CARDIOLOGY/ORIGINAL RESEARCH
A Clinical Prediction Rule for Early Discharge
of Patients With Chest Pain
Jim Christenson, MD
Grant Innes, MD
Douglas McKnight, MD
Christopher R. Thompson, MD
Hubert Wong, PhD
Eugenia Yu, MSc
Barb Boychuk, BSc
Eric Grafstein, MD
Frances Rosenberg, MD, PhD
Kenneth Gin, MD
Aslam Anis, PhD
Joel Singer, PhD
From the University of British Columbia (Christenson, Innes, McKnight, Thompson, Wong,
Yu, Boychuk, Grafstein, Rosenberg, Gin, Anis, Singer); St. Paul’s Hospital (Christenson,
Innes, Thompson, Boychuk, Grafstein, Rosenberg); Vancouver Hospital (McKnight, Gin); and
the Center for Health Evaluation and Outcome Sciences (Christenson, Innes, Grafstein,
Wong, Yu, Anis, Singer) Vancouver, British Columbia, Canada.
Study objective: Current risk stratification tools do not identify very-low-risk patients who can be safely
discharged without prolonged emergency department (ED) observation, expensive rule-out protocols, or
provocative testing. We seek to develop a clinical prediction rule applicable within 2 hours of ED arrival
that would miss fewer than 2% of acute coronary syndrome patients and allow discharge within 2 to
3 hours for at least 30% of patients without acute coronary syndrome.
Methods: This prospective, cohort study enrolled consenting eligible subjects at least 25 years old at a
single site. At 30 days, investigators assigned a diagnosis of acute coronary syndrome or no acute
coronary syndrome according to predefined explicit definitions. A recursive partitioning model included
risk factors, pain characteristics, physical and ECG findings, and cardiac marker results.
Results: Of 769 patients studied, 77 (10.0%) had acute myocardial infarction and 88 (11.4%) definite
unstable angina. We derived a clinical prediction rule that was 98.8% sensitive and 32.5% specific.
Patients have very low risk of acute coronary syndrome if they have a normal initial ECG, no previous
ischemic chest pain, and age younger than 40 years. In addition, patients at least 40 years old and
with a normal ECG result, no previous ischemic chest pain, and low-risk pain characteristics have very
low risk if they have an initial creatine kinase-MB (CK-MB) less than 3.0 mg/L or an initial CK-MB greater
than or equal to 3.0 mg/L but no ECG or serum-marker increase at 2 hours.
Conclusion: The Vancouver Chest Pain Rule for early discharge defines a group of patients who can be
safely discharged after a brief evaluation in the ED. Prospective validation is needed. [Ann Emerg Med.
2006;47:1-10.]
0196-0644/$-see front matter
Copyright ª 2006 by the American College of Emergency Physicians.
doi:10.1016/j.annemergmed.2005.08.007
INTRODUCTION
Background
Approximately 15% to 25% of patients who present to
emergency departments (EDs) with undifferentiated chest pain
prove to have acute coronary syndrome within 30 days. US data
suggest that 2.1% of patients with acute myocardial infarction
and 2.3% of patients with unstable angina are initially
misdiagnosed,1 whereas a recent Canadian study identified 11
Volume 47, no. 1 : January 2006
of 241 (4.6%) missed cases of acute myocardial infarction
and 10 of 157 (6.4%) cases of missed unstable angina.2
Chest pain units reduce the rate of missed myocardial infarction
but do so in part by including very-low-risk patients in
extensive rule-out myocardial infarction protocols.3-5
Many investigators have developed chest pain risk
stratification tools. Goldman et al6 developed a clinical/ECG
algorithm that identified patients with less than 7% risk of acute
Annals of Emergency Medicine 1
Christenson et al
Clinical Prediction Rule for Early Discharge of Chest Pain Patients
Editor’s Capsule Summary
What is already known on this topic
Although several algorithms and computer programs
have been developed for emergency department (ED)
patients with potential acute coronary syndrome, none
has achieved sufficient accuracy to allow safe and rapid
discharge of ED chest pain patients.
What question this study addressed
This study developed a clinical decision rule to identify
low-risk ED chest pain patients who can be safely
discharged within 2 hours of presentation. This study did
not prospectively evaluate this rule.
What this study adds to our knowledge
Not surprisingly, this study found that patients younger
than 40 years and with normal initial ECG results and no
previous ischemic chest pain were at very low risk of acute
coronary syndrome. Additionally, patients older than
40 years had low-risk if they had a normal ECG result, no
previous ischemic chest pain, low risk chest pain
characteristics, and an initial creatine kinase-MB
(CK-MB) less than 3.0 ug/L or an initial CK-MB greater
than 3.0 ug/L but without an increase or ECG changes
within 2 hours.
How this might change clinical practice
This derivation study should not change clinical practice
because it has not yet been validated. It does, however,
provide support for the general belief that young patients
with a normal ECG result and no cardiac risk factors are
at very low risk of an adverse cardiovascular event.
myocardial infarction. Limkakeng et al7 subsequently found
that 4.9% of patients who had low-risk Goldman criteria
and a negative initial troponin I assay result experienced death,
acute myocardial infarction, or revascularization within 30 days.
Pozen et al8,9 developed a 7-item predictive equation that
reduced coronary care unit admissions but not inappropriate
discharges. Selker et al10-12 modified this to create the acute
cardiac ischemia time-insensitive predictive instrument
(ACI-TIPI), which defined low risk as less than 10% chance
of acute coronary syndrome. The Erlanger protocol13 is an
intense 2-hour assessment that includes serial ECGs and
creatine kinase-MB (CK-MB) and troponin measurements,
but it does not define a subset of patients who can forgo
nuclear stress testing. The American Heart Association/Agency
for Health Care Policy and Research guidelines suggest early
discharge only for patients with ‘‘evidence’’ of an alternate
diagnosis.14 Unfortunately, few patients clearly fall into this
category. Our own study of patients with chest pain at 2
Vancouver hospitals identified that 5.4% of patients with
acute coronary syndrome were discharged from the ED without
a diagnosis or planned investigations and only 30% of those
2 Annals of Emergency Medicine
without acute coronary syndrome were discharged less than
three hours after ED admission.2
Importance
Clinical prediction rules are decisionmaking tools for
clinicians that contain elements of the medical history, physical
examination, and simple diagnostic tests.15 An objective clinical
prediction rule to identify very-low-risk patients with chest
pain who can be safely discharged without prolonged ED
observation, expensive rule-out protocols, or provocative
testing is needed. Such a rule would help reduce emergency
crowding, minimize patient inconvenience, and improve
cost-effectiveness of acute coronary syndrome diagnostic testing.
Goals of This Investigation
Our specific objective was to develop a clinical prediction
rule that would improve on current practice by identifying
patients with chest pain who are safe for discharge after 2 hours
of ED evaluation. The rule will miss fewer than 2% of acute
coronary syndrome patients and allow discharge within 2 to
3 hours of at least 30% of patients without acute coronary
syndrome.
MATERIALS AND METHODS
Study Design
Using established methodology for clinical prediction
rules,15-17 this prospective cohort study was conducted in 2
separate periods between June 29, 2000, and January 24, 2003,
when research assistants were funded.
Setting and Selection of Participants
Patients presenting to St. Paul’s Hospital, an urban tertiary
care ED, with a primary complaint of anterior or lateral chest
pain were eligible for the study. Research assistants obtained
informed consent and enrolled eligible patients between 7 AM
and 10 PM 7 days per week. Patients were excluded if they were
younger than 25 years, had a clear traumatic or radiologically
evident cause, were enrolled in the study in the previous 30
days, had a terminal noncardiac illness, had severe
communication problems, were without a fixed address in the
province of British Columbia, or were without available
telephone contact. The Providence Health Care Research Ethics
Board approved the study protocol.
Data Collection and Processing
This report conforms to the published standard reporting
guidelines for studies evaluating risk stratification of ED
patients with potential acute coronary syndromes,18 although
they were developed before this publication. A total of 123
potential predictor variables were screened (Figure 1). During
the first 2 hours of the index visit, research assistants
documented vital signs every 30 minutes, initiated ST-segment
monitoring, and ensured that serial ECGs (0, 1, and 2 hours)
were performed. Cardiac markers at 0 and 2 hours were
evaluated on several platforms: creatine phosphokinase (Access,
Volume 47, no. 1 : January 2006
Christenson et al
Pain and Associated Symptom Characteristics
Age*
Sex
Chest pain present on arrival?
Chest pain continue at rest?
Number of episodes in previous 24 hours*
Number of episodes in previous week*
Maximum duration of previous episode
Diaphoresis
Weakness
Shortness of breath
Nausea
Vomiting
Palpitations
Syncope
Dizziness/lightheadedness
Radiation to left arm/shoulder*
Radiation to right arm/shoulder*
Radiation to back
Radiation to neck*
Radiation to jaw*
Radiation to abdomen
Radiation to lower limb
Pain center of chest*
Pain in left lateral chest
Pain in right anterior chest
Pain in right lateral chest
Most prominent character of pain*
Heavy chest discomfort
Maximum discomfort level*
Pain diagnosed in past as AMI*
Pain diagnosed in past as angina*
Pain diagnosed in past as either AMI or angina
Pain worse with deep breaths*
Pain worse with movement
Pain worse in different positions
Did pain improve within 5 minutes of nitroglycerin?*
Did pain improve within 5 minutes of oxygen?
Did pain improve within 5 minutes of liquid antacid?
During 2 hours, did pain relieve and not return?*
During 2 hours, did pain continue?*
During 2 hours, did pain come and go?
Clinical Prediction Rule for Early Discharge of Chest Pain Patients
Hx hyperlipidemia*
Parent or sibling with heart attack \65 years
Ratio of weight (kg) over height (cm)
Recreational substance use
Alcohol use in past 24 hours
Cocaine use in 48 hours
ECG Findings
First ECG result normal*
First ECG with only T-wave flattening*
Aggregate of ECGs no ischemic changes*
2-h ST segment monitoring (any ST elevation or ST depression)*
Laboratory Markers
0-h CPK (Standard)*
2-h CPK (Standard)*
Difference in CPK (2 h–0 h) (Standard)*
0-h Tn I or T (Standard)*
2-h Tn I or T (Standard)*
Difference in Tn I or T (2 h–0 h) (Standard)*
0-h CRP
0-h CK-MB (Biosite)*
2-h CK-MB*
Difference in CK-MB (2 h–0 hour) (Biosite)*
0-h TnI (Biosite)*
2-h TnI (Biosite)*
Difference in TnI (2 h–0 h) (Biosite)*
0-h Myoglobin (Biosite)*
2-h Myoglobin (Biosite)*
Difference in myoglobin (2 h–0 h) (Biosite)*
Cardiovascular History
Previous
Previous
Previous
Previous
Previous
Previous
Previous
Previous
Previous
Previous
Previous
Previous
Previous
angina*
MI*
cardiac arrest
angiogram yes and previous lesion O50%*
angiogram and no lesion O50%
PCI*
peripheral vascular surgery
CABG*
CVA
TIA
ventricular arrhythmia
atrial arrhythmia
unknown arrhythmia
Physical Examination
Medications
Chest clear
Crackles at bases
Crackles to scapulae
Wheezes
Heart sounds normal
Systolic murmur
Diastolic murmur
S3
S4
Carotid pulses equal with no bruit
Radial pulses equal
Chest tenderness reproducing pain*
Right arm systolic pressure, left arm systolic pressure
Systolic pressure \90 mm Hg in 2 hours
Bradycardia \50 beats/min in 2 hours*
Tachycardia O100 beats/min in 2 hours
Jugular venous pressure*
Previous
Previous
Previous
Previous
Previous
Previous
Previous
Previous
Previous
Previous
Previous
Previous
Previous
Risk Factors
Regular smoker
Hx diabetes*
Hx hypertension*
Volume 47, no. 1 : January 2006
ASA*
other antiplatelets
calcium-channel blockers
ACE inhibitor
digitalis
b-blocker
Coumadin
low-molecular-weight heparin
nitrates*
lipid-lowering agents
antiarrhythmics
diuretics*
antihypertensives
Figure 1. List of all predictors collected for possible inclusion in
the prediction rule. Those meeting criteria for inclusion in the
model are identified with an asterisk.
AMI, Acute myocardial infarction; Hx, history; MI, myocardial
infarction; PCI, percutaneous coronary intervention; CABG, coronary
artery bypass graft; CVA, cerebrovascular accident; TIA, transient
ischemic attack; ASA, acetyl salicylic acid; ACE, angiotensinconverting enzyme.
Annals of Emergency Medicine 3
Christenson et al
Clinical Prediction Rule for Early Discharge of Chest Pain Patients
Table 1. Baseline characteristics and outcomes of patient
cohort (n=769).
Characteristic
Mean age (yGSD)
Age range, y
% Male
% With diabetes
% Smokers
% With hypertension
% With hyperlipidemia
Median time from symptom onset
to arrival in minutes (quartiles)
Mean systolic BP on arrival (mm HgGSD)
Mean pulse rate on arrival (beats/minGSD)
Mean respiratory rate on arrival (breaths/minGSD)
% With past MI
% With past angina
% With abnormal initial ECG result
% With pain similar to past MI or angina
% With pain radiating to left arm, jaw, or neck
% With pain increasing with a deep breath
% With chest pressure reproducing pain
% with CK-MB \3.0 mg/L on arrival
% With any increase in Tn during 2 hours
% With any increase in CK-MB during 2 hours
30-Day outcomes (%)
Definite AMI
Definite unstable angina
Possible unstable angina
AE but no acute coronary syndrome
No acute coronary syndrome or adverse event
Overall 30-day mortality
Mortality with AMI (%)
Mortality with definite unstable angina (%)
Mortality with possible unstable angina (%)
Mortality with no acute coronary syndrome
but an adverse event (%)
Mortality with no acute coronary syndrome
and no adverse event (%)
Value
58G14
25–93
62.0
16.6
30.4
38.2
36.2
104 (52.5, 262)
142G27
81G20
19G3
24.2
35.9
30.1
27.0
59.0
32.8
16.6
63.5
19.5
41.8
77 (10.0)
88 (11.4)
29 (3.7)
32 (4.2)
543 (70.6)
9 (1.2)
5 (6.5)
1 (1.0)
0
2 (6.3)
1 (0.2)
Beckman Coulter, Mississauga, Ontario, Canada), troponin I
(Access and Biosite Triage, San Diego, CA), troponin T
(Roche Elecsys, Hoffman LaRoche, LaValle, Quebec, Canada),
myoglobin (Biosite Triage), and CK-MB (Biosite Triage).
Serum markers ordered by research protocol but not for
clinical purposes were unavailable to clinicians. They acquired
chest pain characteristics, patient risk factors, and medical
history independently of clinicians and recorded physical
findings determined by emergency physicians, provisional
emergency diagnosis, admission and discharge times,
consultations, cardiac medications administered, other cardiac
investigations, hospital discharge diagnoses, and predefined
adverse events.
Adverse events included death, tachycardia requiring
intervention, bradycardia requiring intervention, assisted
ventilations, proven pulmonary thromboembolism, proven
aortic aneurysm or dissection, new congestive heart failure
requiring intravenous therapy, hypotension requiring
4 Annals of Emergency Medicine
Table 2. Univariate predictive statistics for predictors used in
the rule.
Predictor
N (all patients)
First ECG result
not normaly
Age R40 y
Previous MI
Previous angina
Previous nitrate use
Previous angina, MI,
or nitrate use
Pain radiates to neck
Pain radiates to jaw
Pain radiates
to left arm
Pain does not
increases with
deep breath
Chest tenderness
does not
reproduce pain
Pain radiates to
neck, jaw, or left
arm AND pain
does not increase
with deep breath
or palpation
0-h CK-MB R3.0 mg/L
CK-MB increase
during 2 h
Tn increase during 2 h
CK-MB or Tn increase
during 2 h
Possible ischemic
changes in ECGs
during 2 h in all
patients
ST-T changes in 1- or
2-h ECG in
patients with
initial normal
ECG (n=539)
Acute
No Acute
Coronary Coronary
Syndrome Syndrome
Odds
Ratio*
95% CI
for Odds
Ratio
604
22.6%
4.65
3.2–6.7
99.4%
35.8%
53.9%
48.5%
66.7%
87.9%
21.0%
31.0%
23.8%
40.6%
22.55
2.09
2.61
3.01
2.93
35.2%
24.9 %
57.0%
24.3%
12.3 %
43.4%
1.69
2.37
1.73
1.1–2.5
1.5–3.7
1.2–2.5
78.2%
64.2%
1.99
1.3–3.1
94.3%
80.5%
4.04
2.0–9.3
48.4%
25.6%
2.73
1.9–4.0
56.3%
46.7%
31.2%
40.4%
2.84
1.29
2.0–4.1
0.89–1.9
35.5%
60.5%
15.1%
48.2%
3.10
1.65
2.0–4.7
1.1–2.4
68.9%
33.5%
1.22
0.44–2.9
42.9%
20.7%
2.87
1.7–4.8
165
57.6 %
3.8–908.0
1.4–3.1
1.8–3.8
2.1–4.4
2.0–4.3
*Predictors with an odds ratio O1 indicate that the predictor is associated with an
increased risk of acute coronary syndrome.
y
‘‘ECG normal’’ is defined as no ST-segment deviation and no T-wave inversion
regardless of conditions such as left ventricular hypertrophy or left bundle-branch
block. T-wave flattening was not defined as abnormal.
intervention, chest compressions, percutaneous intervention,
and coronary artery bypass grafting.
Interobserver reliability was estimated by having a second
emergency physician blindly evaluate 95 patients.
At 30 days, research staff telephoned patients to review
current medications, physician visits, and hospital visits. Reports
of all diagnostic tests performed during the 30 days in any
setting were obtained, and relevant physicians were contacted
for confirmation of any new diagnoses. In cases in which patient
contact was lost, other city hospital records were reviewed to
Volume 47, no. 1 : January 2006
Christenson et al
Clinical Prediction Rule for Early Discharge of Chest Pain Patients
Figure 2. Decision tree.
identify hospital visits, procedures, and new diagnoses, and the
study list was cross-referenced against the British Columbia
Death Registry.
Outcome Measures
Investigators used predefined explicit criteria to assign a
30-day outcome diagnosis of acute myocardial infarction, definite
unstable angina, possible unstable angina, or no acute coronary
syndrome. The primary outcome was definite acute coronary
syndrome (acute myocardial infarction or definite unstable
angina). The diagnosis of acute myocardial infarction required
any one of the following: (1) a CK-MB increase definite for
acute myocardial infarction based on the specific hospital
criteria or troponin I level greater than or equal to1.0 mg/L;
(2) a troponin I increase (O0.1 mg/L but \1.0 mg/L) with
dynamic ECG changes consistent with ischemia, a coronary
angiogram with greater than 70% lesion, a positive stress test
result (radionuclide scan, echocardiogram, or ECG), or urgent
revascularization; (3) ECG evolution consistent with acute
myocardial infarction; (4) fibrinolytic therapy or primary
angioplasty and a clinical diagnosis of acute myocardial
infarction; or (5) death with no other definite cause found in
Volume 47, no. 1 : January 2006
pre- or postmorbid investigation. The diagnosis of definite
unstable angina required rest pain greater than or equal to 20
minutes and at least 1 of the following related to the presenting
symptoms: (1) troponin increase 0.1 to 0.99 mg/L but no other
acute myocardial infarction criteria, (2) dynamic ECG changes
consistent with ischemia in 2 contiguous leads (dynamic ST
depression O0.5 mm or dynamic deep T-wave inversion) but no
persistent ST elevation, (3) a coronary angiogram with greater
than 70% lesion plus hospital admission for acute coronary
syndrome, or (4) a positive stress test result (radionuclide scan,
echo, or ECG stress test) plus admission for acute coronary
syndrome.
If a diagnosis could not be assigned or if an investigator
assigned a diagnosis of ‘‘no acute coronary syndrome’’ despite
objective criteria (eg, positive troponin assay result), or if an
investigator assigned a diagnosis of ‘‘definite unstable angina’’
based only on a positive troponin assay result, 2 cardiologist
coinvestigators independently reviewed all clinical data and
assigned an adjudicated outcome diagnosis. If necessary, final
diagnosis was based on agreement of any 2 adjudicators and the
principal investigator. If all 3 disagreed, the final outcome was
the most significant diagnosis.
Annals of Emergency Medicine 5
Clinical Prediction Rule for Early Discharge of Chest Pain Patients
Christenson et al
Figure 3. Chest pain early discharge rule.
Investigators reviewed all ECGs in a structured format to
identify T-wave flattening, T-inversion in 2 contiguous leads,
ST depression (R0.05 mV) in 2 contiguous leads, ST-elevation
(R0.1 mV in limb leads or R0.2 mV in precordial leads), and
the presence of left bundle-branch block, paced rhythm, or left
ventricular hypertrophy. In addition, investigators blinded to
final outcome reviewed all ECGs for each patient and recorded
their impression of the likelihood of ischemia. Evaluation of all
available ECGs was used for outcome determination.
Evaluation of ECGs up to 2 hours was used as predictors
in the rule.
Primary Data Analysis
Data were collected on specific study data forms and entered
into an Oracle database (Oracle Corporation, Redwood City,
CA, USA). Proportions, medians, means, and SDs were
calculated and multiple clinical prediction rules were derived
using the S-Plus (Insightful Corp., Seattle, WA) ‘‘rpart’’
recursive partitioning module. The algorithm was run using all
of the default options except that a 10:1 loss was specified for
missed acute coronary syndrome patients versus missed nonacute coronary syndrome patients. This higher penalty on
missed acute coronary syndrome patients encouraged the
algorithm to choose splits that maximized sensitivity and so
reduce the complexity of the model. Cut points for continuous
6 Annals of Emergency Medicine
predictors were chosen initially by the algorithm and then
rounded to convenient values by the investigators, and the
patients were reclassified based on rounded values. Predictor
variables were allowed in the model if they were univariately
associated with the primary outcome (P\.2). In addition,
variables subjected to interpretation (medical history and
physical examination findings) were included if they were
reliable (lower bound of confidence [95%] was at least 0.4 for
the reliability measure; prevalence and bias adjusted k
coefficient for dichotomous variables and intraclass correlation
coefficient for continuous variables).
Investigators reviewed each derived rule for clinical relevance.
Nodes with small numbers of patients were reaggregated and
recursive partitioning was reinstituted to simplify the branch
points. This modification, although not previously described,
makes clinical sense and reduces the chance of selecting a
nonoptimal splitting variable when partitioning groups with
smaller sample sizes. The simplest, most specific prediction
rule that met the goal of sensitivity greater than 98% was
chosen.
We do not report confidence intervals (CI) around the
sensitivity and specificity values, because these values arise as a
consequence of the rule development process, not as an
evaluation of a fixed rule. Specifically, the sampling variability
in the estimated sensitivity cannot be assessed because different
Volume 47, no. 1 : January 2006
Christenson et al
Clinical Prediction Rule for Early Discharge of Chest Pain Patients
samples lead to different rules (all of which are forced to satisfy
the ‘‘at least 98% sensitive’’ criterion). Meaningful CIs will be
obtained in the validation phase.
The sample size of 800 patients was chosen so that equal
numbers would be used in rule development and in subsequent
rule validation. In the validation phase, the aim is to confirm
that the sensitivity of the rule exceeds current practice (ie, to
obtain a lower bound on the 95% CI for sensitivity of acute
coronary syndrome detection that is no less than 0.95).
Assuming that the true sensitivity is of the rule is 0.98, roughly
187 positive cases (800 patients in total) are needed to achieve
this lower bound with 90% probability.
RESULTS
We enrolled 819 patients and subsequently excluded
50 patients from analysis, 31 who met exclusion criteria and 19
who were lost to follow-up. Of the 19 lost to follow-up, 14 were
known to be alive at subsequent hospital visits, and the other 5
did not appear in the provincial death registry. Table 1
summarizes baseline characteristics and outcomes for the 769
patients analyzed showing 30-day diagnosis was acute
myocardial infarction in 77 (10.0%) and definite unstable
angina in 88 (11.4%). Documented adverse events related or
unrelated to acute coronary syndrome included tachycardia
(35), bradycardia (14), assisted ventilations (4), proven
pulmonary thromboembolism (5), proven thoracic aortic
aneurysm or dissection (0), new congestive heart failure
requiring intravenous therapy (12), hypotension requiring
intervention (9), and chest compressions (2). Eighty-three
patients had percutaneous intervention and 31 had coronary
artery bypass grafting within 30 days.
Of the 123 possible predictor variables screened, 48 were
retained for entry into the recursive partitioning algorithm.
Table 2 lists the variables selected for the rule by diagnostic
groups with respective univariate statistics and P values. These
variables include a normal ECG result (which could include
T-wave flattening but no ST deviation or T-wave inversion), age,
previous ischemic pain, 3 pain descriptors, 1 physical finding,
and both CK-MB and troponin results. Reliability measures
(95% CI) for clinical subjective variables were pain radiates to
neck, 0.68 (95% CI 0.54 to 0.83); pain radiates to jaw, 0.81
(95% CI 0.69 to 0.93); pain radiates to left arm, 0.79 (95% CI
0.67 to 0.91); and chest tenderness reproduces pain, 0.63 (95%
CI 0.45 to 0.80).
Figure 2 outlines the classification tree, and Figure 3 is a
simplified clinical algorithm of the chest pain prediction rule.
In this development cohort, the rule is 98.8% sensitive,
identifying all but 2 patients with unstable angina, and would
have allowed the appropriate discharge of 32.5% of patients
without acute coronary syndrome using data collected within 2
hours of arrival (Table 3). Of those identified for discharge,
89.4% would be determined with information available at
presentation, most without any laboratory testing, and the
remainder would be determined safe for discharge based on
repeat ECG, CK-MB, and troponin at 2 hours. This potentially
Volume 47, no. 1 : January 2006
Table 3. Diagnostic accuracy of the rule.*
Unsuitable for early
discharge
Meet rule criteria for
early discharge
Definite Acute
Coronary
Syndrome
No Definite
Acute Coronary
Syndrome
163
408
571
2
196
198
165
604
769
*Sensitivity 98.8%; specificity 32.5%; positive predictive value 28.5%; negative
predictive value 99.0%.
allows the early discharge process to be divided into
2 sequential stages.
The 2 missed patients were subsequently diagnosed with
unstable angina. A 66-year-old woman was diagnosed 8 days
later by outpatient nuclide stress testing, with no subsequent
admission or adverse event. The other was a 52-year-old patient
with a clear history of new and accelerating angina before rest
pain who was treated medically for a single critical lesion not
amenable to intervention. During validation, we will clarify the
definition of past angina to include a recent history of typical
angina of effort whether diagnosed or not.
LIMITATIONS
Although we used accepted and validated methodology,15
the Vancouver Chest Pain Prediction Rule was developed in
1 cohort of patients in a single center. We are currently
beginning validation of the rule, and it will subsequently
require evaluation of implementation in multiple centers. Our
criterion standard for definite acute coronary syndrome was
not based on mandated investigations but on a careful
evaluation of all tests and visits during 30 days. It is possible that
we missed some patients with silent events. The cohort was
collected during day and evening hours and will require
validation in patients presenting at all hours of the day. The rule
is modestly complex and may be difficult to remember, which
could limit its use by clinicians. However, the rule is largely
intuitive, and simple memory aids (eg, pocket cards or PDA
algorithms) will enhance implementation once the rule is
validated. It is important to note that although our focus was on
identifying patients with very low risk of acute coronary
syndrome, no patients were missed with pulmonary embolism
or aortic dissection. Suspicion of other life-threatening
conditions requires appropriate, independent investigations.
In retrospect, this rule development should have included
patients 24 hours per day. Ideally, the study would be
strengthened by mandating initial and 30-day imaging
studies to improve the criterion standard of acute coronary
syndrome. The cost implications of the Vancouver Chest
Pain Rule are not yet known.
DISCUSSION
Patients who present to the ED with chest pain fall into
3 categories: those with objective ischemia who need admission
Annals of Emergency Medicine 7
Clinical Prediction Rule for Early Discharge of Chest Pain Patients
and treatment, those with a clear noncardiac cause, and those
who require a diagnostic process to rule out acute coronary
syndrome and other life-threatening conditions. The Vancouver
Chest Pain Rule identifies 32.5% of patients who do not have
acute coronary syndrome using information available in the
first 2 hours of the ED visit. These very-low-risk patients are
identified with minimal error, in fact including only 1.2% of
those who do have unstable angina or acute myocardial
infarction. This information can be integrated into the decision
framework of the emergency physician who continually
evaluates patients for safe and appropriate disposition.
This prediction rule is different from previous risk
stratification tools. In 1988, Goldman et al6 developed a
clinical/ECG algorithm that identified patients with less than
7% risk of acute myocardial infarction. This was not designed
to detect patients with unstable angina and it did not improve
disposition decisions relative to experienced clinician judgment
alone. Limkakeng et al7 later reported that 4.9% of patients
with low-risk Goldman criteria and a negative initial troponin
I level experienced death, acute myocardial infarction, and
revascularization within 30 days. Lee et al19 and Goldman et al20
subsequently developed a strategy to identify patients
unlikely to have acute myocardial infarction and criteria to
identify patients who need coronary care unit admission. These
criteria required 12 to 24 hours of observation and did not
facilitate early ED disposition decisions.21 Pozen et al8
developed and tested a predictive equation using 7 clinical
and ECG variables. After application of this instrument,
inappropriate coronary care unit admissions fell from 24% to
17%, but the 3% inappropriate discharge rate was unchanged.
Selker et al10-12 modified the Pozen et al8 equation and linked
it to ECG findings, creating an electronic tool called the
ACI-TIPI. This tool increased the rate of appropriate
discharges by unsupervised residents but did not change
the performance of attending physicians. Of note,
ACI-TIPI defines low risk as less than 10% chance of acute
myocardial infarction or unstable angina, and few clinicians will
discharge patients with this level of risk.22
In 1992, Baxt23 and Baxt and Skora24 ‘‘trained’’ neural
networks to recognize acute myocardial infarction and, in 1993,
Jonsbu et al25 developed a sensitive computer-derived
algorithm. Doyle et al26 studied 9 algorithms but found none
that improved clinical decisionmaking. The recently described
Thrombolysis in Myocardial Infarction (TIMI) Risk score27 is
an excellent stratification tool but was derived from studies of
patients with objective evidence of acute coronary syndrome
and has not been validated in undifferentiated ED patients with
chest discomfort. The Erlanger protocol28 is an intense 2-hour
assessment that includes serial ECGs, 2 CK-MB and troponin
measurements, clinical judgment, and nuclear stress testing.
It is the most detailed published protocol designed to facilitate
early discharge but requires nuclear imaging to increase
sensitivity from 80% to 99%. None of the above stratification
tools help clinicians identify patients who are safe to discharge
after a brief ED evaluation.
8 Annals of Emergency Medicine
Christenson et al
Recent American College of Cardiology/American Heart
Association14 guidelines recommend that all patients with a
compatible history have serial marker testing until 9 hours after
symptom onset, followed by provocative testing. Applying this
guideline in the ED, where most patients have ill-defined
chest discomfort, could greatly increase diagnostic and ED bed
utilization without a corresponding improvement in patient
outcomes. The ED/AHCPR guidelines and the Toronto Heart
Centre ACS Guidelines29 recommend excluding patients with
discomfort because of other causes from the diagnostic pathway,
but neither suggests a process for identifying these patients.
The Vancouver Chest Pain Rule is unique in its ability to
identify very-low-risk patients who are suitable for early
discharge. It is also unique in that it identifies patients in a
clinically logical stepwise process with a minimum of
investigations. First, age, initial history, and ECG findings
identified 56 of 198 patients who ultimately meet rule criteria.
A further 121 patients were identified based on initial CK-MB
and low-risk pain characteristics. The last 21 patients were
identified by the absence of changes in the ECG or serum
markers during 2 hours. If disposition according to the rule was
implemented in this stepwise manner, ED lengths of stay and
investigational costs could be reduced substantially in many
settings.
Some would argue that the age criteria might result in
missing younger patients with myocardial infarction, but age is
in fact a powerful predictor, and the low-risk group younger
than 40 years is also defined by a near-normal ECG and the
absence of past myocardial infarction or angina. Marsan et al30
recently studied 1032 patients younger than 40 years and
presenting with chest pain and found that those with no
cardiac history and either no cardiac risk factors or a normal
ECG result had less than 0.5% 30-day risk of acute coronary
syndrome.
We found that CK-MB at a low threshold was consistently a
better discriminator than any level of troponin, surprising
because many studies suggest troponin is the superior marker.
Our findings are not unique, and at least 1 other investigator
has reported that CK-MB was more predictive of acute coronary
syndrome than troponin early in the patient’s presentation.31
Laboratory technicians used the Triage (Biosite) point-of-care
device, and we attribute the greater discriminatory power of
CK-MB over troponin to its better analytic performance within
the normal range. The reported 99th percentile for troponin
on the Triage point-of-care device is actually the detection
limit (0.19 mg/L), whereas the 95th percentile for CK-MB
(usual cutoff) is well above the detection limit (0.75 mg/L).
de Winter et al32 determined that an increase in CK-MBmass
greater than 2 mg/L represents the critical difference between
2 results for the detection of myocardial damage. We
hypothesize that it is the absence of a critical increase over
baseline value (median 1.5 mg/L) that underlies the
discriminatory power of a CK-MB level less than 3 mg/L. In
comparison, present troponin methods do not permit the
detection of a change within the normal range.
Volume 47, no. 1 : January 2006
Christenson et al
Clinical Prediction Rule for Early Discharge of Chest Pain Patients
An early discharge strategy such as the one developed here
defines patients safe for early discharge and reduces the number
who require more definitive testing but also defines the
companion group of patients who are not clearly ‘‘safe’’ and
require further investigations. To maximize diagnostic
efficiency, a more prolonged ‘‘chest pain unit’’ rule-out strategy
should be applied to patients without objective ischemia who do
not meet early discharge criteria.
In summary, the Vancouver Chest Pain Rule defines a group
of chest pain patients who can be safely discharged from the
ED in the first few hours after arrival. If validated in an
independent cohort, it will improve the process of diagnostic
decisionmaking in patients presenting with chest pain.
The authors are grateful for the support and data supplied by the
physicians and nurses at St. Paul’s Hospital, Vancouver, Canada,
and for the dedication of Patti Lawson, BA, and Courtney Barker.
Supervising editor: Judd E. Hollander, MD
Author contributions: DM, HR, EY, EG, FR, AA, and JS helped
design the study, analyzed the data, and reviewed the
manuscript. BB helped design the study, collected data, and
reviewed the manuscript. JMC, GI, and CRT conceived and
helped design the study, collected data, and wrote the
manuscript. JMC takes responsibility for the paper as a whole.
Funding and support: Canadian Institutes of Health research
grant MOP 53102 and Heart and Stroke Foundation of British
Columbia and Yukon Grant. Materials for analysis of some
laboratory testing were donated by Biosite.
Publication dates: Received for publication March 17, 2005.
Revision received June 13, 2005. Accepted for publication July
13, 2005. Available online October 19, 2005.
Reprints not available from the authors.
Address for correspondence: Jim Christenson, MD, Department
of Emergency Medicine, St. Paul’s Hospital, 1081 Burrard
Street, Vancouver, British Columbia V6Z 1Y6, Canada; 604806-8985, fax 604-806-8798; E-mail jimchris@interchange.
ubc.ca.
REFERENCES
1. Pope J, Aufderheide T, Ruthazer R, et al. Missed diagnoses of acute
cardiac ischemia in the emergency department. N Engl J Med.
2000;342:1163-1170.
2. Christenson J, Innes G, McKnight D, et al. Safety and efficiency of
emergency department assessment of chest discomfort. CMAJ.
2004;170:1803-1807.
3. Anderson HJ. Hospitals battle heart attacks with new chest pain
centers. Hospitals. 1991;65:36-37.
4. Zalenski RJ, Rydman RJ, Ting S, et al. A national survey of
emergency department chest pain centers in the United States.
Am J Cardiol. 1998;81:1305-1309.
5. Graff L, Joseph T, Andelman R, et al. American College of
Emergency Physicians information paper: chest pain units in
emergency departments: a report from the Short-Term
Observation Services Section. Am J Cardiol. 1995;76:
1036-1039.
Volume 47, no. 1 : January 2006
6. Goldman L, Cook E, Brand D. A computer protocol to predict
myocardial infarction in emergency department patients with chest
pain. N Engl J Med. 1988;318:797-803.
7. Limkakeng A Jr, Gibler WB, Pollack C, et al. Combination of
Goldman risk and initial cardiac troponin I for emergency
department chest pain patient risk stratification. Acad Emerg Med.
2001;8:696-702.
8. Pozen MW, D’Agostino RB, Selker HP, et al. A predictive instrument to
improve coronary-care-unit admission practices in acute
ischemic heart disease: a prospective multicenter clinical trial. N Engl J
Med. 1984;310:1273-1278.
9. Pozen MW, D’Agostino RB, Mitchell JB, et al. The usefulness of a
predictive instrument to reduce inappropriate admissions to the
coronary care unit. Ann Intern Med. 1980;92:238-242.
10. Selker HP, Beshansky JR, Griffith JL, et al. Use of the acute cardiac
ischemia time-insensitive predictive instrument (ACI-TIPI) to assist
with triage of patients with chest pain or other symptoms
suggestive of acute cardiac ischemia: a multicenter, controlled
clinical trial. Ann Intern Med. 1998;129:845-855.
11. Selker H, D’Agostino R, Laks M. A predictive instrument for acute
ischemic heart disease to improve coronary care unit admission
practices: a potential on-line tool in a computerized
electrocardiograph. J Electrocardiol. 1988;(21 suppl):S11-S17.
12. Selker H, Griffith J, Patil S, et al. A comparison of performance of
mathematical predictive methods for medical diagnosis: identifying
acute cardiac ischemia among emergency department patients.
J Investig Med. 1995;43:468-476.
13. Fesmire FM, Hughes AD, Fody EP, et al. The Erlanger chest pain
evaluation protocol: a one-year experience with serial 12-lead ECG
monitoring, two-hour delta serum marker measurements, and
selective nuclear stress testing to identify and exclude acute
coronary syndromes. Ann Emerg Med. 2002;40:584-594.
14. Braunwald E, Antman E, Beasley J, et al. ACC/AHA: a guideline
update for the management of patients with unstable angina and
non-ST-elevation myocardial infarction: a report of the American
College of Cardiology/American Heart Association task force on
practice guidelines (committee on the management of patients
with unstable angina). Available at: http://www.acc.org/
clinicalguidelines/unstable/unstable.pdf. Accessed July 10, 2005.
15. Laupacis A, Sekar N, Stiell IG. Clinical prediction rules: a review and
suggested modifications of methodological standards. JAMA.
1997;277:488-494.
16. Wasson J, Sox H, Neff R, et al. Clinical prediction rules: application
and methodologic standards. N Engl J Med. 1985;313:793-799.
17. Stiell IG, Wells GA. Methodologic standards for the development of
clinical decision rules in emergency medicine. Ann Emerg Med.
1999;33:437-447.
18. Hollander JE, Blomkalns AL, Brogan GX, et al. Standardized
reporting guidelines for studies evaluating risk stratification of ED
patients with potential acute coronary syndromes. Acad Emerg
Med. 2004;11:1331-1340.
19. Lee TH, Juarez G, Cook EF, et al. Ruling out acute myocardial
infarction: a prospective multicenter validation of a 12-hour
strategy for patients at low risk. N Engl J Med. 1991;324:
1239-1246.
20. Goldman L, Cook EF, Johnson PA, et al. Prediction of the need for
intensive care in patients who come to the emergency departments
with acute chest pain. N Engl J Med. 1996;334:1498-1504.
21. Reilly B, Durairaj L, Husain S, et al. Performance and potential
impact of a chest pain decision prediction rule in a large public
hospital. Am J Med. 1999;106:285-291.
22. Davis MA, Keerbs A, Hoffman JR, et al. Admission decisions
in emergency department chest pain patients at low risk for
myocardial infarction: patient versus physician preferences.
Ann Emerg Med. 1996;28:606-611.
Annals of Emergency Medicine 9
Clinical Prediction Rule for Early Discharge of Chest Pain Patients
23. Baxt WG. Analysis of the clinical variables driving decision in
an artificial neural network trained to identify the presence of
myocardial infarction. Ann Emerg Med. 1992;21:1439-1444.
24. Baxt WG, Skora J. Prospective validation of artificial neural network
trained to identify acute myocardial infarction. Lancet. 1996;347:
12-15.
25. Jonsbu J, Aase O, Rollag A, et al. Prospective evaluation of
an EDB-based diagnostic program to be used in patients
admitted to hospital with acute chest pain. Eur Heart J. 1993;
14:441-446.
26. Doyle Y, Conroy R, Mulcahy R. Can we improve on the diagnostic
accuracy of chest pain in accident and emergency departments?
Irish Med J. 1988;81:21-24.
27. Holper E, Antman E, McCabe C, et al. A simple, readily available
method for risk stratification of patients with unstable angina and
non-ST elevation myocardial infarction. Am J Cardiol. 2001;87:
1008-1010.
10 Annals of Emergency Medicine
Christenson et al
28. Fesmire F, Hughes A, Fody E, et al. The Erlanger chest pain
evaluation protocol: a one-year experience with serial 12-lead ECG
monitoring, two-hour delta serum marker measurements, and
selective nuclear stress testing to identify and exclude acute
coronary syndromes. Ann Emerg Med. 2002;40:584-594.
29. Fitchett D, Goodman S, Langer A. New advances in the management
of acute coronary syndromes. CMAJ. 2001;164:1309-1316.
30. Marsan R, Shaver K, Sease K, et al. Validation of a clinical decision
rule for young adult chest pain patients. Acad Emerg Med. 2004;
11:507.
31. Svensson L, Isaksson L, Axelsson C, et al. Predictors of myocardial
damage prior to hospital admission among patients with acute
chest pain or other symptoms raising a suspicion of acute coronary
syndrome. Coron Artery Dis. 2003;14:225-231.
32. de Winter RJ, Koster RW, van Straalen JP, et al. Critical difference
between serial measurements of CK-MB mass to detect myocardial
damage. Clin Chem. 1997;43:338-343.
Volume 47, no. 1 : January 2006