Abstract 11109: Electrocardiogram-Based Deep Learning to Predict Peak Oxygen Consumption During Exercise
Shaan Khurshid,Timothy Churchill,Nathaniel Diamant,Paolo Di Achille,Christopher Reeder,Pulkit Singh,Meagan M Wasfy,Jennifer E Ho,Aaron L Baggish,Puneet Batra,Steven A Lubitz,James S Guseh,Shaan KhurshidTimothy ChurchillNathaniel DiamantPaolo Di AchilleChristopher ReederPulkit SinghMeagan M WasfyJennifer E HoAaron L BaggishPuneet BatraSteven A LubitzJames S GusehCardiac Electrophysiology,Massachusetts General Hosp,Boston,MAMASSACHUSETTS GENERAL HOSPITAL,Boston,MABerkeley,CABroad Institute,Medford,MABroad Institute,Cambridge,MABroad Institute,Somerville,MACardiology,Beth Israel Deaconess Med Cntr,Boston,MACardiology,Massachusetts General Hosp,Boston,MACambridge,MAMassachusetts General Hosp,Boston,MA
DOI: https://doi.org/10.1161/circ.146.suppl_1.11109
IF: 37.8
2022-11-01
Circulation
Abstract:Circulation, Volume 146, Issue Suppl_1, Page A11109-A11109, November 8, 2022. Introduction:Peak oxygen consumption (V̇O2 Peak) is a robust marker of cardiovascular health. However, V̇O2 Peak assessment requires specialized exercise testing.Methods:We studied individuals who underwent cardiopulmonary exercise testing (CPET) with available resting 12-lead electrocardiogram (ECG) within 1 year. We developed a deep learning model to generate 320-dimensional representations of ECG waveforms. We trained and validated 3 penalized linear models to estimate V̇O2 Peak using 1) basic clinical factors (age, sex, and body mass index), 2) clinical factors plus routine ECG parameters (e.g., PR interval), and 3) clinical factors plus deep learning- ECG embeddings ("Deep ECG-V̇O2"). We assessed model performance by comparing Pearson correlation (r) and mean absolute error (MAE) against true V̇O2 Peak. Within a separate ambulatory sample, we assessed associations between Deep ECG-V̇O2-estimated V̇O2 Peak and incident outcomes using Cox proportional hazards models adjusted for age, sex, and body mass index.Results:The analysis set included 2,339 individuals (age 46±19 years, 37% female) who underwent CPET and was divided into development (n=1,891) and test (n=448) sets. Mean V̇O2 Peak was 33.6±14.4 mL/kg/min. In the test set, correlation and agreement with true V̇O2 Peak were favorable using Deep ECG-V̇O2 versus the comparison models (Figure). Within a separate set of 84,718 individuals, estimated V̇O2 Peak <14 mL/kg/min (a clinical threshold for substantially impaired fitness) was associated with higher risks of incident atrial fibrillation (hazard ratio [HR] 1.36, 95% CI 1.21-1.54), myocardial infarction (HR 1.21, 95% CI 1.02-1.45), heart failure (HR 1.67, 95% CI 1.49-1.88), and all-cause death (HR 1.84, 95% CI 1.68-2.03) (Figure).Conclusions:Information encoded within 12-lead ECGs can estimate V̇O2 Peak. Estimated V̇O2 Peak demonstrates strong inverse relationships with incident cardiovascular disease and all-cause mortality.
cardiac & cardiovascular systems,peripheral vascular disease