Professional Documents
Culture Documents
research-article2017
ACC0010.1177/2048872617710788European Heart Journal: Acute Cardiovascular CareVan Den Berg and Body
EUROPEAN
SOCIETY OF
Review CARDIOLOGY ®
The HEART score for early rule 2018, Vol. 7(2) 111–119
© The European Society of Cardiology 2017
Reprints and permissions:
out of acute coronary syndromes sagepub.co.uk/journalsPermissions.nav
https://doi.org/10.1177/2048872617710788
DOI: 10.1177/2048872617710788
in the emergency department: a journals.sagepub.com/home/acc
Abstract
Aims: The objective of this systematic review was to summarise the current evidence on the diagnostic accuracy of the
HEART score for predicting major adverse cardiac events in patients presenting with undifferentiated chest pain to the
emergency department.
Methods and results: Two investigators independently searched Medline, Embase and Cochrane databases between
2008 and May 2016 identifying eligible studies providing diagnostic accuracy data on the HEART score for predicting
major adverse cardiac events as the primary outcome. For the 12 studies meeting inclusion criteria, study characteristics
and diagnostic accuracy measures were systematically extracted and study quality assessed using the QUADAS-2 tool.
After quality assessment, nine studies including data from 11,217 patients were combined in the meta-analysis applying a
generalised linear mixed model approach with random effects assumption (Stata 13.1). In total, 15.4% of patients (range
7.3–29.1%) developed major adverse cardiac events after a mean of 6 weeks’ follow-up. Among patients categorised as
‘low risk’ and suitable for early discharge (HEART score 0–3), the pooled incidence of ‘missed’ major adverse cardiac
events was 1.6%. The pooled sensitivity and specificity of the HEART score for predicting major adverse cardiac events
were 96.7% (95% confidence interval (CI) 94.0–98.2%) and 47.0% (95% CI 41.0–53.5%), respectively.
Conclusions: Patients with a HEART score of 0–3 are at low risk of incident major adverse cardiac events. As 3.3%
of patients with major adverse cardiac events are ‘missed’ by the HEART score, clinicians must ask whether this risk is
acceptably low for clinical implementation.
Keywords
HEART score, acute coronary syndromes, diagnostic accuracy, sensitivity and specificity, emergency medicine, clinical
decision rules
Background
Chest pain is one of the most common reasons for emer- hospitals.3 The HEART score is a simple tool that can be
gency hospital admission.1 Hospital and emergency depart- used at the time of a patient’s first presentation to the ED. It
ment (ED) crowding is a growing problem and is associated
with higher patient mortality.2 This highlights the pressing 1Maastricht University, Maastricht, The Netherlands
need for diagnostic strategies that can rapidly ‘rule out’ 2The University of Manchester, Manchester, UK
acute coronary syndromes (ACSs) and avoid unnecessary 3Central Manchester University Hospitals Foundation NHS Trust,
Table 1. The HEART score.4 not allowing for sufficient assessment of the methodologi-
cal approach. Likewise, publications in languages other
HEART score
than English, Dutch and German were excluded.
History Highly suspicious 2
Moderately suspicious 1
Slightly suspicious 0 Index test and outcome measures
ECG Significant ST-depression 2 The HEART score is defined as a composite score ranging
Non-specific repolarisation disturbance 1 from 0 to 10 points. Each of the five acronym domains (his-
Normal 0 tory, electrocardiogram (ECG), age, risk factors and tro-
Age ⩾65 years 2 ponin) is scored with 0 to 2 points according to the original
>45–<65 years 1 definition (Table 1).
⩽45 years 0
The primary outcome of MACEs was defined as a com-
Risk factors ⩾3 risk factors,a or history of 2
MIDAS and METANDI commands utilising a generalised those two studies was based on a significant deviation from
linear mixed model approach with random effects assump- the original HEART score definition in combination with a
tion to generate pooled estimates of diagnostic accuracy. vaguely defined reference standard allocated using rela-
Between-study heterogeneity was assessed with the tively old troponin assays as a result of early study periods.
Cochrane Q chi-square test and the I2 statistics. With diag- Both authors, based on the QUADAS-2 assessment, agreed
nostic test accuracy studies heterogeneity is to be expected, that both studies were at significant risk of allocating inter-
therefore requiring a random effects model approach by mediate risk patients wrongly to the low-risk group. A sum-
default.11 We generated a summary receiver operating char- mary of the quality assessment of all 12 initially eligible
acteristic (ROC) curve, which allowed calculation of the area studies with the QUADAS-2 tool is shown in Figure 1. The
under the curve (AUC) as a global measure of diagnostic test methodological quality assessment of included studies
performance as well as a visual evaluation of heterogeneity identified that most had important limitations. A detailed
and any potential threshold effect.12 To determine the poten- assessment of each individual study is available in the
tial effects of publication bias a Deeks funnel plot was Supplementary Appendix. The major methodological limi-
obtained, with P<0.10 for the slope coefficient considered to tations identified in various degrees included deviating
indicate significant asymmetry and therefore a high likeli- from the original HEART score definition, frequent
hood of publication bias.13 unblinded determination of the HEART score or patient
outcome, the use of outdated definitions and insufficient or
lacking information on the timing of reference standard tro-
Results
ponin testing, contributing to potential verification bias.
Our literature search identified 218 articles of which 18 The study by Carlton et al. was excluded from the meta-
were considered potentially eligible for inclusion based on analysis as this study did not provide data about the inci-
title and abstract screening. After independent review 12 dence of MACEs.20 This yielded a total of nine studies for
studies met the inclusion criteria for this review providing meta-analysis (Figure 2).
the necessary data allowing for computation of diagnostic These nine studies included data from 11,217 patients.
accuracy measures.8,9,14–23 General study characteristics are The pooled prevalence of MACEs was 15.4% (95% confi-
summarised in Table 2. A more detailed summary of inclu- dence interval (CI) 14.8–16.1%, range 7.3–29.1%) at a
sion and exclusion criteria, precise definitions of outcomes, mean follow-up time of 6 weeks. Across the various studies
the exact definitions utilised for the calculation of the 4101 (36.6%, range 28.2–60.1%) had a HEART score of
HEART score, as well as the cardiac troponin assays used 0–3 and would therefore have been potentially suitable for
and the timing of troponin testing, is provided in the immediate discharge. In this low-risk group a combined
Supplementary Appendix. Test characteristics for each total of 1.6% (95% CI 1.2–2.0%, range 0.9–5.9%) of the
individual study including 2×2 tables, sensitivity, specific- patients would have had a missed MACE. We found no evi-
ity, positive predictive value (PPV), negative predictive dence of publication bias (P=0.58; see Supplementary
value (NPV), as well as the proportion of patients identified Appendix).
as low risk (HEART score 0–3) are presented in Table 3. The pooled sensitivity estimate of the HEART score for
After quality assessment, two further studies were predicting MACEs in the nine studies included was 96.7%
excluded from the final meta-analysis.17,22 The exclusion of (95% CI 94.0–98.2%) as summarised in Figure 3. The
114 European Heart Journal: Acute Cardiovascular Care 7(2)
Table 3. Diagnostic and predictive indices of the HEART score for predicting major adverse cardiac events or acute myocardial
infarction (95% confidence intervals in parentheses).
N: number of patients; NPV: negative predictive value; PPV: positive predictive value; MACEs: major adverse cardiac events; AMI: acute myocardial
infarction; hs-cTnT: high-sensitivity cardiac troponin T; hs-cTnI: high-sensitivity cardiac troponin I.
aIn patients with HEART score of 0–3.
pooled specificity for the HEART score was 47.0% (95% CI NPV was fairly constant between 97.4% and 100%, while
41.0–53.5%). The positive likelihood ratio (LR+) and nega- the PPV was more variable between 17.3% and 41.2%.
tive likelihood ratio (LR–) were 1.82 (95% CI 1.62–2.06) On summary ROC curve analysis, the AUC of the
and 0.07 (95% CI 0.04–0.13), respectively. Depending on HEART score was 0.81 (95% CI 0.77–0.84), as shown in
the prevalence of MACEs in the various study cohorts the Figure 4. While overall heterogeneity was relatively high,
Van Den Berg and Body 115
12% was estimated to be attributed to a threshold effect Forest plot, summary ROC curve and Deeks funnel plot for
with a low between-study variation of 14% for sensitivity the sensitivity analysis are provided in the Supplementary
and 5% for specificity. This was visually confirmed by Appendix.
inspection of the summary ROC curve.
Sensitivity analysis with the HEART score at
Sensitivity analysis a lower ⩽2 points cut-off
In a sensitivity analysis pooling all studies eligible for the Five studies provided data for a potential lower 2 points or
review, irrespective of the quality assessment, data from less cut-off for the HEART score including 6397 patie
27,724 patients were combined. The sensitivity of the nts.14–16,18,21 The pooled sensitivity of the HEART score for
HEART score for predicting MACEs varied across individ- predicting MACEs at the lower 2 points or less cut-off was
ual studies from 75.5% to 100%. In comparison to the pri- higher at 99.4% (95% CI 96.8–99.9%) at the cost of a lower
mary analysis, the pooled sensitivity was lower at 95.1% pooled specificity of 22.0% (95% CI 14.2–32.5%). The
(95% CI 90.5–97.5%) with a very high degree of heteroge- likelihood ratios were respectively a LR+ of 1.28 (95% CI
neity (I2=96.09%, 95% CI 94.74–97.45%). The pooled 1.13–1.44) and a LR– of 0.03 (95% CI 0.00–0.18).
specificity was slightly higher at 49.3% (95% CI 42.9–
55.8%). The likelihood ratios were respectively calculated
HEART score with high-sensitivity troponin
at LR+ of 1.88 (95% CI 1.68–2.10) and LR– of 0.10 (95%
assays
CI 0.05–0.18). The area under the summary ROC curve was
lower, at 0.75 (95% CI 0.71–0.78) with a more scattered Three studies evaluated the HEART score with hs-cTn
appearance representing higher between-study variability. A assays, only two of which reported data for MACEs at 30
116 European Heart Journal: Acute Cardiovascular Care 7(2)
Discussion
Our findings demonstrate that the HEART score could be
used to identify patients with a suspected diagnosis of ACS
Figure 4. Summary receiver operating characteristic curve of who are at low probability (1.6%) of developing MACEs
the HEART score for prediciting major adverse cardiac events.
and who could potentially be discharged immediately from
the ED. However, this risk may not be acceptable to clini-
days (which rendered a separate meta-analysis inappro- cians. Indeed, only 40% of emergency physicians would be
priate). The prospective study of Visser et al. used the willing to discharge a patient from the ED if the probability
Roche Elecsys hs-cTnT assay, reporting a sensitivity of of MACEs exceeds 1%.24 Furthermore, the post-test prob-
93.3% (95% CI 84.1–97.8%) for MACEs at 6 weeks.18 ability of MACEs is heavily dependent on prevalence.
Conversely, Santi et al. using the same troponin assay in Sensitivity may therefore be a superior measure of diagnos-
a recent retrospective study reported 100% sensitivity tic accuracy, particularly on a meta-analysis of cohorts with
Van Den Berg and Body 117
varying prevalence. We found that the HEART score has a validate the HEART score prospectively with hs-cTn
sensitivity of 96.7%. In our sensitivity analysis (which assays.
included two studies with lower methodological quality), Other diagnostic strategies to rule out ACS following a
the pooled sensitivity was lower still, at 95.1%. In an addi- single blood test have also been described. For example,
tional sensitivity analysis of five studies at a lower 2 points the computer-based Manchester acute coronary syndromes
or less cut-off for the HEART score, suggested by the study (MACS) rule. The MACS rule was shown effectively to
of Six et al.,15 suggested the pooled sensitivity of 99.4% rule out MACEs within 30 days with high sensitivity
was raised to what is considered an acceptable risk of miss- (97.8% and 100%) in two external observational studies,
ing a MACE in the low-risk group. However, this increase identifying around 20% of patients as ‘very low risk’ and
in sensitivity comes with a drastically decreased pooled therefore eligible for immediate discharge.26,27 Similar to
specificity of only 22.0%. the HEART score, the MACS rule could be used not only to
Our main analysis, although excluding two studies with rule out ACS following a single blood test but also to risk
with MACEs are ‘missed’ by the HEART score, clinicians 13. Deeks JJ, Macaskill P and Irwig L. The performance of tests
must ask whether this risk is acceptably low for clinical of publication bias and other sample size effects in system-
implementation, and should be carefully guided by local atic reviews of diagnostic test accuracy was assessed. J Clin
circumstances influencing diagnostic performance. Future Epidemiol 2005; 58: 882–893.
14. Fesmire FM, Martin EJ, Cao Y, et al. Improving risk stratifi-
work should focus on robust comparison with alternative
cation in patients with chest pain: the Erlanger HEARTS(3)
strategies.
score. Am J Emerg Med 2012; 30: 1829–1837.
15. Six AJ, Cullen L, Backus BE, et al. The HEART score for
Conflict of interest the assessment of patients with chest pain in the emergency
Richard Body has previously undertaken research involving dona- department: a multinational validation study. Crit Pathw
tion of reagents without charge by Roche, Abbott, Alere, Siemens Cardiol 2013; 12: 121–126.
and Randox. Richard Body has accepted the provision of econ- 16. Melki D and Jernberg T. HEART score: a simple and useful
omy class travel and accommodation to present findings unrelated tool that may lower the proportion of chest pain patients who
T high-sensitive assay for diagnosis of acute myocardial patients with acute chest pain for early discharge. Circ
infarction in emergency department: systematic review and Cardiovasc Qual Outcomes 2015; 8: 195–203.
meta-analysis. BMJ 2015; 350: h15. 31. Than MP, Pickering JW, Aldous SJ, et al. Effectiveness of
29. Body R, Burrows G, Carley S, et al. High-sensitivity car- EDACS versus ADAPT accelerated diagnostic pathways for
diac troponin t concentrations below the limit of detection to chest pain: a pragmatic randomized controlled trial embed-
exclude acute myocardial infarction: a prospective evalua- ded within practice. Ann Emerg Med 2016; 68: 93–102.e1.
tion. Clin Chem 2015; 61: 983–989. 3 2. Mokhtari A, Borna C, Gilje P, et al. A 1-h combination algo-
30. Mahler SA, Riley RF, Hiestand BC, et al. The HEART
rithm allows fast rule-out and rule-in of major adverse car-
Pathway randomized trial: identifying emergency department diac events. J Am Coll Cardiol 2016; 67: 1531–1540.