Adversarial MACE Prediction After Acute Coronary Syndrome Using Electronic Health Records

Area of Science:

• Cardiology
• Medical Informatics
• Machine Learning

Background:

• Acute coronary syndrome (ACS) is a major global health concern, leading to significant mortality and disability.
• Current major adverse cardiac event (MACE) prediction models for ACS lack granularity in distinguishing between ACS subtypes.
• There is a need for improved MACE prediction that accounts for the unique characteristics of different ACS subtypes.

Purpose of the Study:

• To develop and validate a novel multi-task learning (MTL) model for predicting MACE in ACS patients, specifically addressing different ACS subtypes.
• To leverage heterogeneous electronic health records (EHRs) for enhanced MACE prediction.
• To improve the fine-grained prediction of MACE by mining shared and private knowledge across ACS subtypes.

Main Methods:

• Proposed a multi-task learning (MTL) framework to predict MACE for distinct ACS subtypes simultaneously.
• Incorporated adversarial learning to disentangle shared and private latent features specific to each ACS subtype.
• Validated the model using a clinical dataset of 2,863 ACS patients from a Chinese hospital.

Main Results:

• The proposed MTL model demonstrated significant improvements in MACE prediction performance compared to single-subtype prediction models.
• Adversarial learning effectively mitigated interference between shared and private feature spaces of different ACS subtypes.
• The model successfully utilized heterogeneous EHR data for multi-subtype-oriented MACE prediction.

Conclusions:

• The developed MTL model offers a more accurate and nuanced approach to MACE prediction in ACS patients by considering subtype-specific information.
• This approach holds promise for improving early prevention and intervention strategies for ACS.
• Leveraging advanced machine learning techniques on EHR data can enhance clinical decision-making in cardiovascular medicine.