Using machine learning to improve risk prediction in durable left ventricular assist devices.

Area of Science:

• Cardiology
• Medical Informatics
• Machine Learning

Background:

• Traditional risk models for left ventricular assist device (LVAD) therapy show moderate predictive performance for mortality.
• Accurate risk stratification is crucial for optimizing patient selection and management in LVAD therapy.

Purpose of the Study:

• To evaluate the efficacy of machine learning (ML) models in improving mortality risk prediction for patients receiving primary durable LVADs.
• To compare the performance of ML models against traditional logistic regression.

Main Methods:

• Utilized data from the Interagency Registry for Mechanically Assisted Circulatory Support (2006-2016) for 16,120 patients.
• Developed and compared logistic regression and extreme gradient boosting (ML) models for 90-day and 1-year mortality prediction.
• Employed bootstrapping for performance evaluation and analyzed model concordance.

Main Results:

• Extreme gradient boosting demonstrated a statistically significant improvement in the C-index for both 90-day (0.740 vs 0.707) and 1-year (0.714 vs 0.691) mortality prediction compared to logistic regression.
• Net reclassification index analysis showed significant improvements of 48.8% (90-day) and 36.9% (1-year) with extreme gradient boosting.
• Model concordance between logistic regression and extreme gradient boosting substantially enhanced predictive accuracy for both methods.

Conclusions:

• Machine learning, specifically extreme gradient boosting, offers superior risk prediction performance for durable LVAD therapy compared to traditional methods.
• ML models can be used independently or in conjunction with logistic regression to refine mortality risk assessment in LVAD patients.
• These findings support the integration of ML into clinical practice for enhanced LVAD patient management and outcomes.