 (16.5%) 120 (16.5%) 0.7

Lipid lowering drugs 315 (22.5%) 148 (20.3%) 0.25

Vitamin K antagonists 328 (23.4%) 167 (22.9%) 0.8

Continued

L. Manzano et al.530

at Aston U


Dow

nloaded from



hospitalization (time to first event; 175 and 532, respectively, tomake up the primary endpoint). A total of 361 patients died (sec-ondary endpoint). Table 1 compares the baseline variables thatwere considered as candidates for inclusion in the prognosticmodels for the derivation and validation datasets. Although conco-mitant medications were not considered for the prognosticmodels, they are included in the table in order to compare thetwo datasets. Only one variable, right bundle branch block(RBBB), showed a statistically significant difference between thetwo groups. There was also no significant difference betweenthe groups for either of the two outcomes that were used inthe modelling process. The overall rates of mortality were 15.5%in the derivation and 17.7% in the validation dataset, confirmingthat these cohorts are comparable.

Predicting all-cause mortality orcardiovascular hospitalizationTable 2 gives univariate Cox model results of all-cause mortality orCV hospital admission for the 1400 patients in the derivationdataset. During the 200 iterations of the model building process,a total of 2557 variables were chosen, with New York HeartAssociation (NYHA) selected in 198 of the 200 models. The vari-ables selected for the final model (in order of the statistical associ-ation) were NYHA, prior myocardial infarction (MI), LA dimension,peripheral arterial disease (PAD), uric acid, body mass index (BMI),number of years with HF, RBBB, diabetes mellitus, and orthopnoea.These 10 variables accounted for 1585 of the total numberselected during the 200 model building procedures. The hazardratios, bootstrap confidence intervals and model coefficients forthe final model are given in Table 2.

Predicting all-cause mortalityTable 3 gives the univariate Cox model results of all-cause mor-tality for the 1400 patients in the derivation dataset. During the200 iterations of the model building process, a total of 2286 vari-ables were chosen, with prior MI selected in 186 of the 200models. The variables selected for the final model were priorMI, uric acid, BMI, LA dimension, NYHA, serum creatinine, 6 minwalk test distance, prior history of coronary artery disease(CAD) and age. These 9 variables accounted for 1344 of the

total number selected during the 200 model building procedures.The hazard ratios, bootstrap confidence intervals and model coef-ficients for the final model are given in Table 3. Note that lowerBMI was associated with a higher risk, as was shorter walkingdistance.

Predicting an individual’s riskThe multivariate models in Tables 2 and 3 can be used to predictthe risk of each outcome for an individual. The risk score is calcu-lated as a linear combination of the observed variable values mul-tiplied by the corresponding coefficients. The risk score is thenused to adjust the baseline survival estimate for each outcome,in order to estimate the probability of observing an event.Examples of how to use the risk score for risk prediction in indi-vidual patients are presented in the Supplementary materialonline, Appendix. Both scores were normally distributed.Figure 1A shows both the distribution of the risk score for the com-posite of all-cause mortality or CV hospitalization (vertical bars)and the relationship between risk score and estimated probabilityof a primary event within 2 years of follow-up by the solid line.Figure 1B shows the corresponding distribution and probabilitycurve for the all-cause mortality risk score. For both outcomes,the risk score was validated externally using the validationcohort of 728 patients. The distribution of risk was found to becomparable in the validation dataset for both outcomes (meanrisk score 1.23 vs. 1.22 of primary outcome for derivation andvalidation cohorts, respectively). Figure 2 further illustrates the pre-dictive power of each of the two risk scores, by showing Kaplan–Meier plots for all-cause mortality or CV hospitalization (Figure 2A)and for all-cause mortality alone (Figure 2B) for patients classifiedinto tertiles of risk score. For both outcomes there is clear separ-ation of the curves by 12 months. Similar figures were obtained forthe validation cohort and found to be comparable (data notshown). The discriminatory properties of the risk model werealso confirmed by the C statistic. For the primary endpoint, theC statistic had a value of 0.68 (derivation dataset) and 0.66 (vali-dation dataset) and for all-cause mortality 0.72 (derivation data)and 0.69 (validation data).

Figure 3 demonstrates the predictive accuracy for the compositeof death or CV hospital admission, comparing the observed andpredicted probabilities in the derivation and validation cohorts

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

Table 1 Continued

Characteristic Derivation set Validation set P-value

Aspirin 727 (51.9%) 378 (51.9%) 0.998

Calcium channel antagonists 188 (13.4%) 93 (12.8%) 0.673

Survival comparison

Primary outcome event rate 466 (33.3%) 241 (33.1%) 0.933

Primary outcome hazard ratio 0,99 95% CI (0.85, 1.17) 1 (reference) 0.974

All-cause mortality event rate 248 (17.7%) 113 (15.5%) 0.201

All-cause mortality hazard ratio 1.14 95% CI (0.91, 1.43) 1 (reference) 0.242

Data are number of patients (%) or mean (standard deviation); NYHA, New York Heart Association; MDRD, Modification of Diet in Renal Disease; ACE, angiotensin-convertingenzyme.


at Aston U


Dow

nloaded from





. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

Table 2 Univariate and multivariate predictors of all-cause mortality or cardiovascular hospital admission in thederivation cohort

Univariate predictors

Variable HR 95% CI P-value Number of times selected

Age (year) [75] 1.039 1.021 1.059 ,0.001 73

Female 0.757 0.623 0.920 0.005 26

Years with heart failure [0] 1.028 1.011 1.045 0.001 128

Hypertension 0.917 0.761 1.105 0.363 8

Diabetes mellitus 1.375 1.131 1.672 0.001 120

Coronary artery disease 0.915 0.754 1.110 0.367 23

Prior myocardial infarction 1.647 1.373 1.976 ,0.001 195

Idiopathic dilated cardiomyopathy 0.950 0.739 1.221 0.688 16

Atrial fibrillation 1.334 1.109 1.605 0.002 85

Peripheral arterial disease 1.979 1.414 2.769 ,0.001 187

Chronic obstructive pulmonary disease 1.143 0.813 1.606 0.442 7

Cancer 0.866 0.508 1.474 0.595 45

NYHA 1.703 1.463 1.983 ,0.001 198

Body mass index (Kg/m2) [25] 0.960 0.937 0.983 0.001 165

Systolic BP (per 10mmHg) [140] 0.927 0.885 0.971 0.001 14

Diastolic BP (per 10 mmHg) [80] 0.858 0.788 0.933 ,0.001 51

Heart rate (per 10 bpm) [80] 1.051 0.986 1.119 0.126 46

Orthopnoea 1.732 1.374 2.182 ,0.001 111

6MWT distance (per 10 m) [300] 0.985 0.977 0.993 0.001 91

Haemoglobin (g/dL) [13] 0.928 0.873 0.987 0.017 94

MDRD (per 10 units) [65] 0.892 0.849 0.938 ,0.001 16

Creatinine (per 10 mmol/L) [100] 1.075 1.049 1.101 ,0.001 76

Sodium (per 10 mmol/L) [140] 0.664 0.518 0.852 0.001 87

Cholesterol (mmol/L) [5] 0.911 0.840 0.989 0.026 12

Uric acid (mmol/L) (per 10 units) [400] 1.025 1.017 1.032 ,0.001 172

Left bundle branch block 1.366 1.104 1.691 0.004 91

Right bundle branch block 1.583 1.209 2.072 0.001 121

Left ventricular hypertrophy 1.034 0.788 1.356 0.809 21

Left ventricular ejection fraction (%) [35] 0.991 0.983 0.999 0.034 32

Left atrial dimension (cm) [4] 1.369 1.231 1.523 ,0.001 188

On nebivolol 0.902 0.752 1.082 0.267 58

Multivariate predictors

Variable HR 95% CI Coefficient

NYHA 1.451 1.220 1.766 0.373

Prior myocardial infarction 1.545 1.264 1.905 0.435

Left atrial dimension (cm) [4] 1.277 1.142 1.453 0.244

Peripheral arterial disease 2.037 1.524 2.811 0.711

Uric acid (mol/L) (per 10 units) 1.017 1.010 1.025 0.017

Body mass index (Kg/m2) [25] 0.957 0.933 0.974 20.044

Years with heart failure [0] 1.026 1.017 1.045 0.026

Right bundle branch block 1.532 1.334 2.004 0.427

Diabetes mellitus 1.366 1.230 1.639 0.312

Orthopnoea 1.460 1.310 1.780 0.378

HR, hazard ratio; CI, confidence interval; NYHA, New York Heart Association; BP, blood pressure; bpm, beats per minute; 6MWT, six minute walk test.When estimating the risk score, continuous variables were centred around an approximate mean value (zero for years with HF). This value is given in square brackets in the modelresult tables.


at Aston U


Dow

nloaded from



for patients classified into quintiles of risk score. We observed avery strong gradient in risk with patients in the top quintile ofthe risk score, having �4 times the risk of patients in thebottom quintile of risk.

The CHARM risk model was also tested in the SENIORS cohortand found to perform well. For the primary outcome, the riskmodel developed in this study performed better, and when bothrisk models were analysed together the CHARM one was no

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

Table 3 Univariate and multivariate predictors of all-cause mortality in the derivation cohort

Variable HR 95% CI P-value Number of times selected

Univariate predictors

Age (year) [75] 1.057 1.032 1.084 ,0.001 111

Female 0.591 0.446 0.783 ,0.001 83

Years with heart failure [0] 1.027 1.004 1.050 0.022 86

Hypertension 0.811 0.630 1.045 0.106 6

Diabetes mellitus 1.228 0.936 1.611 0.138 53

Coronary artery disease 0.706 0.547 0.913 0.008 115

Prior myocardial infarction 1.800 1.400 2.314 ,0.001 186

Idiopathic dilated cardiomyopathy 1.138 0.819 1.582 0.441 36

Atrial Fibrillation 1.175 0.910 1.516 0.216 17

Peripheral arterial disease 1.385 0.834 2.298 0.208 36

COPD 1.325 0.855 2.053 0.208 23

Cancer 0.972 0.480 1.967 0.937 28

NYHA 1.699 1.379 2.093 ,0.001 156

Body mass index (Kg/m2) [25] 0.932 0.902 0.964 ,0.001 175

Systolic BP (per 10 mmHg) [140] 0.874 0.819 0.933 ,0.001 26

Diastolic BP (per 10 mmHg) [80] 0.792 0.705 0.890 ,0.001 38

Heart rate (per 10 bpm) [80] 1.049 0.964 1.142 0.269 56

Orthopnoea 1.590 1.160 2.179 0.004 43

6MWT distance (per 10m) [300] 0.985 0.973 0.996 0.009 118

Haemoglobin (g/dL) [13] 0.914 0.841 0.993 0.033 27

MDRD (per 10 units) [65] 0.827 0.772 0.887 ,0.001 35

Creatinine (per 10 mmol/L) [100] 1.120 1.087 1.154 ,0.001 140

Sodium (per 10 mmol/L) [140] 0.570 0.409 0.794 0.001 69

Cholesterol (mmol/L) [5] 0.901 0.806 1.008 0.068 20

Uric Acid (mmol/L) (per 10 units) [400] 1.036 1.027 1.046 ,0.001 178

Left-bundle branch block 1.475 1.110 1.961 0.007 51

Right-bundle branch block 1.691 1.183 2.418 0.004 89

Left ventricular hypertrophy 0.800 0.533 1.202 0.284 35

Left ventricular ejection fraction (%) [35] 0.988 0.977 0.999 0.039 38

Left atrial dimension (cm) [4] 1.428 1.239 1.647 ,0.001 165

On nebivolol 0.895 0.698 1.148 0.382 47

Multivariate predictors

Variable HR 95% CI Coefficient

Prior MI 1.743 1.374 2.464 0.556

Uric acid (umol/L) (per 10 units) 1.023 1.012 1.036 0.022

Body mass index (Kg/m2) [25] 0.934 0.899 0.964 20.069

Left atrial dimension (cm) [4] 1.322 1.185 1.542 0.279

NYHA 1.477 1.237 1.869 0.390

Creatinine (per 10 mmol/L) [100] 1.094 1.043 1.215 0.090

6MWT distance (per 10 m) [300] 0.980 0.972 0.987 20.020

Coronary artery disease 0.630 0.484 0.742 20.462

Age (year) [75] 1.041 1.026 1.059 0.040

When estimating the risk score, continuous variables were centred on an approximate mean value (zero for years with HF). This value is given in square brackets in the abovemodel result table.HR, hazard ratio; CI, confidence interval; COPD, chronic obstructive pulmonary disease; NYHA, New York Heart Association; BP, blood pressure; bpm, beats per minute; 6MWT,six minute walk test; MI, myocardial infarction; MDRD, modification of diet in renal disease.


at Aston U


Dow

nloaded from



longer significant. Regarding all-cause mortality our model was stillbetter, but the effect was not as pronounced as for the primaryoutcome.

DiscussionTo our knowledge this is the first study to propose a risk model forelderly HF patients managed in the outpatient setting using infor-mation from routine baseline clinical and laboratory measure-ments. Five factors (NYHA class, prior MI, LA dimension, uricacid, and BMI) were common for the prediction of the compositeof all-cause mortality or CV hospitalization, and all-cause mortalityalone. In addition to these five variables, PAD, years with HF,RBBB, diabetes mellitus, and orthopnoea featured in the compositeoutcome model, while creatinine, 6 min walk test, CAD, and agefeatured in the mortality model. Our model is also the first todemonstrate the prognostic importance of LA dimension anduric acid in elderly HF population.

The profile of the final multivariable analysis of the SENIORSdataset differs from that reported so far in previous risk modelswhich were designed retrospectively or included younger patientsor those admitted with acute HF.3 –13 Several strong predictors

identified in other studies such as anaemia, left ventricular EF,blood pressure, or hyponatraemia were not significant indepen-dent factors in the SENIORS cohort, whereas uric acid and LAdimension were found to be relevant predictors. Our resultsalso show that in elderly HF patients, the risk score modelderived from SENIORS data performs better than that proposedin the CHARM trial, particularly for the combined endpoint of all-cause mortality or CV hospitalization.

Left atrial size is a simple echocardiographic parameter to assessleft ventricular filling pressures over time, independent of theEF.17,18 Evidence for a prognostic role for LA enlargement topredict incident HF is emerging, and recently it has been proposedas a marker for the diagnosis of HF with preserved EF.18 Inaddition, convincing evidence now exists demonstrating that LAenlargement is a strong predictor of CV outcomes both in patientswith HF and other conditions such as left ventricular dysfunction,atrial arrhythmias, acute MI, and valve disease.16,19 It has alsobeen shown to independently predict death in the general popu-lation.20 Recently, in the MUSIC study, LA size was found to bea very important predictor, both for total and CV mortality, in arisk model based on a cohort of ambulatory HF patients withboth preserved and reduced EF.13 Left atrial size is closelyrelated to ventricular filling pressure, rather like B-type natriureticpeptide (BNP), and thus could be applied as a prognostic factorwhich is independent of EF.

Figure 1 Distribution of risk scores and association of risk withprobability of events. Distribution of risk scores for (A) all-causemortality or cardiovascular hospital admission and (B) all-causemortality, and their relation to probabilities of an event occurringwithin 2 years. Histograms represent percentage of patients witha particular risk score, and the solid line represents probability ofan event over the 2-year follow-up period for a particular riskscore.

Figure 2 Kaplan–Meier curves of event rates stratified by ter-tiles of risk. Kaplan–Meier plots for (A) all-cause mortality or car-diovascular hospital admission and (B) all-cause mortality, bytertiles of risk score in the derivation model; tertile 1, lowerrisk; tertile 2, intermediate risk; tertile 3, higher risk.


at Aston U


Dow

nloaded from



The prognostic significance of uric acid has been reported in afew observational studies in HF,21,22 and even in the general popu-lation.23 In addition, uric acid was identified as a predictor in themultiparametric Seattle HF Model.5 It is important to note thatboth in SENIORS and the Seattle HF models, the predictivevalue of uric acid was independent of renal function. Indeedrenal function was not a predictor in the Seattle HF model. Thisfinding raises the question of whether uric acid is a more importantprognostic factor than conventionally estimated renal function, andwhether it should become a routine additional assessment in theprognostic assessment of HF patients. Some recent experimentalevidence suggests that this molecule could mediatepro-inflammatory immune activation.21 Alternatively, uric acidcould be a mere marker of disease. In fact, uric acid levels were sig-nificantly associated with diuretic treatment, which is also depen-dent on severity of HF (data not shown). In a very recent HFtrial, it has been shown that oxypurinol (a xanthine oxidase inhibi-tor) may provide benefit to a subgroup of patients with high serumuric acid levels (.9.5 mg/dL), and the degree of uric acid reductioncorrelated with clinical outcome.24 So, serum uric acid may serveas a valuable biomarker to target xanthine oxidase inhibition in HFwhich remains to be confirmed by further studies.

Other predictors of both endpoints in the SENIORS cohortwere NYHA class, prior MI, and BMI (higher BMI associated withlower risk), which are consistent with other studies. Our studyconfirms the relevance of these prognostic factors and expandsits significance to elderly ambulatory patients. Interestingly, lowerBMI is a fairly consistent predictor of worse outcome andfurther supports the cachexia theory of HF where weight loss isan adverse prognostic marker.25 On the other hand, PAD, yearswith HF, RBBB, diabetes mellitus and orthopnoea appeared toonly be predictors in the model for the composite of all-causemortality or CV hospital admission, whereas creatinine, 6 minwalk test, CAD and age were predictors only in the all-cause mor-tality model. Interestingly age did not appear as a particularlystrong predictor of risk in this elderly population when otherfactors were taken into account. This could be explained by thenarrow and elderly age range of SENIORS patients. It is alsoworth noting that several prognostic factors reported in otherstudies were not statistically significant in our cohort of elderlypatients in the multivariate analysis including anaemia, left ventricu-lar EF, blood pressure and hyponatraemia. There are two possibleexplanations. First, inclusion of LA size and uric acid measurementsmay have resulted in some other factors being excluded from thefinal model and second, prognostic predictors in elderly ambulat-ory patients may differ from those in other populations. In thisrespect, it has previously been reported that anaemia does notpredict mortality in elderly hospitalized patients with HF which isconsistent with our findings.26 Another important differencebetween the SENIORS cohort and other cohorts is that about athird of patients had EF.40%, which may also have affected theinclusion of EF in the final models, in addition to the fact thatthe prognostic predictors in HF with preserved and withreduced systolic function may be quite different.

Our study has some limitations. Subjects were participating in aclinical trial and it is unknown how representative they are ofpatients seen in routine clinical practice even though the demo-graphics of SENIORS patients do reflect those of community-based HF studies better than most other trials. Patients withco-morbidities such as renal or hepatic failure were excludedfrom the trial. The study did not include routine measurement ofBNP, that has emerged as an important prognostic biomarker inelderly patients, and could affect the variables in our models.27

When the SENIORS study was being planned, BNP was an emer-ging risk factor and measurements were costly. We did not evalu-ate the influence of treatment on outcome mainly because there isno clear agreement on modelling post-entry variables. Finally, wevalidated our model using patients from the same datasetwhereas external validation from an independent dataset wouldhelp to confirm the external validity of our model.

In conclusion, we have developed and validated two novel riskmodels for elderly patients with HF, based on routinely availableclinical and laboratory variables. The risk models include the non-traditional variables of uric acid and LA dimension and provide areliable estimate of death or hospital admission rates over a2-year follow-up period. Use of these models should help toidentify higher risk patients who may benefit from more intensivetreatment and multidisciplinary follow-up and possibly supportdecisions about palliative care.

Figure 3 Observed vs. predicted risk at 2 years. Observed(dark grey bar) and predicted (pale grey bar) rates using therisk prediction model of all-cause mortality or cardiovascular hos-pital admission after 2 years for patients from (A) derivation and(B) validation cohorts.


at Aston U


Dow

nloaded from



Supplementary materialSupplementary material is available at European Journal of HeartFailure online.

FundingThe SENIORS study was funded by a grant from Menarini RicercheSpA. The Clinical Trials and Evaluation Unit received a grant fromMenarini Ricerche to support statistical analyses and preparation ofsecondary manuscripts, however all analyses and their interpret-ation were carried independent of any funding source.

Conflict of interest: S.D.A., D.Jv.V., A.C.S., A.J.C., and M.F. havereceived honoraria and consultancy from Menarini but L.M., M.R.,D.B., M.S., S.G., and P.P.W. have no financial disclosures to declare.

References1. Dickstein K, Cohen-Solal A, Filippatos G, McMurray JJ, Ponikowski P,

Poole-Wilson PA, Stromberg A, van Veldhuisen DJ, Atar D, Hoes AW,Keren A, Mebazaa A, Nieminen M, Priori SG, Swedberg K, Vahanian A,Camm J, De Caterina R, Dean V, Dickstein K, Filippatos G, Funck-Brentano C,Hellemans I, Kristensen SD, McGregor K, Sechtem U, Silber S, Tendera M,Widimsky P, Zamorano JL, Tendera M, Auricchio A, Bax J, Bohm M, Corra U,della Bella P, Elliott PM, Follath F, Gheorghiade M, Hasin Y, Hernborg A,Jaarsma T, Komajda M, Kornowski R, Piepoli M, Prendergast B, Tavazzi L,Vachiery JL, Verheugt FW, Zamorano JL, Zannad F. ESC guidelines for the diag-nosis and treatment of acute and chronic heart failure 2008: the task force forthe diagnosis and treatment of acute and chronic heart failure 2008 of the Euro-pean Society of Cardiology. Developed in collaboration with the Heart FailureAssociation of the ESC (HFA) and endorsed by the European Society of IntensiveCare Medicine (ESICM). Eur J Heart Fail 2008;10:933–989.

2. Curtis LH, Whellan DJ, Hammill BG, Hernandez AF, Anstrom KJ, Shea AM,Schulman KA. Incidence and prevalence of heart failure in elderly persons,1994–2003. Arch Intern Med 2008;168:418–424.

3. Aaronson KD, Schwartz JS, Chen TM, Wong KL, Goin JE, Mancini DM. Develop-ment and prospective validation of a clinical index to predict survival in ambulat-ory patients referred for cardiac transplant evaluation. Circulation 1997;95:2660–2667.

4. Brophy JM, Dagenais GR, McSherry F, Williford W, Yusuf S. A multivariate modelfor predicting mortality in patients with heart failure and systolic dysfunction. Am JMed 2004;116:300–304.

5. Levy WC, Mozaffarian D, Linker DT, Sutradhar SC, Anker SD, Cropp AB,Anand I, Maggioni A, Burton P, Sullivan MD, Pitt B, Poole-Wilson PA,Mann DL, Packer M. The Seattle Heart Failure Model: prediction of survival inheart failure. Circulation 2006;113:1424–1433.

6. Pocock SJ, Wang D, Pfeffer MA, Yusuf S, McMurray JJ, Swedberg KB, Ostergren J,Michelson EL, Pieper KS, Granger CB. Predictors of mortality and morbidity inpatients with chronic heart failure. Eur Heart J 2006;27:65–75.

7. Scardovi AB, De Maria R, Coletta C, Aspromonte N, Perna S, Cacciatore G,Parolini M, Ricci R, Ceci V. Multiparametric risk stratification in patients withmild to moderate chronic heart failure. J Card Fail 2007;13:445–451.

8. Kerzner R, Gage BF, Freedland KE, Rich MW. Predictors of mortality in youngerand older patients with heart failure and preserved or reduced left ventricularejection fraction. Am Heart J 2003;146:286–290.

9. Lee DS, Austin PC, Rouleau JL, Liu PP, Naimark D, Tu JV. Predicting mortalityamong patients hospitalized for heart failure: derivation and validation of a clinicalmodel. JAMA 2003;290:2581–2587.

10. Huynh BC, Rovner A, Rich MW. Long-term survival in elderly patients hospital-ized for heart failure: 14-year follow-up from a prospective randomized trial.Arch Intern Med 2006;166:1892–1898.

11. Jong P, Vowinckel E, Liu PP, Gong Y, Tu JV. Prognosis and determinants of survivalin patients newly hospitalized for heart failure: a population-based study. ArchIntern Med 2002;162:1689–1694.

12. Jones RC, Francis GS, Lauer MS. Predictors of mortality in patients with heartfailure and preserved systolic function in the Digitalis Investigation Group trial.J Am Coll Cardiol 2004;44:1025–1029.

13. Vazquez R, Bayes-Genis A, Cygankiewicz I, Pascual-Figal D,Grigorian-Shamagian L, Pavon R, Gonzalez-Juanatey JR, Cubero JM, Pastor L,Ordonez-Llanos J, Cinca J, de Luna AB, on behalf of the MUSIC Investigators.The MUSIC Risk score: a simple method for predicting mortality in ambulatorypatients with chronic heart failure. Eur Heart J 2009;30:1088–1096.

14. Flather MD, Shibata MC, Coats AJ, Van Veldhuisen DJ, Parkhomenko A, Borbola J,Cohen-Solal A, Dumitrascu D, Ferrari R, Lechat P, Soler-Soler J, Tavazzi L,Spinarova L, Toman J, Bohm M, Anker SD, Thompson SG, Poole-Wilson PA,SENIORS Investigators. Randomized trial to determine the effect of nebivololon mortality and cardiovascular hospital admission in elderly patients withheart failure (SENIORS). Eur Heart J 2005;26:215–225.

15. Cohen-Solal A, Kotecha D, van Veldhuisen DJ, Babalis D, Bohm M, Coats AJ,Roughton M, Poole-Wilson P, Tavazzi L, Flather M, SENIORS Investigators. Effi-cacy and safety of nebivolol in elderly heart failure patients with impaired renalfunction: insights from the SENIORS trial. Eur J Heart Fail 2009;11:872–880.

16. van Veldhuisen DJ, Cohen-Solal A, Bohm M, Anker SD, Babalis D, Roughton M,Coats AJ, Poole-Wilson PA, Flather MD, on behalf of the SENIORS Investigators.Beta-blockade with nebivolol in elderly heart failure patients with impaired andpreserved left ventricular ejection fraction. J Am Coll Cardiol 2009;53:2150–2158.

17. Abhayaratna WP, Seward JB, Appleton CP, Douglas PS, Oh JK, Tajik AJ, Tsang TS.Left atrial size: physiologic determinants and clinical applications. J Am Coll Cardiol2006;47:2357–2363.

18. Paulus WJ, Tschope C, Sanderson JE, Rusconi C, Flachskampf FA, Rademakers FE,Marino P, Smiseth OA, De Keulenaer G, Leite-Moreira AF, Borbely A, Edes I,Handoko ML, Heymans S, Pezzali N, Pieske B, Dickstein K, Fraser AG,Brutsaert DL. How to diagnose diastolic heart failure: a consensus statementon the diagnosis of heart failure with normal left ventricular ejection fractionby the Heart Failure and Echocardiography Associations of the EuropeanSociety of Cardiology. Eur Heart J 2007;28:2539–2550.

19. Manzano L, Cornide-Santos L. Left atrium size as marker of ventricular dysfunc-tion. Med Clin (Barc) 2007;129:454–455.

20. Kizer JR, Bella JN, Palmieri V, Liu JE, Best LG, Lee ET, Roman MJ, Devereux RB.Left atrial diameter as an independent predictor of first clinical cardiovascularevents in middle-aged and elderly adults: the Strong Heart Study (SHS). AmHeart J. 2006;151:412–418.

21. Doehner W, von Haehling S, Anker SD. Uric acid as a prognostic marker in acuteheart failure—new expectations from an old molecule. Eur J Heart Fail 2007;9:437–439.

22. 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 in chronic heartfailure: validation and application in metabolic, functional, and hemodynamicstaging. Circulation 2003;107:1991–1997.

23. Meisinger C, Koenig W, Baumert J, Doring A. Uric acid levels are associated withall-cause and cardiovascular disease mortality independent of systemic inflam-mation in men from the general population: the MONICA/KORA cohort study.Arterioscler Thromb Vasc Biol 2008;28:1186–1192.

24. Hare JM, Mangal B, Brown J, Fisher C Jr, Freudenberger R, Colucci WS, Mann DL,Liu P, Givertz MM, Schwarz RP, for the OPT-CHF Investigators. Impact of oxypur-inol in patients with symptomatic heart failure. Results of the OPT-CHF Study.J Am Coll Cardiol 2008;51:2301–2309.

25. von Haehling S, Lainscak M, Springer J, Anker SD. Cardiac cachexia: a systematicoverview. Pharmacol Ther 2009;121:227–252.

26. Kerzner R, Gage BF, Rich MW. Anemia does not predict mortality in elderlypatients with heart failure. Am J Geriatr Cardiol 2007;16:92–96.

27. Alehagen U, Dahlstrom U, Lindahl TL. Cystatin C and NT-proBNP, a powerfulcombination of biomarkers for predicting cardiovascular mortality in elderlypatients with heart failure: results from a 10-year study in primary care. Eur JHeart Fail 2009;11:354–360.


at Aston U


Dow

nloaded from





predictors of clinical outcomes in elderly patients with heart failure

Documents