Risk Stratification and Therapeutic Decision Making in Acute Coronary Syndromes

Management of acute coronary syndromes has advanced considerably during recent years, similar to the evolution in treatment of acute myocardial infarction (MI) in the 1980s. Then, fibrinolytic and aspirin therapy were shown to reduce mortality substantially.¹ Although these therapies first were applied to all patients with suspected MI,¹ those with ST-segment elevation or left bundle-branch block were shown to derive most, if not all, of the benefit.² Accelerated tissue-type plasminogen activator (tPA) then was found to be superior to streptokinase in the Global Utilization of Streptokinase and tPA for Occluded Coronary Arteries (GUSTO-I) trial, with a relative 14% reduction in 30-day mortality.³ Mortality models suggested that tPA treatment independently predicted better outcomes, consistent across patient subgroups regardless of absolute risk.⁴ Without a heterogeneous treatment effect, the interpretation was that almost all patients with MI would benefit from tPA treatment.

Risk models for mortality have been central to current understanding of what is appropriate reperfusion therapy for MI. It was understood early that some factors—age, infarction location, and signs of heart failure—were highly predictive of early and late mortality.^{4 - 6} Most of the variables were relatively easy to characterize and, thus, could be used in simple prediction rules. Age, however, appeared to follow a nonlinear relationship with mortality, suggesting a disproportionately higher risk in very elderly vs slightly younger patients (a 30-day mortality of 30.3% for patients >85 years old and 9.5% for patients 65-74 years old).⁴ Moreover, some of the most powerful predictors of mortality were continuous, objective hemodynamic measures, such as blood pressure and heart rate.⁴ Their nonlinear relationships with mortality made simple calculations difficult, although several scoring systems have incorporated such variables.^{6 - 7}

Physicians know that it is impossible to remember all the facts of medicine. Clinicians are taking advantage of new, handheld computer-based technologies to add programs such as pharmacopeias and clinical management guidelines to daily practice. With these types of technology, more complex and accurate scoring systems will be universally available and user-friendly. The question arises, however, about the standards for such systems.

First, to apply to practice, the cohort from which a risk score is derived must represent a general population with acute coronary syndromes. Second, the risk model must provide accurate predictions. Predictive accuracy typically has been described by the C (for concordance) statistic, which defines how well a model or prediction rule can discriminate between patients who do and do not have an event.⁴ For a binary end point, the C statistic is the proportion of all pairs of patients, 1 with and 1 without the outcome, in which the patient with the outcome had the higher predicted probability of an event. Thus, the C statistic is a measure of how well a clinical prediction rule can correctly rank-order patients by risk. A model that accurately discriminates patients 85% of the time would have a C statistic of 0.85; with completely random predictions, as in a coin toss, the C statistic would be 0.5; and a model that discriminates perfectly between patients with and without events would have a C statistic of 1.0.⁸ For a clinical prediction rule, it is generally considered that a C statistic of less than 0.6 has no clinical value, 0.6 to 0.7 has limited value, 0.7 to 0.8 has modest value, and greater than 0.8 has discrimination adequate for genuine clinical utility. Of note, a risk score may have a statistically significant association with a clinical outcome, but the relationship may not discriminate enough to allow clinicians to accurately and reproducibly separate patients who will and will not have the outcome. Furthermore, the C statistic value almost always is higher when assessing predictive accuracy in the patient data set used to develop the model rather than independent sets of patients.

Models have been developed to predict the risks of death; death or nonfatal MI; and death, nonfatal MI, or severe recurrent ischemia in patients with acute coronary syndromes without ST-segment elevation, a larger, more heterogeneous group than patients with ST-segment elevation.⁹ Models from general populations include those from the Platelet Glycoprotein IIb/IIIa in Unstable Angina: Receptor Suppression Using Integrilin (Eptifibatide) Therapy (PURSUIT) trial of 9461 patients.¹⁰ The best predictors of mortality within 30 days included advanced age, female sex, increased heart rate, lower systolic blood pressure, severity of prior angina, ST-segment depression, and signs of heart failure. The model for 30-day mortality from the GUSTO-IIb trial of 12,142 patients¹¹ contained similar variables (age, systolic blood pressure, ST-segment changes, and heart failure), but some were unique to GUSTO-IIb, including hypertension, peripheral arterial disease, and smoking. In both models, cardiac enzymes (creatine kinase or its MB fraction) interacted with predictor variables to identify a higher-risk cohort.

A recent predictive model for 30-day mortality or nonfatal MI from the Efficacy and Safety of Subcutaneous Enoxaparin in Unstable Angina and Non-Q-Wave MI trial (ESSENCE) reflected data from 3171 such patients.¹² Its most predictive variables were almost identical to those in the PURSUIT and GUSTO-IIb models (age, ST-segment depression, signs of heart failure, severity of prior angina, and elevated cardiac markers), with the addition of enrollment region.

The ability of the models to predict mortality was excellent. The C statistic value ranged from 0.81 to 0.84 in the PURSUIT and GUSTO-IIb models. The ability to predict the composite of death or MI was less in all trials; in PURSUIT and ESSENCE, the C statistic values were 0.67 and 0.71. The ability to predict the triple end point (death, nonfatal MI, or recurrent ischemia) was only modest when tested in ESSENCE (C statistic, 0.65). These findings highlight the challenges in predicting nonfatal ischemic events.

In this issue of THE JOURNAL, Antman and colleagues¹³ present a simple scoring system to predict all-cause mortality, nonfatal MI, and severe recurrent ischemia requiring urgent revascularization within 2 weeks after presentation among patients with an acute coronary syndrome without ST-segment elevation. The system is based on 7081 patients from the ESSENCE¹⁴ and Thrombolysis in Myocardial Infarction (TIMI-11B) trials.¹⁵ The system was modestly predictive of the triple end point (C statistic, 0.63), whereas its ability to predict mortality was better (C statistic, 0.72-0.78). Variables used to develop the scoring system included age 65 years or older, 3 or more risk factors for coronary disease (family history, hypertension, diabetes, current smoking, or hypercholesterolemia), prior coronary stenosis of at least 50%, ST-segment deviation, severe anginal symptoms, aspirin use within the previous week, and elevated creatine kinase or its MB fraction. The lowest score (0 or 1) correlated with a 4.7% risk of the triple end point, whereas the highest score (6 or 7) correlated with a 40.9% risk. Thus, the scoring system was highly successful in allocating patients into low- and high-risk categories.

This study is consistent with previous analyses in several ways. First, several of the baseline predictors of death or nonfatal MI are common to all studies of patients with non–ST elevation acute coronary syndromes: age, ST-segment changes, and elevated cardiac enzymes. Severity of angina also was shown to predict adverse events in ESSENCE and PURSUIT. The addition of other risk factors, such as those for coronary disease and prior significant coronary stenosis, is consistent with guidelines for management of unstable angina.⁹

The main difference between the TIMI scoring system reported by Antman et al and previous models is the absence of a heart failure variable and objective hemodynamic measures. Heart failure has been a consistent predictor of adverse events in previous trials, including ESSENCE, ever since a simple system for its classification was described in 1967 for patients with MI.¹⁶ The exclusion of patients with planned cardiac intervention in TIMI-11B may have created a bias against enrolling higher-risk patients, including those with heart failure. This would lead to a lower prevalence of heart failure and reduced power to test its predictive ability, perhaps explaining its absence from the scoring system. The exclusion of objective, continuous hemodynamic measures, such as heart rate and blood pressure, most likely reflects the desired simplicity of the system. Such measures can increase the accuracy of mortality predictions (C statistic for mortality, 0.81 in PURSUIT vs 0.72-0.78 in the TIMI score), but at the cost of reducing user-friendliness.

The TIMI risk score offers a new way to explore treatment effects among subgroups. For example, platelet glycoprotein (Gp) IIb/IIIa inhibitors and low-molecular-weight heparins (LMWHs) are becoming standard treatments for non–ST elevation acute coronary syndromes. A recent meta-analysis has suggested that LMWHs are at least as effective as intravenous heparin in reducing the composite of death or nonfatal MI,^{17 - 18} whereas a meta-analysis of the only 2 enoxaparin trials noted a reduction in this end point with enoxaparin vs heparin.¹⁹ Antman et al, however, reported interactions between TIMI risk score and enoxaparin treatment in the prediction of all-cause mortality (P = .02) and all-cause mortality or nonfatal MI (P = .15). This finding may represent a chance observation, as with any subgroup analysis, but the TIMI-11B results suggest that the advantage of enoxaparin over heparin in reducing mortality is limited to patients with a TIMI risk score of more than 4 (about 17% of study patients). For death or nonfatal MI, the reduction with enoxaparin was evident only in patients with a TIMI score of more than 3, or 46% of the population. For patients with lower risk, then, there may be little advantage of enoxaparin over heparin.

The exception to this pattern may be patients with an elevated troponin level. In the Fragmin in Unstable Coronary Artery Disease (FRISC) trial of dalteparin vs placebo treatment of non–ST elevation acute coronary syndromes, troponin T elevation and dalteparin treatment interacted significantly on the composite of all-cause mortality or nonfatal MI.²⁰ The PURSUIT, ESSENCE, and TIMI-11B trials did not include troponin substudies. Given that baseline troponin elevation is a powerful risk marker in this population, patients with such elevations might derive more benefit with LMWHs than patients otherwise at lower risk.

Meta-analysis suggests a consistently reduced rate of all-cause mortality or nonfatal MI with intravenous GpIIb/IIIa inhibitors and a trend toward reduced mortality alone.²¹ Troponin status may be a key identifier of an enhanced therapeutic response with this drug class. Patients in the c7E3 Fab Antiplatelet Therapy in Unstable Refractory Angina (CAPTURE) trial who were randomly assigned to receive abciximab and who were troponin T–positive had substantially reduced mortality or nonfatal MI at 6 months, but not abciximab-treated, troponin T–negative patients.²² Positive troponin status also was linked with angiographically detected thrombus,²³ and troponin-positive patients given abciximab again were more likely to have a reduced thrombus burden with 24 hours of therapy.

The link between baseline troponin status and therapeutic response to GpIIb/IIIa inhibition has been extended to patients receiving tirofiban²⁴ or lamifiban.²⁵ In the former study, there was lower 30-day mortality with tirofiban vs heparin among patients who were either troponin T– or I–positive.²⁴ Thus the admission troponin level can be viewed as a key marker of therapeutic response to GpIIb/IIIa inhibitors, similar to the presence of ST-segment elevation with regard to fibrinolytic therapy for MI.

Patients with non–ST-elevation acute coronary syndromes are a diverse population. Risk stratification with either a simple scoring system or troponin status on admission offers the best way to identify patients who would benefit most from either GpIIb/IIIa inhibitors or LMWHs. For patients who are troponin-positive, more evidence exists for the use of GpIIb/IIIa inhibitors, whereas the use of a risk score may be more appropriate for LMWHs. Future studies should focus on the combination of GpIIb/IIIa inhibitors and LMWHs and on patients at the highest risk. The TIMI scoring system, as well as troponin status, provide a way to explore combination therapies, using heightened risk to identify the greatest opportunity for therapeutic benefit. The ability to select optimal therapy will be enhanced further when genetic testing becomes mature enough to be useful in practice. This will result in more complex models, but also, hopefully, in truly tailored therapy.