Author + information
- Received February 21, 2010
- Revision received May 20, 2010
- Accepted June 21, 2010
- Published online January 4, 2011.
- Peter Damman, MD,
- Marcel A.M. Beijk, MD,
- Wichert J. Kuijt, MD,
- Niels J.W. Verouden, MD,
- Nan van Geloven, MSc,
- José P.S. Henriques, MD, PhD,
- Jan Baan, MD, PhD,
- Marije M. Vis, MD,
- Martijn Meuwissen, MD, PhD,
- Jan P. van Straalen,
- Johan Fischer, PhD,
- Karel T. Koch, MD, PhD,
- Jan J. Piek, MD, PhD,
- Jan G.P. Tijssen, PhD and
- Robbert J. de Winter, MD, PhD* ()
- ↵*Reprints requests and correspondence:
Dr. Robbert J. de Winter, Department of Cardiology, B2–137, Academic Medical Center, Meibergdreef 9, 1105 AZ Amsterdam, the Netherlands
Objectives We investigated whether multiple biomarkers improve prognostication in ST-segment elevation myocardial infarction (STEMI) patients undergoing primary percutaneous coronary intervention.
Background Few data exist on the prognostic value of combined biomarkers.
Methods We used data from 1,034 STEMI patients undergoing primary percutaneous coronary intervention in a high-volume percutaneous coronary intervention center in the Netherlands and investigated whether combining N-terminal pro-brain natriuretic peptide, glucose, C-reactive protein, estimated glomerular filtration rate, and cardiac troponin T improved the prediction of mortality. A risk score was developed based on the strongest predicting biomarkers in multivariate Cox regression. The additional prognostic value of the strongest predicting biomarkers to the established prognostic factors (age, body weight, diabetes, hypertension, systolic blood pressure, heart rate, anterior myocardial infarction, and time to treatment) was assessed in multivariable Cox regression.
Results During follow-up (median, 901 days), 120 of the 1,034 patients died. In Cox regression, glucose, estimated glomerular filtration rate, and N-terminal pro-brain natriuretic peptide were the strongest predictors for mortality (p < 0.05, for all). A risk score incorporating these biomarkers identified a high-risk STEMI subgroup with a significantly higher mortality when compared with an intermediate- or low-risk subgroup (p < 0.001). Addition of the 3 biomarkers to established prognostic factors significantly improved prediction for mortality, as shown by the net reclassification improvement (0.494, p < 0.001) and integrated discrimination improvement (0.0295, p < 0.01).
Conclusions Our data suggest that addition of a multimarker to a model including established risk factors improves the prediction of mortality in STEMI patients undergoing primary percutaneous coronary intervention. Furthermore, the use of a simple risk score based on these biomarkers identifies a high-risk subgroup.
Early primary percutaneous coronary intervention (PPCI) is the preferred treatment for patients presenting with ST-segment elevation myocardial infarction (STEMI). Despite the improvement in both morbidity and mortality in these patients, groups at high risk of complications and adverse clinical events remain. Patients at high risk may benefit from additional mechanical and pharmacological measures obtained during and after PPCI. The ability to differentiate between patients at high and low risk may be a valuable tool to optimize the use of adjunctive therapies, which may improve outcomes.
Several demographic, electrocardiographic, and percutaneous coronary intervention (PCI)-related characteristics have been identified as important prognostic factors regarding mortality in patients with STEMI (1). Furthermore, novel biomarkers such as N-terminal pro-brain natriuretic peptide (NT-proBNP) (2), glucose (3), C-reactive protein (CRP) (4), creatinine or estimated glomerular filtration rate (eGFR) (5), and cardiac troponin T (cTnT) (6) have been reported to be associated with an increased risk for mortality in STEMI patients. These biomarkers reflect left ventricular dysfunction, glucose metabolism, inflammation status, renal function, and myocardial cell damage, respectively.
Consequently, a multimarker approach may yield valuable prognostic information for STEMI patients undergoing PPCI, and a simple multimarker risk score would be of particular value for fast, early assessment of baseline risk. We investigated whether multiple biomarkers, either alone or in combination with established risk factors, improved prognostication in STEMI patients who underwent PPCI using a large database from a high-volume PCI center in the Netherlands.
Source population and procedures
We used data from consecutive STEMI patients who underwent PPCI in our center between January 1, 2005, and January 5, 2007. The PPCI and adjunctive pharmacological treatment was performed according to the American College of Cardiology/American Heart Association and European Society of Cardiology guidelines. In general, patients were eligible for PPCI if they had ischemic chest pain, onset of symptoms no later than 12 h, and at least 1 mm of ST-segment elevation in 2 contiguous leads on the 12-lead electrocardiogram. Patients received aspirin (500 mg), clopidogrel (300 to 600 mg), and unfractionated heparin (5,000 IU). Glycoprotein IIb/IIIa inhibitors were used at the discretion of the operator. If a coronary stent was implanted, clopidogrel was prescribed for 1 month or more to patients with a bare-metal stent and for 6 months or more after a drug-eluting stent.
Blood samples were obtained before PPCI as part of routine clinical care. Blood samples were drawn immediately after insertion of the arterial sheath before PPCI for assessment of cTnT, CRP, glucose, NT-proBNP, and plasma creatinine. Blood samples were centrifuged without undue delay and analyzed. Both cTnT and NT-proBNP were measured using a Hitachi modular E-170 analyzer (Roche Diagnostics GmbH, Mannheim, Germany). CRP was measured with an immunoturbidimetric assay on a Hitachi modular P-800 (Roche Diagnostics GmbH). Glucose and plasma creatinine were measured with an enzymatic assay on a Hitachi modular P-800 analyzer (Roche Diagnostics GmbH). The eGFR was calculated according to the Cockcroft and Gault formula (7).
Our local catheterization laboratory database was consulted for information about patients' demographics and about procedural and angiographic characteristics that had been collected prospectively and entered by interventional cardiologists and specialized nurses. Patients were surveyed at 1 year after PPCI using a mailed questionnaire.
For the current analysis, we included all STEMI patients who underwent PPCI between January 1, 2005, and January 5, 2007, and for whom complete and valid laboratory measurements obtained before PCI were available. Only the first PPCI was included in the case of a patient with multiple PPCIs within the study period. We excluded STEMI patients with cardiogenic shock (n = 85) and patients undergoing rescue PCI after failed thrombolysis (n = 10) (8).
Main outcome measure
The main outcome measure for our current analysis was all-cause mortality before November 6, 2008. Information on vital status obtained from the institutional database was synchronized with the Dutch national population registry and was verified until November 6, 2008.
Normally distributed continuous variables were compared with the Student t test and skewed-distributed with the Wilcoxon rank-sum test. Categorical variables were compared with the chi-square test. The prognostic value of the biomarkers was assessed by investigating the relationship between mortality and biomarkers in 2 sets of Cox proportional-hazards analyses: univariate analyses of single biomarkers and multivariate analysis including all biomarkers (multimarker). The univariate analyses were used to determine cutoff values for the biomarkers. These cutoff values were based on the quartiles of the individual biomarker levels. When hazard did not differ significantly between adjacent quartiles, as indicated by the p value from the Cox model, these were merged.
A multivariate model initially was developed incorporating the 5 biomarkers. Biomarkers with p < 0.05 by the Wald test were identified by backward selection. From the final model, a simplified score was obtained by putting weights to individual biomarkers proportional to the hazard ratio (HR) coefficients. The multimarker risk score was defined as the sum of these points, with higher points indicating a higher mortality risk. Cutoffs for low-, intermediate-, and high-risk groups were determined according to thirds of risk among those who died. Cumulative event rates were estimated with the Kaplan-Meier method and were compared with the log-rank test.
The additional prognostic value of the biomarkers in the multimarker risk score to established prognostic factors was assessed in a multivariate Cox proportional-hazards model. This model included the following established prognostic factors derived from the TIMI score: age, body weight (body mass index), history of diabetes or hypertension, systolic blood pressure and heart rate, anterior myocardial infarction (MI), and time to treatment (symptom onset to first balloon inflation) (1,9). The increased discriminative value after addition of the biomarkers to the established prognostic factors was estimated using 3 measures: the Harrell's C index, net reclassification improvement (NRI), and integrated discrimination improvement (IDI). The Harrell's C index is defined as the proportion of usable patients pairs in which the predictions and outcomes are concordant (10). The NRI and IDI were calculated by analyzing the differences in patients' individual estimated probability of mortality after addition of a single or multiple biomarkers to a model containing the aforementioned established prognostic factors (11). Because no prior risk categories exist for 2-year mortality, we chose Harrell's implementation of the NRI, where the combination of any improved and worsened probability results in the categoryless NRI (Fig. 1). Thus, the reported NRI indicates relatively how many patients improve their predicted probability for mortality. The IDI considers the change in the estimated prediction probabilities as a continuous variable and represents the average improvement in predicted probability. The NRI and IDI were calculated in patients with complete follow-up at 2 years.
The proportional-hazards assumptions of all analyses were assessed with Schoenfeld's tests. No relevant violations were observed.
A total of 1,340 consecutive STEMI patients underwent PPCI at our institution between January 1, 2005, and January 5, 2007. We excluded 306 patients because of missing or incomplete biomarkers. The baseline characteristics of the 1,034 included patients who had complete and valid biomarker measurements are shown in Table 1. The mean age in our study population was 62 years, and 73% percent were male. The baseline characteristics of the included and excluded patients were comparable. There were clinically small, but statistically significant, differences in hypercholesterolemia, history of coronary artery bypass grafting, and prevalence of anterior MI. During a median follow-up time of 901 days (interquartile range: 758 to 1,122 days), 120 patients died (cumulative event rate: 14.1%). All patients were identified in the national population registry.
In the univariate analyses, death rates increased significantly with increasing levels of the biomarkers cTnT, CRP, glucose, NT-proBNP, and eGFR (Table 2). The details of determining the cutoff values are shown in the Online Appendix. Generally, the lowest 2 quartiles were merged because of comparable HR coefficients.
The multivariate analyses are shown in Table 3. cTnT and CRP were not associated with an increased mortality and were excluded from the final model. Within the final model, an eGFR of <90 ml/min, NT-proBNP of 150 ng/l or more, and glucose of 8 mmol/l or more all were associated with significantly higher mortality hazards (p < 0.05, for all). Based on the HR coefficients of the final multivariate model (Table 3), glucose of 8 to 9 mmol/l, NT-proBNP of 150 to 600 ng/l, and an eGFR of 60 to 89 ml/min were assigned 2 points; glucose of 10 mmol/l or more and NT-proBNP of 600 ng/l or more were assigned 3 points; and an eGFR of <60 ml/min was assigned 4 points. The multimarker risk score is shown in Table 4. The Kaplan-Meier curves of the low-risk (multimarker score ≤4), intermediate-risk (multimarker score 5 or 6), and the high-risk (multimarker score >6) groups are shown in Figure 2. Higher mortality at the end of follow-up was observed in the high-risk group (42.0%) when compared with the intermediate-risk (17.4%) or low-risk (5.8%) groups (p < 0.001, log-rank test for all pair wise comparisons).
Biomarkers and established prognostic factors
After adjustment for established risk factors, eGFR, NT-proBNP, and glucose remained significant predictors for mortality (Table 3). An eGFR of <60 ml/min was associated with a 2.80 increased mortality hazard (HR: 2.80; 95% confidence interval: 1.28 to 6.09; p < 0.01). Glucose of 10 mmol/l or more (HR: 2.42; 95% confidence interval: 1.36 to 4.33; p < 0.01) was associated with higher mortality at the end of follow-up. Finally, a 2-fold increase in mortality was observed with NT-proBNP values of 600 ng/l or more (p = 0.02). Adding eGFR, NT-proBNP, and glucose to the established risk factors improved the prediction of mortality, as shown by the increase in the Harrell's C index (Table 5). Reclassification of patients who died or were alive at follow-up is presented by the NRI. Addition of each single biomarker or the 3 markers from our score significantly improved the reclassification of patients (p < 0.001). The integrated discrimination improved significantly after addition of NT-proBNP, eGFR, or the 3 markers.
Because we hypothesized that the relationship between admission glucose levels and mortality differed between diabetics and nondiabetics, we assessed the interaction between glucose and diabetes by testing the significance of this relation in the Cox proportional-hazards model with established risk factors. No interaction was observed (p = 0.11).
Several inferences can be drawn from the current study. First, within multivariate analyses, glucose, NT-proBNP, and eGFR were the strongest predicting biomarkers of mortality. In contrast, cTnT and CRP were not predictive in the multivariate analyses. Incorporation of the 3 predictive biomarkers in the multimarker risk score yields important information regarding baseline risk and mortality. Importantly, as can be seen from Figure 2, the increased mortality occurs early, and additional therapies to improve outcome should start as early as possible, preferably already during or immediately after PPCI. Finally, in our real-world cohort of STEMI patients undergoing PPCI, the addition of eGFR, NT-proBNP, and glucose to a model including established prognostic factors provided incremental prognostic information regarding mortality. This improvement was indicated to be statistically significant by the increase in the NRI and IDI.
Previous multimarker studies
To the best of our knowledge, this is one of the first studies to evaluate a multimarker approach for the prediction of baseline risk of long-term death in patients undergoing PPCI for STEMI. In another study of 298 patients receiving revascularization or thrombolysis for MI, a multimarker approach with NT-proBNP, CRP, matrix metalloproteinase-9, pregnancy-associated plasma protein A, myeloperoxidase, soluble CD40 ligand, and fibrin monomer rendered no additional prognostic information beyond conventional risk stratification tools (12). A key factor in explaining these different results may be the timing of biomarker assessment. In our study, blood samples were obtained before PPCI. In the study by Brügger-Andersen et al. (12), blood samples were obtained 4 to 6 days after the index event. It is likely that biomarker levels have been influenced by the index MI and treatment.
In a community-based cohort of elderly men with or without cardiovascular disease, a multimarker approach substantially improved the prediction of death (13). In this study, addition of individual biomarkers to established risk factors did not result in an improved prediction of mortality. However, addition of a multimarker significantly increased the prediction of mortality, supporting our rationale of adding a combination of biomarkers to improve risk stratification.
eGFR, glucose, and NT-proBNP
Our data emphasize the importance of renal disease, estimated by GFR, as a risk factor for mortality after MI. An eGFR of <60 ml/min was associated with an approximately 3- to 4-fold higher mortality hazard. Possible mechanisms by which renal dysfunction increases mortality risk are progressive renal decline, a high prevalence of coronary risk factors among patients with chronic kidney disease, and so-called therapeutic nihilism (5). Further research may focus on the effect of reperfusion therapy on renal function, as well as the efficacy and safety of cardiovascular medication in these patients. This is particularly important when realizing that renal dysfunction has been an exclusion criterion in most clinical trials.
Numerous studies have identified admission glucose as a predictor of adverse outcomes after acute coronary syndromes (14). However, uncertainty remains regarding whether glucose is a direct mediator of adverse outcomes or an indicator of greater disease severity. Possible pathophysiological mechanisms that may be responsible for the increased mortality with hyperglycemia are direct detrimental effects on ischemic myocardium, microvascular dysfunction, increased inflammation, a prothrombotic state, and an impaired myocardial glucose utilization (14). The prothrombotic state with hyperglycemia seems to result from a combination of increased thrombin generation and platelet activation as well as inhibited fibrinolysis (15). A better understanding of the above-mentioned mechanisms could lead to specific intervention to improve the prognosis regarding mortality.
Finally, NT-proBNP levels, determined by left ventricular function and extent of myocardial ischemia, have been shown to predict mortality better then established risk factors (2). Importantly, previous MI, being a determinant of left ventricular function, was not included in our adjusted multimarker model. We included prior MI in exploratory analyses to make a fairer evaluation of NT-proBNP, and this did not materially alter the HRs. Moreover, brain natriuretic peptide has been shown to predict long-term mortality independently of the presence of clinical evidence of heart failure (16).
cTnT and CRP
Admission cTnT and CRP were not predictive in the multivariate analyses. This in part may be the result of redundancy between several biomarkers. Regarding cTnT, a major limitation of this assay is the low sensitivity at admission because of a delayed increase in circulating levels (17), typically peaking several hours after revascularization. An elevation in CRP levels can be discerned approximately 16 h after the onset of symptoms (18). Complicating the assessment of CRP is the observation that higher admission CRP values are observed in patients with preinfarction unstable angina, compared with those with an unheralded MI (19). We appreciate that assessment of cTnT and CRP at admission does not fully represent the potential prognostic value provided by later assessments. However, our current model is based on admission data and provides valuable prognostic information at the start of PPCI.
Our multimarker score stratified STEMI patients into low-, intermediate-, and high-risk subgroups concerning long-term mortality. Patients were admitted a mean of 2 h after onset of symptoms. Although biomarkers were measured at a later time point offline in our study, such results could have been available during the PPCI procedure with rapid bedside tests. The main advantage of this multimarker score is its simplicity and strong discriminative capacity. When confirmed by other independent cohorts (20), our multimarker score may be useful in identifying patients eligible for adjuvant therapies during or after PPCI. This is particularly important in view of the number of new treatments investigated in STEMI patients, such as ticagrelor and delta-protein kinase C inhibitors (21,22), potentially leading to resource-limited treatment options.
Finally, the use of coronary intervention, fibrinolytic or antithrombotic therapy, and secondary prevention have led to a low mortality rate after STEMI. This means that very large clinical trials would need to be powered adequately to show a mortality benefit for promising new adjunctive therapies. Focusing on the high-risk subgroup may assist in designing modestly size, but adequately powered, trials.
Several limitations of the current study deserve mention. First, there is a potential selection bias because of 306 excluded patients with missing or incomplete biomarkers. We assume that this limitation has not significantly influenced our results, because no major differences were observed in baseline characteristics of included and excluded patients. In the adjusted multimarker, 279 patients were excluded because of missing established risk factor data. However, there were no statistically significant differences between included and excluded patients with the exception of more prior coronary artery bypass grafting procedures in the excluded patients. Second, the single use of data from our PCI center could have caused further selection of patients. Third, the established prognostic factors used in our analyses were derived from the thrombolysis in myocardial infarction risk score (1). Although the thrombolysis in myocardial infarction risk score was derived from a clinical trial population instead of a real-world population, this score has been validated in a community-based STEMI population (23). Finally, information on vital status was obtained from the Dutch national population registry, wherein information on the cause of death is not available.
Our data suggest that the sole use or addition of a multimarker to a model including established risk factors improves the prediction of mortality in STEMI patients undergoing PPCI.
The authors thank all interventional cardiologists and catheterization laboratory nurses for data collection.
For a table on the hazard ratios of the quartiles per biomarker in univariable Cox regression, please see the online version of this article.
The authors have reported that they have no relationships to disclose.
- Abbreviations and Acronyms
- C-reactive protein
- cardiac troponin T
- estimated glomerular filtration rate
- hazard ratio
- integrated discrimination improvement
- myocardial infarction
- net reclassification improvement
- N-terminal pro-brain natriuretic peptide
- percutaneous coronary intervention
- primary percutaneous coronary intervention
- ST-segment elevation myocardial infarction
- Received February 21, 2010.
- Revision received May 20, 2010.
- Accepted June 21, 2010.
- American College of Cardiology Foundation
- Morrow D.A.,
- Antman E.M.,
- Charlesworth A.,
- et al.
- Khan S.Q.,
- Quinn P.,
- Davies J.E.,
- Ng L.L.
- Kosiborod M.,
- Rathore S.S.,
- Inzucchi S.E.,
- et al.
- Ortolani P.,
- Marzocchi A.,
- Marrozzini C.,
- et al.
- De Luca G.,
- Suryapranata H.,
- Ottervanger J.P.,
- Antman E.M.
- Deedwania P.,
- Kosiborod M.,
- Barrett E.,
- et al.
- Undas A.,
- Wiek I.,
- Stepien E.,
- Zmudka K.,
- Tracz W.
- Anderson J.L.,
- Adams C.D.,
- Antman E.M.,
- et al.
- de Winter R.J.,
- Fischer J.C.,
- de Jongh T.,
- van Straalen J.P.,
- Bholasingh R.,
- Sanders G.T.
- Liuzzo G.,
- Baisucci L.M.,
- Gallimore J.R.,
- et al.
- Hlatky M.A.,
- Greenland P.,
- Arnett D.K.,
- et al.
- Bates E.,
- Bode C.,
- Costa M.,
- et al.