Comprehensive Summary
This study by Feeny et. al investigates the effectiveness of machine learning (ML) on predicting cardiac resynchronization therapy (CRT) response compared to currently existing guidelines. The researchers collected CRT patient data from Johns Hopkins and Cleveland Clinic—splitting them into training and testing cohorts—and used com ML classification from different combinations of clinical and device variables. Among 925 patients, the best ML model demonstrated was the Bayes classifier, which used nine variables. Increasing the number of clinical variables did not improve performance. Overall, ML incrementally improved the prediction of echocardiographic CRT response and survival beyond current guidelines.
Outcomes and Implications
The research is important because CRT is beneficial for left ventricular (LV) remodeling and improved outcomes in patients with heart failure. ML can leverage this clinical data, as better predictive outcomes can optimize patient selection and avoid unnecessary interventions. Clinically, incorporating ML can refine CRT candidate selection beyond current guidelines. However, there is no timeline for clinical implementation as further validation is required. Limitations of the study were that it is unknown how a model with a larger training size may perform. Furthermore, the researchers did not assess the benefit of ML usage compared with a non-CRT comparison group, as all patients in the study received CRT.