Predicting Survival of End-Stage Heart Failure Patients Receiving HeartMate-3: Comparing Machine Learning Methods

Area of Science:

• Cardiology
• Medical Devices
• Data Science

Background:

• The HeartMate 3 is the sole durable left ventricular assist device (LVAD) currently implanted in the United States.
• Predictive modeling is crucial for managing patient outcomes after LVAD implantation.

Purpose of the Study:

• To develop and compare predictive models for 1-year mortality in HeartMate 3 LVAD patients.
• Evaluate the performance of standard statistical techniques against machine learning algorithms for outcome prediction.

Main Methods:

• Analysis of adult patients from the STS-INTERMACS registry who received a primary HeartMate 3 implant (2017-2019).
• Inclusion of epidemiological, clinical, hemodynamic, and echocardiographic data.
• Comparison of standard logistic regression with machine learning models (elastic net, neural network) for predicting 1-year survival.

Main Results:

• 3,853 patients were included; 12.8% experienced 1-year mortality.
• Logistic regression identified age, MELD-XI score, RA pressure, INTERMACS profile, heart rate, and HF etiology as key predictors (AUC: 0.72).
• The standard logistic regression model demonstrated non-inferiority to elastic net and neural network models.

Conclusions:

• Standard statistical techniques are as effective as machine learning algorithms for predicting 1-year survival post-HeartMate 3 implantation.
• The utility of machine learning may be dataset-dependent for predicting LVAD outcomes.