Matches in SemOpenAlex for { <https://semopenalex.org/work/W3162183883> ?p ?o ?g. }
Showing items 1 to 59 of
59
with 100 items per page.
- W3162183883 abstract "HomeCirculation: Arrhythmia and ElectrophysiologyVol. 14, No. 5Impact of ECG Characteristics on the Performance of an Artificial Intelligence Enabled ECG for Predicting Left Ventricular Dysfunction Free AccessLetterPDF/EPUBAboutView PDFView EPUBSections ToolsAdd to favoritesDownload citationsTrack citationsPermissions ShareShare onFacebookTwitterLinked InMendeleyRedditDiggEmail Jump toFree AccessLetterPDF/EPUBImpact of ECG Characteristics on the Performance of an Artificial Intelligence Enabled ECG for Predicting Left Ventricular Dysfunction Julio Perez-Downes, DO Patrick Fitzgerald, MD Demilade Adedinsewo, MD, MPH Rickey E. Carter, PhD Peter A. Noseworthy, MD Fred KusumotoMD Julio Perez-DownesJulio Perez-Downes Correspondence to: Julio Perez-Downes, DO, Department of Cardiovascular Disease, Mayo Clinic Florida, 4500 San Pablo Rd. S, Jacksonville, FL 32204. Email E-mail Address: [email protected] https://orcid.org/0000-0001-9675-2896 Department of Cardiovascular Diseases (J.P.-D., D.A., F.K.), Mayo Clinic, Jacksonville, FL. Search for more papers by this author , Patrick FitzgeraldPatrick Fitzgerald https://orcid.org/0000-0003-3191-7308 Internal Medicine (P.F.), Mayo Clinic, Jacksonville, FL. Search for more papers by this author , Demilade AdedinsewoDemilade Adedinsewo https://orcid.org/0000-0002-8629-2029 Department of Cardiovascular Diseases (J.P.-D., D.A., F.K.), Mayo Clinic, Jacksonville, FL. Search for more papers by this author , Rickey E. CarterRickey E. Carter https://orcid.org/0000-0002-0818-273X Department of Health Sciences Research (R.E.C.), Mayo Clinic, Rochester, MN. Search for more papers by this author , Peter A. NoseworthyPeter A. Noseworthy https://orcid.org/0000-0002-4308-0456 Department of Cardiovascular Diseases (P.A.N.), Mayo Clinic, Rochester, MN. Search for more papers by this author , Fred KusumotoFred Kusumoto https://orcid.org/0000-0002-8300-6277 Department of Cardiovascular Diseases (J.P.-D., D.A., F.K.), Mayo Clinic, Jacksonville, FL. Search for more papers by this author Originally published17 May 2021https://doi.org/10.1161/CIRCEP.121.009871Circulation: Arrhythmia and Electrophysiology. 2021;14:e009871The use of artificial intelligence (AI) is rapidly expanding in clinical care,1 including the use of neural networks for mortality predictions and identification of cardiac pathologies based on an ECG.2 The ubiquity of the ECG holds much promise as a screening tool with the goal of enhancing patient care. An ideal setting for utilization of this technology is the emergency department (ED), specifically among patients with dyspnea. In light of this, the AI-enabled ECG algorithm has received emergency use authorization by the Food and Drug Administration to screen patients with confirmed or suspected coronavirus disease 2019 (COVID-19) disease for left ventricular dysfunction (LVD).3,4 Our objective was to evaluate the effect of ECG characteristics on AI-ECG performance among ED patients.A retrospective review of patients who presented to the ED at Mayo Clinic, Florida, who had an ECG and transthoracic echocardiogram within 48 hours of one another between May 1, 2020 and December 31, 2020, was performed and 1300 unique patients were selected for study inclusion. Demographics, left ventricular ejection fraction (EF), and ECG parameters (PR interval, QT duration, corrected QT duration, presence of right bundle branch block, left bundle branch block, and ventricular pacing) were obtained. Details of the AI-ECG algorithm have been previously published.2 The diagnostic performance of the AI-ECG in predicting LVD (based on left ventricular EF assessment by echocardiography) was evaluated using area under the receiver operating characteristic curve (AUROC), sensitivity, specificity, negative, and positive predictive values. A numeric predicted probability between 0 and 1 is generated by the AI model for each ECG analyzed and the threshold value used to indicate a positive screen was ≥0.256 as determined from the initial study.Consistent with the STARD criteria (Standards for Reporting of Diagnostic Accuracy Studies),5 estimates of diagnostic accuracy are provided with 95% confidence intervals. Statistical analysis was performed with R Statistical Software (version 3.6.2; R Foundation for Statistical Computing, Vienna, Austria). The data that support the findings of this study are available from the corresponding author upon reasonable request. The study was approved by the Mayo Clinic Florida Institutional Review Board.Median age of patients was 69 years (Q1: 58 years Q3: 78 years), and 43.3% were female. Overall, for detection of LVD at EF threshold ≤35%, the AI-ECG had an AUROC of 0.893, accuracy of 82.7%, a sensitivity of 78.6%, a specificity of 83.1%, a positive predictive value of 30.4%, and a negative predictive value of 97.6%. Within a prespecified subset of patients, the AI-ECG performance remained stable with an AUROC of 0.822 among those with ventricular pacing, 0.857 for right bundle branch block, 0.784 for left bundle branch block, and 0.859 for wide QRS (Figure).Download figureDownload PowerPointFigure. Forrest plot depicting the performance of the artificial intelligence–ECG algorithm for detection of left ventricular dysfunction (defined as ejection fraction ≤35%). Percentage values in parenthesis represent 95% exact CIs while the fraction shows the number of subjects in each group. For sensitivity: true positive (TP)/(TP +false negative); for specificity true negative (TN)/(TN+false positive). AUROC indicates area under the receiver operator characteristic curve; LBBB, left bundle branch block; and RBBB, right bundle branch block.To detect LVD at EF threshold <50%, the AI-ECG had an AUROC of 0.841, accuracy of 81.8%, sensitivity of 59.9%, specificity of 87.3%, positive predictive value of 54.3%, and negative predictive value of 89.6%. The AI-ECG algorithm had an AUROC of 0.752 among those with ventricular pacing, 0.825 for right bundle branch block, 0.834 for left bundle branch block, and 0.832 for wide QRS.The effectiveness of the AI-ECG tool is further highlighted in our study, specifically among patients with conduction abnormalities on a resting ECG. The AI-ECG performs well independent of ventricular pacing, wide QRS, or right bundle branch block at baseline. However, a modestly reduced performance is observed among patients with left bundle branch block (at an EF threshold of ≤35%) or ventricular pacing (at an EF threshold of < 50%). Although the AUROC values in our patient population appear slightly lower than in the original study, the performance of the AI-ECG remained strong and similar to subsequent validation studies of this AI model among ED patients (AUROC, 0.89).1 Changes in the AUROC value are likely due to expected statistical variation in patient data and a small sample size. Our findings suggest that the presence of a specific wide QRS pattern, pacer spikes, or QRS shape might be important variables in the AI-ECG’s determination of an individual’s probability of LVD.Our findings identify an important caveat for clinicians to be aware of when utilizing this tool in daily practice. Reporting AI model results without a measure of the certainty of the model may make it hard for clinicians to incorporate the results into their decision-making. Our study suggests that models should be reported not only as a standalone value but also in the context of other features that can be discerned from the model. We note that, in developing and training these models, we generally wish to include as broad a sample as possible (so that the model can be more broadly applied in practice), but it is important to be aware of the caveats that may inform model implementation. Our study was retrospective in nature with a relatively small sample size derived from ED where echocardiography was deemed to be clinically indicated. Efforts are currently underway to validate the AI-ECG tool in larger patient populations.Nonstandard Abbreviations and AcronymsAIartificial intelligenceEDemergency departmentLVDleft ventricular dysfunctionSources of FundingNone.Disclosures None.FootnotesFor Sources of Funding and Disclosures, see page 534.Correspondence to: Julio Perez-Downes, DO, Department of Cardiovascular Disease, Mayo Clinic Florida, 4500 San Pablo Rd. S, Jacksonville, FL 32204. Email perez-downes.[email protected]eduReferences1. Adedinsewo D, Carter RE, Attia Z, Johnson P, Kashou AH, Dugan JL, Albus M, Sheele JM, Bellolio F, Friedman PA, et al.. Artificial intelligence-enabled ECG algorithm to identify patients with left ventricular systolic dysfunction presenting to the emergency department with dyspnea.Circ Arrhythm Electrophysiol. 2020; 13:e008437. doi: 10.1161/CIRCEP.120.008437LinkGoogle Scholar2. Attia ZI, Kapa S, Lopez-Jimenez F, McKie PM, Ladewig DJ, Satam G, Pellikka PA, Enriquez-Sarano M, Noseworthy PA, Munger TM, et al.. Screening for cardiac contractile dysfunction using an artificial intelligence-enabled electrocardiogram.Nat Med. 2019; 25:70–74. doi: 10.1038/s41591-018-0240-2CrossrefMedlineGoogle Scholar3. Ladejobi AO, Cruz J, Attia ZI, van Zyl M, Tri J, Lopez-Jimenez F, Noseworthy PA, Friedman PA, Kapa S, Asirvatham SJ. Digital health innovation in cardiology.Cardiovasc Digit Health J. 2020; 1:6–8. doi: 10.1016/j.cvdhj.2020.07.003CrossrefMedlineGoogle Scholar4. US FDA. Emergent use of the Eko electrocardiogram Low Ejection Fraction Tool (ELEFT) during the COVID-19 pandemic.May 11, 2020.Google Scholar5. Bossuyt PM, Reitsma JB, Bruns DE, Gatsonis CA, Glasziou PP, Irwig L, Lijmer JG, Moher D, Rennie D, de Vet HC, et al.; STARD Group. STARD 2015: an updated list of essential items for reporting diagnostic accuracy studies.BMJ. 2015; 351:h5527. doi: 10.1136/bmj.h5527CrossrefMedlineGoogle Scholar Previous Back to top Next FiguresReferencesRelatedDetails May 2021Vol 14, Issue 5Article InformationMetrics Download: 162 © 2021 American Heart Association, Inc.https://doi.org/10.1161/CIRCEP.121.009871PMID: 33993719 Originally publishedMay 17, 2021 Keywordsechocardiographyartificial intelligencedyspneaCOVID-19patientsPDF download SubjectsArrhythmiasMachine Learning and Artificial IntelligenceElectrophysiologyHeart Failure" @default.
- W3162183883 created "2021-05-24" @default.
- W3162183883 creator A5008299307 @default.
- W3162183883 creator A5020600761 @default.
- W3162183883 creator A5020655045 @default.
- W3162183883 creator A5023936819 @default.
- W3162183883 creator A5043658621 @default.
- W3162183883 creator A5088770382 @default.
- W3162183883 date "2021-05-01" @default.
- W3162183883 modified "2023-09-24" @default.
- W3162183883 title "Impact of ECG Characteristics on the Performance of an Artificial Intelligence Enabled ECG for Predicting Left Ventricular Dysfunction" @default.
- W3162183883 cites W2901226889 @default.
- W3162183883 cites W3020992444 @default.
- W3162183883 cites W3047391569 @default.
- W3162183883 cites W3082446445 @default.
- W3162183883 doi "https://doi.org/10.1161/circep.121.009871" @default.
- W3162183883 hasPubMedId "https://pubmed.ncbi.nlm.nih.gov/33993719" @default.
- W3162183883 hasPublicationYear "2021" @default.
- W3162183883 type Work @default.
- W3162183883 sameAs 3162183883 @default.
- W3162183883 citedByCount "1" @default.
- W3162183883 countsByYear W31621838832023 @default.
- W3162183883 crossrefType "journal-article" @default.
- W3162183883 hasAuthorship W3162183883A5008299307 @default.
- W3162183883 hasAuthorship W3162183883A5020600761 @default.
- W3162183883 hasAuthorship W3162183883A5020655045 @default.
- W3162183883 hasAuthorship W3162183883A5023936819 @default.
- W3162183883 hasAuthorship W3162183883A5043658621 @default.
- W3162183883 hasAuthorship W3162183883A5088770382 @default.
- W3162183883 hasBestOaLocation W31621838831 @default.
- W3162183883 hasConcept C126322002 @default.
- W3162183883 hasConcept C154945302 @default.
- W3162183883 hasConcept C164705383 @default.
- W3162183883 hasConcept C41008148 @default.
- W3162183883 hasConcept C71924100 @default.
- W3162183883 hasConceptScore W3162183883C126322002 @default.
- W3162183883 hasConceptScore W3162183883C154945302 @default.
- W3162183883 hasConceptScore W3162183883C164705383 @default.
- W3162183883 hasConceptScore W3162183883C41008148 @default.
- W3162183883 hasConceptScore W3162183883C71924100 @default.
- W3162183883 hasIssue "5" @default.
- W3162183883 hasLocation W31621838831 @default.
- W3162183883 hasOpenAccess W3162183883 @default.
- W3162183883 hasPrimaryLocation W31621838831 @default.
- W3162183883 hasRelatedWork W2008851126 @default.
- W3162183883 hasRelatedWork W2011347913 @default.
- W3162183883 hasRelatedWork W2049397185 @default.
- W3162183883 hasRelatedWork W2073151595 @default.
- W3162183883 hasRelatedWork W2074833529 @default.
- W3162183883 hasRelatedWork W2125804349 @default.
- W3162183883 hasRelatedWork W2159512267 @default.
- W3162183883 hasRelatedWork W2304633692 @default.
- W3162183883 hasRelatedWork W2355498105 @default.
- W3162183883 hasRelatedWork W2399063111 @default.
- W3162183883 hasVolume "14" @default.
- W3162183883 isParatext "false" @default.
- W3162183883 isRetracted "false" @default.
- W3162183883 magId "3162183883" @default.
- W3162183883 workType "article" @default.