We studied consecutive patients who were diagnosed with acute PE at the University Hospital of Goettingen (Goettingen, Germany) between October 2004 and March 2007 and gave informed consent. In accordance with existing guidelines (12–14), all patients presenting at the emergency room with a high clinical probability of PE, or with low/intermediate probability and a positive (>500 μg/L) D-dimer ELISA test (15), underwent contrast-enhanced computed tomography, pulmonary angiography, ventilation–perfusion lung scan, and/or sonographic or phlebographic examination of the leg veins to confirm the diagnosis. Overall, the diagnosis of acute venous thromboembolism was confirmed by contrast-enhanced computed tomography in 95 patients (77% of the study population), ventilation–perfusion lung scan in 21 (17%), and sonographic or phlebographic examination of the leg veins in 60 (49%). Sixty patients underwent two or more imaging procedures.

The study protocol strongly recommended a transthoracic echocardiogram within 2 hours of PE diagnosis. Of 112 patients (91% of the study population) who underwent cardiac ultrasound, 47 (42%) were diagnosed with right ventricular dysfunction. The latter finding was prospectively defined as dilatation of the right ventricle (end-diastolic diameter greater than 30 mm from the parasternal view, or a right ventricle:left ventricle diameter ratio greater than 1.0 from the subcostal or apical views), combined with right atrial hypertension (absence of the inspiratory collapse of the inferior vena cava) in the absence of left ventricular or mitral valve disease (9, 16).

According to the study protocol, and as described elsewhere (17), preliminary consent was obtained from all patients who were admitted with clinical suspicion of acute PE and/or a positive D-dimer test, and blood was drawn immediately for measurement of baseline (admission) biomarker levels before further diagnostic workup. After confirmation of the diagnosis, patients were asked to sign the informed consent form. Subsequently, complete data on baseline clinical, hemodynamic, and laboratory parameters were obtained using a standardized questionnaire (9, 16) by investigators unaware of the patient's biomarker levels. Cardiogenic shock on admission was defined by at least one of the following conditions: need for cardiopulmonary resuscitation, systolic blood pressure less than 90 mm Hg or drop of systolic blood pressure by at least 40 mm Hg for at least 15 min with signs of end-organ hypoperfusion, or need for catecholamine administration to maintain adequate organ perfusion and a systolic blood pressure of 90 mm Hg or more. Patients were excluded from analysis and their blood samples were discarded if they refused to give consent, if PE could not be confirmed by imaging studies, or if PE was an accidental finding obtained at diagnostic workup for another suspected disease.

Thirty-day clinical follow-up data were obtained from all patients included in the study. A complicated 30-day outcome was defined as PE-related death or at least one of the following major adverse events: need for catecholamine administration (except for dopamine at a rate of 5 μg/kg/min or less) to maintain adequate tissue perfusion and prevent or treat cardiogenic shock, endotracheal intubation, or cardiopulmonary resuscitation. The cause of death (PE related vs. non–PE related) was adjudicated by two of the investigators (M.L. and S.K.) who reviewed the patients' records and the results of autopsy, when performed. Long-term survival was assessed by means of a clinical examination or a telephone conversation with the patient or his/her treating physician at 6-month intervals after PE diagnosis.

The study design was observational and biomarker levels were not used to guide patient management or to monitor the effects of treatment during the initial hospital stay or at any time during the follow-up period. The study protocol was approved by the independent Ethics Committee of the University of Goettingen.

Venous plasma and serum samples were collected on admission and immediately stored at −80°C. Samples were later analyzed in batches after a single thaw. Plasma levels of GDF-15 were measured by a new immunoradiometric assay as described (18). The assay has a linear range from 200 to 50,000 ng/L (18). Cardiac troponin T (cTnT) and NT-proBNP levels were determined in plasma samples using quantitative electrochemiluminescence immunoassays (Elecsys 1010/2010 analyzer; Roche Diagnostics, Mannheim, Germany) as described (9, 16). The cTnT assay has a lower detection limit of 0.01 μg/L. As proposed by the manufacturer and reported previously (9, 16), a level of 0.04 μg/L or more was used to define elevated cTnT concentrations. The NT-proBNP assay has 20 ng/L as the lower detection limit. Creatinine measurements were performed at the Department of Clinical Chemistry, University of Goettingen, using standard laboratory techniques. The glomerular filtration rate was estimated using the Modification of Diet in Renal Disease study equation, and renal dysfunction was defined as a glomerular filtration rate below 60 ml/minute/1.73 m² body surface area (19, 20). All investigators who determined biomarker levels were unaware of patient baseline parameters or clinical course.

The modified Kolmogorov-Smirnov test (Lilliefors test) was used to test for a normal distribution of continuous variables. Continuous variables are expressed as medians with corresponding 25th and 75th percentiles and were compared by using the Mann-Whitney U test. Categorical variables were compared using Fisher's exact test. Spearman rank correlation was used to identify baseline variables related to the levels of GDF-15. The prognostic relevance of GDF-15 levels and other baseline parameters with regard to 30-day outcome was estimated using logistic regression analysis after logarithmic transformation of the continuous variables. Odds ratios and 95% confidence intervals were calculated. To identify predictors of long-term mortality, Cox's proportional hazards regression analysis was performed using Wald's test; the results are presented as hazard ratios with corresponding 95% confidence intervals. Prognostically relevant cutoff values of the biomarkers were derived from receiver operating characteristic curve analysis, which also was used to determine the areas under the curve (c-statistic). On the basis of the calculated cutoff values of GDF-15 (4,600 ng/L) and NT-proBNP (1,000 ng/L), and the previously validated (9, 16) reference level for cTnT (0.04 μg/L), survival rates were estimated by the Kaplan-Meier method, and statistical comparison was performed using the log-rank test. All tests were two sided and used a significance level of 0.05. All analyses were performed with SPSS 14.0 software (SPSS, Chicago, IL).

Of 137 screened patients with clinical suspicion of acute PE and/or a positive D-dimer test, 7 were excluded from analysis because they refused to give consent, and 7 because the diagnostic workup failed to provide definite confirmation of PE. Thus, the study population comprised 123 patients. Their baseline characteristics and biomarker levels are summarized in Table 1. GDF-15 concentrations on admission ranged from 553 to 47,274 ng/L with a median value of 2,196 (25th to 75th percentiles, 1,333 to 3,457) ng/L. Overall, 101 patients (82% of the study population) had a GDF-15 level above 1,200 ng/L, which has been reported as the upper limit of normal in apparently healthy elderly individuals (18). As shown in Table 1, patients with GDF-15 levels above the median were older and more likely to present in cardiogenic shock, or to have a diagnosis of chronic heart failure, diabetes mellitus, cancer, or renal dysfunction. They also had higher NT-proBNP concentrations on admission.

During the first 30 days, 17 patients (14%) developed major complications; 14 of those died of acute PE. As shown in Figure 1, baseline plasma levels of GDF-15 were significantly higher in patients with a complicated 30-day outcome (median, 6,039 [25th to 75th percentiles, 2,778 to 19,722] ng/L) compared with those with an uncomplicated course (median, 2,036 [25th to 75th percentiles, 1,279 to 3,176] ng/L; P < 0.001). When compared with other biomarkers such as cTnT and NT-proBNP, less overlap was found between the GDF-15 levels of patients with an uncomplicated outcome and those with a complicated outcome (Figure 1; compare left with middle and left with right, respectively). Of note, none of the 22 patients (18%) with GDF-15 levels below the reference value of 1,200 ng/L developed major complications during the first 30 days.

Figure 1. Baseline biomarker levels in patients who developed major complications as opposed to those with an uncomplicated 30-day outcome. Levels of growth differentiation factor (GDF)-15, cardiac troponin T (cTnT), and N-terminal propeptide of B-type natriuretic peptide (NT-proBNP) are expressed as box (25th percentile, median, and 75th percentile) and whisker (10th and 90th percentiles) plots. Note same values for 25th and 50th percentiles in the left box and whisker plot in the middle panel.

Download Figure | Download Powerpoint

[More] [Minimize]

Univariable logistic regression analysis (Table 2, left) indicated a 5.0-fold increase in the risk of a complicated 30-day outcome for each increase by 1 standard deviation of ln (natural log) GDF-15. Cardiac TnT and NT-proBNP also predicted a complicated 30-day outcome in the above-described model, albeit with a less pronounced effect (odds ratios, 1.7 and 1.4, respectively). By multivariable analysis (Table 2, right), baseline levels of GDF-15 and cTnT, and cardiogenic shock on admission, emerged as independent predictors of a complicated 30-day outcome. To further assess the role of GDF-15 as a prognostic biomarker, we performed sensitivity analyses (logistic regression), in which we compared GDF-15 levels on admission with each one of the other baseline parameters univariably associated with a complicated course. In all of these tests (comparing only two prognostic variables each), GDF-15 remained a significant independent predictor of a complicated 30-day outcome (data not shown).

Receiver operating characteristic curve analysis further illustrated that GDF-15 is a strong indicator of 30-day outcome in acute PE (Figure 2). The calculated area under the curve (c-statistic) for GDF-15 was 0.84 (95% confidence interval, 0.76 to 0.90), which compared favorably with the c-statistics for cTnT (0.72; 95% confidence interval, 0.63 to 0.80; P = 0.18 vs. GDF-15) and NT-proBNP (0.65; 95% confidence interval, 0.56 to 0.73; P = 0.022 vs. GDF-15).

Figure 2. Prognostic sensitivity and specificity of growth differentiation factor (GDF)-15, cardiac troponin T (cTnT), and N-terminal propeptide of B-type natriuretic peptide (NT-proBNP). Receiver operating characteristic curves showing the prognostic sensitivity and specificity of GDF-15, cTnT, and NT-proBNP levels on admission with regard to a complicated 30-day outcome. AUC = area under the receiver operating characteristic curve.

Download Figure | Download Powerpoint

[More] [Minimize]

A concentration of 4,600 ng/L was identified by receiver operating characteristic curve analysis as the best cutoff level for GDF-15 in our study population. This value was associated with a prognostic sensitivity of 0.71, a specificity of 0.90, a positive predictive value of 0.52, and a negative predictive value of 0.95. Of the 23 patients presenting with GDF-15 concentrations above the calculated cutoff value, 12 (52%) developed complications during the first 30 days, as opposed to only 5 of 100 patients (5%) with GDF-15 levels less than 4,600 ng/L (P < 0.001).

Next, we investigated whether GDF-15 can provide prognostic information additive to that of the previously studied biomarkers, cTnT and NT-proBNP, during the acute phase of PE. As shown in Table 3, elevated levels of cTnT or NT-proBNP on admission were associated with a significant but modest increase in the risk of a complicated 30-day outcome (odds ratios of 3.7 and 4.4, respectively). Importantly, however, the risk of complications predicted by either biomarker was much higher when elevation of the GDF-15 level was also taken into account (odds ratios of 17.7 and 17.3, respectively). In contrast, the combination of cTnT with NT-proBNP did not appear to provide additive prognostic information as compared with either biomarker alone (Table 3).

As biomarkers and imaging procedures may complement each other in the risk stratification of acute PE (7), we also tested the (additive) prognostic value of GDF-15 in combination with echocardiographic findings and compared it with that of cTnT and NT-proBNP. None of the 55 patients with GDF-15 levels less than 4,600 ng/L on admission and absence of right ventricular dysfunction on ultrasound developed a major complication during the first 30 days (Figure 3, top). The additive prognostic value of the other biomarkers was almost as good, as only 2 of 53 patients with baseline cTnT levels less than 0.04 μg/L and no right ventricular dysfunction, and only 1 of 36 patients with NT-proBNP levels less than 1,000 ng/L and no right ventricular dysfunction, developed PE-related complications during the acute phase (Figure 3). At the other end of the risk spectrum, 7 of 11 patients with GDF-15 levels equal to or exceeding 4,600 ng/L and right ventricular dysfunction had a complicated 30-day course (corresponding to an odds ratio of 15.9) (Table 4 and Figure 3). By comparison, the odds ratio for cTnT elevation equal to or exceeding 0.04 μg/L in combination with right ventricular dysfunction was 3.0, and the odds ratio for NT-proBNP elevation equal to or exceeding 1,000 ng/L in combination with right ventricular dysfunction was 3.7 (Table 4 and Figure 3).

Figure 3. Combination of growth differentiation factor (GDF)-15, cardiac troponin T (cTnT), or N-terminal propeptide of B-type natriuretic peptide (NT-proBNP) elevation with evidence of right ventricular dysfunction by echocardiography. The number of patients with complications and the overall number of patients are given, along with percentages, for each column. Biomarker levels were dichotomized, and elevated concentrations were defined as those equal to or exceeding 4,600 ng/L for GDF-15, equal to or exceeding 0.04 μg/L for cTnT, and equal to or exceeding 1,000 ng/L for NT-proBNP. RV = right ventricular.

Download Figure | Download Powerpoint

[More] [Minimize]

Long-term follow-up data on survival were available for all but one patient (n = 122). The median follow-up time was 287 (25th to 75th percentiles, 188 to 826) days; 29 patients (24%) died during this period. Overall, 15 deaths (48%) were due to the initial or recurrent episodes of PE, 7 (24%) were due to malignancies, and 5 (17%) were due to heart failure; in 2 cases, the cause of death could not be identified with certainty. Patients who died had higher baseline levels of GDF-15 (median, 4,977 [25th to 75th percentiles, 2,954 to 14,265] ng/L) as compared with survivors (median, 1,808 [25th to 75th percentiles, 1,238 to 2,939] ng/L; P < 0.001). Univariable Cox regression analysis (Table 5, left) indicated a 2.8-fold increased risk of death for each increase by 1 SD of ln GDF-15 (the corresponding values were 1.4-fold for either cTnT or NT-proBNP). By multivariable analysis (Table 5, right), baseline levels of GDF-15 (but not cTnT or NT-proBNP) and chronic heart failure emerged as independent predictors of long-term mortality. Additional sensitivity analyses confirmed that GDF-15 remained a significant predictor of long-term mortality when it was directly compared with each of the other baseline parameters associated with long-term mortality by univariable analysis (data not shown).

Kaplan-Meier analysis further illustrated that patients with GDF-15 levels of 4,600 ng/L or more on admission had a significantly decreased probability of long-term survival (Figure 4, top). The survival curves of patients with GDF-15 levels less than 4,600 ng/L, versus 4,600 ng/L or more, separated early after admission to the hospital and continued to diverge beyond the acute phase of PE. A similar pattern, albeit with a less pronounced difference, was observed for NT-proBNP (Figure 4, bottom). By contrast, cTnT elevation at presentation did not appear to predict mortality beyond the acute phase of PE (Figure 4, middle).

Figure 4. Probability of long-term survival in patients with or without elevation of growth differentiation factor (GDF)-15, cardiac troponin T (cTnT), or N-terminal propeptide of B-type natriuretic peptide (NT-proBNP). Biomarker levels were dichotomized, and elevated concentrations were defined as those equal to or exceeding 4,600 ng/L for GDF-15, equal to or exceeding 0.04 μg/L for cTnT, and equal to or exceeding 1,000 ng/L for NT-proBNP. P values were calculated by the log-rank test.

Download Figure | Download Powerpoint

[More] [Minimize]

GDF-15 is emerging as a new biomarker in patients with cardiovascular disease. GDF-15 was found to be a strong predictor of adverse events in patients with non–ST-elevation acute coronary syndrome (5) or with chronic heart failure associated with left ventricular systolic dysfunction (6). The present study shows that the prognostic utility of GDF-15 may extend to patients with acute PE. In our study population of 123 consecutive patients with confirmed PE, elevated levels of GDF-15 on admission were strongly and independently related to an increased risk of death or major complications during the first 30 days after diagnosis. Importantly, the prognostic information provided by GDF-15 appeared to be additive to that of the established biomarkers cTnT and NT-proBNP, and to echocardiographic findings of right ventricular dysfunction. Moreover, GDF-15 emerged as an independent predictor of long-term mortality.

The strategies for risk stratification of acute PE continue to evolve. Data on normotensive patients with acute PE support the value of echocardiographic findings in the diagnosis of acute right ventricular dysfunction (21), and evidence derived from retrospective studies (22–24) suggests a similar prognostic role for the detection of right ventricular enlargement on the four-chamber axial view of a computed tomography scan. However, imaging modalities are not ubiquitously available and probably not sensitive enough for detecting minor injury or compromise of right ventricular function. Furthermore, and importantly, cardiac ultrasound or computed tomography may not account for the impact of comorbidities on short-term and, particularly, long-term prognosis after acute PE. The cardiac biomarkers cTnT and NT-proBNP were previously found to possess high prognostic sensitivity and negative predictive value in acute PE (8, 9). Thus, they appear useful for ruling out potentially life-threatening right ventricular injury or dysfunction. At the other end of the severity spectrum, biomarker elevation may assist echocardiography in identifying patients with a high risk of death or complications (9, 25–28). However, cardiac troponin elevation may not be present on admission, because it does not occur until 6 to 12 hours after the onset of symptoms (9, 29). In addition, cutoff levels for defining a prognostically relevant elevation of NT-proBNP remain to be determined (9, 30, 31). It also needs to be mentioned that existing biomarkers do not integrate the prognostic impact of coexisting extracardiac disease; moreover, there is lack of evidence that their predictive value regarding survival extends beyond the acute (in-hospital) phase of PE.

Experimental studies in mice found that myocardial expression of GDF-15 rapidly increases during pressure overload and remains upregulated in the hypertrophied and failing heart (4). Accordingly, the circulating levels of GDF-15 may be related, at least in part, to acute and chronic increases in right or left ventricular load. On the other hand, GDF-15 is not exclusively expressed in cardiomyocytes, and upregulation of the cytokine has also been observed in chronic inflammatory disease states and cancer (3, 32, 33). In the present study, chronic heart failure and cardiogenic shock on admission, but also extracardiac diseases including cancer, diabetes, and renal dysfunction, were associated with elevated levels of GDF-15. This is consistent with previous observations in patients with non–ST-elevation acute coronary syndrome, in whom the levels of GDF-15 also were closely related to preexisting heart failure, diabetes, and renal dysfunction (5). Consequently, the myocardial specificity of GDF-15 is probably lower than that of NT-proBNP, the cardiac troponins, or heart-type fatty acid–binding protein (17), and elevated levels of GDF-15 cannot be interpreted as exclusively indicating right ventricular overload or dysfunction. Interestingly, however, this apparent lack of specificity may represent a strength rather than a limitation of this biomarker, because GDF-15 appears capable of integrating several clinical and biochemical indicators of severe disease and/or poor prognosis. In the present study, this was reflected by the relatively large area under the receiver operating characteristics curve, and the high, independent prognostic value of baseline GDF-15 levels at logistic regression analysis. At long-term follow-up, the Kaplan-Meier survival curves of patients with elevated versus low GDF-15 levels on admission continued to separate beyond the acute phase of PE, a pattern that was similar to (but more pronounced than) that of NT-proBNP. In contrast, the prognostic impact of cTnT elevation appeared to be confined to the first few days after PE diagnosis, supporting the role of cTnT as an “acute phase” indicator (8).

Studies in patients with acute coronary syndromes suggest that a multimarker approach may improve risk stratification by providing information about pathophysiologically distinct processes (34, 35). Our results in patients with acute PE indicate that GDF-15 adds prognostic information to previously studied biomarkers. In particular, the likelihood of a complicated outcome was increased when elevation of either cTnT or NT-proBNP was combined with GDF-15 levels. Notably, and in contrast to a previous report (36), combination of cTnT with NT-proBNP did not appear to improve prognostic assessment. Finally, our data suggest that measurement of GDF-15 levels may be a helpful step in risk stratification algorithms that combine echocardiography with biomarker testing in acute PE. In this regard, the information added by GDF-15 in patients with evidence of right ventricular dysfunction appeared to be superior to that of the established biomarkers cTnT or NT-proBNP. Of note, the cutoff concentration (4,600 ng/L) used to demonstrate the prognostic utility of GDF-15, alone or in combination with other biomarkers or imaging tests, was derived from receiver operating characteristic curve analysis of the study population. Prospective validation in an external cohort will be necessary to confirm the positive and negative predictive value of this cutoff level in acute PE.

The results of our study point to GDF-15 as a promising new biomarker for risk stratification of patients with acute PE. On the basis of these findings, we believe that the prognostic value of GDF-15 in this setting deserves further investigation. Prognostic algorithms using multimarker approaches, or the combination of GDF-15 with imaging procedures (echocardiography or computed tomography), need to be validated in larger numbers of patients with PE. From the therapeutic perspective, the critical question is whether GDF-15, being a sensitive global indicator of poor outcome in acute pulmonary embolism, can assist imaging studies or myocardium-specific biomarkers in identifying (1) normotensive high-risk patients with right ventricular dysfunction who may benefit from early thrombolytic or surgical treatment, and (2) patients in whom closer long-term follow-up may help prevent late deaths.