Cardiac Magnetic Resonance Ventricular Function and Hemodynamic Multivariable Model Predictors of Exercise Performance in Tetralogy of Fallot Using Machine Learning and Multivariable Modelling: A Substudy of the Single Center Cardiac Magnetic Resonance Outcomes Registry-Tetralogy of Fallot

Background: There are no robust predictors of exercise performance in repaired tetralogy of Fallot (rTOF) using both ventricular function and hemodynamics. Aim: To utilize machine learning and multivariable modeling to determine cardiac magnetic resonance (CMR) and other metrics predictive of exercise performance. Methods: We retrospectively reviewed data of all rTOF pts undergoing CMR from 2005-2020. Those who had cardiopulmonary exercising testing (CPET) within 6 months of CMR were analyzed using the machine learning technique LASSO to choose candidate variables for subsequent testing with multivariable modeling. CMR parameters included right (RV) and left ventricular (LV) function and hemodynamics. Significance P<0.05. Results: 266 of 761 pts with rTOF met the inclusion criteria (age at CMR 16.1 + 9.7 years old, age at repair 0.36 years old, 56% male). Higher RV mass, RV global function index and lower pulmonary regurgitant fraction (PR) were independently associated with higher % predicted (%P) maximum oxygen consumption (VO 2 max). In addition, females and non-syndromic pts were also associated with higher %P VO 2 max (table). Those with a higher LV end-diastolic volume (EDV), females and pts without pulmonary atresia performed better at %P work. Those with a higher RV mass/volume demonstrated a worse ventilatory equivalent of carbon dioxide (VE/VCO 2 ). See table. Conclusion: RV ventricular functional parameters, PR (hemodynamics), gender and syndromic status are all independently associated with %P VO 2 max. Those with larger LV volumes were able to perform more work and those with a higher RV mass/volume had a worse VE/VCO 2 . Our findings shed some light into the etiology of poor rTOF exercise performance.