Predicting Stroke and Mortality in Mitral Regurgitation: A Machine Learning Approach

Area of Science:

• Cardiology
• Medical Informatics
• Biostatistics

Background:

• Mitral regurgitation (MR) poses significant risks for mortality and cerebrovascular events.
• Accurate prediction of these outcomes is crucial for patient management.
• Existing risk stratification models may not fully leverage diverse clinical and imaging data.

Purpose of the Study:

• To develop and validate an interpretable gradient boosting machine (GBM) model for predicting mortality and cerebrovascular events (CVEs) in patients with MR.
• To identify key predictors of high-risk patients using the GBM model.
• To compare the predictive performance of GBM against other machine learning models.

Main Methods:

• A cohort of 706 patients with MR from a tertiary center was analyzed.
• An interpretable gradient boosting machine (GBM) model was developed incorporating comorbidities, P-wave duration (PWD), and echocardiographic measurements.
• Model performance was evaluated against logistic regression, decision tree, random forest, support vector machine, and artificial neural networks using precision, sensitivity, c-statistic, and F1-score.

Main Results:

• The GBM model identified significant predictors for TIA/stroke including age, blood pressure, albumin, mean PWD, MR regurgitant volume, LVEF, LADs, VTI, and effective regurgitant orifice.
• Predictors for all-cause mortality included age, sodium, urea, albumin, platelet count, mean PWD, LVEF, LADs, LVDs, and VTI.
• GBM demonstrated superior predictive performance compared to other evaluated models.

Conclusions:

• An interpretable GBM model integrating clinical, electrocardiographic, and echocardiographic data significantly enhances risk prediction for mortality and CVEs in MR patients.
• The GBM approach effectively identifies critical indicators for adverse outcomes in MR.
• This methodology offers a more robust tool for risk stratification and personalized management in MR.