Risk Scores in Acute Coronary Syndrome and PCI
The GRACE and TIMI scores are the most popular and validated ACS prediction models. Their performances in other populations as well as against lesser known ACS risk scores are assessed. The characteristics of the derivation studies, with details of C-statistics in test and internal validation cohorts, are presented in Table I.
Separate GRACE scores were devised from a large unbiased multinational registry to predict in-patient mortality and post-discharge 6-month mortality, respectively, for the entire spectrum of ACS. These preceded the simplified GRACE nomogram designed by Cox regression analysis to predict cumulative 6-month risk of mortality or myocardial infarction (MI), with predictor variables including age, heart rate (HR), systolic blood pressure (SBP), creatinine level, Killip class of heart failure, cardiac arrest at admission, ST-segment deviation and cardiac enzymes. C-statistics for predicting mortality and mortality or MI from admission to 6 months in the validation cohort were 0.81 and 0.73 respectively. The GRACE nomogram is primarily used at initial presentation to identify those patients most likely to benefit from an invasive strategy. Such approach is recommended by contemporary guidelines, supported by the fact that patients with GRACE scores of more than 140 were found to benefit more from coronary angiography within 24 hours than did patients with lower scores with respect to the composite of death, MI, or stroke.
A recent meta-analysis identified 22 validation studies for the GRACE in-patient mortality model and demonstrated its good discriminatory performance across all types of ACS for short and long term prognosis (C-statistics ≥0.80). However, the utility of the GRACE models in predicting risks in patients after treatment with emergent or urgent PCI remains unsettled. This is not surprising considering that only 26.6% patients were treated with PCI in the development cohort of the GRACE 6-month post-discharge mortality model. In a Spanish multicentre registry, the GRACE 6-month post-discharge mortality score performed less well in patients undergoing PCI for ACS compared to patients receiving medical therapy only (C-statistics 0.73 versus 0.76, P < .004). In the PCI subgroup of the ACUITY trial, the GRACE score had less than satisfactory C-statistics for the prediction of mortality and important ischemic endpoints compared to other scores incorporating both clinical and angiographic variables. Interestingly, the GRACE score had the best discriminative performance when applied to the whole population of the ACUITY trial. With regard to the STEMI population treated with primary PCI, the GRACE models predicted mortality well up to 5 years (C-statistics >0.74) in 2 geographically distinct centres. However, one study found poor discrimination (C-statistic = 0.47) of the GRACE in-patient mortality model in determining 30-day and 1-year mortality following primary PCI although patients with cardiogenic shock were excluded.
Like the GRACE scores, the TIMI risk score for UA/NSTEMI has been extensively validated. It can be rapidly determined as the sum of the number of variables present to predict the composite of all-cause mortality, new or recurrent MI and severe recurrent ischemia requiring urgent revascularization within 14 days of event. It was derived from 1957 patients assigned to receive unfractionated heparin in the TIMI 11B trial and validated in 3 separate groups: the enoxaparin cohort from TIMI 11B (n = 1953), the unfractionated heparin group (n = 1564) and the enoxaparin group (n = 1607) from the ESSENCE trial. The predictor variables were age ≥65, at least 3 risk factors for coronary artery disease, prior coronary stenosis of 50% or more, ST-segment deviation on electrocardiogram at presentation, at least 2 anginal events in the past 24 hours, use of aspirin in the prior 7 days, and elevated serum cardiac markers. Each variable had a similar logistic regression coefficient so that a value of 1 was assigned when a variable was present, underpinning the model's simplicity. The C-statistics of the model in the derivation and validation cohorts ranged between 0.59 and 0.65 and the increase in the rates of events with increasing TIMI scores was statistically significant in all cohorts. Limitations of the TIMI risk score include those inherent to trial based scores including the exclusion of high-risk patients. While improving usability, the lack of weighting for the risk factors reduced its discriminatory performance. Despite these shortcomings, the simplicity of the TIMI score is recognized by current guidelines. It has also been used in key studies to demonstrate the benefit of clopidogrel across all levels of risk and to show graded benefits of tirofiban with increasing levels of risk. Of note, a separate TIMI score for STEMI patients eligible for fibrinolysis was previously derived from the INTIME II trial and included age 65 to 74 (2 points), age ≥75 (3 points), diabetes/hypertension or angina (1 point), SBP <100 (3 points), HR >100 (2 points), Killip II to IV (2 points), weight <67 kg (1 point), anterior STEMI or left bundle branch block (1 point), and time to treatment >4 hours (1 point). The score showed good prognostic capacity (C-statistics 0.727–0.87) for 30-day to 5-year mortality following primary PCI in contemporary cohorts.
The variables independently associated with 30-day mortality and composite of mortality/MI in the PURSUIT Risk Score were age (0–14 points), male sex (1 point), worse Canadian Cardiovascular Society Angina class in previous 6 weeks (0–2 points), admission HR (0–5 points), and SBP (0–2 points), ST-segment depression (1–3 points), and signs of heart failure (2–3 points). Despite showing almost comparable discriminative performance to the GRACE models in other registries, the PURSUIT risk score is rarely used in practice.
In the EMMACE risk model, age, SBP, and HR were identified as the most significant and practical predictors of 30-day mortality.
The Simple Risk Index (SRI) includes the most important and consistently proven predictors of mortality in the STEMI population, namely age, admission SBP and HR. Based on the relationship between its variables and 30-day mortality, the risk index is calculated as HR × [age/10])/SBP.
The GUSTO Risk Score was developed from the GUSTO trial, a randomized study of four thrombolytic agents with aspirin and beta blockers. Simplified nomograms derived with and without angiographic data predict 1-year survival in 30-day survivors of STEMI with C-statistics of 0.75 and 0.79 respectively. These included the most significant prognosticators namely age, HR, ejection fraction measured at ventriculography, previous infarction and in-hospital congestive heart failure or pulmonary edema.
Comparisons between GRACE and TIMI UA/NSTEMI risk scores in unselected ACS cohorts at various time points have generally favored the former with regards to discriminative performance.
In a Canadian multicenter registry of 1728 patients with NSTEMI, the GRACE and PURSUIT scores outperformed the TIMI UA/NSTEMI score in predicting in-hospital mortality (C-statistics 0.81 versus 0.80 versus 0.68, P < .001) and 1-year mortality (C-statistics 0.79 versus 0.77 versus 0.69, P < .0001). This is partly explained by the fact that PURSUIT and GRACE models include hemodynamic parameters and creatinine levels (GRACE only) which are powerful independent prognosticators in ACS. Another study of 2753 NSTEMI/UA patients demonstrated the superiority of the GRACE score over the TIMI UA/NSTEMI score in predicting in-hospital (C-statistics 0.85 versus 0.54, P = .01) and 6-month mortality (C-statistics 0.79 versus 0.56, P = .01), with the observed discriminative deficit resulting from the omission of hemodynamic variables and heart failure from the TIMI score.
The GRACE scores were also compared to the PURSUIT, GUSTO, SRI and EMMACE models in the MINAP, a large unselected ACS UK-based registry of 100 686 patients. Each model was assessed over the time for which they were designed and the C-statistics obtained matched the values from the original derivation cohorts (in-patient GRACE score: 0.80 versus 0.83, 6-month GRACE score: 0.80 versus 0.75, PURSUIT score: 0.79 versus 0.81, GUSTO score: 0.80 versus 0.75, SRI: 0.79 versus 0.78, EMMACE score: 0.78 versus 0.79). Simple models such as EMMACE and SRI had comparable performance to more complex risk models, emphasizing the importance of easily obtainable parameters such as age, HR, and SBP.
Risk Scores in ACS
The GRACE and TIMI scores are the most popular and validated ACS prediction models. Their performances in other populations as well as against lesser known ACS risk scores are assessed. The characteristics of the derivation studies, with details of C-statistics in test and internal validation cohorts, are presented in Table I.
GRACE Risk Scores
Separate GRACE scores were devised from a large unbiased multinational registry to predict in-patient mortality and post-discharge 6-month mortality, respectively, for the entire spectrum of ACS. These preceded the simplified GRACE nomogram designed by Cox regression analysis to predict cumulative 6-month risk of mortality or myocardial infarction (MI), with predictor variables including age, heart rate (HR), systolic blood pressure (SBP), creatinine level, Killip class of heart failure, cardiac arrest at admission, ST-segment deviation and cardiac enzymes. C-statistics for predicting mortality and mortality or MI from admission to 6 months in the validation cohort were 0.81 and 0.73 respectively. The GRACE nomogram is primarily used at initial presentation to identify those patients most likely to benefit from an invasive strategy. Such approach is recommended by contemporary guidelines, supported by the fact that patients with GRACE scores of more than 140 were found to benefit more from coronary angiography within 24 hours than did patients with lower scores with respect to the composite of death, MI, or stroke.
A recent meta-analysis identified 22 validation studies for the GRACE in-patient mortality model and demonstrated its good discriminatory performance across all types of ACS for short and long term prognosis (C-statistics ≥0.80). However, the utility of the GRACE models in predicting risks in patients after treatment with emergent or urgent PCI remains unsettled. This is not surprising considering that only 26.6% patients were treated with PCI in the development cohort of the GRACE 6-month post-discharge mortality model. In a Spanish multicentre registry, the GRACE 6-month post-discharge mortality score performed less well in patients undergoing PCI for ACS compared to patients receiving medical therapy only (C-statistics 0.73 versus 0.76, P < .004). In the PCI subgroup of the ACUITY trial, the GRACE score had less than satisfactory C-statistics for the prediction of mortality and important ischemic endpoints compared to other scores incorporating both clinical and angiographic variables. Interestingly, the GRACE score had the best discriminative performance when applied to the whole population of the ACUITY trial. With regard to the STEMI population treated with primary PCI, the GRACE models predicted mortality well up to 5 years (C-statistics >0.74) in 2 geographically distinct centres. However, one study found poor discrimination (C-statistic = 0.47) of the GRACE in-patient mortality model in determining 30-day and 1-year mortality following primary PCI although patients with cardiogenic shock were excluded.
TIMI Risk Scores
Like the GRACE scores, the TIMI risk score for UA/NSTEMI has been extensively validated. It can be rapidly determined as the sum of the number of variables present to predict the composite of all-cause mortality, new or recurrent MI and severe recurrent ischemia requiring urgent revascularization within 14 days of event. It was derived from 1957 patients assigned to receive unfractionated heparin in the TIMI 11B trial and validated in 3 separate groups: the enoxaparin cohort from TIMI 11B (n = 1953), the unfractionated heparin group (n = 1564) and the enoxaparin group (n = 1607) from the ESSENCE trial. The predictor variables were age ≥65, at least 3 risk factors for coronary artery disease, prior coronary stenosis of 50% or more, ST-segment deviation on electrocardiogram at presentation, at least 2 anginal events in the past 24 hours, use of aspirin in the prior 7 days, and elevated serum cardiac markers. Each variable had a similar logistic regression coefficient so that a value of 1 was assigned when a variable was present, underpinning the model's simplicity. The C-statistics of the model in the derivation and validation cohorts ranged between 0.59 and 0.65 and the increase in the rates of events with increasing TIMI scores was statistically significant in all cohorts. Limitations of the TIMI risk score include those inherent to trial based scores including the exclusion of high-risk patients. While improving usability, the lack of weighting for the risk factors reduced its discriminatory performance. Despite these shortcomings, the simplicity of the TIMI score is recognized by current guidelines. It has also been used in key studies to demonstrate the benefit of clopidogrel across all levels of risk and to show graded benefits of tirofiban with increasing levels of risk. Of note, a separate TIMI score for STEMI patients eligible for fibrinolysis was previously derived from the INTIME II trial and included age 65 to 74 (2 points), age ≥75 (3 points), diabetes/hypertension or angina (1 point), SBP <100 (3 points), HR >100 (2 points), Killip II to IV (2 points), weight <67 kg (1 point), anterior STEMI or left bundle branch block (1 point), and time to treatment >4 hours (1 point). The score showed good prognostic capacity (C-statistics 0.727–0.87) for 30-day to 5-year mortality following primary PCI in contemporary cohorts.
PURSUIT Risk Score
The variables independently associated with 30-day mortality and composite of mortality/MI in the PURSUIT Risk Score were age (0–14 points), male sex (1 point), worse Canadian Cardiovascular Society Angina class in previous 6 weeks (0–2 points), admission HR (0–5 points), and SBP (0–2 points), ST-segment depression (1–3 points), and signs of heart failure (2–3 points). Despite showing almost comparable discriminative performance to the GRACE models in other registries, the PURSUIT risk score is rarely used in practice.
EMMACE Risk Score
In the EMMACE risk model, age, SBP, and HR were identified as the most significant and practical predictors of 30-day mortality.
Simple Risk Index
The Simple Risk Index (SRI) includes the most important and consistently proven predictors of mortality in the STEMI population, namely age, admission SBP and HR. Based on the relationship between its variables and 30-day mortality, the risk index is calculated as HR × [age/10])/SBP.
GUSTO Risk Score
The GUSTO Risk Score was developed from the GUSTO trial, a randomized study of four thrombolytic agents with aspirin and beta blockers. Simplified nomograms derived with and without angiographic data predict 1-year survival in 30-day survivors of STEMI with C-statistics of 0.75 and 0.79 respectively. These included the most significant prognosticators namely age, HR, ejection fraction measured at ventriculography, previous infarction and in-hospital congestive heart failure or pulmonary edema.
Comparison of ACS Risk Scores
Comparisons between GRACE and TIMI UA/NSTEMI risk scores in unselected ACS cohorts at various time points have generally favored the former with regards to discriminative performance.
In a Canadian multicenter registry of 1728 patients with NSTEMI, the GRACE and PURSUIT scores outperformed the TIMI UA/NSTEMI score in predicting in-hospital mortality (C-statistics 0.81 versus 0.80 versus 0.68, P < .001) and 1-year mortality (C-statistics 0.79 versus 0.77 versus 0.69, P < .0001). This is partly explained by the fact that PURSUIT and GRACE models include hemodynamic parameters and creatinine levels (GRACE only) which are powerful independent prognosticators in ACS. Another study of 2753 NSTEMI/UA patients demonstrated the superiority of the GRACE score over the TIMI UA/NSTEMI score in predicting in-hospital (C-statistics 0.85 versus 0.54, P = .01) and 6-month mortality (C-statistics 0.79 versus 0.56, P = .01), with the observed discriminative deficit resulting from the omission of hemodynamic variables and heart failure from the TIMI score.
The GRACE scores were also compared to the PURSUIT, GUSTO, SRI and EMMACE models in the MINAP, a large unselected ACS UK-based registry of 100 686 patients. Each model was assessed over the time for which they were designed and the C-statistics obtained matched the values from the original derivation cohorts (in-patient GRACE score: 0.80 versus 0.83, 6-month GRACE score: 0.80 versus 0.75, PURSUIT score: 0.79 versus 0.81, GUSTO score: 0.80 versus 0.75, SRI: 0.79 versus 0.78, EMMACE score: 0.78 versus 0.79). Simple models such as EMMACE and SRI had comparable performance to more complex risk models, emphasizing the importance of easily obtainable parameters such as age, HR, and SBP.