Box plots of Thrombolysis in Myocardial Infarction (TIMI) (A), Platelet Glycoprotein IIb/IIIa in Unstable Angina: Receptor Suppression Using Integrilin (PURSUIT) (B), and Global Registry of Acute Cardiac Events (GRACE) (C) risk scores by patient risk categories according to the treating physician. Horizontal bars indicate median; bottom and top edges of boxes, 25th and 75th percentiles, respectively; top and bottom whiskers, smallest and largest values that are not outliers, respectively; and circles outside the whiskers, outliers (1.5 to 3 box lengths from the edge of the box).
Yan AT, Yan RT, Huynh T, Casanova A, Raimondo FE, Fitchett DH, Langer A, Goodman SG, . Understanding Physicians' Risk Stratification of Acute Coronary SyndromesInsights From the Canadian ACS 2 Registry. Arch Intern Med. 2009;169(4):372-378. doi:10.1001/archinternmed.2008.563
Copyright 2009 American Medical Association. All Rights Reserved. Applicable FARS/DFARS Restrictions Apply to Government Use.2009
An important treatment-risk paradox exists in the management of acute coronary syndromes (ACSs). However, the process of risk stratification by physicians and its relationship to the management of ACS have not been well studied. Our objective was to examine patient risk assessment by physicians in relation to treatment and objective risk score evaluation and the underlying patient characteristics that physicians consider to indicate high risk.
The prospective Canadian ACS 2 Registry recruited 1956 patients admitted for non-ST-segment elevation ACS in 36 hospitals in October 2002 to December 2003. We recorded patient risk assessment by the treating physician and case management on standardized case report forms and calculated the Thrombolysis in Myocardial Infarction (TIMI), Platelet glycoprotein IIb/IIIa in Unstable angina: Receptor Suppression Using Integrilin Therapy (PURSUIT), and Global Registry of Acute Cardiac Events (GRACE) risk scores.
Of the 1956 patients with ACS, 347 (17.8%) were classified as low risk, 822 (42.0%) as intermediate risk, and 787 (40.2%) as high risk by their treating physicians. Patients considered as high risk were more likely to receive aggressive medical therapies and to undergo coronary angiography and revascularization. However, there were only weak correlations between risk assessment by physicians and all 3 validated risk scores. In multivariable analysis, history of stroke, worse Killip class, presence of ST-segment deviation, T-wave inversion, and positive cardiac biomarker status were all independently associated with high-risk categorization by the treating physician, while advanced age and previous coronary bypass surgery were independent negative predictors. There was no significant association between the high-risk category and several established prognosticators, such as history of heart failure, hemodynamic variables, and creatinine level.
Contemporary risk stratification of ACS appears suboptimal and may perpetuate the treatment-risk paradox. Physicians may not recognize and incorporate the most powerful adverse prognosticators into overall patient risk assessment. Routine use of validated risk score may enhance risk stratification and facilitate more appropriate tailoring of intensive therapies toward high-risk patients.
Because non-ST-segment elevation acute coronary syndromes (ACSs) encompass a broad spectrum of clinical conditions, accurate risk stratification is a critical step in optimal patient care.1,2 The overriding principle is to selectively target more intensive medical and interventional treatment toward high-risk patients.
Recent studies have demonstrated important treatment disparities, showing that high-risk patients paradoxically receive less aggressive therapies.3- 8 This “treatment-risk paradox” has been based on objective risk assessment. Although validated risk scores may supersede and provide incremental prognostic information beyond subjective risk assessment,9- 12 they have not been widely adopted in routine clinical practice.13 Indeed, physicians' underestimation of patient risk may contribute to the observed treatment-risk paradox.14 Nevertheless, the process of risk stratification and its relationship to patient care in the “real world” have not been elucidated.
Accordingly, our objective was to examine (1) patient risk assessment by physicians in relation to treatment and objective risk score evaluation and (2) the underlying patient characteristics that physicians consider to indicate high risk. A better understanding of the current risk stratification scheme and its shortcomings may be crucial to improve the quality of care for patients with ACS.
The study rationale and design of the Canadian ACS Registries have been previously described.7,15 Briefly, the Canadian ACS 2 Registry was a prospective, multicenter, observational study of patients 18 years or older who were hospitalized for suspected non-ST-segment elevation ACS (symptoms consistent with acute cardiac ischemia within 24 hours of onset), which was not accompanied or precipitated by a serious concurrent illness (eg, trauma or gastrointestinal tract bleeding). There were no other specific exclusion criteria, and we encouraged all participating hospitals to enroll consecutive patients to reduce patient selection bias. At each site, the designated physician or study coordinator collected data on patient demographics, clinical features, in-hospital treatment, and outcome on standardized case report forms, which were then forwarded to the Canadian Heart Research Centre and scanned into an electronic database (Teleform, version 7.0; Cardiff, San Diego, California). The appendix of the case report form included a brief summary of the contemporary Canadian practice guidelines on the management of ACS.16 The Canadian Heart Research Centre performed central data checks and sent queries to sites for data correction. The research ethics board at each participating hospital approved the study protocol, and the local study coordinator obtained informed consent from all patients who were followed up after discharge.
Between October 2002 and December 2003, the ACS 2 Registry recruited 2359 patients with suspected ACS from 36 participating hospitals (33% had on-site cardiac catheterization laboratories and 42% had teaching affiliations) in Canada; 1956 (82.9%) had a final diagnosis of ACS (unstable angina or non-ST-segment elevation myocardial infarction) according to the most responsible physician. These 1956 patients constituted the present study cohort. Cardiologists were the most responsible physicians for the majority of patients (72.8%), followed by general internists (26.0%) and others (1.2%).
On the case report form, we asked the treating physician to categorize each patient into low-, intermediate-, and high-risk groups, based on overall risk assessment of medical history (eg, age), physical examination (eg, heart failure or hemodynamic instability), and laboratory investigations (eg, electrocardiogram, cardiac biomarker).16 The early risk stratification approach based on initial clinical presentation, according to the Canadian ACS guidelines (similar to the American College of Cardiology/American Heart Association [AHA/ACC] guidelines), was recommended in the case report form. However, physicians were not required to prospectively complete this section within a certain time frame after admission. For comparison with systematic risk assessment, we calculated the Platelet Glycoprotein IIb/IIIa in Unstable Angina: Receptor Suppression Using Integrilin Therapy (PURSUIT), Thrombolysis in Myocardial Infarction (TIMI), and Global Registry of Acute Cardiac Events (GRACE) risk scores for each patient, according to the published nomograms.17- 19 All 3 risk scores were determined using clinical characteristics at presentation. To calculate the TIMI risk score for patients without prior coronary angiography, we assigned 1 point if there was a history of myocardial infarction or coronary revascularization in substitution for known coronary artery stenosis of 50% or greater.20,21 Although the proportions of missing data were small (<2% for most variables, except 8.7% for Killip class), we could not calculate the TIMI, PURSUIT, and GRACE risk scores for 9 (0.5%), 195 (10.0%), and 220 (11.2%) patients, respectively.
Continuous variables are presented as medians with interquartile ranges and discrete variables as percentages. We used the Kendall τ-b test to examine trends and correlations. The TIMI, PURSUIT, and GRACE risk scores were analyzed as continuous variables. For presentation of the patients' baseline characteristics, we stratified the study cohort into 3 groups based on the physicians' risk assessments (low, intermediate, and high risk).
Because evidence-based practice guidelines recommend early aggressive therapies (eg, glycoprotein IIb/IIIa inhibitors and an early invasive strategy) specifically for high-risk patients, we focused on clinical features that distinguish the high-risk group from the low- and intermediate-risk groups combined. To determine the patient characteristics that independently influenced the physicians' risk assessments, we developed a multivariable logistic regression model to predict the high-risk category. Candidate predictor variables included all relevant baseline characteristics (listed in Table 1) that were associated with high-risk category on bivariate analysis (P < .25) and the components of the TIMI, PURSUIT, and GRACE risk scores. Because the ACS 2 Registry only recruited few (5.8%) very elderly (aged ≥85 years) patients, we stratified age into 3 groups using the following recommended cut points: younger than 65 years, 65 to 74 years, and 75 years or older.22 We also considered the following variables in the multivariable model: physician's discipline (cardiologist vs others), teaching vs nonteaching hospitals, and the presence vs absence of cardiac catheterization facilities on site. A parsimonious model was constructed by backward elimination (for P > .05) of the candidate predictor variables. We used generalized estimating equations to account for within-hospital correlations,23 since patients admitted to the same hospital were more likely to be similar. We tested for selected interaction terms (with age groups) in the final model. To confirm the robustness of our results, we performed additional analyses. First, we repeated the analysis to include all enrolled patients with suspected ACS (ie, including those with a final non-ACS diagnosis). Second, because physicians might be less adept at integrating continuous variables (heart rate, systolic blood pressure, and creatinine level) into overall risk assessment, they were also analyzed as dichotomous variables (using the following predefined cut points for abnormal: >100 beats/min, <100 mm Hg, and >1.36 mg/dL [to convert to micromoles per liter, multiply by 88.4], respectively) in a separate multivariable model. Third, the intermediate- and high-risk groups were combined in a sensitivity analysis, with low-risk patients serving as the reference group. Finally, we excluded patients who had in-hospital death and/or myocardial (re-) infarction because these events might have had an undue impact on physician's risk assessment. Model discrimination and calibration were evaluated by the C statistic and Hosmer-Lemeshow goodness-of-fit test, respectively. We performed data analysis using SPSS version 15.0 (SPSS Inc, Chicago, Illinois) and considered 2-sided P values <.05 as statistically significant.
Of the 1956 patients with ACS in this study, 347 (17.8%) were classified as low risk, 822 (42.0%) as intermediate risk, and 787 (40.2%) as high risk by their treating physicians. Table 1 summarizes the baseline characteristics of the patients stratified by the risk categories. Patients considered to be high risk were more likely to have diabetes, worse Killip class, ST-segment deviation, T-wave inversion, and positive biomarker status on presentation. Age was not different across the 3 risk groups.
Table 2 summarizes the use of antiplatelet and antithrombin medications within 24 hours of hospitalization. In general, there were significant positive trends in the use of these drugs across the higher risk groups. Compared with the low- and intermediate-risk groups, the glycoprotein IIb/IIIa inhibitor administration rate was almost 10 times and 2.5 times greater, respectively, in the high-risk group.
Table 3 demonstrates a similar positive relationship between the use of invasive cardiac procedures and the risk category. High-risk patients more frequently underwent cardiac catheterization, percutaneous coronary intervention, and coronary bypass surgery during index hospitalization (P for trend, all <.001). Furthermore, the time to these procedures was significantly shorter in the high-risk group.
Although there were positive significant trends (P < .001 for all) with increasing TIMI, PURSUIT, and GRACE risk scores across the higher risk groups (Table 1), the correlations were weak. The Kendall τ-b correlation coefficients with the physician-assigned risk groups were 0.08, 0.10, and 0.14 for TIMI, PURSUIT, and GRACE risk scores, respectively.
The Figure illustrates the relatively wide scatter in the calculated TIMI, PURSUIT, and GRACE risk scores within each group. There was also substantial overlap among the 3 risk groups, with some low-risk patients having high TIMI, PURSUIT, and GRACE risk scores. Conversely, some high-risk patients had relatively low calculated TIMI, PURSUIT, and GRACE risk scores.
In multivariable analysis, history of stroke, worse Killip class, presence of ST-segment deviation, T-wave inversion, and positive cardiac biomarker status were all independent predictors of the high-risk category according to the treating physician, while elderly patients and those with previous coronary bypass surgery were less likely to be regarded as high risk (Table 4). History of myocardial infarction or heart failure, heart rate, systolic blood pressure, and creatinine level were not significantly associated with the high-risk category. The model C statistic was 0.74, and the Hosmer-Lemeshow P value was .58, indicating good discrimination and calibration, respectively. Physician's specialty and hospital type (teaching and on-site cardiac catheterization facilities) were not retained in the final model (all P > .25) and exhibited no significant interactions with age groups. Similar results were obtained when heart rate, systolic blood pressure, and creatinine level were analyzed as dichotomous variables or when all patients with suspected ACS were included. Sensitivity analysis revealed that age, previous myocardial infarction or heart failure, and hemodynamic variables were also not independent predictors of the intermediate- or high-risk group. Finally, exclusion of patients who died or had a myocardial (re-)infarction during admission did not alter the main results, except that previous stroke and previous coronary bypass surgery were no longer significant independent predictors (P = .07 and .16, respectively).
We further explored possible age-related differences by examining interaction terms. The adjusted odds ratios for ST-segment deviation were 2.61 (95% CI, 1.83-3.72; P < .001) for age younger than 65 years, 1.70 (95% CI, 1.10-2.62; P = .02) for age between 65 and 74 years, and 1.91 (95% CI, 1.16-3.16; P = .01) for age 75 years or older; the adjusted odds ratios for abnormal biomarker were 4.23 (95% CI, 3.00-5.96; P < .001), 5.77 (95% CI, 3.25-10.26; P < .001), and 6.72 (95% CI, 3.84-11.76; P < .001), respectively.
In this “real-world” study of a broad spectrum of patients with non-ST-segment elevation ACS, we found that treatment intensity was commensurate with the patient's risk as estimated by the treating physician. However, there was a poor correlation between physicians' risk assessments and validated risk scores. Furthermore, several well-established and powerful prognosticators did not have an impact on the physician's estimation of the patient's risk. To our knowledge, the present study is the first to closely examine the relationship between validated risk scores and physicians' risk assessments in practice.
Current ACC/AHA practice guidelines state that “optimal risk stratification requires accounting for multiple prognostic factors simultaneously by a multivariable approach”1(p165) and specifically endorse the use of the TIMI, PURSUIT, and GRACE risk scores as potentially useful risk assessment tools (class IIa recommendation). These risk scores have been developed and extensively validated in various clinical trial and registry patient populations, with adequate discriminatory performance for routine clinical use.12,18- 20,24- 29 More important, these risk scores are superior to risk assessment by physicians in predicting long-term outcome in the Canadian ACS 2 Registry, implying that they may be a valuable adjunct to clinical judgment in medical decision making.12 Despite this, risk scores seem to be underused in the real world.13
The present study demonstrates a weak correlation between risk assessment by physicians and all 3 validated risk scores. As expected, we observed a direct relationship between treatment intensity and patient risk as perceived by the treating physician—patients deemed to be high risk were more likely to receive potent antithrombotic therapies and to undergo coronary angiography and revascularization. In contrast, when patients with ACS were stratified by risk scores, a treatment-risk paradox became evident and was consistent in the Canadian ACS Registries and several other large observational studies.3- 8 Moreover, in the Canadian ACS 2 Registry, of the patients who were not referred for cardiac catheterization because they were thought to be “not at high enough risk,” the majority were at intermediate or high risk based on their TIMI risk score.14 Thus, our findings strongly suggest that the current treatment-risk paradox is at least partly mediated by ineffective risk stratification.
Of particular concern is the paradoxical assignment of elderly patients with ACS to the lower risk group by their treating physicians. Physicians may hold the misguided belief that ACS in a young patient must portend a rapidly progressive course of disease, while failing to recognize that severe coronary disease and left ventricular dysfunction are in fact more prevalent among elderly patients.14,22 In the GRACE risk model for postdischarge 6-month mortality, which was derived from more than 15 000 unselected patients with ACS, age alone (Wald χ2 = 228.6) accounts for most of the prognostic information and is much stronger than most other well-recognized adverse prognosticators, such as positive biomarker status (Wald χ2 = 33.3) or ST depression (Wald χ2 = 19.2).30 Even in randomized controlled trials that preferentially excluded older patients with serious comorbidities, age remains the single most powerful predictor of death and the composite end point of death or myocardial (re-)infarction.17,31- 34 Because an early invasive strategy is more effective in high-risk ACS, elderly patients may stand to gain the greatest therapeutic benefits.22 In TACTICS (Treat Angina With Aggrastat and Determine Cost of Therapy With an Invasive or Conservative Strategy)-TIMI 18, an early invasive strategy conferred an absolute risk reduction in death or myocardial (re-)infarction at 6 months that was most pronounced in the elderly group (aged >75 years).35 Although treatment decisions must be carefully individualized by taking the patient's overall health context into consideration, the mere underrecognition of risk in elderly patients may deprive them of substantial improvement in survival and quality of life.
Physicians may tend to focus on dichotomous findings (normal vs abnormal) in risk assessment. Both ST-segment deviation on the electrocardiogram and elevated cardiac biomarker are strong independent predictors of the high-risk category in the present study. While this is concordant with guideline recommendations,1,2,16 it is noteworthy that other readily available clinical data, such as previous heart failure, heart rate, and blood pressure on presentation, carry at least as much prognostic significance as these laboratory findings17,19,30 but are underrepresented in the physicians' risk assessment. These results lend further credence to the notion that without the aid of risk scores, accurate and comprehensive integration of numerous prognostic factors is a daunting task, which may prove to be overwhelming in day-to-day practice.9- 13
Several study limitations should be noted. Although encouraged in this registry, we could not confirm consecutive patient enrollment, and participating hospitals were not a true population-based random sample. Risk assessment by physicians might be influenced by unusual high-risk features (eg, very strong family history of premature coronary artery disease or known left main coronary artery disease) not captured on the case report form, although overall, it remained inferior to risk score assessment in predicting outcome.12 Furthermore, physicians' risk assessments may have been biased or enhanced by subsequent test results and events during index hospitalization. Nevertheless, we have previously shown that the prognostic accuracy of risk scores (based only on initial presentation) was superior to risk assessment by physicians for 1-year outcome. Clinical decision making in the management of ACS is complex, and accurate risk stratification is only one of its key components. We did not collect detailed information on physician characteristics such as training and experience, which may be related to proficiency in patient risk estimation. Nonetheless, our results reflect real-world practice. Finally, the previous iteration of ACS treatment guidelines in 200236 did not specifically support the use of risk scores for risk stratification when the ACS 2 Registry was conducted. Yet, this should not detract from our conclusion that global subjective risk assessment often neglects to incorporate all the key prognosticators.
In conclusion, contemporary risk stratification of ACSs appears suboptimal and may promote the existing treatment-risk paradox. Physicians often fail to identify the most powerful adverse prognosticators and to effectively integrate them into overall risk assessment. Our findings underscore the challenges of optimal tailoring of evidence-based therapies in the real world, and the potential utility of validated risk scores to better guide individual patient care. Coupled with sound clinical judgment, systematic application of validated risk scores may present a novel and promising strategy to eradicate the treatment-risk paradox in the management of ACSs.
Correspondence: Andrew T. Yan, MD, Division of Cardiology, St Michael's Hospital, Queen 6-030, 30 Bond St, Toronto, ON M5B 1W8, Canada (email@example.com).
Accepted for Publication: September 8, 2008.
Author Contributions:Study concept and design: A. T. Yan and Goodman. Acquisition of data: Raimondo, Langer, and Goodman. Analysis and interpretation of data: A. T. Yan, Huynh, Casanova, Raimondo, Fitchett, Langer, and Goodman. Drafting of the manuscript: A. T. Yan and R. T. Yan. Critical revision of the manuscript for important intellectual content: A. T. Yan, R. T. Yan, Huynh, Casanova, Raimondo, Fitchett, Langer, and Goodman. Statistical analysis: A. T. Yan, R. T. Yan, Casanova, and Langer. Obtained funding: Goodman. Administrative, technical, and material support: Goodman. Study supervision: Fitchett and Goodman.
Financial Disclosure: Dr A. T. Yan has received research grant support and honoraria from Sanofi Aventis and Bristol-Myers Squibb. Dr Huynh has received research grant support from Sanofi Aventis, Bristol-Myers Squibb, and GlaxoSmithKline. Dr Fitchett has received speaker and consulting honoraria and research grant support from Key Pharmaceuticals, Bristol-Myers Squibb, Sanofi Aventis, Pfizer, and Key Schering. Dr Langer has received research grant support and/or honoraria for educational activities and/or served as a consultant to Astra Zeneca, Bayer, Biovail, BMS, Boston Scientific, Cordis (J&J), DuPont, Eli Lilly, Fournier, GlaxoSmithKline, Guidant, Medtronic, Merck Schering, Novartis, Oryx, Pfizer, Roche, Sanofi Aventis, and Servier. Dr Goodman has received speaker and consulting honoraria and research grant support from Key Pharmaceuticals, Bristol-Myers Squibb, Sanofi Aventis, Pfizer, and Key Schering.
Funding/Support: The Canadian ACS 2 Registry was sponsored by the Canadian Heart Research Centre (a federally incorporated not-for-profit academic research organization), Pfizer Canada Inc, Sanofi Aventis Canada Inc, and Bristol-Myers Squibb Canada Inc. Dr A. T. Yan is supported by the Canadian Institutes of Health Research and the New Investigator Award from the Heart and Stroke Foundation of Canada. Dr R. T. Yan is a recipient of the Research Fellowship Award from the Canadian Institutes of Health Research and the Detweiler Travelling Fellowship from the Royal College of Physicians and Surgeons of Canada.
Role of the Sponsors: The industrial sponsors had no involvement in the study conception or design; collection, analysis, and interpretation of data; in the writing, review, or approval of the manuscript; and in the decision to submit the manuscript for publication.
Additional Contributions: Sue Francis provided secretarial assistance. We are indebted to all the study investigators, coordinators, and patients who participated in the Canadian ACS Registries.