A Machine Learning Approach to Predictive Modeling of Cardiovascular Events.

Area of Science:

• Cardiology
• Machine Learning in Healthcare
• Predictive Analytics

Background:

• Acute coronary syndromes (ACS) are a significant cause of death and illness.
• Predicting major adverse cardiovascular events (MACE) is crucial for patient care and resource management.
• Current prediction methods require enhancement for improved accuracy and timeliness.

Purpose of the Study:

• To develop and validate a Random Forest (RF) model for predicting MACE in ACS patients.
• To assess the model's performance at various time points: 30 days, 1 year, 2 years, and 3 years post-admission.
• To identify key predictors of MACE in the ACS population.

Main Methods:

• Utilized data from 2,721 ACS patients (2018-2024) at the Heart Hospital, Qatar.
• Employed a Random Forest algorithm, incorporating demographics, medical history, and clinical data, with NLP for text processing.
• Implemented rigorous methods to prevent data leakage and ensure reliable model estimation.

Main Results:

• Cumulative MACE prevalence reached 58.1% by 3 years.
• The RF model demonstrated strong predictive performance with AUC values from 0.817 to 0.865.
• Key predictors included higher age, lower ejection fraction, and elevated troponin and creatinine levels.

Conclusions:

• The developed RF model accurately predicts MACE in ACS patients across multiple time horizons.
• This predictive tool can aid in optimizing patient management, resource allocation, and cost reduction.
• The study highlights the potential of machine learning in enhancing cardiovascular care outcomes.