Portal de Periódicos da CAPES

Machine Learning Algorithms for Prediction of Permanent Pacemaker Implantation After Transcatheter Aortic Valve Replacement

2021; Lippincott Williams & Wilkins; Volume: 14; Issue: 3 Linguagem: Inglês

10.1161/circep.120.008941

1941-3149

Takahiro Tsushima, Sadeer Al‐Kindi, Fahd Nadeem, Guilherme F. Attizzani, Yakov Elgudin, Alan Markowitz, Marco A. Costa, Daniel I. Simon, Maurício Arruda, Judith A. Mackall, Sergio Thal,

Infective Endocarditis Diagnosis and Management

HomeCirculation: Arrhythmia and ElectrophysiologyVol. 14, No. 3Machine Learning Algorithms for Prediction of Permanent Pacemaker Implantation After Transcatheter Aortic Valve Replacement Free AccessReview ArticlePDF/EPUBAboutView PDFView EPUBSections ToolsAdd to favoritesDownload citationsTrack citationsPermissions ShareShare onFacebookTwitterLinked InMendeleyRedditDiggEmail Jump toSupplementary MaterialsFree AccessReview ArticlePDF/EPUBMachine Learning Algorithms for Prediction of Permanent Pacemaker Implantation After Transcatheter Aortic Valve Replacement Takahiro Tsushima, MD Sadeer Al-Kindi, MD Fahd Nadeem, MD Guilherme F. Attizzani, MD Yakov Elgudin, MD, PhD Alan Markowitz, MD Marco A. Costa, MD, PhD Daniel I. Simon, MD Mauricio S. Arruda, MD Judith A. Mackall, MD Sergio G. ThalMD Takahiro TsushimaTakahiro Tsushima https://orcid.org/0000-0003-0190-7555 Department of Medicine (T.T.), Case Western Reserve University, Harrington Heart and Vascular Institute, and University Hospitals Cleveland Medical Center, OH. , Sadeer Al-KindiSadeer Al-Kindi https://orcid.org/0000-0002-1122-7695 Division of Cardiology, Department of Medicine (S.A.-K., F.N., G.F.A., M.A.C., D.I.S., M.S.A., J.A.M., S.G.T.), Case Western Reserve University, Harrington Heart and Vascular Institute, and University Hospitals Cleveland Medical Center, OH. , Fahd NadeemFahd Nadeem https://orcid.org/0000-0001-9473-6243 Division of Cardiology, Department of Medicine (S.A.-K., F.N., G.F.A., M.A.C., D.I.S., M.S.A., J.A.M., S.G.T.), Case Western Reserve University, Harrington Heart and Vascular Institute, and University Hospitals Cleveland Medical Center, OH. , Guilherme F. AttizzaniGuilherme F. Attizzani Division of Cardiology, Department of Medicine (S.A.-K., F.N., G.F.A., M.A.C., D.I.S., M.S.A., J.A.M., S.G.T.), Case Western Reserve University, Harrington Heart and Vascular Institute, and University Hospitals Cleveland Medical Center, OH. , Yakov ElgudinYakov Elgudin Division of Cardiac Surgery, Department of Surgery (Y.E., A.M.), Case Western Reserve University, Harrington Heart and Vascular Institute, and University Hospitals Cleveland Medical Center, OH. , Alan MarkowitzAlan Markowitz https://orcid.org/0000-0003-0101-5047 Division of Cardiac Surgery, Department of Surgery (Y.E., A.M.), Case Western Reserve University, Harrington Heart and Vascular Institute, and University Hospitals Cleveland Medical Center, OH. , Marco A. CostaMarco A. Costa Division of Cardiology, Department of Medicine (S.A.-K., F.N., G.F.A., M.A.C., D.I.S., M.S.A., J.A.M., S.G.T.), Case Western Reserve University, Harrington Heart and Vascular Institute, and University Hospitals Cleveland Medical Center, OH. , Daniel I. SimonDaniel I. Simon https://orcid.org/0000-0002-3386-7650 Division of Cardiology, Department of Medicine (S.A.-K., F.N., G.F.A., M.A.C., D.I.S., M.S.A., J.A.M., S.G.T.), Case Western Reserve University, Harrington Heart and Vascular Institute, and University Hospitals Cleveland Medical Center, OH. , Mauricio S. ArrudaMauricio S. Arruda Division of Cardiology, Department of Medicine (S.A.-K., F.N., G.F.A., M.A.C., D.I.S., M.S.A., J.A.M., S.G.T.), Case Western Reserve University, Harrington Heart and Vascular Institute, and University Hospitals Cleveland Medical Center, OH. , Judith A. MackallJudith A. Mackall https://orcid.org/0000-0003-4324-0226 Division of Cardiology, Department of Medicine (S.A.-K., F.N., G.F.A., M.A.C., D.I.S., M.S.A., J.A.M., S.G.T.), Case Western Reserve University, Harrington Heart and Vascular Institute, and University Hospitals Cleveland Medical Center, OH. , Sergio G. ThalSergio G. Thal Correspondence to: Sergio G. Thal, MD, Electrophysiology, University Hospitals Cleveland Medical Center, Clinical Associate Professor, Department of Medicine, Case Western Reserve University, 11100 Euclid Ave Cleveland, OH 44106. Email E-mail Address: [email protected] Division of Cardiology, Department of Medicine (S.A.-K., F.N., G.F.A., M.A.C., D.I.S., M.S.A., J.A.M., S.G.T.), Case Western Reserve University, Harrington Heart and Vascular Institute, and University Hospitals Cleveland Medical Center, OH. Originally published9 Mar 2021https://doi.org/10.1161/CIRCEP.120.008941Circulation: Arrhythmia and Electrophysiology. 2021;14:e008941Atrioventricular block requiring permanent pacemaker (PPM) implantation remains an important complication after transcatheter aortic valve replacement (TAVR), and the risk stratification is essential to identify the subset of patients requiring the new PPM implantation beforehand. However, an accurate risk prediction is not established yet. Recently, machine learning (ML) technique which is a scientific discipline focusing on pattern recognitions is utilized for developing prediction models in clinical medicine, and the previously reported ML-based models demonstrated significantly high predictive accuracy.1,2 The aim of this study is to evaluate the performance of ML-based algorithms for predicting post-TAVR PPM implantation.This is a single-center retrospective study of consecutive patients who underwent TAVR from March 10, 2011 to October 8, 2018 (derivation cohort, group A) and a prospective cohort of TAVR patients between October 9, 2018 and November 9, 2019 (validation cohort, group B), at University Hospitals Cleveland Medical Center. This study utilized data extracted from TAVR research registry that was approved by an institutional review board at University Hospitals Cleveland Medical Center. All patients provided signed informed consent for the data collection. Patients with preexisting cardiac implantable electronic device were excluded from this study. The detailed information of ML analysis was summarized in the Data Supplement. In patients with post-TAVR PPM implantation who had available data on right ventricular pacing burden (n=132), we also evaluated whether these ML models (trained on PPM need) can predict significant right ventricular burden (≥40%) at 1 month by combining training and testing datasets. We considered right ventricular pacing burden ≥40% at 1 month after cardiac implantable electronic device implantations to be significant based on prior literature.3 The data that support the findings of this study are available from the corresponding author upon reasonable request.A total of 888 patients were ultimately included in group A, and 272 patients were in group B. In group A, 184 patients (20.7%) required new PPM, and the major indications were complete heart block in 70.1% and new left bundle branch block with subsequent high-grade atrioventricular block in 23.4%. In group B, 38 patients (14.0%) required PPM similarly for complete heart block in 71.1% and new left bundle branch block in 26.3%, respectively. The baseline characteristics of patients were summarized in Table I in the Data Supplement. Both preprocedural right bundle branch block and atrioventricular block were significantly associated with the new PPM implantation in both groups.Regarding the ML-model performances, Figure shows the classifier accuracy in both groups, and Table II in the Data Supplement summarized other model parameters. In group A, the model accuracy ranged from 59% to 69%, with sequential minimal optimization, simple logistic regression (SLR), and locally weighted learner (LWL)–based models demonstrating highest results (69%, 68%, and 68%, respectively). In group B, the model accuracy ranged between 55% and 75%, with SLR, LWL, and sequential minimal optimization–based classifiers demonstrating the best performance (75%, 74%, and 73%, respectively). Both SLR and LWL-based models achieved the highest area under curve receiver operating characteristics (AUCROC), 0.82. In group B, we also evaluated the performance of our previously reported prediction model that was extracted with the conventional multivariate logistic regression analysis and it also showed a high diagnostic accuracy (AUCROC, 0.81).3 We also found both SLR and LWL ML models modestly predicted significant right ventricular pacing burden at 1 month (AUCROC, 0.62 and 0.66, accuracy 61% and 67%, respectively).Download figureDownload PowerPointFigure. Accuracy of various machine learning classifiers in derivation and validation cohort (group A and B). LWL indicates locally weighted learner; REP, reduced error pruning; and SMO, sequential minimal optimization.Two unpublished studies reported the performance ok ML models to predict post-TAVR PPM implantation.4,5 Agasthi et al4 used 964 patients, and Gradient Boosting classifier showed modest discrimination (AUCROC of 0.66). Truong et al5 also used 701 patients, and the Random Forest demonstrated high prediction (balanced accuracy, 79%; F1 score, 0.62; and AUCROC, 0.88). In comparison to these studies, we utilized larger patient sample (n=1390) and ML-based classifiers (n=14). The internally validated results further supported ML-based classifiers can predict the incidence of the post-TAVR PPM accurately. However, SLR is one of the classical methods, and most ML algorithms did not outperform conventional methods in our present study. The current ML-algorism is still an imperfect science and clinicians should use appropriate ML-classifiers based on each dataset characteristic.There are some limitations. First, this is a single- center retrospective study, and the prediction models were extracted from our older cohort. Second, the indication for post-TAVR PPM was not established clearly in the beginning of TAVR era, and the limited experience may cause unnecessary PPM implantations. However, such a phenomenon was only for the early cases in our institution and it should not affect the diagnostic accuracy of our ML models entirely. Finally, the difference between utilized TAVR valves and preprocedural risk of adult cardiac surgery might affect the outcome.In conclusion, ML algorithms can predict the risk of post-TAVR PPM implantation accurately and both SLR and LWL-based classifiers achieved high performance in this study. A prospective or multicenter external validation should be undertaken.Nonstandard Abbreviations and AcronymsAUCROCarea under curve receiver operating characteristicsLWLlocally weighted learnerMLmachine learningPPMpermanent pacemakerSLRsimple logistic regressionTAVRtranscatheter aortic valve replacementSources of FundingNone.Disclosures Dr Attizzani is a consultant and is on the advisory board of Medtronic. Dr Simon has received honoraria for work as a course director from Medtronic. Dr Mackall has received consulting honoraria from Abbott. The other authors report no conflicts.Footnotes*T. Tsushima and S. Al-Kindi contributed equally as first authorsThe Data Supplement is available at https://www.ahajournals.org/doi/suppl/10.1161/CIRCEP.120.008941.For Sources of Funding and Disclosures, see page 371.Correspondence to: Sergio G. Thal, MD, Electrophysiology, University Hospitals Cleveland Medical Center, Clinical Associate Professor, Department of Medicine, Case Western Reserve University, 11100 Euclid Ave Cleveland, OH 44106. Email sergio.[email protected]orgReferences1. Deo RC. Machine learning in medicine.Circulation. 2015; 132:1920–1930. doi: 10.1161/CIRCULATIONAHA.115.001593LinkGoogle Scholar2. Hernandez-Suarez DF, Kim Y, Villablanca P, Gupta T, Wiley J, Nieves-Rodriguez BG, Rodriguez-Maldonado J, Feliu Maldonado R, da Luz Sant'Ana I, Sanina C, et al.. Machine learning prediction models for in-hospital mortality after transcatheter aortic valve replacement.JACC Cardiovasc Interv. 2019; 12:1328–1338. doi: 10.1016/j.jcin.2019.06.013CrossrefMedlineGoogle Scholar3. Tsushima T, Nadeem F, Al-Kindi S, Clevenger JR, Bansal EJ, Wheat HL, Kalra A, Attizzani GF, Elgudin Y, Markowitz A, et al.. Risk prediction model for cardiac implantable electronic device implantation after transcatheter aortic valve replacement.JACC Clin Electrophysiol. 2020; 6:295–303. doi: 10.1016/j.jacep.2019.10.020CrossrefMedlineGoogle Scholar4. Agasthi P, Mookadam F, Venepally N, Girardo M, Buras M, Khetarpal BK, Mulpuru SK, Eleid M, Greason K, Beohar N, et al.. Abstract 15572: Machine learning helps predict permanent pacemaker requirement post transcatheter aortic valve replacement.Circulation. 2019; 140:A15572. doi: 10.1161/circ.140.suppl_1.15572LinkGoogle Scholar5. Truong VT, Wigle M, Bateman E, Pallerla A, Ngo TNM, Beyerbach D, Kereiakes D, Shreenivas S, Tretter J, Palmer C, et al.. Pacemaker imlantation following TAVR: using machine learning to optimize risk stratification.JACC. 2020; 75:1478–1478. doi: 10.1016/S0735-1097(20)32105-7CrossrefGoogle Scholar Previous Back to top Next FiguresReferencesRelatedDetails March 2021Vol 14, Issue 3Article InformationMetrics Download: 169 © 2021 American Heart Association, Inc.https://doi.org/10.1161/CIRCEP.120.008941PMID: 33685208 Originally publishedMarch 9, 2021 Keywordsmachine learningatrioventricular blockrisktranscatheter aortic valve replacementpatientspacemakerPDF download SubjectsAortic Valve Replacement/Transcatheter Aortic Valve ImplantationArrhythmiasPacemaker