Author + information
- Received December 27, 2000
- Revision received May 3, 2001
- Accepted June 25, 2001
- Published online October 1, 2001.
- Daniel H Solomon, MD, MPH∗,* (, )
- Peter H Stone, MD, FACC†,
- Robert J Glynn, PhD, ScD∗,
- David A Ganz, MD, MPH∗,
- C.Michael Gibson, MD, MSc, FACC‡,
- Russell Tracy, PhD§ and
- Jerry Avorn, MD∗
- ↵*Reprint requests and correspondence:
Dr. Daniel H. Solomon, Division of Pharmacoepidemiology and Pharmacoeconomics, Brigham and Women’s Hospital, 221 Longwood Avenue, Suite 341, Boston, Massachusetts 02115
This study was designed to determine whether patient characteristics collected at presentation can identify which patients benefit from immediate coronary angiography and revascularization.
Risk stratification may offer a method for identifying which patients with unstable angina or non–Q-wave myocardial infarction (NQMI) are likeliest to benefit from invasive management strategies.
The analysis was based on data from a randomized controlled trial that enrolled 1,473 patients presenting with unstable angina or NQMI who were randomly assigned to an early invasive or early conservative (medical) management strategy. We constructed a risk-stratification score for each patient based on adjusted odds ratios for clinical variables likely to predict adverse outcomes. We stratified all trial subjects by their risk scores and studied the rates of death or myocardial infarction (MI) of the early invasive management strategy in each stratum.
The final multivariate model included older age, ST segment depression on presentation, history of complicated angina before presentation, and elevation in baseline creatine kinase-MB fraction. Although patients with a higher risk score had an increased rate of death or MI within 42 days and 365 days (p < 0.001) in both management strategies, early invasive management for patients in the high and very high risk categories was associated with a lower rate of death or MI within 42 days compared with conservative management. No such benefit was seen in patients in the larger group of patients in the very low, low or moderate risk categories (p = 0.03 for the interaction between risk category and management assignment).
Risk stratification may be an effective method for identifying those patients with unstable angina or NQMI most likely to benefit from early invasive management. Selective use of early invasive management can have a substantial impact in reducing morbidity and mortality in higher risk patients, but may not be warranted in lower risk patients.
While the management of unstable angina rapidly evolves, it remains important to identify which patients require routine early coronary angiography and revascularization if indicated, versus more conservative management. At least three
published studies have examined the efficacy of early catheterization for unstable angina or non–Q-wave myocardial infarction (NQMI) (1–3). In unselected patients enrolled in current trials, this approach has failed to show a morbidity or mortality benefit over medical management (1,2). A European trial did find an advantage with an early invasive strategy; patients in this study were required to have electrocardiographic changes or cardiac enzyme abnormalities upon presentation, so that population may not reflect the broader spectrum of patients presenting with unstable angina or NQMI (3). Furthermore, the invasive strategy in that trial cannot be considered “early” because it was implemented up to three days after treatment with heparin or low-molecular-weight heparin. Another recent study suggests that an early invasive strategy may be useful in patients with unstable angina who have received a glycoprotein (Gp) IIb/IIIa agent (4).
The inconsistent findings of these large clinical trials might be explained by different enrollment criteria or treatment protocols; another possibility is that the populations for each trial differed in baseline risk factor profiles. Not all patients with unstable angina or NQMI are at equal risk for cardiac events. The Braunwald criteria for unstable angina (5), as well as the Unstable Angina Clinical Practice Guidelines (6), suggest that patients with unstable angina can be categorized into low, moderate and high-risk groups. Clinical factors at initial presentation may help categorize such patients and possibly determine the optimal management for each (6–8). Preliminary data indicate that newer therapies, such as Gp IIb/IIIa inhibitors and low-molecular-weight heparin (3,9), are beneficial only in high risk patients, but risk stratification has not been applied to determine which patients should be treated with an early invasive versus early conservative management strategy.
According to this concept of risk stratification to guide management, some subgroups of patients with unstable angina or NQMI might be more likely to benefit from early invasive management, whereas others might not (10–14). We risk-stratified patients with unstable angina or NQMI on the basis of clinical data collected at presentation and tested whether high risk subgroups differed in the effect of an early invasive strategy.
The analysis was based on data from patients enrolled in the Thrombolysis In Myocardial Infarction (TIMI) IIIB trial; its methods and participants have been described in detail previously (1). Briefly, TIMI IIIB was a randomized controlled trial of patients with unstable angina or NQMI, designed to determine whether outcomes could be improved with an early invasive management strategy versus conservative (medical) management. Eligible patients had experienced chest pain at rest in the prior 24 h lasting ≥5 min but ≤6 h that was judged to be caused by cardiac ischemia, as evidenced by several criteria: new electrocardiogram (ECG) evidence of ischemia in at least two contiguous leads (≥0.1 mV ST segment elevation lasting <30 min, transient or persistent ≥0.1 mV ST segment depression, or T-wave inversion during an episode of pain at rest within the prior seven days), history of prior myocardial infarction (MI), ≥70% stenosis on a previous coronary arteriogram or a positive exercise thallium scintigram. Patients were excluded if they had any of the following: a treatable cause of unstable angina, such as thyrotoxicosis; an MI within the prior 21 days; a history of coronary arteriography within 30 days; percutaneous transluminal coronary angioplasty (PTCA) within six months or coronary artery bypass grafting (CABG) at any time in the past.
The trial enrolled 1,473 patients, who were randomly assigned to early conservative or early invasive therapy. The patients were mostly white, middle-aged and men (Table 1). Most had a history of angina, and one-third experienced symptoms severe enough to place them in Canadian Cardiovascular Society Class II or higher (angina with walking stairs resulting in mild functional limitations or worse). About a third of patients reported an MI ≥21 days before enrollment. Cardiac risk factors were common. The outcomes of interest included death or MI. All MIs were reviewed and classified by the Central Classification Committee; a creatine kinase-MB fraction (CK-MB) exceeding normal or, if no CK-MBs were available, a total CK greater than twice normal was considered evidence of infarction (1). In the group as a whole, the proportion of subjects who died or experienced an MI by 42 and 365 days was similar regardless of management assignment (1).
The conservative treatment protocol of TIMI IIIB included bed rest, oxygen, a beta-adrenergic blocking agent (metoprolol 50 mg orally every 12 h), a calcium channel blocking agent (diltiazem 30 mg orally every 6 h), a long-acting nitrate (isosorbide dinitrate 10 mg orally every 8 h), sublingual and/or intravenous nitroglycerin, heparin and aspirin. Patients were continued on anti-ischemic medicines after discharge. Patients in this arm underwent cardiac catheterization and revascularization only if indicated by evidence of ongoing ischemia despite maximal medical therapy (see Outcomes section).
By contrast, patients randomized to the early invasive strategy underwent cardiac catheterization, left ventricular angiography and coronary arteriography 18 to 48 h after randomization. Angiograms were reviewed immediately to determine whether a culprit coronary artery lesion could be identified; if so, revascularization by PTCA or CABG was performed.
Patients were also randomized to tissue plasminogen activator (t-PA) versus placebo. No effect of this treatment was observed in the original trial findings.
Development of the risk score
The primary outcome of interest used in developing the risk-stratifying score described here was death or MI within 42 days of randomization. The rate of death or MI at 365 days was examined as a secondary end point. The relationship of the candidate variables with the primary end point was examined using data for patients in both management arms of TIMI IIIB.
On the basis of the existing literature, we considered several variables available at presentation that could predict poor outcomes for patients with unstable angina and NQMI. These included patient factors that could be easily accessed at the bedside by physicians, including age, gender, characteristics of prior angina, signs of congestive heart failure, ECG abnormalities, CK-MB, cardiac troponin I (cTnI) and C-reactive protein (CRP) (Table 2).
Age and gender have long been known to be important predictors of death or MI (15–19). Several characteristics of a patient’s angina portend poor outcomes, including episodes lasting >20 min, nocturnal episodes and an accelerated pattern (20,21). Congestive heart failure is also associated with a poor prognosis in these patients (6). ST segment depression or elevation on the baseline ECG has been found to predict subsequent acute MI and death (18,22,23), as have serum markers such as CK-MB, cTnI and CRP (9,18,24–31).
To determine whether to assess predictive factors as continuous, ordinal or categorical variables, we examined their relationships with the primary outcome, death or MI at 42 days. Age was first examined as a continuous variable, then in five-year increments and finally collapsed into decades. In the original trial, patients were queried about several aspects of their angina in the two months preceding presentation, including the presence of episodes >20 min, accelerated episodes of angina and any history of angina with pain >20 min. We assessed the relationship between these aspects of patients’ angina and rates of outcomes. An ordinal scale (one point for each angina characteristic present) predicted the outcome as well as did the individual characteristics considered separately, and no one characteristic was more predictive than another. Hence, angina was considered “complicated” if any of the above characteristics were present.
Congestive heart failure was considered present if an S3or rales were heard on the initial physical examination. Of the baseline ECG findings, only ST segment depression ≥0.1 mV was significantly associated with the outcome of interest. Cardiac troponin I was considered in several ways: as a continuous variable; in deciles; and, based on prior work (25), trichotomized at <0.4 ng/ml, 0.4 to 3.99 and ≥4.0. Each definition had a similar association with the outcome, and the three-level definition was used in subsequent analyses. We treated CK-MB in a similar fashion with similar results; a variable dichotomized at 5 IU/ml performed similarly to other definitions and was chosen for further analyses. C-reactive protein testing was performed according to recent protocols (32), and values ranged from 0.1 to 260.09 mg/dl. This variable was examined in a continuous fashion, by deciles, and dichotomized at 1.55 mg/dl on the basis of prior work (29). The two-level CRP variable performed as well as a continuous or ten-level variable; because it would be easier to use at the bedside, we used the dichotomous CRP for further analyses.
All categories and cut points were identified before any assessment of outcomes, and in a manner designed to enhance ease of use at the bedside.
To develop the risk-stratification rule, unadjusted logistic equations were first used to examine the relationship of each of the predictive factors with the primary outcome, death or MI within 42 days. These crude relationships were assessed in the total TIMI IIIB population. Predictive factors with p values ≤0.2 were considered potential predictors of death or MI and were entered in a multivariate model. Variables with p values >0.2 in these adjusted models were removed. We chose this cut point because the goal of our model was to develop a prediction rule rather than to determine epidemiologic association. The prediction rule was statistically validated by sampling 1,000 subgroups of patients from the cohort (bootstrap procedure) to determine the internal consistency of the factors determined on multivariate modeling. We compared the predictive ability of the original and bootstrapped multivariate models by examining the area under the receiver operating characteristic curves (C statistic) (33). The odds ratios (ORs) derived from the final adjusted model were used to define the weights for each predictor in the risk-stratification rule.
We tested for the presence of confounding by study hospital or assignment to the t-PA arm by examining the changes in multivariate ORs of the original predictors using a conditional logistic model (PROC PHREG in SAS); the point estimates changed by <7% and thus we did not include study hospital or t-PA in the final model. Additionally, we assessed whether patient risk factors were balanced within centers (design effect) by examining the changes in standard errors in the conditional logistic models; the standard errors varied <8%, suggesting an insignificant design effect.
To facilitate such an approach at the bedside, a risk score and risk categories were created. Variables with ORs from 1.01 to 1.5 were assigned 1 point, 1.51 to 2.0 were assigned 2 points, 2.01 to 2.5 were assigned 3 points and so on. We summed the points to create five categories for risk-stratification score: very low risk (0 to 2 points), low risk (3 to 4 points), moderate risk (5 to 6 points), high risk (7 to 8 points) and very high risk (9 to 10 points). The rates of death or MI by 42 days in each risk category were then calculated. This was done separately for the study sample as a whole as well as for the patients in each randomly assigned management group. A chi-square test of trend was performed on these values. A multivariate model that included an interaction term between treatment assignment (early invasive vs. early conservative) and risk category as a predictor of death or MI within 42 days was then created. Finally, we assessed the association between risk category and treatment assignment in relation to the secondary outcome of death or MI within 365 days, using similar models. Chi-square tests were performed to determine the significance of differences between rates of outcomes in individual risk categories for the two management strategies.
The predictive ability of the risk score was also compared with that of the full multivariate model by examining the C statistic for each. We compared the observed proportion of patients who died or had an MI to the expected rates using the Hosmer-Lemeshow goodness-of-fit test (34). All analyses were run using SAS Statistical Software (SAS Institute Inc. SAS Statistical Software, Cary, North Carolina).
In unadjusted analyses examining the relationship between potential predictive factors and death or MI within 42 days, the following variables all had p values ≤0.2 and were included in the adjusted models: older age, by decade; a history of complicated angina; S3or rales at presentation; ST segment depression on the initial ECG; and a CK-MB fraction >5 IU/ml (Table 3). However, CRP and cTnI had p values >0.2 and were therefore not included in further analyses. The variable indicating the presence of S3or rales had a p value >0.2 in adjusted models and was likewise removed; all other variables remained in the model. The ORs from the final model were used to determine the predictive factor weights (Table 4), with a total possible score of 10. As expected, the rate of death or MI by 42 days rose significantly for patients with greater scores in progressively higher risk categories, as did the rate of death or MI within 365 days (both p values < 0.001) (Table 5).
Figure 1Apresents the rates of death or MI within 42 days for patients in each risk category. Outcomes are presented separately for those randomized to early conservative versus early invasive therapy. Patients in the very low, low and moderate risk groups (0 to 6 points) had similar rates of death or MI at 42 days, irrespective of management assignment (5% of 594 patients assigned to the conservative management arm versus 6% of 601 patients assigned to the early invasive strategy). However, in the high and very high-risk groups, patients managed with early catheterization had lower rates of death or MI than did similar patients managed conservatively (19% in the conservatively managed patients versus 12% in the invasively managed group; risk difference = 7%, 95% confidence interval [CI] −1% to 15%).
Overall, in the higher risk categories, patients treated with conservative management had a higher rate of death or MI than patients managed invasively. There was a 92% adjusted increase in risk of death or MI by 42 days with each higher category of risk score for patients randomized to the conservative arm (95% CI, 67% to 116%). By contrast, patients treated with the early invasive strategy had an adjusted increase per category of only 31% (95% CI, 4% to 67%). The interaction effect on outcomes between risk score and management assignment was significant (p = 0.03). Thus, as illustrated in Figure 1A, patients in the high or very high risk categories had lower rates of death or MI by 42 days if treated invasively rather than conservatively; no such benefit was seen in patients in the lower-risk categories.
The trend for the secondary outcome, death or MI by 365 days, was very similar to the primary endpoint findings (Fig. 1B). Patients in the high or very high risk categories had fewer deaths or MIs if randomized to early invasive therapy, whereas those in the very low, low and moderate risk groups had similar rates of outcome regardless of management assignment; however, the interaction between increments of risk and management assignment did not reach statistical significance at the 365-day point (p = 0.3).
The risk score and full multivariate model both had moderate predictive power; the C statistic for the final multivariate model was 0.66; for the one-variable logistic model with risk category, C = 0.64 (95% CI 0.60 to 0.70). The C statistic for the bootstrapped multivariate model was 0.65. The predicted outcomes of the risk score were not significantly different from the observed outcomes (Hosmer-Lemeshow goodness-of-fit p = 0.4).
Unstable angina and NQMI are very common problems throughout the industrialized world; in the U.S. alone, they accounted for more than 1.4 million hospital admissions in one recent year (35). However, controversy persists over the optimal management strategy for this syndrome, particularly since the advent of several new invasive clinical approaches. Although some have speculated that high-risk patients are most likely to benefit from an early invasive strategy (6,10,11), this concept has remained controversial. Even in trials that demonstrate overall benefit from invasive strategies (3,4), it is not clear whether most or all of this benefit is realized in a subset of the study population at highest risk of poor outcome. The present analysis demonstrates that baseline patient factors known to be markers of poor prognosis, identified at the time of presentation, can be used to target a group of patients with unstable angina or NQMI who are most likely to benefit from an early invasive management strategy. Older age, prior complicated angina, ST segment depression, and CK-MB > 5 IU/ml were all associated with a higher risk of death or MI at 42 days. A score developed from these variables proved to be a useful predictor of benefit from invasive management: patients in the high or very high risk categories who were randomized to the early invasive strategy had lower rates of adverse outcomes than did similar-risk patients who were medically treated. This finding was statistically significant for the primary outcome, death or MI by 42 days, and there was a similar trend for death or MI within 365 days.
This study demonstrates that the process of risk stratification of patients with unstable angina or NQMI can identify those who are most likely to benefit from an early invasive strategy. These findings are consistent with a close examination of the VANQWISH and FRISC II trials. In the VANQWISH trial, patients with NQMI who were treated with initial medical therapy had better outcomes than those treated with early invasive management (2). However, patients in VANQWISH entered the trial up to 72 h after presentation; thus, the patients with the most severe coronary artery disease, who were most likely to benefit from an early cardiac catheterization, might have not been entered into the trial. The FRISC II trial investigators enrolled only higher risk patients, that is, those with ST segment deviation or elevated CK-MB (3). Hence, the benefits of an early invasive management in FRISC II are consistent with our results in the higher risk categories.
Preliminary studies indicate that only high risk patients benefit from the administration of either Gp IIb/IIIa inhibitors or low-molecular-weight heparin (3,9), but prior studies have not addressed the use of risk stratification to select an early invasive versus early conservative management strategy—a choice with much larger implications for both risk and cost. Early risk stratification may be of value not only to select management strategy, but also specific pharmacologic agents within a strategy.
For example, a patient aged 65 (3 out of a possible 5 points), with chest pain suspected of being unstable angina, but no evidence for “complicated” angina (0 of 2 points), no depression in the ST segment on the initial ECG (0 of 1 point) and no elevation in CK-MB (0 of 1 point) would have a total score of 3. This patient would be categorized as low risk and would not have had any benefit from an early invasive approach. On the other hand, another 65-year-old patient (3 of 5 points) with ST segment depression ≥0.1 mV (2 of 2 points) and nocturnal angina (“complicated” angina) in two months before presentation (2 of 2 points), but with a normal CK-MB (0 of 1 point) would have a total score of 7. Seven points would put this patient in the high-risk category, suggesting that an early invasive approach may provide a reduction in the risk of death or MI by 42 days.
Cardiac troponin I and CRP were not significantly associated with death or MI within 42 days in these patients. Previously, Antman et al. (25)found that cTnI was associated with death at 42 days using the same data. We were able to reproduce this finding but did not find a similar relationship when MI was added to the composite end point. Although some researchers have found that CRP is associated with death or MI, others have not (29–31,36,37). We were not able to confirm this relationship. It is possible that this reflects variations in laboratory assay techniques or differences in patient populations.
Although the technology of coronary intervention has evolved since this trial was conducted, the approach presented provides a useful model for the evaluation of any management strategy for unstable angina. The randomized nature of the TIMI IIIB trial and its careful follow-up make it well suited for the analyses we performed. However, some important limitations of our findings must be considered. First is the post-hoc nature of the subgroup definition. Although not originally planned in the trial protocol, our analyses were hypothesis-driven and grounded in prior literature. Nonetheless, as a result, these findings will need to be confirmed in other populations. Second, the risk score was developed and tested in the same population. However, development of the score pooled all patients, and testing examined how well the score predicted outcomes in each group separately; our finding is unlikely to result from such an artifact. Furthermore, the predictive power of the model did not decrease substantially on application of the model to repeated samples of the study population using the bootstrap method, confirming the validity of the risk-stratification determination. Nonetheless, the risk score and the implications noted above will require replication in another population of patients with unstable angina or NQMI. The purpose of our risk score was not to generate a definitive risk-stratification process per se, but to allow us to investigate the differential treatment effects of the two management strategies based on accepted risk criteria.
Another important limitation of TIMI IIIB trial data is the absence of use of Gp IIb/IIIa inhibitors and low-molecular-weight heparin in the medical management strategy, as well as coronary stents in those who underwent cardiac catheterization; all patients did, however, receive aspirin and heparin. Glycoprotein IIb/IIIa inhibitors, low-molecular-weight heparin, and coronary stents are important additions to the management of unstable angina, and their use could influence the behavior of the risk-stratification approach described here. As well, diltiazem was part of the treatment protocol in the TIMI IIIB trial. This is not part of standard care, but we do not believe it influenced these results.
These findings have potentially important implications for the management of patients with unstable angina or NQMI. If confirmed in other populations, they suggest that higher risk patients are more likely than lower risk patients to benefit from early invasive management, as they are likely to have better outcomes when treated invasively than if treated with initial conservative therapy. Of equal importance is the finding that the relatively larger group of lower risk patients did not demonstrate a reduction in rates of death or MI if randomized to early invasive management. Even though TIMI IIIB was conducted before the use of coronary stents and Gp IIb/IIIa inhibitors, the underlying principle of risk stratification is likely to apply in a wide variety of therapeutic settings. If replicated in other settings and with other approaches (such as coronary stents and Gp IIb/IIIa inhibitors), such a risk-stratification scheme could provide a more evidence-based approach to the management of the very large number of patients with unstable angina and NQMI. It could also result in a better allocation of invasive therapies to patients most likely to benefit from them, sparing others the risk and cost of such interventions.
☆ The TIMI III Trial was supported by NHLBI grants R01-HL-42311, HL-42419 and HL-42428. The present study was supported by a core grant to the Division of Pharmacoepidemiology and Pharmacoeconomics from the Brigham and Women’s Hospital. Dr. Solomon is the recipient of an Arthritis Foundation (Atlanta, Georgia) Investigator Award.
- coronary artery bypass grafting
- confidence interval
- creatine kinase-MB fraction
- C-reactive protein
- cardiac troponin I
- myocardial infarction
- non–Q-wave myocardial infarction
- odds ratio
- percutaneous transluminal coronary angioplasty
- Thrombolysis In Myocardial Infarction
- tissue plasminogen activator
- Received December 27, 2000.
- Revision received May 3, 2001.
- Accepted June 25, 2001.
- American College of Cardiology
- TIMI IIIB Investigators
- ↵Cannon CP, Weintraub WS, Demopoulos L, et al. Comparison of early invasive and conservative strategies in patients with unstable coronary syndromes treated with the glycoprotein IIb/IIIa inhibitor tirofiban. N Engl J Med 2001;344:1879–87.
- Braunwald E
- ↵Braunwald E, Mark DB, Jones RH, et al. Clinical Practice Guideline Number 10: Unstable Angina: Diagnosis and Management. 86th ed. Rockville, MD: U.S. Department of Health and Human Services, Agency for Health Care Policy and Research, 1994. AHCPR Publication 94-0602.
- Iliadis E.A,
- Klein L.W,
- Vandenberg B.J,
- et al.
- Fuchs S,
- Kornowski R,
- Mehran R,
- et al.
- Jacobs D.R Jr.,
- Kroenke C,
- Crow R,
- et al.
- Boersma E,
- Pieper K.S,
- Steyerberg E.W,
- et al.
- Galvani M,
- Ottani F,
- Ferrini D,
- et al.
- Fuchs S,
- Kornowski R,
- Mehran R,
- et al.
- Toss H,
- Lindahl B,
- Siegbahn A,
- Wallentin L
- Morrow D.A,
- Rifai N,
- Antman E.M,
- et al.
- Macy E,
- Hayes T,
- Tracy R
- Hosmer D.W Jr.,
- Lemeshow S
- ↵National Hospital Discharge Survey, 1996, Hyattsville MD, National Center for Health Statistics 1998;13(138) Vital Health Stat.
- Montalescot G,
- Philippe F,
- Ankri A,
- et al.