Author + information
- Received May 16, 2016
- Revision received July 7, 2016
- Accepted September 14, 2016
- Published online March 27, 2017.
- Alberto Aimo, MDa,∗ ( )(, )
- Giuseppe Vergaro, MDa,b,
- Claudio Passino, MDa,b,
- Andrea Ripoli, EngDb,
- Bonnie Ky, MDc,
- Wayne L. Miller, MD, PhDd,
- Antoni Bayes-Genis, MD, PhDe,
- Inder Anand, MD, DPhil (Oxon)f,
- James L. Januzzi, MDg and
- Michele Emdin, MD, PhDa,b
- aScuola Superiore Sant’Anna, Pisa, Italy
- bFondazione Toscana G. Monasterio, Pisa, Italy
- cPenn Cardiovascular Institute, University of Pennsylvania School of Medicine, Philadelphia, Pennsylvania
- dMayo Clinic College of Medicine, Rochester, Minnesota
- eInstitut del Cor, Hospital Universitari Germans Trias i Pujol, Barcelona, Spain
- fUniversity of Minnesota Medical School, Minneapolis, Minnesota
- gMassachusetts General Hospital and Harvard Clinical Research Institute, Boston, Massachusetts
- ↵∗Address for correspondence:
Dr. Alberto Aimo, Scuola Superiore Sant’Anna, Piazza Martiri della Libertà, 33, Pisa 56127, Italy.
Objectives The purpose of this study was to perform the first meta-analysis of currently available data.
Background Soluble suppression of tumorigenesis 2 (sST2) plasma concentration is elevated in chronic heart failure (CHF) and helps to predict prognosis in this setting, although the evidence is limited.
Methods Three databases (Medline, Cochrane Library, and Scopus) were searched. Inclusion criteria were: follow-up studies; papers published in English; enrollment of CHF outpatients; available data on hazard ratio (HR) for the log2 ST2 (so that the reported HRs represent the risk per doubling of sST2) and 95% confidence interval (CI) for all-cause death, and possibly also for cardiovascular (CV) death; and use of standardized sST2 assay. Exclusion criteria were: sST2 considered only as an element of a prognostic score, and studies on patients with end-stage HF.
Results Seven studies were finally included for all-cause death, with a global population of 6,372 patients; data on CV death were available for 5 studies, totaling 5,051 patients. The HR was 1.75 (95% CI: 1.37 to 2.22) for all-cause death and 1.79 (95% CI: 1.22 to 2.63) for CV death (both p < 0.001). Significant heterogeneity among studies was detected in the quantification of sST2 predictive value, attributable to marked differences in pharmacological treatment among trials. The predictive power of sST2 was greater when patients were managed according to present guideline-recommended medical treatment.
Conclusions sST2 is a predictor of both all-cause and CV death in CHF outpatients. The present meta-analysis supports the use of sST2 for risk stratification in patients with stable CHF.
Soluble suppression of tumorigenesis 2 protein (sST2) is the circulating form of the cellular receptor ST2L, expressed by cardiomyocytes and vascular endothelial cells together with its ligand interleukin-33 after cardiovascular injury. Binding between interleukin-33 and ST2L is likely to inhibit myocardial hypertrophy and fibrosis and mitigate adverse cardiac remodeling (1,2). sST2 competes with ST2L for interleukin-33 binding, thus likely blunting the cardiovascular protective effects exerted by the interleukin-33/ST2L interaction (1,2).
sST2 has elicited interest as a potential aid to predict prognosis and manage therapy of chronic heart failure (CHF) (3,4). Although the characterization of the interleukin-33/ST2L axis in CHF is currently incomplete (4), sST2 plasma concentrations have been found to be generally higher in CHF patients than in healthy subjects (5). Furthermore, several studies have suggested that higher plasma sST2 concentrations also predict a worse prognosis in CHF patients (6). Accordingly, recent clinical practice guidelines support the use of sST2 assay for risk stratification in CHF outpatients, with a Class IIb, Level of Evidence: B, recommendation (7). Of note, the sST2 assay for risk stratification of acute HF patients was considered Level of Evidence: B (7), reflecting the fact that the prognostic role of sST2 assay in CHF is less well defined than in acute HF (2,6,8), because of the limited number of clinical trials on this topic and the lack of systematic reviews or meta-analyses (7). To fill this gap, we set out to perform a meta-analysis on the prognostic role of sST2 in stable CHF.
Data sources, search strategy, and eligibility criteria
The meta-analysis was conducted according to the Preferred Reporting Items for Systematic reviews and Meta-Analyses statement (9). In March 2016, 2 reviewers (A.A. and G.V.) independently searched 3 databases (Medline, Cochrane Library, and Scopus). The search terms, deliberately generic, were: “ST2” OR “sST2” AND “heart failure” OR “cardiac failure” OR “cardiac dysfunction” OR “cardiac insufficiency” OR “left ventricular dysfunction.”
Inclusion criteria were:
• Follow-up studies, including post hoc analyses of randomized clinical trials;
• Papers in English;
• Enrollment of CHF outpatients (with either reduced or preserved left ventricular ejection fraction);
• Available data, either from papers or from authors, relative to univariate analysis providing median hazard ratio (HR) and to the corresponding 95% confidence interval (CI) for all-cause death, and, possibly, also for cardiovascular (CV) death; and
• The use of last generation, high-sensitivity sST2 assay (Presage ST2 assay, Critical Diagnostics, San Diego, California). At present, this is the only assay that meets the needs of quality specifications of laboratory medicine to have received the Conformitè Européenne Mark and the Food and Drug Administration approval (10).
The exclusion criteria were:
• sST2 not considered per se, but only as an element of a prognostic score; and
• Studies on patients with end-stage HF.
Two reviewers (A.A and G.V.) independently evaluated the studies to be included in the meta-analysis; in case of disagreement, the opinion of a third reviewer (M.E.) was considered.
When a study could not be included only because of lack of crude data on HR and 95% CI, a request was made to the corresponding author to provide the data.
The following data were extracted from each study: first author’s last name, publication year, country where the study was conducted, study period, number of patients enrolled, left ventricular ejection fraction (LVEF) criteria, follow-up duration, outcomes evaluated, and all demographic and clinical baseline data provided. The data were extracted by a reviewer (A.A.), and were checked for accuracy by a second reviewer (G.V.).
Statistical analysis was performed using the R statistical software (version 3.2.3, 2015, metaphor package, R Foundation for Statistical Computing, Vienna, Austria) (11).
The HRs for all-cause and CV death with the corresponding 95% CI were recorded; in all cases, the HRs were calculated for the log2ST2, so that the reported HRs represented the risk per doubling of sST2 (12). Because of the small study number, and the consequent skewed distribution of HRs, the log10 transformation of HRs was performed (11). The summary of effect size was estimated by using both the fixed-effects model and the random-effects model. These models rely on different assumptions (fixed-effects model: same mean effect size, zero between-study variance; random-effects model: different mean effect size, between-study variance), and the use of both has been recommended (11).
The summary of effect size allowed to calculate global log(HR) with its 95% CI. The z statistic was used to evaluate the null hypothesis (sST2 values are not predictive of either all-cause or CV death). The corresponding p value was calculated; a p value <0.05 was considered significant.
The heterogeneity analysis was compatible only with the random-effects model. Due to the small number of studies, the Cochran’s Q test was not employed (13). The following measures were calculated: tau2 (estimated amount of total heterogeneity), tau (square root of tau2), I2 (total heterogeneity/total variability), H2 (total variability/sampling variability), and H (square root of H2); the 95% CI was calculated for all these measures. The jackknife or “leave-one-out” analysis was performed to identify the study contributing most to global HR and the studies with low heterogeneity (I2 <30%), as well as to assess whether the exclusion of any study changed the main conclusion about the predictive value of sST2 (11).
Meta-regression analysis was then performed to search for population characteristics that were unevenly distributed among studies, thus potentially contributing to the different estimations of sST2 predictive value. Finally, funnel plot analysis was used to verify the presence of a “file-drawer problem,” that is, the lack of publication of studies with negative results. The “trim-and-fill” method was used to estimate the number of supposedly unpublished studies (11).
Published data search and study selection
The search process is summarized in Figure 1. Seven studies were finally included (4 reporting HR and 95% CI values in the published paper [12,14–16], and 3 after the authors’ consent to provide unpublished data [17–19]). The included studies had been published between 2011 and 2016, and ranged from 151  to 1,650 patients , with a final population of 6,372 patients. Five studies were subgroup analyses of clinical trials: the Val-HeFT (Valsartan Heart Failure Trial) (17), the Controlled Rosuvastatin Multinational Trial in Heart Failure (16), the Heart Failure: A Controlled Trial Investigating Outcomes of Exercise Training (12), the Use of NT-proBNP Testing to Guide Heart Failure Therapy in the Outpatient Setting trial (18), and the Penn Heart Failure Study (16). Six studies enrolled only patients with reduced LVEF (12,14,15,17–19), whereas one considered also patients with preserved LVEF (16). For 1 paper, the cumulative outcome of all-cause death or heart transplantation was considered, because the author was unable to provide the data relative to all-cause death alone (16). Data for CV death were available for 5 studies, with an overall population of 5,051 patients (12,14,15,17,18). The main population characteristics are summarized in Table 1.
Summary of effect size
When using the fixed-effects model, the global log(HR) was 0.008 (95% CI: 0.007 to 0.010); the global HR was 1.019 (1.016 to 1.023). The z value was 9.77, and the null hypothesis was rejected (p < 0.001). With the random-effects model, the global log(HR) was 0.242 (95% CI: 0.138 to 0.346) (Figure 2); the global HR was 1.75 (95% CI: 1.37 to 2.22). The z value was 4.57. The null hypothesis was then rejected (p < 0.001). When excluding the study by Ky et al. (16), for which only cumulative data on all-cause death and heart transplantation were available, HR values were not affected, being 1.019 (95% CI: 1.015 to 1.023) with the fixed-effects model, and 1.77 (95% CI: 1.32 to 2.37) with the random-effects model.
The studies reporting median sST2 values above (14,19) and below (12,15–17) the established cut-off (35 ng/ml) (10) were then compared. In the studies reporting median sST2 above 35 ng/ml, the global log(HR) was 0.214 (95% CI: 0.177 to 0.250), the global HR was 1.64 (95% CI: 1.50 to 1.78), and the z value was 11.41, with both the fixed- and random-effects model. The null hypothesis was then rejected (p < 0.001). When considering the studies with median sST2 below 35 ng/ml, the global log(HR) was 0.008 (95% CI: 0.006 to 0.010), the global HR was 1.02 (95% CI: 1.01 to 1.02), and the z value was 9.26; the null hypothesis was then rejected with p < 0.001 (fixed-effects model). With the random-effects model, the global log(HR) was 0.254 (95% CI: 0.099 to 0.408), the global HR was 1.79 (95% CI: 1.26 to 2.56), and the z value was 3.22; therefore, the null hypothesis was rejected (p < 0.001).
With the fixed-effects model, the global log(HR) was 0.008 (95% CI: 0.006 to 0.010); therefore, the global HR was 1.019 (95% CI: 1.015 to 1.024). The z value was 8.66, and the null hypothesis was rejected (p < 0.001). With the random-effects model, the global log(HR) was 0.253 (95% CI: 0.086 to 0.420) (Figure 3); the global HR was 1.79 (95% CI: 1.22 to 2.63). The z value was 2.97. Therefore, the null hypothesis was rejected (p < 0.001).
In the studies reporting median sST2 above 35 ng/ml (14,19), the global log(HR) was 0.179 (95% CI: 0.124 to 0.234), the global HR was 1.51 (95% CI: 1.33 to 1.71), and the z value was 6.41, with both fixed- and random-effects model. The null hypothesis was then rejected (p < 0.001). In the studies reporting sST2 below 35 ng/ml (12,15–17) the global log(HR) was 0.008 (95% CI: 0.006 to 0.010); the global HR was then 1.02 (95% CI: 1.01 to 1.02) with the fixed-effects model; the z value was 8.44, and the null hypothesis was rejected (p < 0.001). When using the random-effects model, the global log(HR) was 0.283 (95% CI: 0.062 to 0.503); the global HR was then 1.92 (95% CI: 1.15 to 3.18); the z value resulted 2.51, and the null hypothesis was rejected (p < 0.01).
Meta-regression analysis, assessment of publication bias
The following measures of heterogeneity were calculated with the random-effects model: tau2 = 0.0163 (95% CI: 0.0057 to 0.1140); tau = 0.1278 (95% CI: 0.0753 to 0.3377); I2 = 96.90% (95% CI: 91.56% to 99.54%); H2 = 32.24 (95% CI: 11.85 to 219.31); H 5.68 (95% CI: 3.44 to 14.81). The value of H (significant heterogeneity when the CI lower limit is >1) and I2 (high heterogeneity when >75%) (11) suggested a significant heterogeneity among studies, which was confirmed by the dispersion of studies in the Forest plot (Figure 2). The discrepancy between the HR values calculated through the fixed- and random-effects models further confirmed the heterogeneity among studies; in turn, such heterogeneity justified the preferential consideration of HR values calculated through the random-effects model (11).
At jackknife analysis, the Val-HeFT substudy (17) was the main contributor to the global HR value; no study showed low heterogeneity (I2 <30%), and the exclusion of each study changed the main conclusion about the predictive value of sST2. At meta-regression analysis, all population variables reported in at least 5 studies were considered: publication year of the original study; age; sex; ethnic group; body mass index; New York Heart Association functional classes I/II or III/IV; heart rate; left ventricular ejection fraction; estimated glomerular filtration rate; systolic arterial pressure; prevalence of arterial hypertension; diabetes mellitus; ischemic etiology; follow-up duration; and treatment with angiotensin-converting enzyme inhibitors or angiotensin receptor blockers (ACEi/ARBs), beta-blockers (BBs), mineralocorticoid receptor antagonists (MRAs), diuretic agents, digoxin, plasma N-terminal fraction of pro-brain natriuretic peptide (NT-proBNP), and sST2. The variables emerging as significant sources of heterogeneity were: publication year of the original study (p < 0.01); hypertension (p < 0.001); and therapy with ACEis/ARBs (p < 0.01), BBs (p < 0.001), MRAs (p < 0.001), and digoxin (p < 0.001).
A publication bias was searched only with regard to all-cause death. The presence of such bias was suggested by funnel plot analysis (fixed-effects model). The estimated number of unpublished studies was 4 (p < 0.001).
As in the case of all-cause death, a significant heterogeneity among studies was detected: tau2 = 0.0307 (95% CI: 0.0088 to 0.3488); tau = 0.1752 (95% CI: 0.0940 to 0.5906); I2 = 97.35% (95% CI: 91.34% to 99.76%); H2 = 37.68 (95% CI: 11.55 to 417.66); H 6.14 (95% CI: 3.40 to 20.44). A visual representation of the heterogeneity with respect to the prediction of CV death is provided in Figure 3. The jackknife analysis was repeated, confirming that the study by Anand et al. (17) was the main contributor to the global HR value, that no study had low heterogeneity (I2 <30%), and that the exclusion of each study changed the main conclusion about the predictive value of sST2.
All population characteristics reported in at least 4 studies were considered: publication year of the original study; age; sex; ethnic group; body mass index; New York Heart Association functional classes I/II or III/IV; heart rate; estimated glomerular filtration rate; systolic arterial pressure; hypertension; diabetes; ischemic etiology; follow-up duration; and therapy with ACEi/ARBs, BBs, MRAs, diuretic agents, digoxin, NT-proBNP, and sST2. The following sources of heterogeneity were identified: hypertension (p < 0.01) and treatment with ACEi/ARBs (p < 0.01), BBs (p < 0.01), MRAs (p < 0.05), and digoxin (p < 0.01).
A limited number of studies has explored the clinical value of sST2 assay in ambulatory patients with CHF (6). An exhaustive search of the published data allowed the identification of 7 follow-up studies providing the HR and 95% CI for all-cause death, 5 of which also reported the corresponding data on CV mortality; the global population assessed was quite large (n = 6,372 for all-cause death, and n = 5,051 for CV death). The present meta-analysis indicates that sST2 is an independent predictor of all-cause and CV death, with good performance (median HR values 1.75 and 1.79, respectively).
Research studies in animals have established that the interleukin-33/ST2 signaling axis protects the myocardium against maladaptive hypertrophy, fibrosis, and cardiomyocyte apoptosis, thereby reducing cardiac dysfunction and improving survival (1,4). By sequestering interleukin-33, sST2 exerts negative effects on the myocardium (20,21). In other damaged organs, sST2 attenuates immune responses, thus proving beneficial (as reviewed in ). On this premise, clinical research has addressed the value of the sST2 assay for risk stratification. The results on sST2 prognostic value for all-cause and CV death in the general population are contradictory (22,23). Conversely, in disease settings characterized by local or systemic inflammation/sepsis, a predictive value of plasma sST2 for all-cause mortality is well established (24,25).
sST2 has been established as a biomarker of inflammation and fibrosis (1,2). These processes are active in CHF, underlying adverse ventricular remodeling (23), which has been associated with higher mortality rates (20,24). Our results demonstrate that sST2 predicts both all-cause and CV mortality in CHF, and its prognostic role is independent from the clinical or biohumoral data included in the prognostic scores that have been recently proposed for CHF outpatients (26,27).
In addition to and in combination with other clinical (20) and biohumoral data (21), sST2 emerges as an independent risk indicator in CHF outpatients, with very similar HRs for all-cause and CV death, probably reflecting the fact that CV causes accounted for the majority (75%) of deaths.
Median sST2 values were around the upper reference limit (35 ng/ml) (10) in all included studies. It is worthwhile to note that the predictive value of sST2 is unaffected by different sST2 concentrations: indeed, similar global HR values were found in the 2 subsets reporting median sST2 values either below or above the pre-specified cut-point of 35 ng/ml (10). Finally, HR values for all-cause and CV death were very similar; this result was quite unexpected, given that sST2 exerts opposite effects in the heart and in other organs, but possibly mirrors the fact that CV causes accounted for around 75% of all deaths.
Finally, the use of guideline-recommended pharmacological treatment (ACEi/ARBs, BBs, and MRAs) as well as of digoxin were identified as pivotal sources of heterogeneity. This suggests that sST2 predictive value is greater when currently recommended neurohormonal antagonism is pursued (23). Nevertheless, dedicated studies would be necessary to clarify the effect of therapy on sST2 efficacy as a risk predictor.
The small number of studies, their high heterogeneity, and the potential presence of unpublished studies represent obvious limitations to the exact quantification of the prognostic value of sST2: further studies might allow a more accurate estimation of sST2 efficacy for the prediction of fatal outcome.
The present meta-analysis shows that sST2 is a predictor of both all-cause and CV death in CHF outpatients, further supporting current American College of Cardiology Foundation/American Heart Association recommendations, which have indicated that sST2 assay is a valuable tool for risk stratification in this setting.
COMPETENCY IN MEDICAL KNOWLEDGE: The present meta-analysis provides the advised conceptual support to sST2 assay for risk stratification of patients with stable CHF. The risk of all-cause and CV death increases with plasma sST2 levels. Therefore, sST2 assay can assist the clinician in the identification of CHF outpatients who will benefit from an enhanced therapeutic effort.
TRANSLATIONAL OUTLOOK: The present meta-analysis focuses on the prognostic value of sST2 in current clinical practice, thus not using a “bench-to-bedside” approach. However, a better understanding of sST2 biology could clarify why higher plasma sST2 levels are predictors of worse prognosis, and possibly disclose new therapeutic approaches to cardiac dysfunction.
Dr. Bayes-Genis has received lecture honoraria from Critical Diagnostics. Dr Januzzi has received grant support from Roche Diagnostics, Siemens, Singulex, and Prevencio; consulting income from Roche Diagnostics, Critical Diagnostics, Sphingotec, Phillips, and Novartis; and participates in clinical end point committees for Novartis, Amgen, Janssen, and Boehringer Ingelheim. All other authors have reported that they have no relationships relevant to the contents of this paper to disclose.
John R. Teerlink, MD, served as Guest Editor for this paper.
- Abbreviations and Acronyms
- angiotensin-converting enzyme inhibitor or angiotensin receptor blocker
- chronic heart failure
- confidence interval
- hazard ratio
- left ventricular ejection fraction
- mineralocorticoid receptor antagonist
- soluble suppression of tumorigenesis 2
- Received May 16, 2016.
- Revision received July 7, 2016.
- Accepted September 14, 2016.
- 2017 American College of Cardiology Foundation
- Weinberg E.O.,
- Shimpo M.,
- Hurwitz S.,
- Tominaga S.,
- Rouleau J.L.,
- Lee R.T.
- Yancy C.W.,
- Jessup M.,
- Bozkurt B.,
- et al.
- Viechtbauer W.
- Felker G.M.,
- Fiuzat M.,
- Thompson V.,
- et al.
- Higgins J.P.,
- Thompson S.G.,
- Deeks J.J.,
- Altman D.G.
- Ky B.,
- French B.,
- McCloskey K.,
- et al.
- Anand I.S.,
- Rector T.S.,
- Kuskowski M.,
- Snider J.,
- Cohn J.N.
- Gaggin H.K.,
- Szymonifka J.,
- Bhardwaj A.,
- et al.
- Miller W.L.,
- Saenger A.K.,
- Grill D.E.,
- Slusser J.P.,
- Bayes-Genis A.,
- Jaffe A.S.
- Weinberg E.O.,
- Shimpo M.,
- De Keulenaer G.W.,
- et al.
- Chen L.Q.,
- de Lemos J.A.,
- Das S.R.,
- Ayers C.R.,
- Rohatgi A.
- Hughes M.F.,
- Appelbaum S.,
- Havulinna A.S.,
- et al.
- Lupón J.,
- Sanders-van Wijk S.,
- Januzzi J.L.,
- et al.