ORIGINAL CONTRIBUTION
The Association Between Pretest Probability of Coronary Artery Disease and Stress Test Utilization and Outcomes in a Chest Pain Observation Unit Anthony M. Napoli, MD
Abstract Objectives: Cardiology consensus guidelines recommend use of the Diamond and Forrester (D&F) score to augment the decision to pursue stress testing. However, recent work has reported no association between pretest probability of coronary artery disease (CAD) as measured by D&F and physician discretion in stress test utilization for inpatients. The author hypothesized that D&F pretest probability would predict the likelihood of acute coronary syndrome (ACS) and a positive stress test and that there would be limited yield to diagnostic testing of patients categorized as low pretest probability by D&F score who are admitted to a chest pain observation unit (CPU). Methods: This was a prospective observational cohort study of consecutively admitted CPU patients in a large-volume academic urban emergency department (ED). Cardiologists rounded on all patients and stress test utilization was driven by their recommendations. Inclusion criteria were as follows: age > 18 years, American Heart Association (AHA) low/intermediate risk, nondynamic electrocardiograms (ECGs), and normal initial troponin I. Exclusion criteria were as follows: age older than 75 years with a history of CAD. A D&F score for likelihood of CAD was calculated on each patient independent of patient care. Based on the D&F score, patients were assigned a priori to low-, intermediate-, and high-risk groups (90%, respectively). ACS was defined by ischemia on stress test, coronary artery occlusion of ≥70% in at least one vessel, or elevations in troponin I consistent with consensus guidelines. A true-positive stress test was defined by evidence of reversible ischemia and subsequent angiographic evidence of critical stenosis or a discharge diagnosis of ACS. An estimated 3,500 patients would be necessary to have 1% precision around a potential 0.3% event rate in low-pretest-probability patients. Categorical comparisons were made using Pearson chi-square testing. Results: A total of 3,552 patients with index visits were enrolled over a 29-month period. The mean ( standard deviation [SD]) age was 51.3 (9.3) years. Forty-nine percent of patients received stress testing. Pretest probability based on D&F score was associated with stress test utilization (p < 0.01), risk of ACS (p < 0.01), and true-positive stress tests (p = 0.03). No patients with low pretest probability were subsequently diagnosed with ACS (95% CI = 0 to 0.66%) or had a true-positive stress test (95% CI = 0 to 1.6%). Conclusions: Physician discretionary decision-making regarding stress test use is associated with pretest probability of CAD. However, based on the D&F score, low-pretest-probability patients who meet CPU admission criteria are very unlikely to have a true-positive stress test or eventually receive a diagnosis of ACS, such that observation and stress test utilization may be obviated. ACADEMIC EMERGENCY MEDICINE 2014;21:401–407 © 2014 by the Society for Academic Emergency Medicine
From the Department of Emergency Medicine, Warren Alpert Medical School of Brown University, Providence, RI. Received July 15, 2013; revisions received October 17 and November 15, 2013; accepted November 19, 2013. Presented at the Society for Academic Emergency Medicine Annual Meeting, Atlanta, GA, May 2013. The authors have no relevant financial information or potential conflicts of interest to disclose. Supervising Editor: Alan Jones, MD. Address for correspondence and reprints: Anthony M. Napoli, MD; e-mail:
[email protected].
© 2014 by the Society for Academic Emergency Medicine doi: 10.1111/acem.12354
ISSN 1069-6563 PII ISSN 1069-6563583
401 401
402
C
ardiovascular disease remains the leading cause of death among men and women in the United States, representing greater than 25% of allcause mortality.1 Chest pain is the second most common chief complaint of patients presenting to the emergency department (ED) for treatment, although only 15% of these patients are actually found to have acute coronary syndrome (ACS).2,3 Despite the focus on early identification and treatment of patients with ACS, studies demonstrate emergency physicians (EPs) continue to “miss” 2% of both acute myocardial infarction (AMI) and unstable angina (UA).4,5 These patients have a mortality rate twice as high as those correctly identified and hospitalized.4 As such, balancing the need for avoiding missed AMI and UA with the high cost of admissions and negative evaluations is an important area of clinical research. Chest pain observation units (CPUs) offer the promise of a protocol-based, safe, and cost-effective management of undifferentiated chest pain.6–9 Components of the accelerated diagnostic protocols with CPUs vary, but often include some form of stress testing in those without initial evidence of myocardial ischemia or infarction.10 The success of CPUs has generally been predicated on safe, efficient identification of ACS while preventing 30-day major adverse cardiovascular events (MACE). We demonstrated a MACE rate of 0.3% in a CPU.9 Notably, this low rate was achieved with stress test use rates much lower than most traditional CPUs. While American College of Cardiology and American Heart Association (ACC/AHA) guidelines recommend stress utilization in the ED CPU or within 3 days, there is increasing support for alternatives to immediate stress testing.11–13 To balance the need for readily available, appropriate stress test utilization, and the risks and costs associated with overuse of stress testing, an alternative, more objective criteria to safely reduce or defer stress test use in the lowest risk subgroups would be helpful in CPU patient management. The Diamond and Forrester (D&F) score is an estimate of pretest probability of CAD based on age, sex, and presenting symptoms. It was originally developed and validated based on patients undergoing coronary angiography.14 It has recently been validated15 and is widely used, including as the basis for assessing stress test use in guidelines recommending appropriateness of stress testing.16 Although the score is validated against coronary angiography at predicting the likelihood of a 50% obstructive lesion, its application with stress testing has been much the same—predicting the likelihood of flow-limiting obstructive coronary artery disease (CAD). Thus, unlike the Thrombolysis in Myocardial Infarction (TIMI) risk score, which predicts the risk of short-term adverse events, this score predicts the likelihood of CAD. It has the potential to also objectively risk-stratify the likelihood of CAD in CPU patients. Despite this, one study of hospitalized, non-CPU patients found that stress testing did not correlate well with pretest probability of CAD based on the D&F score.17 Patients with low pretest probabilities were less likely to have positive stress tests, but no less likely to get stress tests. This suggests that there is room for improvement in stress test use, if these results can be
Napoli • PRETEST PROBABILITY OF CAD
replicated in a CPU. One benefit of a CPU may be better subjective risk assessment, such that it will correlate better with an objective score, like the D&F score. Additionally, the D&F score has the potential to augment clinical decision-making by identifying a low-risk population in which no further testing would be necessary. As such, it was hypothesized that subjective utilization of stress testing would be associated with the D&F score, that the D&F score would be associated with the likelihood of a true-positive stress test, and that there would be little benefit to stress testing in patients categorized by D&F score as having a low probability of disease. METHODS Study Design This was an observational prospective trial of consecutive chest pain patients. The institutional review board of the participating hospital approved this study with waiver of consent. Study Setting and Population Patients admitted to the CPU of an urban, academic ED with an annual census of approximately 104,000 visits were eligible for inclusion. This seven-bed CPU is contained within the ED; is open 7 days a week, 24 hours a day; and is operated by ED staff. All patients admitted to the CPU were included for analysis if they had D&F scores calculated. CPU inclusion criteria consisted of age older than 18 years, AHA low-to-intermediate risk,18 electrocardiogram (ECG) nondiagnostic for ACS, and a negative initial troponin I (ACS:Centaur troponin I, Bayer, Tarrytown, NY).19 Patients excluded from the CPU were those who left against medical advice, patients age older than75 years with a history of CAD, those with active comorbid medical problems requiring admission, or those unable to obtain stress testing in the ED for any reason. Study inclusion criteria, therefore, were largely equivalent to the criteria for admission to the CPU with the exception of age. The D&F criterion has been validated for an age range of 30 to 70 years of age. Therefore, patients included in this study were between 30 and 70 years of age and otherwise met CPU inclusion and exclusion criteria. Study Protocol All patients admitted to this unit undergo a 6-hour observational protocol including serial ECG, serial biomarkers, and continuous telemetry. Attending EPs provide patient oversight with midlevel providers (physician assistants and nurse practitioners). Cardiologists consult on all patients upon completion of the protocol and provide recommendations on the necessity and best type of stress testing. There is no formal use of the D&F score in this decision-making. Moreover, cardiologists were not provided these scores at the time of patient evaluation. Available tests include exercise stress testing, stress nuclear imaging, stress echocardiography, and computed tomography coronary angiography. Demographic, historical, and physical examination features were collected prospectively on each patient
ACADEMIC EMERGENCY MEDICINE • April 2014, Vol. 21, No. 4 • www.aemj.org
using a standardized data collection form. Collected data included demographics, clinical characteristics of pain, cardiac risk factors, vital signs, significant physical examination findings, ED treatments, and the ECG. The primary study hypothesis, inclusion and exclusion criteria, study outcome definitions, and desired data variables were defined prior to data abstraction. All patient data were entered into a Microsoft Access database (Microsoft, Redmond, WA) and a trained data abstractor, who was blinded to the study hypothesis, with extensive experience working with this database, extracted and deidentified the data prior to analysis. The study author confirmed that the correct filters had been applied to the data prior to data analysis. Patients were followed throughout their CPU or hospital stay (if admitted from the CPU) and the results of cardiac testing or cardiac interventions (including angioplasty, stenting, or coronary artery bypass grafting), the date of discharge, and the ultimate diagnoses were recorded. The consulting cardiologists were blinded to the ongoing purpose and nature of the study and therefore the D&F score. All ECGs were coded as normal, nondiagnostic, dynamic, or ischemic at the time of cardiology consultation. Criteria for definition of an abnormal ECG were as follows: a left bundle-branch block, pathologic Q-waves consistent with prior myocardial infarction, ST elevation or depression of more than 1 mm in at least two leads, any T-wave inversion in at least two leads, an ST elevation of less than 1 mm in at least two leads, a transient ST elevation of more than 2 mm, symmetric precordial T-wave inversions, and any other ST depression. A D&F score for likelihood of CAD14 was calculated on each patient on arrival to the CPU based on demographic and historical information obtained on arrival and initial evaluation in the ED. The treating EP was not asked to categorize patients and thus was blinded to the ongoing nature of the study. AMI was defined in accordance with the consensus guidelines by the European Society of Cardiology and ACC.20,21 ACS was defined as documented ischemia on stress test, coronary artery occlusion of ≥70% in at least one vessel on cardiac catheterization with subsequent percutaneous coronary intervention, a discharge diagnosis of ACS, or elevations in troponin I consistent with the aforementioned consensus guidelines.22 A positive stress test was based on the presence of reversible ischemia. A true-positive stress test was defined by subsequent angiographic evidence of critical stenosis with intervention or without intervention but a discharge diagnosis consistent with ACS. An indeterminate stress test was considered negative unless detailed review of the hospital record demonstrated further confirmatory testing or a discharge diagnosis to the contrary. A false-positive stress test was defined by a normal angiogram and the absence of 30-day MACE. Elective coronary angiography was defined as catheterization done in absence of an elevated troponin or dynamic ECG changes without a preceding stress test. The D&F score for the pretest probability of CAD is calculated based on patient age, sex, and the character of the chest pain.14 Briefly, three aspects were assessed to rate the chest pain character as typical angina, atyp-
403
ical angina, or nonanginal chest pain (asymptomatic is also an option but was not included for this study as all patients were symptomatic by definition). These characteristics were: 1) substernal chest pain or discomfort that is 2) provoked by exertion or emotional stress and 3) relieved by rest and/or nitroglycerin. Chest pain that had all three of these characteristics was classified as typical angina. Chest pain that had two of these criteria was categorized as atypical angina, and pain that had none or one of these criteria was categorized as nonanginal pain. Interrater reliability was not assessed, as the D&F score was calculated by the midlevel provider based on the ED record and patient evaluation at the time of evaluation in the CPU. Pretest probability of CAD was calculated and then categorized as low if the calculated risk was below 10%, intermediate for 10% to 90%, and high for over 90%, based on previous literature.14,17,23 Thirty-day follow-up was obtained by contacting the patients directly 30 days after admission. We attempted to contact each patient three times. If unsuccessful, in patients with primary care providers, those providers were contacted for further history of recent events. Follow-up questions included whether the patient had sought care for chest pain in the past 30 days, had followed-up with the primary care provider, or had continued symptoms. If further care was provided, follow-up results of any stress testing, catheterization, or admissions (whether ultimately from a cardiac or noncardiac etiology) were obtained. Data Analysis We have previously described a 0.3% MACE rate in our patient population.9 Based on previous experience with patients enrolled in the CPU, we estimated approximately 10% of patients would have a low probability D&F score. As a result, an estimated 3,500 patients would be necessary to have 1% precision around a potential 0.3% event rate in low pretest probability patients. Categorical comparisons were made using Pearson chi-square testing. Chi-square and t-tests were used for univariate comparisons of demographics, cardiac comorbidities, and risk scores. Chi-square tests were used to individually test the association between D&F score and stress test utilization, stress test positivity rate, and ACS. RESULTS A total of 3,918 patients were enrolled over a 29 month period; 3,552 cases were index visits that met the inclusion/exclusion criteria. The mean (SD) age was 51.3 (9.3) years, and 55% were female (Table 1). A total of 1,758 patients (49%) received stress testing (55% nuclear stress, 37% stress echocardiogram, 4% exercise, and 4% other; Figure 1). The overall ACS rate was 1.7% (95% CI = 1.3% to 2.2%). The 30-day MACE rate was 0.1% (95% CI = 0 to 0.2%): four patients underwent revascularization. A total of 48 patients (1.3%) underwent elective coronary angiography with 31% of those patients (n = 15) subsequently diagnosed with ACS. Of the 1,754 patients undergoing stress testing, the results were abnormal in 29 (1.7%). Twenty of these tests were
404
Napoli • PRETEST PROBABILITY OF CAD
Table 1 Demographic, Historical, and Risk Stratification Features Baseline Characteristics Age (yr) Female sex Race White Black Hispanic Hypertension Hyperlipidemia Diabetes Smoker TIMI score 0 1 2 3 4 D&F score
All (N = 3,552)
Stressed (n = 1,758)
51.3 (9.3) 55 (52–56)
51.9 (9) 51 (49–54)
60 11 25 52 37 19 26
64 9 23 53 39 20 24
2,472 738 271 66 5 37
(59–62) (10–12) (24–27) (50–54) (35–38) (18–20) (24–27) (70) (21) (8) (1) (0.1) (24)
243 359 122 28 2 39
(62–66) (8–11) (21–25) (50–55) (36–41) (18–22) (21–25) (71) (20) (7) (2) (0.1) (24)
Data are reported as mean (SD), percent (95% CI), or n (%). D&F = Diamond and Forrester; TIMI = Thrombolysis in Myocardial Infarction.
3918 Chest pain unit admissions
3554 Index chest pain unit visits
1754 (49%) Had stress tests
9 (0.6%) False positives
48 (1.3%) Elective coronary angiogram
20 (1.1%) True positives
15 (0.4%) Underwent PCI
Figure 1. Patients who received stress testing.
ultimately categorized as true positives, while nine were categorized as false positives. Pretest probability based on D&F score, that was subsequently recategorized into the probability categories low, medium, and high, was associated with stress utilization, risk of ACS, as well as true-positive stress tests, respectively (v2(2) = 11.6, p < 0.01; v2(2) = 49.4, p < 0.01; and v2(2) = 7.7, p = 0.02). No patients with low pretest
probabilities were subsequently diagnosed with ACS (95% CI = 0 to 0.66%) or had true-positive stress tests (95% CI = 0 to 1.6%; Table 2). While the a priori definition of low pretest probability was below 10% based on prior studies, expanding the low pretest probability to below 20%, and using it to stop testing in patients where stress testing had been planned by the cardiologist, would have had 100% sensitivity (95% CI = 83% to 100%) and 23% specificity (95% CI = 22% to 25%) for identifying true-positive stress tests (area under the receiver operator curve = 0.63, 95% CI = 0.52 to 0.75, p = 0.01; Table 3). DISCUSSION A consensus approach to the low-risk chest pain patient admitted to a CPU has not been established. While many options exist, a consistent approach based on validated criteria of predicting the likelihood of CAD would help standardize guidelines and consistency across institutions, particularly in identifying those patients who are very low risk. Unlike a prior study that failed to find an association between the D&F score and stress test utilization rate,17 we found such an association in our CPU cohort. More importantly, low-pretestprobability patients, by D&F score, were at very low risk of ACS or a true-positive stress test, such that stress testing may be safely deferred or avoided. The most effective and efficient management of lowrisk chest pain patients remains elusive. CPUs have been effective in reducing missed ACS24 and improving adherence to the Centers for Medicare and Medicaid Services Core Measures25 through protocols that offer a combination of serial ECGs, biomarkers, and varying degrees and types of stress testing. While these protocols identify a patient cohort at under 1% risk of MACE at 30 days,9,24 many studies have demonstrated the overall risk of ACS in patients admitted to CPUs is low (2% to 4%).9,26 The role of stress testing in CPUs has evolved. Early CPUs relied on stress testing as a key component of the evaluation of low-risk chest pain patients; nearly all patients in such cases received stress tests. While the addition of stress testing to serial biomarkers and observation has not been shown to influence mortality, ACC/AHA guidelines recommend that stress testing for low-risk chest pain patients occur in CPUs or within 72 hours of discharge.13 However, the recommendation of stress testing within 72 hours has been questioned, and one study demonstrated that it
Table 2 D&F Score and Outcomes of Patients and Diagnostic Testing Score Overall Low probability Intermediate probability High probability p-value*
Number
Stress Rate,%
3,543 451 3,023 69
49 42 50 54
