Abstract 10702: Machine Learning of 12-lead Qrs Waveform Patterns to Identify Cardiac Resynchronization Therapy Patients With Differential Outcomes

• Published in: Circulation (New York, N.Y.) N.Y.), 2019-11, Vol.140 (Suppl_1 Suppl 1), p.A10702-A10702
• Main Authors: Feeny, Albert K, Rickard, John, Trulock, Kevin M, Patel, Divyang, Toro, Saleem, Moennich, Laurie Ann, Varma, Niraj, Niebauer, Mark J, Gorodeski, Eiran Z, Grimm, Richard A, Barnard, John, Madabhushi, Anant, Chung, Mina K
• Format: Article
• Language: English
• ISSN: 0009-7322; 1524-4539
• DOI: 10.1161/circ.140.suppl_1.10702

BackgroundCardiac resynchronization therapy (CRT) improves heart failure outcomes but has significant non-response rates. QRS duration (QRSd) and left bundle branch block (LBBB) are main ECG discriminators for patient selection but have limitations.HypothesisECG waveform machine learning analysis using dimensionality reduction and unsupervised clustering can identify CRT subgroups with differential outcomes.MethodsWe retrospectively studied 946 CRT patients with conduction delay. 539 patients with survival and left ventricular ejection fraction (LVEF) outcomes comprised the Primary Cohort. 301 patients with survival outcomes and 106 patients with LVEF outcomes comprised the validation cohorts. In the Primary Cohort, we used principal components analysis (PCA) dimensionality reduction and k-means clustering to identify two groups from pre-CRT 12-lead QRS patterns. We used Cox regression to examine a composite endpoint of death, left ventricular assist device, or heart transplant, and compared LVEF change in the two groups. Outcomes of the two groups were evaluated in the validation cohorts.ResultsIn the Primary Cohort, PCA and k-means clustering identified two groups. Compared to Group 2 (n=234), Group 1 (n=305) had lower risk for reaching the composite endpoint (HR 0.44, p