Author + information
- Received February 25, 2002
- Revision received May 31, 2002
- Accepted July 24, 2002
- Published online November 20, 2002.
- Mark T Kearney, DM*,* (, )
- Keith A.A Fox, FRCP†,
- Amanda J Lee, MSc‡,
- Robin J Prescott, PhD‡,
- Ajay M Shah, FRCP*,
- Philip D Batin, DM∥,
- Wazir Baig, MD¶,
- Stephen Lindsay, MD#,
- Timothy S Callahan, PhD**,
- William E Shell, MD††,
- Dwain L Eckberg, MD‡‡,
- Azfar G Zaman, MD§§,
- Simon Williams, MRCP∥∥,
- James M.M Neilson, PhD§ and
- James Nolan, MD¶¶
- ↵*Reprint requests and correspondence:
Dr. Mark T. Kearney, Department of Cardiology, GKT School of Medicine, Kings College London, Bessemer Road, Denmark Hill, London SE5 9PJ, UK.
Objectives The aim of this study was to explore the value of noninvasive predictors of death/mode of death in ambulant outpatients with chronic heart failure (HF).
Background Mortality in chronic HF remains high, with a significant number of patients dying of progressive disease. Identification of these patients is important.
Methods We recruited 553 ambulant outpatients age 63 ± 10 years with symptoms of chronic HF (New York Heart Association functional class, 2.3 ± 0.5) and objective evidence of left ventricular dysfunction (ejection fraction <45%, cardiothoracic ratio >0.55, or pulmonary edema on chest radiograph). After 2,365 patient-years of follow-up, 201 patients had died, with 76 events due to progressive HF.
Results Independent predictors of all-cause mortality assessed with the Cox proportional hazards model were as follows: a low standard deviation of all normal-to-normal RR intervals (SDNN); lower serum sodium and higher creatinine levels; higher cardiothoracic ratio; nonsustained ventricular tachycardia; higher left ventricular end-systolic diameter; left ventricular hypertrophy; and increasing age. Independent predictors of death specific to progressive HF were SDNN, serum sodium and creatinine levels. The hazard ratio of progressive HF death for a 10% decrease in SDNN was 1.06 (95% confidence interval [CI], 1.01 to 1.12); for a 2 mmol/l decrease in serum sodium, 1.22 (95% CI, 1.08 to 1.38); and for a 10 μmol/l increase in serum creatinine, 1.14 (95% CI, 1.09 to 1.19) (all p < 0.01).
Conclusions In ambulant outpatients with chronic HF, low serum sodium and SDNN and high serum creatinine identify patients at increased risk of death due to progressive HF.
Over a half million new cases of chronic heart failure (HF) are diagnosed each year in the U.S. (1). Despite advances in understanding of the pathophysiology of chronic HF and improvements in its therapy, mortality in ambulant outpatients with controlled symptoms currently remains ∼10% per year (2,3). A substantial proportion of these deaths are due to a progressive decline in left ventricular function (2,3). It is not fully understood why some patients who have a period of compensated HF after an initial cardiac insult subsequently enter into an irreversible process of declining ventricular function. There is compelling evidence, however, that autonomic and neurohumoral mechanisms play a critical role in mediating this decline in cardiac function (4). Agents such as beta-blockers and angiotensin-converting enzyme (ACE) inhibitors that inhibit specific neurohumoral mechanisms prolong life in chronic HF patients, at least in part, by slowing disease progression (2,5,6). Variables that reflect the magnitude of neurohumoral dysfunction, therefore, may help to identify patients with chronic HF who are at increased risk of entering a decompensated spiral of declining ventricular function, hemodynamic deterioration, and death due to progressive HF.
Several studies have reported variables that might identify patients with chronic HF who are at high risk of death, and their modes of death (7–10). Most of these studies are retrospective analyses of data from trials of therapeutic agents, where risk stratification was not the primary objective (7–9). Additionally, some of these studies were conducted in patients with severe HF (8,9)who may be easier to identify with bedside assessment, and who have annual mortality rates >30% (11).
Treatments aimed at preventing death from progressive HF, as opposed to sudden death, may and currently do require different strategies. Therefore, identifying patients at risk of death specifically from progressive HF may help to tailor further investigation or therapy. The United Kingdom Heart failure Evaluation and Assessment of Risk Trial (UK-HEART) was prospectively designed to assess the value of noninvasive predictors of mortality and mode of death (including noninvasive measurements of neurohumoral function) in ambulant outpatients with chronic HF. We now report on five-year follow-up of our study cohort, with the principal focus of this report on variables that specifically identify patients at increased risk of death due to progressive HF. The current study has the longest follow-up of its type and adds to previous work by evaluating the long-term prognostic utility of both time- and frequency-domain measurements of heart rate variability (HRV).
The UK-HEART was a prospective study carried out in cardiology outpatient clinics of eight United Kingdom general hospitals, between December 1993 and April 1995 (12). Consecutive patients of either gender, age 18 to 85 years, with chronic HF were recruited on the basis of predefined inclusion and exclusion criteria. The ethical committee at each institution approved the protocol, and all patients gave informed consent. Patients were eligible for the trial if they were ambulant outpatients with stable clinical signs and symptoms of chronic HF present for at least three months. We required that symptoms be associated with objective evidence of cardiac dysfunction at rest (pulmonary venous congestion, pulmonary edema, a cardiothoracic ratio >0.55 on at least one chest radiograph, or a radionuclide or echocardiographic ejection fraction [EF] <45%). To avoid possible confounding effects on measurements of neurohumoral activity, patients were excluded if they had a condition associated with impaired autonomic function (diabetes mellitus or chronic renal failure unrelated to HF), a history of alcohol abuse, autonomic neuropathy, or a recent (<3 months) myocardial infarction. Patients with documented constrictive or hypertrophic cardiomyopathy, sustained nonsinus dysrhythmias, atrioventricular conduction defects, or a comorbid noncardiac disease likely to limit survival were also excluded.
Baseline data collection
At the time of recruitment, a case record form detailing baseline clinical and demographic data was completed for each patient. An erect posteroanterior chest radiograph was obtained, and the cardiothoracic ratio was measured. A venous blood sample was taken at rest for measurements of electrolyte concentrations and assessment of renal and liver function. Two-dimensional and M-mode echocardiography were performed according to American Society of Echocardiography recommendations. Left ventricular cavity dimensions, EF, and fractional shortening index were calculated according to standard formulas. All patients were registered with the United Kingdom Office of Population Censuses and Surveys, which notified the Steering Committee of all deaths.
Ambulatory and 12-lead electrocardiography
Twenty-four hour ambulatory electrocardiogram recordings (Tracker, Reynolds Medical, UK) were obtained in all patients during normal, unrestricted out-of-hospital activity. Recordings were analyzed with a Reynolds Medical Pathfinder system by independent technical staff blinded to patient characteristics and outcome data. Nonsustained ventricular tachycardia was defined as three or more consecutive ventricular ectopic beats at a rate >120 beats/min. Left ventricular hypertrophy was assessed using the Sokolow-Lyon voltage (sum of the amplitude of the S-wave on lead V1and the R-wave on V5or V6≥3.5 mV).
Time-domain analyses of HRV
After initial arrhythmia analysis and editing, normal-to-normal RR intervals were identified, and HRV in the time-domain was measured, according to published guidelines (13). Standard deviation of all normal-to-normal RR intervals (SDNN) is an index of total HRV. We have previously demonstrated that SDNN is a better prognostic marker than other time-domain measurements and, therefore, confined our prognostic studies to this index (12).
24-h frequency-domain HRV analyses
Frequency-domain analyses were made for the first 5 min of each hour and averaged. Fast Fourier transform power spectra (with linear interpolation over missing data, resampling at 0.5 Hz, and Hamming windowing) were integrated over all frequencies total power (TP): 0.0033 to 0.04 Hz very low-frequency power (VLFP); 0.04 to 0.15 Hz low-frequency power (LFP); and 0.15 to 0.40 Hz high-frequency power (HFP) (14,15).
Classification of cause of death
Classification criteria for the cause of death were defined before the study commenced. All deaths reported to the Steering Committee were evaluated by at least two senior physicians, who reviewed death certificates, autopsy findings, and hospital and general practitioners’ records. Where these physicians did not agree on the cause of death, the case was adjudicated by the lead investigator. The mode of death was classified as: 1) sudden cardiac, if it occurred within 1 h of a change in symptoms or during sleep, or while the patient was unobserved; 2) progressive HF, if death occurred after a documented period of symptomatic or hemodynamic deterioration; 3) other cardiovascular death, if death did not occur suddenly and was not associated with progression of HF; and 4) noncardiovascular death.
Descriptive group data are given as mean ± SD unless stated otherwise. Estimated glomerular filtration rate (eGFR) was calculated using the Modification of Diet in Renal Disease method (16). Natural logarithms were used for all HRV measurements and cardiothoracic ratios. The Cox proportional hazards regression model was used to determine which measurements were related significantly to mortality during the follow-up period, using a multivariate forward and backward stepping model. The Cox model has advantages over other techniques such as logistic regression. It takes into consideration variable duration of follow-up, censoring of subjects, proportionality of event occurrence, and time to event. For the prediction of progressive HF, observations were censored if the patient died of another cause of death; these patients were included in the analysis only up to the time that they died. The causes of death may not be entirely independent, and this could have an effect on the model. We used receiver operating characteristic (ROC) analysis to generate C-statistics. The ROC curves plot the positive fraction, or sensitivity against the false positive fraction (1-specificity) by varying the threshold value for the test. The ROC curve indicates the probability of a true positive result as a function of the probability of a false positive result for all possible threshold values. A C-statistic of 0.5 indicates that the test results are no better than those obtained by chance, whereas an area of 1.0 indicates a perfectly sensitive and specific test.
For the prediction of progressive HF, variables entered into the model were age, gender, the presence or absence of nonsustained ventricular tachycardia, ventricular hypertrophy, left ventricular end-systolic/diastolic diameters, EF, sodium, potassium, urea, creatinine, and the natural logarithms of the cardiothoracic ratio, SDNN, VLFP, LFP, HFP, and TP. Where appropriate, Kaplan-Meier cumulative mortality curves were plotted to display trends in mortality over time and risk ratio. A value of p < 0.05 was taken as statistically significant. The present study has more events and longer follow-up than our previous report and most other studies in this area and, thus, higher statistical power.
A total of 553 white patients were recruited in cardiology outpatient clinics. Patients’ mean age (and range) was 62.7 ± 9.7 years (18 to 85 years); 76% were male; 2% were categorized as New York Heart Association (NYHA) functional class I, 59% as class II, and 39% as class III. The average NYHA functional class was 2.3 ± 0.5. Mean serum creatinine was 121 ± 40.8 μmol/l (60 to 340 μmol/l), and mean sodium was 140 ± 3.2 mmol/l (122 to 148 mmol/l). Of the patients, 76% had ischemic heart disease as the etiology of their HF, 38% had nonsustained ventricular tachycardia, and 10% had left ventricular hypertrophy.
Most patients (82%) were treated with ACE inhibitors (mean enalapril dose: 12 ± 0.3 mg/day) and loop diuretics (99%, mean furosemide dose: 71 ± 69 mg/day). Of the patients, 19% were taking digoxin at a mean dose of 198.0 ± 7.1 μg/day; 14% of patients were taking amiodarone (all 200 mg/day); and 7.9% took atenolol at a mean dose of 43.7 ± 1.6 mg/day. Information on deaths was recorded up to and including April 2000, allowing five-year survival status to be determined for all patients. Follow-up was complete on all patients. Of the cohort recruited, 433 patients had 24-h Holter recordings suitable for HRV analysis (all had recordings suitable for arrhythmia analysis).
The mean left ventricular EF was 42 ± 17% (6% to 88%), left ventricular end-diastolic diameter was 6.2 ± 1.0 cm (3.2 to 9.4 cm), left ventricular end-systolic diameter was 5.0 ± 1.2 cm (1.7 to 9.2 cm), and cardiothoracic ratio was 0.53 ± 0.07 (0.34 to 0.88). The mean SDNN was 114 ± 41 ms (17 to 264 ms).
An EF of <45% was only one entrance criterion for the present study (see the Methods section). Of the study population, 64% had EFs ≤45%, with a total mortality at five years of 42% (with 16% dying of progressive HF and 15% dying suddenly). A total of 202 patients (36%) had EFs >45%, with a total mortality at five years of 28% (10% progressive HF and 8% sudden death). At 5 years (2,365 patient-years), 201 patients had died (mean annual mortality rate: 7.3%); 67 (33%) of sudden death; 76 (38%) of progressive HF; 24 (12%) of other cardiac causes; and 34 (17%) of noncardiac causes (Figs. 1A to 1D) .
Predictors of all-cause mortality
In multivariate analysis, eight variables provided independent prognostic information that identified patients at increased risk of death from any cause (Table 1):SDNN, serum sodium and creatinine, cardiothoracic ratio, nonsustained ventricular tachycardia, left ventricular end-systolic diameter, the presence of electrocardiographic left ventricular hypertrophy, and increased age (Table 1).The C-statistic for the all-cause mortality model is 0.78. When expressing renal function using eGFR it remained an independent predictor of all-cause mortality, hazards ratio (HR) 1.32 (95% confidence interval [CI] 1.13 to 1.53) for every 10 ml/min/1.73 m2decrement in eGFR (p < 0.001).
Predictors of death due to progressive HF
Seventy-six patients died of progressive HF (Table 1). In multivariate analysis, independent and specific predictors of progressive HF death were SDNN and serum sodium and creatinine (C-statistic: 0.77). When expressing renal function using eGFR, it remained an independent predictor of death due to progressive HF, HR 1.83 (95% CI 1.44 to 2.32) for every 10 ml/min/1.73 m2decrement in eGFR (p < 0.001).
These predictors remained important irrespective of EF and etiology of chronic HF. The power of serum sodium, creatinine, and SDNN to predict all-cause mortality or progressive HF death did not change over the period of follow-up. Figure 1Ashows mode of death in patients with left ventricular EF >45% and <45%. Figures 1B to 1Dshow Kaplan-Meier curves demonstrating the proportion of patients free of death due to HF, with the population dichotomized by median values of SDNN, sodium, and creatinine.
Our principal new findings are that measurements of SDNN and serum sodium and creatinine levels can identify patients with relatively mild disease who are at high risk of death specifically due to progressive HF over a five-year period. In our study, 24-h frequency-domain measurements of HRV did not provide additional prognostic information over the simple time-domain measurement SDNN.
HRV and progressive HF
In an earlier report (12)we showed that SDNN had potential value as an independent and powerful predictor of mortality due to progressive HF during approximately one-year follow-up. The current complete five-year report has substantially greater power and provides novel, clinically important information showing that among the more complex frequency-domain measurements of HRV added to the current analysis, SDNN remained the most important marker of risk for death preceded by progressive HF.
The intriguing observation that SDNN specifically predicts HF progression probably derives from influences of the multiple neurohormonal factors that modulate SDNN, an index of global HRV. In healthy subjects, HRV reflects primarily fluctuations of cardiac autonomic activity. Published evidence makes the linkage between reduction of cardiac autonomic activity, captured in our study by the SDNN measurement, and progression of HF highly plausible. Patients with HF have increased muscle sympathetic nerve activity (17–19), cardiac norepinephrine spillover (an index of sympathetic nerve traffic to the heart (19), plasma catecholamine levels (20), and renin activity (21). Both sympathetic stimulation (22)and angiotensin (23)reduce vagally mediated heart rate fluctuations. (Conversely, increased vagal activity reduces myocardial norepinephrine release.) Reductions of vagally mediated heart rate fluctuations found in patients with HF (24,25)are directly proportional to the level of muscle sympathetic nerve activity and plasma norepinephrine levels (18).
In addition, there is an intimate relationship between the sympathetic and renin-angiotensin-aldosterone systems, such that stimulation of one increases activity in the other (26). Both norepinephrine (27)and angiotensin II (28)are powerful promoters of cardiac myocyte hypertrophy and necrosis, and there is compelling evidence that norepinephrine and angiotensin II play pivotal roles in promoting the progression of clinical HF (4).
Another potential explanation for the ability of SDNN to predict HF progression is that SDNN is a marker for changes of myocardial size and geometry, both of which may be involved with HF progression (29). A recent study shows that chronic sinoatrial node stretch reduces HRV (30), suggesting that SDNN may be a marker for adverse changes of cardiac morphology, a possibility that warrants further study. We have previously demonstrated that ACE inhibitors augment HRV in patients with chronic HF (31). Of the population we studied, over 80% were already stabilized on these agents; despite this, SDNN remained an important predictor of progressive HF death.
Hyponatremia and progressive HF
We have shown, in a prospective study conducted in the ACE inhibitor era, that hyponatremia in ambulant outpatients with chronic HF is a predictor of mortality from pump failure. Lee and Packer (32), in a trial not specifically designed to identify prognostic factors, reported that patients with severe chronic HF (with mean EF ∼17%) and serum sodium levels <137 mmol/l) have a life expectancy half that of patients with serum sodium levels >137 mmol/l. More recently, Aaronson et al. (33)showed in a group of patients referred for cardiac transplantation (with mean EF ∼20%) that serum sodium levels are relatively weak predictors of all-cause mortality, in an already high-risk group of patients. The mechanism by which hyponatremia predicts HF progression may involve neurohumoral activity, including particularly that of the renin-angiotensin-aldosterone system. Although not demonstrated in the current cohort of patients, serum sodium correlates closely with plasma renin activity (5). It is notable that in the study of Lee and Packer (32), prognosis improved in those patients whose serum sodium levels increased in response to ACE inhibition.
Serum creatinine and progressive HF
Alterations of fluid and electrolyte homeostasis are important in the pathophysiology of chronic HF (32). Two recent reports highlight the potential value of impaired renal function as a marker for increased risk of death in chronic HF. Hillege et al. (34), in a retrospective analysis of the Second Prospective Randomized study of Ibopamine on Mortality and Efficacy study, showed that creatinine clearance derived from serum creatinine, body weight, and age is the most powerful predictor of all-cause mortality in patients with NYHA functional class III to IV HF symptoms and EFs <30%. The authors of that study did not attempt to determine the mode of death. Consonant with these findings and those of the present report, Dries et al. (35), in a retrospective analysis of the Studies Of Left Ventricular Dysfunction (SOLVD) database, demonstrated that moderate renal insufficiency is an independent predictor of all-cause mortality and progressive HF. The SOLVD study, however, excluded patients with creatinine >177 μmol/l. The present study shows that across a wide range (60 to 340 μmol/l), each 10 μmol/l increment of creatinine level increases the risk of a progressive HF death by ∼14%.
While we can only speculate on the pathophysiologic mechanisms underlying the correlation between renal impairment and left ventricular dysfunction, our data do not suggest that renal function is simply a marker for impaired cardiac output. We found no correlation between serum creatinine levels and any index of cardiac mechanical performance. Thus, it appears that renal dysfunction per se may contribute to HF progression. The mechanisms underlying this detrimental cardiorenal interaction warrant further investigation.
Predictors of progressive HF in ambulant outpatients
The majority of previous reports discussing risk stratification in chronic HF have focussed on patients with significant systolic dysfunction (10). The present study is the first to assess patients with chronic HF across a wide range of EFs, as one would expect to see in well-controlled population studies of chronic HF (36). Our data provide a novel approach to identify patients with chronic HF with and without preserved left ventricular systolic function, who are at increased risk of death preceded by decompensated HF.
The UK-HEART study was carried out before publication of the landmark trials documenting the benefit of beta-adrenergic blocking drugs and aldosterone antagonists in patients with chronic HF. However, this in itself may be an advantage of the current dataset, in that it allows identification of patients on baseline therapy who warrant more aggressive use of these and possibly other agents. The present study excluded diabetics and, therefore, we cannot extrapolate from our data to the diabetic population. However, when random blood glucose levels were incorporated into our statistical analysis, the predictors of all-cause mortality and mode of death did not change. This suggests that our results may apply to the diabetic population, a possibility that warrants further study.
Determining the cause of death in HF can be difficult. We dealt with this problem by having two senior physicians assess the cause of death, according to strict, predefined criteria. In support of this approach, our data demonstrate levels of SDNN, sodium, and creatinine that were significantly different in patients dying of progressive HF than in the rest of the cohort. Our population was relatively young for patients with chronic HF; therefore, the results of this exploratory analysis should now be assessed and applied in different HF populations.
We report a five-year follow-up of 553 patients presenting with mild-to-moderate HF, who are representative of those seen in cardiology and internal medicine practices. Our study provides insights into the pathophysiology of progressive HF; simple analysis of 24-h ambulatory electrocardiograms and measurement of serum sodium and creatinine levels identify patients at increased risk for terminal decompensated HF. These results refine risk stratification of HF patients and may thereby identify patients who merit aggressive medical treatment or complex therapeutic regimes.
☆ Financial support for this study was by the Chest, Heart, and Stroke Association (Scotland) and the Northern and Yorkshire Research and Development Directorate. Drs. Kearney, Fox, and Shah are supported by the British Heart Foundation.
- angiotensin-converting enzyme
- confidence interval
- ejection fraction
- estimated glomerular filtration rate
- heart failure
- high-frequency power
- hazards ratio
- heart rate variability
- low-frequency power
- New York Heart Association
- receiver operating characteristic
- standard deviation of all normal-to-normal RR intervals
- total power
- United Kingdom Heart failure Evaluation and Assessment of Risk Trial
- very low-frequency power
- Received February 25, 2002.
- Revision received May 31, 2002.
- Accepted July 24, 2002.
- American College of Cardiology Foundation
- Mann D.L.
- Gradman A.,
- Deedwania P.,
- Cody R.,
- et al.
- Singh S.N.,
- Fisher S.G.,
- Carson P.E.,
- et al.
- Teerlink J.R.,
- Jalaluddin M.,
- Anderson S.,
- et al.
- Nolan J.,
- Batin P.D.,
- Andrews R.,
- et al.
- Task Force of the European Society of Cardiology and the North American Society of Pacing and Electrophysiology
- Pomeranz B.,
- Macaulay R.J.B.,
- Caudill M.A.,
- et al.
- Leimbach W.N. Jr..,
- Wallin B.G.,
- Victor R.G.,
- et al.
- Rundqvist B.,
- Elam M.,
- Bergmann-Sverrisdottir Y.,
- et al.
- Dzau V.J.,
- Colucci W.S.,
- Hollenberg N.K.,
- et al.
- Taylor J.A.,
- Myers C.W.,
- Halliwill J.R.,
- et al.
- Taylor J.A.,
- Carr D.L.,
- Myers C.W.,
- et al.
- Nolan J.,
- Flapan A.D.,
- Capewell S.,
- Macdonald T.,
- Neilson J.M.M.,
- Ewing D.J.
- Reid I.A.
- Tan L.B.,
- Jalil J.E.,
- Pick R.,
- Janicki J.S.,
- Weber K.T.
- an International Forum on Cardiac Remodelling,
- Cohn J.N.,
- Ferrari R.,
- Sharpe N.
- Horner S.M.,
- Murphy C.F.,
- Coen B.,
- et al.
- Lee W.H.,
- Packer M.
- Aaronson K.D.,
- Schwartz S.,
- Chen T.-M.,
- et al.
- Hillege H.L.,
- Girbes A.R.J.,
- de Kam P.J.,
- et al.
- Dries D.L.,
- Exner D.V.,
- Domanski M.J.,
- et al.
- Vasan R.S.,
- Larson M.G.,
- Benjamin E.J.,
- Evans J.C.,
- Reiss C.K.,
- Levy D.