Artificial Intelligence for Diagnosis of Acute Coronary Syndromes: A Meta-analysis of Machine Learning Approaches.
Journal
The Canadian journal of cardiology
ISSN: 1916-7075
Titre abrégé: Can J Cardiol
Pays: England
ID NLM: 8510280
Informations de publication
Date de publication:
04 2020
04 2020
Historique:
received:
26
03
2019
revised:
01
09
2019
accepted:
02
09
2019
entrez:
30
3
2020
pubmed:
30
3
2020
medline:
23
10
2020
Statut:
ppublish
Résumé
Machine learning (ML) encompasses a wide variety of methods by which artificial intelligence learns to perform tasks when exposed to data. Although detection of myocardial infarction has been facilitated with introduction of troponins, the diagnosis of acute coronary syndromes (ACS) without myocardial damage (without elevation of serum troponin) remains subjective, and its accuracy remains highly dependent on clinical skills of the health care professionals. Application of a ML algorithm may expedite management of ACS for either early discharge or early initiation of ACS management. We aim to summarize the published studies of ML for diagnosis of ACS. We searched electronic databases, including PubMed, Embase, and Web of Science from inception up to January 13, 2019, for studies that evaluated ML algorithms for the diagnosis of ACS in patients presenting with chest pain. We then used random-effects bivariate meta-analysis models to summarize the studies. We retained 9 studies that evaluated ML in a total of 6292 patients. The prevalence of ACS in the evaluated cohorts ranged from relatively rare (7%) to common (57%). The pooled sensitivity and specificity were 0.95 and 0.90, respectively. The positive predictive values ranged from 0.64 to 1.0, and the negative predictive values ranged from 0.91 to 1.0. The positive and negative likelihood ratios ranged from 1.6 to 33.0 and 0.01 to 0.13, respectively. The excellent sensitivity, negative likelihood ratio, and negative predictive values suggest that ML may be useful as an initial triage tool for ruling out ACS.
Sections du résumé
BACKGROUND
Machine learning (ML) encompasses a wide variety of methods by which artificial intelligence learns to perform tasks when exposed to data. Although detection of myocardial infarction has been facilitated with introduction of troponins, the diagnosis of acute coronary syndromes (ACS) without myocardial damage (without elevation of serum troponin) remains subjective, and its accuracy remains highly dependent on clinical skills of the health care professionals. Application of a ML algorithm may expedite management of ACS for either early discharge or early initiation of ACS management. We aim to summarize the published studies of ML for diagnosis of ACS.
METHODS
We searched electronic databases, including PubMed, Embase, and Web of Science from inception up to January 13, 2019, for studies that evaluated ML algorithms for the diagnosis of ACS in patients presenting with chest pain. We then used random-effects bivariate meta-analysis models to summarize the studies.
RESULTS
We retained 9 studies that evaluated ML in a total of 6292 patients. The prevalence of ACS in the evaluated cohorts ranged from relatively rare (7%) to common (57%). The pooled sensitivity and specificity were 0.95 and 0.90, respectively. The positive predictive values ranged from 0.64 to 1.0, and the negative predictive values ranged from 0.91 to 1.0. The positive and negative likelihood ratios ranged from 1.6 to 33.0 and 0.01 to 0.13, respectively.
CONCLUSIONS
The excellent sensitivity, negative likelihood ratio, and negative predictive values suggest that ML may be useful as an initial triage tool for ruling out ACS.
Identifiants
pubmed: 32220387
pii: S0828-282X(19)31269-3
doi: 10.1016/j.cjca.2019.09.013
pii:
doi:
Types de publication
Journal Article
Meta-Analysis
Systematic Review
Langues
eng
Sous-ensembles de citation
IM
Pagination
577-583Commentaires et corrections
Type : CommentIn
Informations de copyright
Copyright © 2019 Canadian Cardiovascular Society. Published by Elsevier Inc. All rights reserved.