Growth Differentiation Factor 15 Predicts All-Cause Morbidity and Mortality in Stable Coronary Heart Disease

Abstract

BACKGROUND

Higher growth differentiation factor 15 (GDF-15) concentrations are associated with cardiovascular (CV) and non-CV morbidity and mortality. However, information on associations between GDF-15 and the risk of specific CV and non-CV events in stable coronary heart disease (CHD) patients is limited.

METHODS

In 14 577 patients with stable CHD participating in the Stabilization of Atherosclerotic Plaque by Initiation of Darapladib Therapy Trial (STABILITY), GDF-15 and other prognostic biomarkers (N-terminal pro–B-type natriuretic peptide, high-sensitivity troponin T, cystatin C, and high-sensitivity C-reactive protein) were measured. In adjusted Cox regression models, the associations between GDF-15 and the composite CV end point [CV death, myocardial infarction (MI), and stroke], as well as other CV and non-CV events, were assessed.

RESULTS

The median concentration (interquartile range) of GDF-15 at baseline was 1253 (915–1827) ng/L. The hazard ratio for the composite end point for the highest compared to the lowest quartile of GDF-15 was 1.8 (95% CI, 1.5–2.2); for CV death, 2.63 (1.9–3.6); for sudden death, 3.06 (1.9–4.8); for heart failure (HF) death, 4.3 (1.3–14); for cancer death, 2.5 (1.3–4.7); for hospitalization for HF, 5.8 (3.2–10); for MI 1.4 (95% CI, 1.1–1.9); and for stroke, 1.8 (95% CI, 1.1–2.8). After adjustment for other prognostic biomarkers, GDF-15 remained significantly associated with all outcomes except for MI.

CONCLUSIONS

In stable CHD, GDF-15 was independently associated with CV, non-CV, and cancer mortality, as well as with MI and stroke. When also adjusting for other prognostic biomarkers, the associations to all fatal and nonfatal events were maintained except for MI. Information on GDF-15, therefore, might be helpful when assessing the risk of adverse outcomes in patients with stable CHD. ClinicalTrials.gov Identifier: NCT00799903

Growth differentiation factor 15 (GDF-15)¹⁸ is normally secreted by the majority of cell types at low levels; however, secretion is increased as a response to an unfavorable milieu such as ischemia and oxidative stress, and during cellular proliferation (1–7). Higher circulating concentrations of GDF-15 are related to age, diabetes mellitus, renal dysfunction, and smoking, possibly as a consequence of increased expression owing to cellular stress and aging. Yet, its function as a protective or disease-inducing factor is unknown (3–5, 8–10). In the general population and in patients with cardiovascular (CV) disease, especially acute coronary syndrome (ACS), GDF-15 has been found to be associated with several outcomes, such as myocardial infarction (MI), heart failure (HF), CV death, as well as with non-CV and cancer death (11–20). Further, GDF-15 is associated with a wide range of CV and cancer risk factors, as well as other biomarkers indicating increased CV risk (8, 11, 18, 21). However, there is a paucity of information on the association between GDF-15 concentrations and specific CV outcomes and cancer in patients with stable coronary heart disease (CHD). In this prospective study, we investigated the association between GDF-15 and long-term risk of CV and non-CV events, with and without adjustment for clinical and biochemical CV risk factors and prognostic biomarkers in a large global population of patients with stable CHD.

Methods

TRIAL DESIGN

The design, rationale, and results from The Stabilization of Atherosclerotic Plaque by Initiation of Darapladib Therapy Trial (STABILITY) have been published previously (https://clinicaltrials.gov/, NCT00799903) (22, 23). Briefly, the trial evaluated the efficacy of darapladib, an oral inhibitor of lipoprotein-associated phospholipase A₂ (Lp-PLA₂), compared to placebo during a median follow-up of 3.7 years assessing incidence of CV events in 15828 patients with stable CHD on optimal secondary preventive treatment. The trial was performed in 39 countries and sponsored by GlaxoSmithKline, designed and supervised by an executive committee, which also is responsible for the design and performance of the current ancillary study. All patients provided written informed consent, and the relevant ethics committees in each participating country approved the study, in accordance with the Helsinki Declaration.

STUDY POPULATION

Patients were eligible to participate in the trial if they had stable CHD, defined as prior MI, prior coronary revascularization (percutaneous coronary intervention, coronary-artery bypass grafting surgery), or multivessel CHD without revascularization. At least 1 of the following additional enrichment criteria was also required: age ≥60 years; diabetes mellitus requiring pharmacotherapy; HDL cholesterol (HDL-C) <40 mg/dL (<1.03 mmol/L); smoker, defined as ≥5 cigarettes per day at study entry or within 3 months before screening; moderate renal dysfunction [estimated glomerular filtration rate (eGFR)≥30 and <60 mL · min⁻¹ · (1.73 m²)⁻¹ or urinary albumin:creatinine ratio ≥30 mg albumin/g creatinine]; or polyvascular disease. Patients were assessed at baseline, at 1, 3, and 6 months after randomization, and every 6 months thereafter until the end of the trial. The investigators were encouraged to treat patients according to international secondary prevention guidelines for CHD, and standards of care were continuously monitored.

STUDY END POINTS

The primary end point was the composite of CV death, nonfatal MI, or nonfatal stroke. Secondary end points were the individual components of the primary end point (CV death, MI, and stroke) and all-cause death. Hospitalization for HF, the composite of CV death and hospitalization for HF, non-CV death, individual components of CV death (sudden death/arrhythmia and HF death), and cancer death were exploratory end points. The exploratory end points were chosen based on prior data associating them with GDF-15 concentrations and a sufficient number of events. All primary and secondary end points and hospitalization for HF were prespecified and defined previously (23), documented and reported by STABILITY study investigators, and adjudicated by an independent clinical events committee (but for cancers, only gastrointestinal cancers were adjudicated).

LABORATORY TESTING

Blood samples were obtained in 14577 patients at baseline. Plasma aliquots were stored at −70 °C until biochemical analyses. The assays of the biomarkers were performed at the UCR Laboratory at Uppsala University, in Uppsala, Sweden, and were blinded to patient characteristics, outcomes, and assigned treatment. GDF-15 was measured with the GDF-15 precommercial assay on a Cobas e 411 analyzer (Roche Diagnostics), composed of a monoclonal mouse antibody for capture and a monoclonal mouse antibody fragment, [F(ab′)2], for detection in a sandwich assay format. Detection was based on an electrochemiluminescence immunoassay using a ruthenium (II) complex label. The interassay coefficient of variation was 2.3% at 100 ng/L and 1.8% at 17200 ng/L; the intraassay coefficient of variation was 0.8% at 1100 ng/L and 0.9% at 18600 ng/L, with a lower detection limit of 10 ng/L. Combination of intraassay and interassay variabilities resulted in laboratory coefficients of variation of 4.4% at 1500 ng/L and 4.5% at 5900 ng/L.

Cardiac troponin T [high-sensitivity (hs)-troponin T], N-terminal pro–B-type natriuretic peptide (NT-proBNP), and cystatin C were determined by an electrochemiluminescence immunoassays using a Cobas e 601 analyzer (Roche Diagnostics). For hs-troponin T, according to the manufacturer, the limit of detection is 5 ng/L and the coefficient of variation is <10% at 14 ng/L, the 99th percentile upper reference limit for healthy individuals. For NT-proBNP, the analytic range is reported by the manufacturer to be 20–35000 ng/L. The upper reference limit (97.5th percentile) in men and women aged 40–65 years is 184 and 268 ng/L, respectively; and for those aged 66–76 years, it is 269 and 391 ng/L, respectively. The lowest concentration measurable with a coefficient of variation <10% is 30 ng/L. For cystatin C, according to the manufacturer, total imprecision is 2.8% at 1.1 mg/L and 2.09% at 4.35 mg/L. The lower limit of detection is 0.41 mg/L.

Lp-PLA₂ activity was measured in an automated enzyme assay system (PLAC^® Test for Lp-PLA₂ activity, Diazyme) with a colorimetric substrate by the manufacturer. The total imprecision coefficients of variation for each reagent lot and sample were <3%. Linearity with a deviation from linearity of ≤10% was demonstrated from 10–382 μmol · min⁻¹ · L⁻¹, the measuring range of the assay. hs-C-reactive protein (hs-CRP) was analyzed using the CardioPhase® hs-CRP (Dade Behring) 2-site particle-enhanced immunonephelometry sandwich assay. Chronic kidney disease was defined as eGFR <60 mL · min⁻¹ · (1.73 m²)⁻¹, where eGFR was derived from cystatin C using the equation furnished by the assay manufacturer: y = 84.69 × cystatin C (mg/L)^−1.68. All routine biochemical analyses and hs-CRP were performed at a central laboratory with standardized methods (Quest Diagnostics Clinical Laboratories, Inc).

STATISTICAL ANALYSES

Demographics and other baseline characteristics were compared across GDF-15 quartile groups. Continuous variables were presented as means and interquartile ranges (IQRs), and groups compared using Kruskal–Wallis tests. Categorical variables were presented as counts and percentages, and groups compared using χ² tests. GDF-15 was logarithmically transformed (natural logarithmic) due to skewed distribution. The relationship between baseline characteristics and baseline GDF-15 was explored by multivariable log-linear regression (linear regression for log-transformed GDF-15 concentrations), to estimate the relative increase in GDF-15 concentration associated with a change in 1 background factor while keeping other factors fixed. The relation of baseline GDF-15 measurements to each clinical outcome is presented as cumulative Kaplan–Meier curves and analyzed with Cox proportional hazards models [hazard ratios (HRs) with 95% CIs] with GDF-15 both with GDF-15 concentration as a continuous variable and with GDF-15 quartile group (Q1–Q4) as a categorical variable. HRs per quartile increase are expressed relative to the lowest quartile (Q1) as reference. All analyses of outcome events were adjusted for randomized treatment. The 2 models were adjusted accordingly:

Model 1: Clinical background characteristics [age, sex, previous MI, geographic region (North America, Eastern Europe, Western Europe, South America, or Asia/Pacific), body mass index (BMI), systolic blood pressure, previous PCI (percutaneous coronary intervention), previous CABG (coronary-artery bypass grafting), multivessel CHD, diabetes mellitus, smoking (current, former, never), polyvascular disease] and routine clinical chemistry variables [hemoglobin, WBC (white blood cell) count, eGFR, LDL-C, HDL-C, triglycerides].

Model 2 includes model 1, with the addition of hs-troponin T, NT-proBNP, cystatin C (replacing eGFR from model 1), hs-CRP.

The functional form of the relationship between GDF-15 and outcomes was explored using restricted cubic splines (4 knots at the 5th, 35th, 65th, and 95th percentiles) to allow for nonlinearity in the relationship with outcomes and interaction with treatment. The assumption of proportional hazards was assessed by visually inspecting unadjusted Kaplan–Meier plots.

P values <0.05 from 2-sided tests were considered statistically significant. The P values were not adjusted for multiple comparisons due to the exploratory nature of the study evaluating associations between end points and GDF-15.

All analyses were performed at the Uppsala Clinical Research Center, Uppsala University, Uppsala, Sweden, using SAS 9.3 and 9.4 (SAS Institute Inc).

Results

BASELINE CHARACTERISTICS

A total of 14577 patients were included, with a median GDF-15 concentration of 1253 ng/L (IQR, 915–1827 ng/L). Higher GDF-15 concentrations were more common in older patients and were associated with risk factors for CV disease and correlated with higher concentrations of NT-proBNP, hs-troponin T, and cystatin C (Table 1). In multivariable adjusted analysis, a higher GDF-15 concentration was associated with higher age, male sex, current smoking, hypertension, diabetes mellitus, renal dysfunction, polyvascular disease, higher concentrations of triglycerides, higher WBC count, and lower concentrations of hemoglobin and HDL-C. See Table 1 in the Data Supplement that accompanies the online version of this article at http://www.clinchem.org/content/vol63/issue1.

CLINICAL OUTCOMES

Primary end point.

The primary end point, the composite of CV death, nonfatal MI, or nonfatal stroke, occurred in 1454 (10.0%) patients. Higher GDF-15 concentrations at baseline were associated with increased event rates of the primary composite end point (Fig. 1A). The relationship between increasing continuous GDF-15 concentrations and events was linear (Fig. 2), with increased risk for the primary end point (model 1; Fig. 3) and this relationship remained after adjusting for prognostic biomarkers (model 2; see online Supplemental Fig. 1). Model performance improved significantly for the prediction of the composite outcome when plasma GDF-15 was incorporated into model 1 (see online Supplemental Table 2).

Cumulative incidence by outcome and GDF-15 quartile.

3-Year event rates by outcome and continuous GDF-15 concentration.

Association between GDF-15 and outcomes.

CV death.

CV death occurred in 662 (4.5%) patients. A continuous separation of event curves by baseline quartiles of GDF-15 was observed throughout the follow-up (see online Supplemental Fig. 3C). The spline graphs showed a slight increase in CV death event rates at GDF-15 concentrations up to 1000 ng/L, after which the event rates further increased up to and beyond 5000 ng/L (Fig. 2). Higher GDF-15 was associated with increased risk of CV death in continuous analyses, as well as when comparing Q4 with Q1 after adjusting for clinical characteristics (model 1; Fig. 3) and prognostic biomarkers (model 2; see online Supplemental Fig. 1). Model performance improved significantly for the prediction of CV death when plasma GDF-15 was incorporated into model 1 (see online Supplemental Table 2).

Of the patients suffering CV death, 324 (49%) died from sudden death, 66 (10%) from MI, and 73 (11%) from HF death. In model 1, higher GDF-15 was associated with an increased risk of sudden death, HF death, and fatal MI (modeling GDF-15 as a continuous variable; Fig. 4, model 2: see online Supplemental Fig. 2).

Association between GDF-15 and death subcategories.

All-cause death.

All-cause death occurred in 1067 (7.3%) patients. Event rates by baseline quartiles of GDF-15 were stable over time (see online Supplemental Fig. 3G). GDF-15 modeled as a continuous variable showed an increase in event rate up to 1000 ng/L, after which the risk increased more prominently (see online Supplemental Fig. 4). After adjusting both for conventional risk factors (model 1; Fig. 3) and prognostic biomarkers (model 2; see online Supplemental Fig. 1), higher GDF-15 concentrations were associated with increasing hazard for all-cause death and patients in the highest quartile (Q4) had up to a 3-fold increased risk as compared with patients in Q1. Model performance improved significantly for the prediction of all-cause death when plasma GDF-15 was incorporated into model 1 (see online Supplemental Table 2).

MI and stroke.

During follow-up, 702 (4.8%) and 286 (2.0%) patients suffered from a fatal or nonfatal MI and stroke, respectively. Estimates of event rates by quartiles of GDF-15 showed a separation of event curves during follow-up (see online Supplemental Fig. 3, A and B). Modeling continuous GDF-15 concentrations in a spline graph suggested a linear increase in risk for MI and stroke between 200 and 5000 ng/L (Fig. 2). In models adjusted for clinical characteristics, higher GDF-15 concentrations were associated with a marginally higher risk of MI (P = 0.0534 in quartile analyses) and stroke (Fig. 3). After adjustment for other prognostic biomarkers, the association of GDF-15 with MI became attenuated and no longer statistically significant (model 2; see online Supplemental Fig. 1). Also, no significant association of GDF-15 was found with fatal MI (see online Supplemental Fig. 2). Associations between GDF-15 modeled as a continuous variable and stroke persisted even after adjusting for other prognostic biomarkers (model 2; see online Supplemental Fig. 1). Incorporating plasma GDF-15 into model 1 significantly improved the model performance for prediction of MI and stroke (see online Supplemental Table 2).

Hospitalization for HF.

Hospitalization for HF and the composite of CV death and hospitalization for HF occurred in 319 (2.2%), and 896 (6.1%) patients, respectively. Higher concentrations of GDF-15 modeled as a continuous variable, as well as in quartile analyses, were associated with an increased risk for both outcomes (see online Supplemental Fig. 3, D and E). The spline graphs showed a small increase in event rate for both outcomes at GDF-15 concentrations up to 1000 ng/L, after which the event rates further increased (Fig. 2; also see online Supplemental Fig. 4). Higher GDF-15 concentrations were associated in a continuous graded fashion with both higher risk of hospitalization for HF and the composite of CV death and hospitalization for HF after adjusting for clinical characteristics (model 1, Fig. 3) and prognostic biomarkers (model 2; see online Supplemental Figs. 1 and 2). Adding plasma GDF-15 to model 1 significantly improved the performance for predicting hospitalization for HF (see online Supplemental Table 2).

Cancer death.

A total of 167 (1.1%) patients died from cancer. Higher quartiles of GDF-15 were associated with increased risk for cancer death (Fig. 1B). The unadjusted relationship between GDF-15 and cancer death was further supported by models adjusting for clinical characteristics (model 1; Fig. 4) and prognostic biomarkers (model 2; see online Supplemental Fig. 2), with a consistently increasing risk of cancer death with increasing GDF-15 (see online Supplemental Fig. 4). The model performance increment when adding GDF-15 to model 1 is shown in online Supplemental Table 3. Cancer death was the major component (53%) of the 317 (2.2%) patients who suffered a non-CV death. The relations between GDF-15 and non-CV death were similar to cancer death (Fig. 4).

Randomized treatment.

No significant interaction of GDF-15 concentration was found with the effects of the randomized treatment on any outcome (data not shown).

Discussion

In this study of patients with stable CHD, we observed strong associations between GDF-15 and all types of CV events, mainly driven by relations to CV death and hospitalization for HF. The relations between GDF-15 and CV death were mainly driven by associations with HF-related deaths, including sudden death. GDF-15 was also associated with non-CV death mainly driven by an association with cancer death. These observations extend findings from other settings and add information regarding specific causes of death related to the GDF-15 concentration (20). Circulating GDF-15 concentration was also directly related to the concentrations of other biomarkers known to predict CV events, such as NT-proBNP, troponins, cystatin C, and CRP. After adjusting for both established risk factors for CV disease and these other prognostic biomarkers, GDF-15 remained an independent indicator of the risk of all types of death, stroke, and hospitalization for HF, but not for MI. These findings were emphasized by the increased c-indices for most end points, indicating that the addition of plasma GDF-15 measurements to information from clinical characteristics and established CV risk factors might further improve risk stratification.

Our results verify the consistent independent association between GDF-15 and mortality (e.g., all-cause, CV, non-CV, and cancer mortality) observed in other populations (8, 16, 18). The results also validate that GDF-15 is associated with non-CV death (e.g., cancer morbidity and mortality) in patients with and without previous cancer diagnoses (19, 24). GDF-15 has also been reported to be a prognostic indicator of both CV events and CV death in patients with ACS (15), chronic HF (11), and atrial fibrillation (21) and in the community-dwelling individuals with or without cardiac disease (15, 19). However, there are only limited data on the independent prognostic value of GDF-15 for specific CV and non-CV events, adjusting for other prognostic biomarkers (25). The present findings, that GDF-15 is an independent indicator of all types of CV events, CV death, and non-CV death with or without adjustment for clinical characteristics and conventional risk factors, are in agreement with other studies of patients with ACS (15, 17) and chronic HF (11). However, the current results indicate that there may be no independent associations between the GDF-15 concentration and coronary thrombotic events, such as MI, after adjusting for other prognostic biomarkers such as NT-proBNP or troponins.

The study clearly showed that in patients with stable CHD, increased GDF-15 concentrations provided incremental information beyond clinical characteristics on the risk of the progression of myocardial dysfunction, stroke, MI, and the risk for CV and non-CV and cancer mortality. However, the risk of new thrombotic coronary events appeared to be better identified by other cardiac biomarkers collinear with the GDF-15 concentrations. The relation between GDF-15 and death does not seem to be related to an increased risk of new coronary ischemic events, but rather may be due to the progression of underlying myocardial dysfunction with a raised risk of developing HF and sudden death, as well as increased risk of fatal neoplasms (19, 26). Although the primary focus in patients with CHD has been the risk of new thrombotic events, such as MI, the availability of a biomarker indicating other simultaneous and competing risks might be of clinical value for selection of preventive treatments. Therefore, in patients with stable CHD treated with optimal preventive measures for coronary ischemic events, the measurement of GDF-15 seems helpful in estimating the remaining risk for other types of CV events and fatal non-CV events, indicating a need for complementary treatments.

Several mechanisms may explain the association between GDF-15 and subsequent adverse outcomes. Higher GDF-15 concentrations are associated in a graded fashion with increased burden of established and newer risk factors for adverse outcomes. Hence, the higher GDF-15 concentration measured could merely be a reflection of the effects of these other factors rather than exerting a direct effect per se on the progression of disease. However, other data support the current findings that GDF-15 is an independent indicator of cellular damage, stress, and age (1, 2, 5, 27).

This study only includes information on GDF-15 concentrations at the baseline examination, which should be representative, as patients with stable coronary artery disease have fairly stable GDF-15 values over several years (8, 28). The analyses of associations between the GDF-15 concentration and individual events were not corrected for the eventual influences of competing events such as death from other causes, which might lead to a risk of overestimating hazards for individual nonfatal outcomes. Further, no corrections for multiple analyses were performed owing to the exploratory nature of this study. Some caution should be applied when extending these results to a general population with stable CHD owing to the inherent differences of the general population from a clinical trial cohort.

In conclusion, in patients with stable CHD, GDF-15 is an independent risk marker associated with CV and non-CV death, including cancer mortality, and with MI and stroke when adjusting for clinical factors. After adjusting for other prognostic biomarkers, associations of GDF-15 with all fatal and nonfatal events were maintained except for MI. Information on GDF-15, therefore, might be helpful when assessing the overall risk of adverse outcomes in patients with stable CHD.

18 Nonstandard abbreviations

 
• GDF-15

  growth differentiation factor 15

 
• CV

  cardiovascular

 
• ACS

  acute coronary syndrome

 
• MI

  myocardial infarction

 
• HF

  heart failure

 
• CHD

  coronary heart disease

 
• STABILITY

  The Stabilization of Atherosclerotic Plaque by Initiation of Darapladib Therapy Trial

 
• Lp-PLA₂

  lipoprotein-associated phospholipase A₂

 
• HDL-C

  HDL cholesterol

 
• eGFR

  estimated glomerular filtration rate

 
• hs

  high sensitivity

 
• NT-proBNP

  N-terminal pro–B-type natriuretic peptide

 
• CRP

  C-reactive protein

 
• IQR

  interquartile range

 
• HR

  hazard ratio

 
• BMI

  body mass index

 
• WBC

  white blood cell.

Acknowledgments

Ebba Bergman, PhD, and Sanne Carlsson, BA BSc, UCR, Uppsala, Sweden, provided editorial support. The authors would like to posthumously acknowledge STABILITY Steering Committee member Liliana Grinfeld, MD, PhD, University of Buenos Aires, School of Medicine, Argentina, for her contributions to the STABILITY trial and this manuscript.

References