Patients with a non-ST-elevation acute coronary syndrome (NSTE-ACS) encompass a heterogeneous group of individuals with different clinical presentations and varying extents of underlying coronary disease.1 The management of these patients is determined by their perceived risk of progression to non-fatal ischaemic events or death.

In addition to the well-documented events of death and myocardial infarction, ACS patients frequently experience other serious events, including heart failure, arrhythmias, stroke and major bleeding. Patients who have one of these events have a longer length of hospitalisation and a poorer prognosis, and require more intensive medical care.2^–5 NB: Tables have been rearranged and reordered, as per BMJ guidelines.

Existing risk models predict the likelihood of in-hospital mortality or the combination of mortality, myocardial infarction and urgent revascularisation in patients with an ACS.6^–12 These tools have proven most useful for identifying the high-risk population and for directing evidence-based therapies.13 They have also been applied to unselected patients presenting to emergency departments with chest pain, where they have been shown to perform similarly to those seen when applied to clinical trial and registry databases in acute coronary syndromes (ACS).14 15 However, they have not provided comparable clinical utility in non-high-risk patients admitted with an ACS, who currently undergo periods of observation and dynamic risk stratification in hospital.16 We hypothesised that the incorporation of additional adverse events in a risk-prediction model might prove to be an effective method of further risk stratifying these non-high-risk patients.

Using data from a large multinational study—the Global Registry of Acute Coronary Events (GRACE)—we analysed data from patients with non–ST-segment elevation myocardial infarction (NSTEMI) or unstable angina, and incorporated objective clinical predictors derived from patients’ initial clinical history, electrocardiographic findings, and laboratory tests into a model to predict the likelihood of freedom from individual adverse events in hospital. We then developed a tool—the Freedom-from-Event score—to identify ACS patients who had a low risk of any adverse in-hospital event.

METHODS

Study population

Full details of the GRACE methods have been published elsewhere.1 17 GRACE is designed to reflect an unbiased population of patients with ACS, irrespective of geographic region. A total of 123 hospitals located in 14 countries in Europe, North and South America, Australia and New Zealand have contributed data to this observational study. Data from 96 sites in 14 countries were used for this analysis.

Adult patients (⩾18 years) admitted for >24 h with a presumptive diagnosis of ACS at participating hospitals were potentially eligible for this study. Eligibility criteria were a clinical history of ACS accompanied by at least one of the following: electrocardiographic changes consistent with ACS, serial increases in biochemical markers of cardiac necrosis (CK-MB, creatine phosphokinase, or troponin) and documented coronary artery disease. Patients with non-cardiovascular causes for the clinical presentation, such as trauma, surgery or aortic aneurism, were excluded. Patients were followed up at approximately 6 months by telephone, clinic visits, or through calls to their primary care physician to ascertain the occurrence of several long-term outcomes. Where required, study investigators received approval from their local hospital ethics or institutional review board for the conduct of this study.

To enrol an unselected population of patients with ACS, sites were encouraged to recruit the first 10 to 20 consecutive eligible patients each month. Regular audits were performed at all participating hospitals. Data were collected by trained study coordinators using standardised case report forms. Demographic characteristics, medical history, presenting symptoms, duration of prehospital delay, biochemical and electrocardiographic findings, treatment practices and a variety of hospital outcome data were collected. Standardised definitions of all patient-related variables, clinical diagnoses and outcomes were used. All cases were assigned to one of the following categories: ST-segment elevation myocardial infarction (STEMI), NSTEMI or unstable angina.

This analysis was restricted to patients presenting with NSTEMI or unstable angina (NSTE-ACS). Patients were diagnosed as having NSTEMI when they had at least one positive cardiac biochemical marker of necrosis without new ST-segment elevation seen on the index or subsequent electrocardiogram. Unstable angina was diagnosed when serum biochemical markers indicative of myocardial necrosis in each hospital’s laboratory were within the normal range.

Events used to define the composite endpoint from which the Freedom-from-Event score was developed were congestive heart failure or shock, arrhythmia (atrial fibrillation, cardiac arrest, ventricular tachycardia and ventricular fibrillation), myocardial infarction, stroke, major bleed or death. Full definitions can be found on the GRACE website (http://www.outcomes.org/grace).

Statistical analysis

Data are summarised as frequencies and percentages for categorical data. Univariate analyses using the χ² test and Fisher exact test were carried out, as appropriate, to determine the association between these variables and freedom from adverse in-hospital events. Continuous variables are presented as medians and 25th and 75th percentiles, and tested using the Wilcoxon two-sample test.

Two-thirds of the patients were randomly chosen to develop the model, and the remainder to validate it. Variables from table 1 with a p value <0.15 in the univariate analysis were included in a multiple logistic regression analysis of the development group. Factors with α⩽0.05 in the development set whose estimates remained consistent in the validation group were retained in the final model. From the model, we then developed a nomogram of patient risk where points were assigned to each factor based on relative model estimates, then summed to obtain a Freedom-from-Event score.18 Patients in the development set were then grouped by decile of freedom from risk (10 approximately equal groups of patients such that the freedom from risk scores increase montonically from the lowest to the highest decile) to compare actual and predicted rates, as well as freedom from combined death, myocardial infarction and stroke occurring 6 months after admission.

Fig 1A relates the Freedom-from-Event score to the model-estimated probability using a LOESS smoothing algorithm, since a single risk score does not map to a single predicted probability (ie, one number, the Freedom-from-Event score, is not as precise as 15 individual patient characteristics). The loss in precision is, however, not severe.

• Download figure
• Open in new tab
• Download powerpoint

Figure 1

(A) Predicted proportion of patients free from events by risk score; and (B) calibration of Freedom-from-Event score (observed vs predicted proportions free from event) by patient risk decile. The diagonal line represents perfect calibration. bpm, beats per minute; CHF, congestive heart failure; PAD, peripheral arterial disease, SBP, systolic blood pressure.

To assess the relative value of the Freedom-from-Event score compared with the GRACE risk score in predicting freedom from hospital events, we compared c-statistics in logistic regression models having the scores as predictors, first for all patients, then for subgroups with low, medium and high risk. Since a c-value of 0.5 represents a model that performs no better than chance, the relative improvement in the Freedom-from-Event score was measured as the increase in its model c-value compared with the c-value of the risk score, divided by the risk score c-value minus 0.5 (the base c-value).

Odds ratios are reported with 95% confidence intervals. Analyses were conducted using the SAS® software (version 9.1, SAS Institute, Cary, North Carolina).

RESULTS

Patients enrolled between January 2001 and September 2007 were included in this study. Of the 24 097 patients enrolled, 16 127 were randomly selected for model development; 3083 (19.1%) experienced in-hospital myocardial infarction, arrhythmia, congestive heart failure, major bleeding, stroke or death (table 2). The baseline characteristics for the population who remained free from in-hospital events and for those who developed an event are shown in table 1. Patients who remained free from an adverse event were more likely to be younger and male; they were less likely to have a history of hypertension, diabetes, myocardial infarction, congestive heart failure, transient ischaemic attack or stroke, peripheral arterial disease, atrial fibrillation, bleeding and renal dysfunction; to present with a cardiac arrest; to have elevated initial cardiac biomarkers or a diagnosis of NSTEMI within 24 h; to have ST deviation on presentation; and to have been transferred from another hospital. They were more likely to have a history of angina, smoking, hyperlipidaemia, to have undergone prior percutaneous coronary intervention (PCI) and to present in Killip class I. They were less likely to be taking a statin prior to presentation but more likely to be on warfarin, angiotensin-converting enzyme inhibitors and diuretics.

Predictors of freedom-from-events

In multiple logistic regression analysis, variables associated with freedom from adverse in-hospital events were younger age, Killip class I, unstable angina, higher systolic blood pressure, no ST deviation, no cardiac arrest on presentation, normal creatinine levels, no tachycardia, not transferred from another hospital, no history of peripheral arterial disease, diabetes, heart failure or atrial fibrillation, no chronic use of warfarin and chronic use of statins (tables 3, 4).

The model showed good discrimination (c-statistic = 0.77), which was retained in the validation cohort of 6820 patients, of whom 1271 (18.6%) experienced an event (c-statistic also 0.77). This resulted in good calibration between observed and model predicted proportion of patients free from an event (fig 1B). The Freedom-from-Event score card (table 5) reliably stratified patients according to the likelihood of freedom from an adverse coronary event (table 6, fig 1B). Patients with a score >286, ie, about 30% of the patients, constituted a group with >99.5% in-hospital survival, who were >93% free from risk of any in-hospital event. Those with a score <216, ie, the lowest 20%, had >6% in-hospital mortality and were only 42–65% free from an in-hospital event.

The Freedom-from-Event score was also predictive of 6-month outcomes among patients surviving to discharge (c-statistic = 0.77). The same top three deciles with the highest Freedom-from-Event score were >97% free from the hard endpoints of death, myocardial infarction or stroke at 6 months. However, the unscheduled readmission rate for a cardiac event remained frequent among all patients, regardless of the Freedom-from-Event score (⩾15% for all deciles).

Comparison with GRACE risk score

Table 7 compares the GRACE risk score9 for hospital mortality with the Freedom-from-Event score. For the population as a whole, the new score resulted in a modest improvement in predicting freedom from adverse events (c-statistic 0.772 vs 0.737). When patients were stratified according to risk, the Freedom-from-Event score performed better in each group but was most valuable among those classified as intermediate and low risk. For these patients, the new score greatly increased the predictive ability of the mortality risk score.

DISCUSSION

A number of multivariable predictive models have been developed in patients with NSTE-ACS to aid stratification according to risk of in-hospital death, or death and other non-fatal cardiac ischaemic events.

The most widely used of these scores are the TIMI7 and GRACE9 19 20 risk scores. They are of particular value in identifying patients at high risk of an ischaemic event and for directing specific therapies such as low-molecular-weight heparins, glycoprotein IIb/IIIa inhibitors and early invasive strategies to these higher-risk patients.21 22

For patients not at high risk, the existing scoring systems differentiate less well.16 These patients are admitted to coronary or intensive care units for further observation and risk stratification, largely because of the clustering of fatal coronary events early after presentation; these do, however, occur in only a small proportion of these patients.23

There have been few attempts to risk-stratify the non-high-risk cohort of patients with an ACS. Sanchis et al24 developed a risk score to discriminate between patients with chest pain, no ST-segment deviation and normal troponin levels. This score performed better than the TIMI risk score in their small cohort of 646 patients; however, it was applied to consecutive patients with chest pain presenting to an emergency department and not to a cohort of admitted patients with a high likelihood of ACS as was included in the GRACE registry.

Cardiac death and non-fatal ischaemic events constitute only a part of the range of complications that can occur in ACS patients following presentation; they are also at risk of arrhythmias, heart failure, stroke and major bleeding.3 5 11 Patients with NSTE-ACS complicated by ventricular arrhythmias are at much greater risk of in-hospital and 6-month death than those without.25 Similarly, patients with ACS complicated by atrial fibrillation are at increased risk of morbidity and mortality vs ACS patients not experiencing this arrhythmia.4 In addition, ACS patients with heart failure are at higher risk of death, are hospitalised for longer and are more likely to be readmitted for an ischaemic event than those without heart failure.2 Stroke, while uncommon among ACS patients, is associated with high mortality.11 Major bleeding is another common adverse event in the ACS population and is associated with poor outcome.3 5 By incorporating a broader array of events into a predictive tool, we are better able to discriminate among these patients who may be at relatively lower risk of death but are still at risk of other prognostically important adverse events for which intensive monitoring is required.

Our tool incorporates many of the same variables included in the GRACE risk score. Although more complex than other risk algorithms, the GRACE risk score has good discriminative power and has the additional advantage that it includes easy to assess clinical, electrocardiographic and laboratory variables. In a direct comparison of different risk scores in a consecutive NSTE-ACS population admitted to a single centre coronary care unit, the best predictive accuracy was provided by the GRACE score;21 furthermore, it is recommended in international guidelines as one of the preferred classifications to apply on admission in daily clinical practice.26 27

The Freedom-from-Event score includes more factors than the GRACE risk score.9 This is because it is designed with a different purpose: to discriminate between a large group of patients at risk of any adverse event rather than identifying the smaller group at high risk of death. The additional factors include medical history of atrial fibrillation, heart failure, diabetes and peripheral arterial disease, or chronic therapy with warfarin or a statin. Each is collected as a component of the routine admission and therefore does not detract from the ease of assessment of the factors shared with the GRACE risk score.

We have shown previously that prehospital use of aspirin identified a population with less severe presentation and fewer in-hospital complications than patients not taking aspirin.28 That earlier GRACE analysis differed slightly from the current one in that patients with STEMI were included; patients receiving aspirin were less likely to experience STEMI, which was itself associated with a greater rate of in-hospital complications. In this analysis, restricted to the NSTE-ACS population, the association was no longer significant in either univariate (p = 0.15) or adjusted (p = 0.62) analyses. Others have shown that the use of beta-blockers in patients with CAD but no previous coronary events was associated with reduced mortality although not a reduced incidence of myocardial infarction.29 We did not confirm that association in our population (p = 0.1 in univariate and p = 0.21 in adjusted analyses).

Patients at risk of death are also more likely to experience other in-hospital complications, and it is therefore possible that the existing GRACE risk score may be sufficiently robust to identify a cohort at low risk of any in in-hospital event. To determine the added predictive value of the additional factors we did identify in developing the Freedom-from-Event score, the GRACE risk score was applied to predict freedom from an event in this ACS cohort. The difference in c-statistics between the two scores (0.737 for GRACE risk score vs 0.772 for Freedom-from-Event score) represents a 15% increment in predictive power. When the predictive abilities of the GRACE risk score and the Freedom-from-Events score were directly compared among the different risk groups, the overall c-statistic was lower within each group, which is to be expected when looking at subsets of patients who are more homogeneous than the cohort in which the score was derived. Importantly, however, the Freedom-from-Event score proved most valuable in the intermediate- and lower-risk patients, where it provided a 100% or higher improvement in predictive power relative to the GRACE risk score. Thus, this tool achieves our objective of improving risk prediction among non-high-risk patients with NSTE-ACS. These patients constitute the majority of those admitted with an ACS. A recent study based on data from 13 731 patients enrolled in two large clinical trials—CURE (Clopidogrel in Unstable Angina to Prevent Recurrent Events) and PROVE IT-TIMI 22 (Pravastatin or Atorvastatin Evaluation and Infection Therapy–Thrombolysis in Myocardial Infarction 22)—demonstrated that the total cost of healthcare utilisation in the 12 months following new-onset ACS was $309 million ($2312 per patient-month of follow-up); almost three-quarters of the total costs were attributable to hospitalisations.30 Currently these patients contribute a significant burden to healthcare systems; careful use of this tool should provide an opportunity to contain this burden.

Similar to the TIMI7 and GRACE9 risk scores, the Freedom-from-Event score can be computed using a simplified nomogram available for download to computer or personal digital assistant. We suggest that the score will be of most value when applied to patients who have been declared as “non-high risk” on the basis of initial risk assessment. The 30% of patients with the highest Freedom-from-Event score have a very low likelihood of dying in hospital, and a low likelihood of death or myocardial infarction at 6 months following discharge. Their likelihood of any adverse event in hospital is low. This tool should be used to identify ACS patients who can be considered for lower resource-intensive initial strategies of care, which may include outpatient management of their ACS.

Conclusions

Patients admitted with an ACS are at significant risk of a range of adverse events while in hospital. The GRACE Freedom-from-Event score is an easy-to-use tool that can be applied to all patients to predict the risk of an adverse event in hospital. The score identifies a significant subgroup with low in-hospital mortality that is likely to have an uncomplicated clinical course. It is of particular value in risk stratifying the non-high-risk ACS patient, and could be used to identify a cohort in whom lower resource intensive strategies of care may be prospectively evaluated.