Risk Stratification and Outcome Prediction in Heart Failure Patients With Cardiac Implantable Electronic Devices...

Area of Science:

• Cardiology
• Medical Informatics
• Machine Learning

Background:

• Heart failure (HF) prevalence is rising in Japan's aging population.
• Implantable cardioverter defibrillator and cardiac resynchronization therapy use is lower in Japan compared to Western countries.
• The HINODE study prospectively collected data on Japanese patients with cardiac devices.

Purpose of the Study:

• To develop interpretable machine learning (ML) models for improved risk stratification in Japanese HF patients.
• To identify key predictors of adverse outcomes, including HF hospitalization and all-cause mortality.
• To support personalized management strategies for HF patients with cardiac devices.

Main Methods:

• Utilized data from 332 HINODE participants with adequate data.
• Developed predictive models using XGBoost with 5-fold cross-validation.
• Employed Shapley Additive Explanations (SHAP) for feature importance and K-means clustering for risk stratification.

Main Results:

• Models demonstrated strong discrimination for HF events (AUC 0.83) and mortality (AUC 0.85).
• Key predictors identified included QRS duration, QT interval, left ventricular volumes, and medications.
• Two distinct risk clusters were identified: low-risk (n=236) and high-risk (n=86), with significantly different event rates.

Conclusions:

• Interpretable ML models accurately predict risk and enable phenotype-based stratification in Japanese HF patients.
• Findings support the use of ML for personalized management of HF patients with cardiac devices.
• This approach can help optimize treatment strategies in this growing patient population.