Plasma Cytokine Parameters and Mortality in Patients With Chronic Heart Failure
Background—Inflammatory immune activation is an important feature in chronic heart failure (CHF). Little is known about the prognostic importance of tumor necrosis factor-α (TNF-α), soluble TNF-receptor 1 and 2 (sTNF-R1/sTNF-R2), interleukin-6 (IL-6), and soluble CD14 receptors (sCD14) in CHF patients.
Methods and Results—In 152 CHF patients (age 61±1 years, New York Heart Association [NYHA] class 2.6±0.1, peak V̇o2 17.3±0.6 mL · kg−1 · min−1, mean±SEM) plasma concentrations of immune variables were prospectively assessed. During a mean follow-up of 34 months (>12 months in all patients), 62 patients (41%) died. Cumulative mortality was 28% at 24 months. In univariate analyses, increased total and trimeric TNF-α, sTNF-R1, and sTNF-R2 (all P≤0.0001), sCD14 (P=0.0007), and IL-6 (P=0.005) predicted 24-month mortality. With multivariate analysis and receiver operating characteristics, sTNF-R1 emerged among all cytokine parameters as the strongest and most accurate prognosticator in this CHF population, regardless of follow-up duration and independently of NYHA class, peak V̇o2, V̇e/V̇co2 slope, left ventricular ejection fraction, and wasting (P<0.001). The receiver operating characteristic area under the curve for sTNF-R1 was greater than for sTNF-R2 at 6, 12, and 18 months (all P<0.05).
Conclusions—sTNF-R1 was the strongest and most accurate prognosticator, independent of established markers of CHF severity. Assessment of sTNF-R1 may be useful in identifying patients who are at high risk of death and in monitoring patients undergoing anti–TNF-α treatment.
Tumor necrosis factor-α (TNF-α) and other proinflammatory cytokine parameters can be elevated in patients with advanced chronic heart failure (CHF).1 2 3 Little is known, however, about the prognostic importance of these immune markers in CHF. Previous reports on inflammatory cytokines and cytokine receptors are controversial or refer to short-term follow-up only.4 5 6 Soluble TNF receptor 1 and 2 (sTNF-R1 and sTNF-R2) have been found to be particularly high in unstable patients with New York Heart Association (NYHA) class III and IV who died during 1 month of follow-up (P<0.001).7 Plasma concentrations of soluble TNF receptors vary less than those of TNF-α and interleukin-6 (IL-6)8 and appear to reflect the history of inflammatory immune activation. They may therefore more closely relate to the patient’s clinical condition.
Between 1992 and 1998, we consecutively enrolled 152 patients (142 men, 10 women) aged 23 to 85 years into our metabolic study program. The clinical details are given in Table 1⇓. The diagnosis of CHF was based on standard criteria.2 In 135 patients, near-maximal exercise capacity was achieved, as indicated by a respiratory exchange ratio >1.00 during treadmill exercise testing with gas exchange analysis (Amis 2000). All patients were receiving standard medical treatment consisting of diuretics (96%), ACE inhibitors (91%), digoxin, aspirin, oral nitrates, statins, warfarin, calcium antagonists, angiotensin II receptor blockers (7%), and β-adrenoreceptor antagonists (12%) in various combinations. Patients were excluded from the study if they had clinical signs of acute infection, rheumatoid disease, severe renal failure (creatinine >250 μmol/L), or myocardial infarction within the previous 12 months or if they were suspected of having a malignant or a primary wasting disorder. The presence of cardiac cachexia (n=44) was diagnosed as defined previously.9 Sixty (29%) of the 152 patients were studied before June 1995 and were included in a previous mortality study that focused on cachexia.9
We planned a minimum follow-up of 12 months in all patients. This was achieved by outpatient assessments, telephone contact with the patient or his or her local physician, or through the Hospital Information System by October 1999. Of the survivors, 26 patients were censored with a follow-up between 1 and 2 years. Twenty-one patients died between 25 and 75 months and have been censored for the follow-up of interest. The primary end point of the study was all-cause mortality.
Fasting blood samples were collected after supine rest for 20 minutes. The procedures for collecting blood samples and assaying EDTA plasma concentrations of sTNF-R1, sTNF-R2, soluble CD14 (sCD14), and total TNF-α (ie, trimers and fragments10 ) have been described elsewhere.11 Concentrations of trimeric, bioactive TNF-α10 were determined by the high-sensitivity human TNF-α test (R&D Systems, sensitivity 0.18 pg/mL). Plasma concentrations of interleukin-6 (IL-6) were measured by Immulite (sensitivity 1.0 pg/mL, Random Access Immunoassay Analyzer, DPC Biermann).
Data are given as mean±SEM. IL-6 and TNF-α plasma concentrations were log-transformed before analysis. The unpaired Student t test, χ2 test, and Cox proportional hazards analyses were used as appropriate. Hazard ratios (RRs) with 95% CIs and probability values by the likelihood ratio test are given (StatView 5.0, Abacus Concepts).
To compare different predictive values at a particular time point, areas under the curve (AUCs) for sensitivity and specificity were constructed. The best prognostic cutoff for survival status at a given time point was defined as that which gave the highest product of sensitivity and specificity. To contrast prognostic accuracy, statistical comparison of receiver operating characteristics (ROC)12 was performed (MedCalc, version 5.0, MedCalc Software).
Of the 152 CHF patients, 62 (41%) died after 4 to 2286 days (median 404 days). Mean follow-up period of the 90 survivors was 1355±88 days (range 366 to 2747, median 1114 days). Cumulative mortality rate for all patients was 11% (95% CI 6% to 16%) at 6 months (17 deaths), 20% (95% CI 13% to 26%) at 12 months (30 deaths), 26% (95% CI 19% to 33%) at 18 months (39 deaths), and 27% (95% CI 21% to 35%) at 24 months (41 deaths).
Plasma concentrations of the immune markers are depicted in Table 1⇑. Most of the cytokine parameters correlated with each other significantly, except for the relationship of IL-6 with trimeric TNF-α and sCD14 (r<0.10, P>0.40). The strongest relationships were found between sTNF-R1 and sTNF-R2 (r=0.73) and between total TNF-α and trimeric TNF-α (r=0.63, both P<0.0001). The relationship between plasma concentrations of sTNF-R1 and NYHA functional class is illustrated in Figure 1A⇓.
Univariate Survival Analyses
Cox proportional hazards analysis showed that increased concentrations of sTNF-R1 (χ2=26.1), sTNF-R2 (χ2=15.1), both total and trimeric TNF-α (χ2=14.5 and 24.1, respectively, all P≤0.0001), sCD14 (χ2=11.4, P=0.003), and IL-6 (χ2=7.9, P=0.005) predicted 24-month mortality (Table 2⇓). Peak V̇o2 (χ2=25.5, P<0.0001, n=135), NYHA class (χ2=18.3, P<0.0001), serum creatinine (χ2=14.8, P<0.0001), V̇e/V̇co2 slope (χ2=12.1, P=0.0001, n=135), left ventricular ejection fraction (LVEF) (χ2=10.3, P=0.004, n=130, 26 patients with LVEF >40%), age (χ2=7.3, P=0.009), and the presence of cardiac cachexia (χ2=6.5, P=0.008) were also significant prognosticators, whereas serum sodium concentration was not (χ2=3.1, P=0.08). Plasma concentrations of all immune markers within the highest quartile were significantly predictive for impaired 24-month survival (Table 2⇓). Highest quartiles of creatinine (>142 μmol/L, RR=3.64, P<0.0001) and V̇e/V̇co2 slope (>43.9, RR=2.49, P=0.006) and lowest quartiles of peak V̇o2 (≤12.2 mL·kg-1·min-1, RR 3.43, P=0.0002), LVEF (≤17%, RR=2.94, P=0.001), and serum sodium (≤135 mmol/L, RR=2.40, P=0.006) also predicted increased 24-month mortality. No significance was seen for the highest quartile of age (>69.6 years, RR=1.39, P=0.34).
Multivariate Survival Analyses
sTNF-R1 (P<0.0001), trimeric TNF-α (P<0.0001), IL-6 (P=0.006), and total TNF-α (P=0.02, n=85) predicted 24-month mortality independently of age and peak V̇o2. A trend was seen for sCD14 (P=0.06, n=120). In bivariate Cox proportional hazards analyses with all cytokines, sTNF-R1 emerged as the strongest mortality-predicting immune parameter (P<0.001 versus sTNF-R2, total TNF-α, IL-6, and sCD14). Among the cytokine parameters measured, only trimeric TNF-α (P=0.014) predicted mortality independently of sTNF-R1 (P=0.0015). Trimeric TNF-α also predicted 24-month mortality independently of other cytokine parameters studied (P<0.001 in all paired analyses) with the exception of IL-6, which was the only other significant parameter. Quartiles of sTNF-R1 in relation to 24-month mortality are illustrated in Figure 1B⇑. CHF patients in the top quartile had a 12-fold higher risk of death than patients in the lowest quartile (RR=12.34, 95% CI 4.9 to 31.3, P<0.0001). Clinical variables that predicted 24-month mortality independently of sTNF-R1 (LVEF, peak V̇o2, NYHA functional class, and V̇e/V̇co2 slope) are presented in Table 3⇓.
In multivariate analysis with sTNF-R1 and clinical parameters (peak V̇o2, V̇e/V̇co2, LVEF, and NYHA class), only sTNF-R1 (P<0.0001), LVEF (P<0.01), and peak V̇o2 (P<0.01) related to 24-month mortality independently (Table 3⇑, analysis 1). The presence of cachexia predicted 24-month mortality independently of peak V̇o2 and LVEF (Table 3⇑, analysis 2). However, when sTNF-R1 was added to these 3 variables, cachexia no longer had independent prognostic value (analysis 3). The best 3-parameter model for mortality prediction included sTNF-R1, LVEF, and peak V̇o2 (joint χ2=54.4, P<0.0001, analysis 4). Age, serum sodium, and measures of liver and kidney function were not predictive of mortality in multivariate analyses. After adjustment for body wasting, differences in drug therapy, or the dose of diuretics, the principal results did not change. When the group of noncachectic patients was analyzed alone, sTNF-R1 also revealed the strongest prognostic power (P<0.0001) in univariate and multivariate analyses (Table 4⇓).
Receiver Operating Characteristics
Sensitivity and specificity for the cytokine parameters, LVEF, and peak V̇o2 to predict mortality at 6, 12, 18, and 24 months were assessed across a range of cutoff values. At all time points, the ROC AUC for mortality prediction was highest for sTNF-R1 (best cutoff values 1067, 958, 1460, and 1124 pg/mL, respectively, for 6, 12, 18, and 24 months of follow-up). (A table giving detailed results on sensitivity, specificity, and best cutoff values for all parameters can be obtained from the authors on request.) To illustrate the relationships between cytokines, soluble receptors, and mortality, Kaplan-Meier survival curves using the optimal cutoff at 24 months are presented in Figure 2⇓. The best cutoff value for sTNF-R1 at 24-month follow-up had 83% sensitivity (95% CI 68% to 93%) and 73% specificity (95% CI 62% to 82%) to predict mortality (ROC AUC 0.84±0.04, 95% CI 0.78 to 0.91). This was the highest observed ROC AUC. The ROC AUCs for sTNF-R1 were somewhat larger at all respective time points than that of peak V̇o2, but the differences did not reach significance at any specific time point (all P>0.19). Compared with LVEF, prognostic importance of sTNF-R1 increased with time of follow-up (6, 12, 18, 24 months: P=0.44, P=0.19, P=0.03, and P=0.01, respectively). sTNF-R1 predicted mortality significantly better than sTNF-R2 at 6 months (AUC 0.78±0.07 versus 0.62±0.08, P=0.017), 12 months (0.77±0.05 versus 0.63±0.06, P=0.003, Figure 3⇓), and 18 months of follow-up (0.83±0.04 versus 0.73±0.05, P=0.034). There was also a trend at 24 months (P=0.07). At 24 months, the ROC AUC for sTNF-R1 was significantly larger than that for IL-6 (P=0.0001) and sCD14 (P=0.004), but it was not significantly different from that of total TNF-α (P=0.09) and was nearly identical to that of trimeric TNF-α (0.80±0.06, P=0.62). The accuracy of both TNF-α test kits, as assessed by ROC AUCs, was not significantly different at 6, 12, 18, and 24 months (all P>0.20).
Our study reveals that increased plasma concentrations of cytokines and soluble cytokine receptors significantly predict impaired median to longer-term survival in patients with CHF. The best mortality predictive value and accuracy was found for sTNF-R1, which provided the highest sensitivity and specificity among all immune parameters, independently of clinical variables and length of follow-up.
Most studied in CHF is the relationship between mortality and IL-6 plasma concentrations. IL-6 has been linked to CHF severity,4 13 and it has been associated with a poor short-term14 and long-term clinical outcome.6 15 In CHF patients enrolled in PRAISE (Prospective Randomized Amlodipine Survival Evaluation),5 adverse events occurred somewhat more commonly in patients with higher IL-6 levels (P=0.07), whereas a report from the SOLVD treatment trial (Studies Of Left Ventricular Dysfunction) did not find such a relationship in patients with NYHA class I to III functional classification (P=0.72).4 We found, consistent with other studies,16 17 that IL-6 concentrations were increased mainly in patients with NYHA class IV (data not shown) and that IL-6 is a significant prognosticator in univariate analysis (P=0.005). IL-6, however, lost its prognostic power (P=0.33) in multivariate analysis with sTNF-R1 (P<0.0001). Because Cox proportional hazards analysis assumes normally distributed variables, IL-6 required transformation. This was not commented on in the long-term study by Tsutamoto et al.6 Additionally, in that study,6 TNF receptors were not measured, and therefore a comparison between the 2 cytokine parameters is not possible. Because of a relatively high short-term variability of IL-6 concentrations,8 interpretation of results may vary between studies. Different methods of assessing IL-6 plasma concentrations may further complicate comparisons between studies. Finally, a slightly different genetic predisposition of an Asian CHF population may contribute to the impressive prognostic power of IL-6 in the study by Tsutamoto et al.6
Tumor Necrosis Factor-α
For approximately 10 years, it has been known that TNF-α concentrations are raised in patients with severe CHF, particularly in those with cachexia.1 2 3 In the SOLVD study,4 TNF-α (R&D ELISA) has been linked to weakly impaired long-term survival in CHF patients (P=0.07), but NYHA class IV patients were not included in the assessment. Differences between studies may be explained by marked daily and weekly variability of TNF-α,8 its short half-life, and differences between test kits.18 We found a relatively good sensitivity and specificity of both total and trimeric TNF-α for predicting clinical outcome. Both TNF-α indices predicted prognosis independently of peak V̇o2, and the prognostic value of the 2 TNF-α indices did not differ significantly.
Soluble TNF Receptors
The best overall predictive value for increased mortality among all the cytokine parameters, across different time points as well as after restriction of mortality to noncachectic CHF patients, was found for sTNF-R1. For each 1000-pg/mL increase in sTNF-R1 concentrations, the mortality hazard increased by 50% (95% CI 50% to 100%).
Like TNF-α and IL-6, sTNF receptors are highest in patients with severe CHF: in NYHA class IV,7 19 during phases of edematous decompensation,20 and in cachectic CHF patients.21 Moreover, compared with healthy subjects, increased concentrations of sTNF-R1 are already present in stable, noncachectic patients with mild CHF.21 sTNF-R1 appears to be the most powerful and independent immune marker and may well reflect its longer half-life and lower short-term variability.8 sTNF-R1 is also likely to best reflect the history of inflammatory immune activation in patients with CHF. Previous findings on short-term risk stratification in CHF patients suggested elevated sTNF-R2 concentrations to be somewhat better related to poor clinical outcome than those of sTNF-R1.7 However, the number of patients in that study7 was relatively small (n=37, 10 deaths), and the difference has not been tested statistically. In the present study, predictive accuracy of sTNF-R1 was consistently better than that of sTNF-R2 during follow-up between 6 and 24 months.
The primary goals for the treatment of end-stage CHF are improvements in survival and in quality of life. To achieve these goals, treatment needs to be tailored individually. To this end, prognostic markers that accurately define risk groups are needed. In the present study, the best mortality-predicting 3-parameter model included sTNF-R1, LVEF, and peak V̇o2. Our study confirms that cachexia predicts prognosis independently of LVEF and peak V̇o2 as previously shown.9 However, the “bedside marker” cardiac cachexia lost its independent prognostic power in multivariate comparison when the humoral marker sTNF-R1 was included. This was not entirely unexpected, because wasting is closely associated with immune activation.2 21 Moreover, a dichotomous marker is very likely to lose significance when comparison is made with a continuously distributed variable with >4000 degrees of freedom. Additionally, we found a somewhat higher proportion of cachectic patients in the present study (28.9%) than in a previous study (16.4%).9 We believe this does not reflect a selection bias toward patients with wasting, although we have a strong interest in studying cardiac cachexia. It may, however, result from our very strict assessment of cachexia in all CHF patients referred to our center using a semistandardized questionnaire. Because we perform very detailed metabolic assessments, we cannot entirely exclude the possibility that more cachectic patients are being referred to us than to other centers. Most importantly, in this context, the results of the analyses in noncachectic patients alone need to be considered (Table 4⇑). All our findings, particularly with regard to the importance of sTNF-R1, are still valid when cachectic CHF patients are excluded.
Our study adds evidence to recent findings that show neurohormonal and immunologic factors may be of greater prognostic importance than the more conventional assessments of the hemodynamic and clinical status.22 23 This suggests that immunologic abnormalities are not simply an epiphenomenon in patients with CHF but that they carry independent pathophysiological importance that ultimately leads to a prognostic impact on mortality, which is supported by findings of Meldrum and coworkers.24 By specifically targeting TNF-α using a TNF receptor/human IgG-fusion protein, one study has already shown some clinical benefit in a small group of CHF patients.25 Larger clinical trials are currently ongoing to examine the effects of antagonizing TNF-α bioactivity.26 It is hoped that this will demonstrate whether counteracting the TNF-α pathway in patients with moderate to severe CHF in general confers morbidity and mortality benefits. If this is not the case, then a well-reproducible measure of inflammatory immune activation with strong prognostic value (like sTNF-R1) may define patient subgroups likely to benefit from anti-TNF-α therapy.
Dr Rauchhaus was supported by grants of the Commission of the European Communities, Brussels, and of the Deutsche Herzstiftung, Frankfurt. Dr Doehner is supported by a grant of the Verein der Freunde und Förderer der Berliner Charité and a PhD studentship of the National Heart & Lung Institute, London. Dr Francis was supported by the British Heart Foundation. Dr Coats is supported by the Viscount Royston Trust and the British Heart Foundation. Dr Anker was supported by the Ernst- und Bertha-Grimmke-Stiftung in Düsseldorf, Germany, and by a postgraduate fellowship of the Max Delbrück Centrum, Berlin. Support was also received from the Royal Brompton & Harefield NHS Trust Joint Committee for Research, London, UK. The authors thank Dr Aidan Bolger for critically reading the manuscript.
- Received May 12, 2000.
- Revision received July 26, 2000.
- Accepted August 1, 2000.
- Copyright © 2000 by American Heart Association
Anker SD, Chua TP, Ponikowski P, et al. Hormonal changes and catabolic/anabolic imbalance in chronic heart failure and their importance for cardiac cachexia. Circulation. 1997;96:526–534.
Tsutamoto T, Hisanaga T, Wada A, et al. Interleukin-6 spillover in the peripheral circulation increases with the severity of heart failure, and the high plasma level of interleukin- 6 is an important prognostic predictor in patients with congestive heart failure. J Am Coll Cardiol. 1998;31:391–398.
Ferrari R, Bachetti T, Confortini R, et al. Tumor necrosis factor soluble receptors in patients with various degrees of congestive heart failure. Circulation. 1995;92:1479–1486.
Torre-Amione G, Kapadia S, Lee J, et al. Tumor necrosis factor-alpha and tumor necrosis factor receptors in the failing human heart. Circulation. 1996;93:704–711.
Anker SD, Ponikowski PP, Clark AL, et al. Cytokines and neurohormones relating to body composition alterations in the wasting syndrome of chronic heart failure. Eur Heart J. 1999;20:683–693.
Swedberg K, Eneroth P, Kjekshus J, et al. Hormones regulating cardiovascular function in patients with severe congestive heart failure and their relation to mortality: CONSENSUS Trial Study Group. Circulation. 1990;82:1730–1736.
Omland T, Aakvaag A, Bonarjee VV, et al. Plasma brain natriuretic peptide as an indicator of left ventricular systolic function and long-term survival after acute myocardial infarction: comparison with plasma atrial natriuretic peptide and N- terminal proatrial natriuretic peptide. Circulation. 1996;93:1963–1969.
Meldrum DR, Dinarello CA, Shames BD, et al. Ischemic preconditioning decreases postischemic myocardial tumor necrosis factor-alpha production: potential ultimate effector mechanism of preconditioning. Circulation. 1998;98(suppl II):II-214–II-218.
Deswal A, Bozkurt B, Seta Y, et al. Safety and efficacy of a soluble P75 tumor necrosis factor receptor (Enbrel, etanercept) in patients with advanced heart failure. Circulation. 1999;99:3224–3226.
Mann DL. Mechanisms and models in heart failure: a combinatorial approach. Circulation. 1999;100:999–1008.