Post-infarct cardiac remodeling predictions with machine learning.

Area of Science:

• Cardiology
• Medical Informatics
• Machine Learning

Background:

• Left ventricular remodeling (LVR) is a significant concern after myocardial infarction (MI).
• Accurate prediction of LVR is crucial for timely intervention and improved patient outcomes.
• Current prediction methods may lack sufficient sensitivity.

Purpose of the Study:

• To enhance the risk prediction of 3-month LVR occurrence post-MI.
• To leverage a machine learning approach for improved predictive accuracy.
• To identify key variables for predicting LVR using data-driven methods.

Main Methods:

• Utilized data from a prospective cohort study of ST-elevation MI patients.
• Collected clinical, biological, and cardiac magnetic resonance (CMR) imaging data within the first week post-MI.
• Employed a machine learning pipeline with feature selection to identify predictive variables for LVR assessed at 3 months via CMR.

Main Results:

• A neural network model using seven key variables (creatine kinase, mean corpuscular volume, baseline left atrial surface, diabetes history, hypertension history, red blood cell distribution width, creatinine) outperformed a baseline logistic regression model.
• The best predictive model achieved an Area Under the Curve (AUC) of 0.78, with 92% sensitivity and 55% specificity.
• This machine learning model significantly improved upon the baseline model's AUC of 0.71, 67% sensitivity, and 64% specificity.

Conclusions:

• An unbiased, data-driven machine learning approach offers value in predicting post-MI LVR.
• The developed model demonstrates higher sensitivity for predicting 3-month LVR compared to traditional methods.
• This highlights the potential of machine learning in cardiovascular risk stratification.