EAPR: explainable and augmented patient representation learning for disease prediction.

Area of Science:

• Biomedical Informatics
• Machine Learning in Healthcare
• Computational Medicine

Background:

• Patient representation learning encodes Electronic Health Records (EHR) for disease prediction.
• Deep learning enhances EHR representation but requires large datasets and lacks explainability.
• Missing or insufficient patient data hinders robust representation learning.

Purpose of the Study:

• To propose an Explainable and Augmented Patient Representation Learning (EAPR) method for disease prediction.
• To address challenges of limited data and model inexplicability in patient representation learning.
• To improve the performance and interpretability of disease prediction models.

Main Methods:

• EAPR employs confidence interval-controlled data augmentation to enhance patient representations with limited data.
• A two-stage gradient backpropagation technique is utilized to ensure model explainability.
• The approach is validated using real-world clinical data.

Main Results:

• EAPR effectively enhances patient representation learning, particularly with insufficient data.
• The proposed method significantly improves disease prediction model performance.
• Experimental results demonstrate the explainability of the EAPR approach.

Conclusions:

• EAPR offers a robust solution for patient representation learning in disease prediction.
• The method successfully balances data augmentation with model explainability.
• EAPR advances the development of reliable and interpretable AI in healthcare.