In-silico mechanistic interpretation of echocardiographic predictors of reverse remodeling after cardiac resynchronization therapy
L Gegout, J Saleh, J Beaumont, A Hernandez, V Le Rolle, E DonalAbstract
Background
Cardiac resynchronization therapy (CRT) improves outcomes in heart failure with reduced ejection fraction and electrical dyssynchrony, yet up to one-third of guideline-eligible patients fail to exhibit reverse remodeling. While several echocardiographic markers have been associated with CRT response, their mechanistic underpinnings remain incompletely understood. Integrating clinical imaging data with physiologically grounded cardiovascular models may provide novel insight into the determinants of CRT efficacy.
Purpose
To identify key echocardiographic predictors of CRT response using machine learning and to provide a mechanistic interpretation through an in-silico cardiovascular model.
Methods
We studied 152 CRT-eligible patients undergoing baseline speckle-tracking echocardiography. CRT response was defined as ≥15% reduction in LV end-systolic volume at 6 months. Clinical and echocardiographic variables (including septal flash, QRS duration, transmitral inflow parameters, aortic pre-ejection delay, and strain-derived integral features) were analyzed using Random Forest to identify the most predictive markers. An integrated in-silico cardiovascular model incorporating electrical activation, atrial and ventricular mechanics, and systemic-pulmonary circulation was used to perform global sensitivity analysis on the machine-learning-selected predictors, enabling mechanistic interpretation of their physiological determinants.
Results
CRT response occurred in 106 patients (70%). Random Forest analysis demonstrated robust discrimination (AUC 0.83 ± 0.07). The most influential predictors were E/A ratio, septal flash, peak E velocity, diastolic duration, aortic pre-ejection delay, and strain-integral features. Mechanistic simulations consistently highlighted diastolic properties as dominant determinants of response: E/A ratio emerged as the strongest predictor and was primarily governed by left-atrial compliance and LV relaxation parameters. Aortic pre-ejection delay, which also appears as a major predictor of CRT response, was mainly driven by systemic vascular resistance, aortic elastance, and LV contractility. These findings indicate that CRT response is critically modulated by baseline filling pressures, atrioventricular coupling, and ventricular loading conditions rather than electrical dyssynchrony alone.
Conclusion
By combining machine learning with in-silico cardiovascular modeling, this study provides a mechanistic framework linking echocardiographic predictors to CRT response physiology. Diastolic function and loading conditions emerge as central determinants of reverse remodeling, while aortic pre-ejection delay reflects integrated afterload–contractility interaction. This integrative approach supports a shift toward physiology-guided imaging phenotyping to refine CRT patient selection and improve therapeutic yield.