Machine learning for multidimensional response and survival after cardiac resynchronization therapy using features...

Area of Science:

• Cardiology
• Medical Imaging
• Machine Learning

Background:

• Predicting patient response to cardiac resynchronization therapy (CRT) is complex.
• Improved methods are needed to forecast survival and the necessity for advanced therapies in CRT patients.

Purpose of the Study:

• To employ machine learning for characterizing multidimensional CRT response.
• To investigate the relationship between CRT response and long-term survival.

Main Methods:

• Evaluated associations between 39 baseline features (including cardiac magnetic resonance [CMR] findings and glomerular filtration rate [GFR]) and a multidimensional CRT response vector.
• Utilized machine learning to define patient response clusters and assessed their association with 4-year survival via cross-validation.

Main Results:

• Identified associations between CMR dyssynchrony parameters and GFR with multiple response metrics.
• Machine learning defined 3 distinct CRT response clusters with varying survival rates (best: 90.2%, intermediate: 60.0%, worst: 34.4%).
• Incorporating the 6-month response cluster improved 4-year survival prediction accuracy.

Conclusions:

• Machine learning effectively characterizes distinct CRT response clusters.
• These clusters are influenced by CMR features and kidney function, significantly impacting long-term survival.
• A web-based tool was developed for determining these response clusters in future patients.