Multi-modality machine learning approach for risk stratification in heart failure with left ventricular ejection...

Area of Science:

• Cardiology
• Biomedical Engineering
• Data Science

Background:

• Heart failure (HF) involves complex cardiac remodeling, leading to electrical and mechanical dysfunction.
• Accurate risk stratification is crucial for managing HF patients and predicting adverse outcomes.
• Current risk prediction models may not fully capture the complexity of HF progression.

Purpose of the Study:

• To evaluate the efficacy of machine learning (ML) approaches in enhancing risk stratification for adverse outcomes in HF.
• To compare the predictive performance of ML techniques against traditional logistic regression.
• To investigate the utility of multi-modality data, including ECG, echocardiography, NLR, and PNI, in HF risk prediction.

Main Methods:

• A cohort of 312 Chinese HF patients diagnosed between 2010 and 2016 was analyzed.
• Two ML techniques (multilayer perceptron, multi-task learning) were employed.
• Model performance was assessed for predicting incident atrial fibrillation, TIA/stroke, and all-cause mortality, compared to logistic regression.

Main Results:

• Machine learning techniques demonstrated superior prediction performance compared to logistic regression.
• Several clinical, echocardiographic, and laboratory parameters were identified as significant predictors of adverse events.
• Specific ML models showed improved accuracy in predicting new-onset AF, TIA/stroke, and mortality in HF patients.

Conclusions:

• Multi-modality data assessment is vital for effective risk stratification in heart failure.
• Machine learning approaches offer significant added value for improving outcome prediction in HF.
• Integrating diverse data sources via ML can enhance personalized risk assessment and patient management.