Article Text
Abstract
Background: In acute coronary syndrome (ACS), both the Global Registry of Acute Coronary Events (GRACE) score and B-type natriuretic peptide (BNP) predict cardiovascular events. However, it is unknown how BNP compares with GRACE and how their combination performs in ACS.
Methods: The authors recruited 449 consecutive ACS patients and measured admission GRACE score and bedside BNP levels. The main outcome measure was all-cause mortality, readmission with ACS or congestive heart failure (defined as a cardiovascular event) at 10 months from presentation.
Results: Of the 449 patients, 120 patients presented with ST-elevation myocardial infarction (MI) (27%). There were 90 cardiovascular events at 10 months. Both higher GRACE terciles and higher BNP terciles predicted cardiovascular events. There was a significant but only partial correlation between the GRACE score and log BNP (R = 0.552, p<0.001). On multivariate analyses, after adjusting for the GRACE score itself, increasing BNP terciles independently predicted cardiovascular events (second BNP tercile adjusted RR 2.28 (95% CI 1.15 to 4.51) and third BNP tercile adjusted RR 4.91 (95% CI 2.62 to 9.22)). Patients with high GRACE score-high BNP were more likely to experience cardiovascular events at 10 months (RR 6.00 (95% CI 2.40 to 14.83)) compared to those with high GRACE score-low BNP (RR 2.40 (95% CI 0.76 to 7.56)).
Conclusion: In ACS, most but not all of our analyses suggest that BNP can predict cardiovascular events over and above the GRACE score. The combined use of both the GRACE score and BNP can identify a subset of ACS patients at particularly high risk. This implies that both the GRACE score and BNP reflect somewhat different risk attributes when predicting adverse prognosis in ACS and their synergistic use can enhance risk stratification in ACS to a small but potentially useful extent.
Statistics from Altmetric.com
With the advent of new therapeutic approaches for the treatment of patients with acute coronary syndrome (ACS), the need for risk stratification is becoming increasingly important especially to guide key management decisions. Recently, several risk scores have been developed to enable risk stratification on admission. The thrombolysis in myocardial infarction (TIMI)1 and platelet glycoprotein IIb/IIIa in unstable angina: Receptor Suppression Using Integrilin (PURSUIT)2 scores were developed with databases from large clinical trials of NSTE-ACS. The more recent GRACE (Global Registry of Acute Coronary Events) score represents patients across the entire spectrum of ACS and provides robust prediction of the cumulative six-month risk of death or myocardial infarction (MI).3
Another widely acknowledged way of risk stratifying ACS patients that has gained much attention lately is the biomarker B-type natriuretic peptide (BNP).4 BNP and N-amino terminal fragment of the prohormone (NT proBNP) levels sampled up to 7 days after the onset of ACS has been shown to predict both short-term and long-term risk of death, re-infarctions and new congestive heart failure (CHF).5 6 7 8 Previous studies have also demonstrated that BNP predicts mortality independent of certain elements of the GRACE score such as age, ST deviation and cardiac troponin.8 9 However, it is unknown how BNP and GRACE compare and whether adding BNP to the full GRACE score itself would enhance risk prediction in routine ACS patients. In this study, we sought to address this question.
Methods
Study subjects
We deliberately studied patients across the entire spectrum of ACS because the GRACE score was developed and validated in a diverse ACS population. In all, 449 white patients with the diagnosis of ACS (between August 2004 and November 2006) were consecutively recruited from the coronary care unit (CCU) or the cardiology ward, Ninewells Hospital, Dundee. Ethical approval was obtained from the Tayside Committee of Medical Research Ethics and all participating subjects gave written, informed consent. Patients were included if they presented within 72 hours after the onset of ischaemic discomfort and were classified into the following groups:
ST elevation MI (STEMI): ST elevation >1 mm in two limb leads or >2 mm in leads V1-V6 or new left bundle branch block.
Non-ST segment elevation MI (NSTEMI): no ST elevation on electrocardiography (ECG) despite elevated troponin –T >0.03 μg/ml.10
Unstable angina: ischaemic chest pain lasting more than 30 minutes with no evidence of myocyte necrosis or ST elevation but evidence of dynamic ST depression >1.0 mm on ECG.
During the ACS admission, the patients underwent the following:
Clinical history: age, sex, cardiac risk factors, history of ischaemic heart disease or MI and smoking status.
Presenting clinical features: heart rate, blood pressure, Killip class and episodes of cardiac arrest on arrival.
ECG: presence of absence of ST deviation (>0.5 mm).
Bedside BNP assay sampled within 72 hours after the onset of ischaemic chest discomfort. We timed our BNP samples at this time in order to match all previous literature on the predictability of BNP in ACS patients.
Laboratory tests: admission haemoglobin (Hb), estimated glomerular filtration rate (eGFR) using the modification of diet in renal disease (MDRD4) and serum troponin-T level.
Bedside echocardiography: left ventricular systolic dysfunction (LVSD) defined as LV ejection fraction <45% (Simpson’s biplane method).
Calculation of the GRACE score based on clinical history, ECG and laboratory values upon first arrival to the CCU or the acute medical admissions unit.11 We calculated the GRACE score on admission because this is consistent with all previous literature on its predictability despite the BNP being sampled at a different time. We were keen to follow the timing of BNP and the GRACE score in the literature so that any differences in our results from what is in the literature could not be due to differences in timing between our predictors and the literature.
BNP measurement
All samples were collected by venepuncture into ethylene diamine tetra-acitic acid (EDTA) tubes. The blood samples were kept at room temperature and analysed within 4 hours of the draw time. Before analysis, each tube was inverted several times to ensure homogeneity. The whole blood was then analysed with the triage near patient BNP assay (Biosite, USA). The inter-assay percentage CV was 8.8% at 71.3 pg/ml, 11.0% at 629.9 pg/ml and 11.6% at 4088 pg/ml. The detection limit was 5 pg/ml and upper measuring limit was 5000 pg/ml.
Endpoints
The primary endpoint, which is a composite of death from any cause, readmission with ACS or admission with congestive heart failure (CHF), was evaluated at 10 months. The number of endpoints achieved in this study (n = 90) referred to time taken for the occurrence of the first major event. For example, if a patient sustained an ACS event and subsequently died, the endpoint recorded would be the ACS event. We chose to evaluate the endpoints at 10 months because de Lemos et al previously demonstrated that high BNP predicted 10-month mortality in consecutive ACS patients.12 Information on endpoints was collected from the hospital database and patients case notes. The definition of readmission with ACS is as described above (see under Study subjects). Congestive heart failure was defined as hospitalisation for a clinical syndrome involving at least two of the following: paroxysmal nocturnal dyspnoea, orthopnoea, elevated jugular venous pressure, pulmonary crackles, third heart sound and cardiomegaly or pulmonary oedema on chest x-ray. These clinical signs and symptoms must have represented a clear change from the normal clinical status, requiring intravenous diuretics, inotropic support or vasodilator therapy. No patients were lost to follow-up.
Statistical analyses
Patients were divided into terciles of GRACE score and terciles of BNP levels at the time of admission. The mean values and proportion of baseline variables were compared among BNP terciles with ANOVA for continuous variables and χ2 test for categorical variables. BNP was also analysed as a continuous variable when we used logarithmically transformed values because BNP was not normally distributed. The correlation between log BNP and GRACE score was compared using the Pearson’s correlation coefficient. Univariate Cox regression analysis was carried out to look at predictors of mortality or cardiovascular events. Multivariate Cox regression analysis using the backward stepwise method was used to look at the independent predictors of clinical endpoints. We compared the predictive accuracy of the GRACE score, BNP, combined GRACE/BNP score and the TIMI risk index using receiver operating characteristic (ROC) curves, analyses are under the curve (AUC). To do an ROC curve analysis using BNP and GRACE score together, we had to calculate weighted scores for each as follows: (β1 × GRACE score) + (β2 × log BNP), where β1 and β2 denote estimates of β coefficient for the GRACE score and log BNP obtained from the multivariate cox regression model. Event rates for clinical outcomes were also determined using the Kaplan-Meier method and compared using the log-rank test. All statistical analyses were performed using SPSS for Windows version 13.0. A value of p<0.05 was considered to be statistically significant.
Results
Baseline characteristics
The study population consisted of 449 patients: 120 patients presented with STEMI (26.7%), 236 patients with NSTEMI (52.6%) and 93 patients with unstable angina (20.7%). Of the 449 patients, 45% underwent in-patient coronary angiography for which 68% of them then underwent in-patient coronary revascularisation. In the group of STEMI patients who underwent successful thrombolysis (n = 98), only 25% went on to have in-patient coronary angiography. The average duration between index admission and patient undergoing cardiac catheterisation was 48 hours.
The GRACE risk score was calculated from presentation characteristics. The GRACE score ranged from 49 points to 288 points (mean 139 (SD 39) points). Of the 449 patients, 150 were in the low-risk group (first GRACE tercile (<119 points)), 149 in the medium risk group (second GRACE tercile (120–151 points)) and 150 in the high-risk group (third GRACE tercile (>151 points)).
The admission BNP levels ranged from 4 pg/ml to 2390 pg/ml with a mean (SD) of 272 (361) pg/ml and a median of 154 pg/ml. The mean time from the onset of ischaemic symptoms to BNP sampling was 48 (20) hours (median 44 hours). Patients were divided into terciles based on their admission BNP levels (table 1). Patients with ST elevation MI had a significantly higher levels of BNP compared to those who presented with non-ST elevation ACS, 9210 pg/ml (25th–75th percentile: 99.5–417) versus 134 pg/ml (25th–75th percentile: 42.7–304), p<0.001). The prevalence of left ventricular systolic dysfunction was significantly higher in patients presenting with higher BNP levels (p<0.001). Higher BNP terciles was also significantly associated with higher GRACE score (p<0.001).When the GRACE score was plotted against log BNP, a significant but only partial correlation existed between both variables (Pearson correlation of 0.552, p<0.001).
Baseline clinical characteristics according to the terciles of BNP levels
Univariate predictors of clinical endpoints
In total, there were 31 deaths, 41 readmissions with ACS and 18 admissions with CHF at 10 months. We analysed the different clinical endpoints either in isolation or in combination with each other. Table 2 shows the univariate relation between the different GRACE terciles, BNP terciles and clinical endpoints. The highest BNP tercile consistently predicted clinical endpoints. We also analysed log BNP and the GRACE score as a continuous variable in predicting endpoints. As shown in figures 1A and 1B, the distribution of 10-month endpoint rates in the different GRACE score terciles and BNP terciles demonstrated a consistent gradient of risk.
(A) Distribution of 10-month cardiovascular events and mortality according to different terciles of B-type natriuretic peptide (BNP) (p<0.001 for both mortality and cardiovascular events). (B) Distribution of 10-month cardiovascular events and mortality according to different terciles of the GRACE (Global Registry of Acute Coronary Events) score (p = 0.003 for cardiovascular events and p<0.001 for mortality).
Univariate analyses of clinical endpoints
Multivariate predictors of clinical endpoints
We performed two separate multivariate analyses (table 3). On the first analysis, we incorporated the GRACE terciles and BNP terciles into a multivariate model. On the second analysis we included the GRACE score and log BNP as continuous variables in the multivariate model. Regardless of whether BNP was used as a categorical or continuous variable, BNP consistently predicted clinical endpoints independent of the GRACE score in this analysis.
Multivariate analyses of clinical endpoints over and above the GRACE score
Predictive accuracy of the GRACE score and BNP levels (ROC analysis)
We compared the predictive accuracy of the GRACE score, BNP, TIMI risk index (TRI) and the combined use of BNP/GRACE score by using ROC curves. Individual AUC values were obtained for different combinations of clinical endpoints (table 4). BNP demonstrated better discriminatory accuracy in predicting the majority of clinical endpoints compared to the GRACE score or the TRI. This was especially so for ACS/mortality, mortality itself and composite of cardiovascular events. Nevertheless, the AUC values for recurrent ACS/composite CV events were only modest. Interestingly the combined use of BNP and the GRACE score always performed better than GRACE score alone but it only resulted in marginal improvement in risk prediction over BNP alone. However, the differences on this particular analysis were small and not formally statistically significant.
Area under the ROC curve values of clinical endpoints
The GRACE score and BNP as a composite measure
In this analysis, we assessed whether using the GRACE score and BNP synergistically would improve risk prediction. We used a BNP cut-off level of >80 pg/ml, which has been shown in previous ACS studies to predict prognosis.5 For the GRACE score, we used a cut-off of >119 points (upper limit of the first GRACE score tercile) to represent individuals who were at the medium/high risk category in our study population. As demonstrated by the Kaplan-Meier survival curves (fig 2A), patients with a high GRACE score-high BNP on admission were approximately six times more likely to experience a cardiovascular event at 10 months (RR 6.00 (95% CI 2.40 to 14.83)) compared to those with low GRACE score and low BNP levels. Similarly, patients with high GRACE score-high BNP levels on admission were 12 times more likely to die at 10 months (fig 2B) compared to those with low GRACE score-low BNP levels (RR 12.81 (95% CI 1.74 to 94.31)). In addition, patients with high GRACE score-high BNP levels were also more likely to die or experience a cardiovascular event at 10 months compared to those with high GRACE score-low BNP levels (figs 2A and 2B).
(A) Kaplan-Meier survival curves for cardiovascular events. High BNP defined as >80 pg/ml and high GRACE score defined as >119 points. High GRACE-high BNP (RR 6.00 (95% CI 2.40 to 14.83)), low GRACE-high BNP (RR 5.27 (95% CI 1.88 to 14.77)) and high GRACE-low BNP (RR 2.40 (95% CI 0.76 to 7.56)). Log-rank test p<0.001. (B) Kaplan-Meier survival curves for mortality. High BNP defined as >80 pg/ml and high GRACE score defined as >119 points. High GRACE-high BNP (RR 12.81 (95% CI 1.74 to 94.31)), low GRACE-high BNP (RR 4.12 (95% CI 0.37 to 45.39)) and high GRACE-low BNP (RR 3.45 (95% CI 0.31 to 38.07)). Log-rank test p = 0.002. BNP, B-type natriuretic peptide; GRACE, Global Registry of Acute Coronary Events; TIMI, thrombolysis in myocardial infarction.
Reclassification of risk using BNP instead of GRACE score alone
Using GRACE (cut-off 119), 298 patients were classified as high risk and 151 patients as low. If we reclassify risk using BNP alone (80 pg/ml as cut-off), 58 of the high-risk GRACE patients would become low risk, whereas 53 of the low-risk GRACE patients would become high risk—that is, a total of 111 patients will be reclassified representing 25% of the total ACS population. However, if we define the high-risk group as patients with both high GRACE score and high BNP (n = 235), 63 of the 298 patients would be reclassified as low risk, representing 14% of the total population. Alternatively, if high risk is defined as either a high GRACE score or a high BNP, then the number of high-risk patients would increase from 298 to 346—that is, 48 patients are reclassified representing 11% of the total population.
Analyses involving different subgroup of ACS patients
When analysing patients presenting with NSTEACS (n = 325), a more robust association was observed between BNP and clinical endpoints. AUC value for 10-month mortality: GRACE score = 0.822, BNP = 0.848 and combined GRACE−BNP = 0.872. AUC for 10-month CV endpoints: GRACE score = 0.713, BNP = 0.744 and combined GRACE−BNP = 0.760.
When analysing patients presenting with STEMI (n = 120), a non-significant correlation was observed between BNP and clinical endpoints. AUC for 10-month mortality: GRACE score = 0.677, BNP = 0.574 and combined GRACE−BNP = 0.614. AUC for 10-month CV endpoints: GRACE score = 0.429, BNP = 0.557 and combined GRACE−BNP = 0.523.
In the unstable angina subgroup (n = 93), a strong and significant correlation was observed between BNP and clinical endpoints. AUC for 10-month mortality: GRACE score = 0.769, BNP = 0.814 and combined GRACE−BNP = 0.836. AUC for 10-month CV endpoints: GRACE score = 0.707, BNP = 0.744 and combined GRACE−BNP = 0.770.
Discussion
Multivariate analyses and Kaplan-Meier curves suggest that BNP significantly predicts a poor prognosis in ACS patients over and above the GRACE score (although the difference in ROC analyses was not significant). This suggests that both the GRACE score and BNP may reflect somewhat different risk attributes in ACS which is supported by there being only a partial correlation between them. We also demonstrated that a high GRACE score and a high BNP level when used together identified a subset of ACS patients who were at especially high risk of developing future cardiovascular events.
The GRACE score was developed initially to predict in-hospital mortality13 across the entire spectrum of ACS patients but recently its predictive power has also been demonstrated for longer-term risk of death and myocardial infarction in this same patient population.14 15 In addition, the GRACE score was recently found to be superior to TIMI and PURSUIT in non-ST elevation ACS.11 This is the reason why we focused on the GRACE score as well as the fact that the GRACE score is the recommended risk stratifying tool for ACS in our local guidelines.
We deliberately analysed our data in three different but complementary ways because each analysis had its strengths and weaknesses in this complex situation. The multiple regression analysis is best for determining whether BNP independently adds to the GRACE score. The ROC curve helps determine how big an extra contribution BNP makes to the GRACE score’s predictability. Although BNP plus GRACE makes an extra contribution to risk assessment over the BNP alone in multivariate analyses, the ROC shows that this extra contribution is small and non-significant. However, it is worth commenting on a particular weakness of using ROC analysis in this situation. ROC analysis is really designed to compare two single predictors in a head-to-head fashion and is not ideal to compare a combination of predictors versus a single predictor. This limitation can be illustrated by our Kaplan Meier curves where there are clearly three (if not four) different trajectories of risk. Yet ROC analysis insists on describing the data by one dichotomous split which is an artificial constraint in this situation. Weighing each risk marker before the ROC analysis when using two markers together is an attempt to overcome this but it is only a partial solution. These and other limitations of c-statistics and ROC curves have been well described.16 Thus, all three ways of analysing our data should be taken into account with no over-reliance on any single analysis including the ROC. Taking all three analyses into account, the message is that BNP plus GRACE predicts risk better than GRACE alone even if the AUC were not statistically significantly different from each other. In summary, ROC analysis alone suggests the benefit of BNP over GRACE is marginal and non-significant, whereas multivariate analyses and our Kaplan-Meier curves illustrate that BNP plus GRACE can identify risk better than GRACE alone in the majority of patients. This is especially true for the 13% of the ACS patients who have high GRACE but low BNP. The value of adding BNP can also be illustrated by the fact that using BNP in addition to GRACE or instead of GRACE will reclassify risk in 11–25% of all ACS patients, depending on which precise criteria are employed.
Our observation that BNP predicted clinical endpoints independent of the GRACE score in most analyses suggests that BNP’s predictive power reflects additional adverse mechanisms not fully reflected in the GRACE score. One possibly relevant mechanism here is that BNP reflects ischaemia independent of haemodynamics. This is substantiated by the fact that in our study, the benefit of BNP over the GRACE score was slightly greater for the endpoint of ACS/mortality as opposed to the endpoint of heart failure/mortality. This is also consistent with recent studies showing that BNP is higher in patients with multivessel disease, tighter culprit stenosis and left anterior descending involvement.17 18 BNP levels are also known to be higher in those with silent ischaemia.19 20
Our study also demonstrates for the first time that patients with an elevation in only one of BNP or GRACE score have a risk intermediate between those with a high BNP and GRACE score and those with low values for both. This illustrates that risk could be misclassified in some individuals if only one risk predictor is used, especially if the one predictor was the GRACE score.
Before these data, it might have been anticipated that BNP would predict mainly heart failure events while GRACE would predict mainly ACS events. Interestingly, they both appear to predict both ACS and heart failure events. There are possible reasons for this. First, an important component of the GRACE score is the Killip class which obviously reflects heart failure. Second, it has only recently become clear that BNP is manufactured in excess by ischaemic myocardium irrespective of haemodynamic considerations.19 20
The main limitation of this study is patient numbers since the GRACE score did not significantly predict risk independently in every analysis. However, the prognostic role of GRACE score has been firmly established from other studies. Second, both the GRACE score and BNP did not significantly predict CV endpoints in the STEMI subgroup. One possible explanation here is the relatively low event rate in the STEMI group compared to the NSTEACS patients (15% versus 25%). A third limitation worth considering is whether invasive procedures would have influenced the predictability of BNP or GRACE in our study. However, the absolute effect of revascularisation in reducing mortality in high BNP patients is likely to be small—that is, 1.1% in TACTICS TIMI 18, 3.6% in FRISC II and 4.3% in GUSTO IV. Such small absolute changes are unlikely to greatly influence our results. In any case, our main aim was to assess the real world of clinical practice including standard treatments being given for ACS. This will also help future research into tomorrow’s ACS therapies according to the risk remaining after today’s standard ACS treatments are given.
In summary most but not all of our analyses suggest that in a heterogeneous cohort of ACS patients, BNP predicts mortality and cardiovascular events over and above the GRACE score, which implies that BNP and GRACE reflect somewhat different risk attributes in ACS. In addition, we also demonstrated that both BNP and the GRACE score can be used synergistically to identify a subset of ACS patients who are at a particularly high risk of future cardiovascular events. Obviously further data are required to confirm these findings before routine clinical practice should change.
Acknowledgments
The authors would like to thank the British Heart Foundation for generously funding this project.
REFERENCES
Footnotes
Funding British Heart Foundation.
Competing interests ADS is a consultant to a company with a BNP assay.
Provenance and peer review Not commissioned; externally peer reviewed.