Machine Learning Prediction of Cardiac Resynchronisation Therapy Response From Combination of Clinical and
Svyatoslav Khamzin1, Arsenii Dokuchaev1, Anastasia Bazhutina1,2
1Institute of Immunology and Physiology Ural Branch of the Russian Academy of Sciences, Yekaterinburg, Russia.
Frontiers in Physiology
|December 31, 2021
Summary
Predicting cardiac resynchronization therapy (CRT) success is challenging. Combining clinical data with computational modeling significantly improves prediction accuracy for CRT outcomes in heart failure patients.
Area of Science:
- Cardiology
- Biomedical Engineering
- Computational Biology
Background:
- Cardiac resynchronization therapy (CRT) is effective for many heart failure patients, but 30-50% do not respond.
- Patient selection and device optimization for CRT remain significant clinical challenges.
Purpose of the Study:
- To develop a predictive model for CRT outcomes.
- To integrate pre-implantation clinical data with personalized cardiac electrophysiology simulations.
Main Methods:
- Utilized retrospective data from 57 CRT patients.
- Created personalized computational heart models from imaging and ECG data.
- Employed machine learning on a hybrid dataset of clinical and simulation-derived biomarkers.
Main Results:
- The best model combining clinical and simulation data achieved an ROC AUC of 0.82.
- This hybrid model outperformed models using only clinical data.
- Key predictors included pacing site proximity to scar tissue and ventricular activation patterns.
Conclusions:
- Integrating computational modeling with clinical data enhances the accuracy of CRT outcome prediction.
- This approach offers a promising strategy for patient stratification and treatment optimization.
