Causal Transformer for Learning Embeddings from Structured Medical History Records and Multi-Source Data Integration...

Area of Science:

• Computational biology
• Medical informatics
• Genomics

Background:

• Traditional disease risk models often use limited data, impacting accuracy.
• Integrating comorbidities and medical history is crucial for comprehensive risk prediction.
• Existing models struggle to leverage complex medical history data effectively.

Purpose of the Study:

• To develop a novel framework, Multi-source Integration for Disease Risk Prediction (MIDRP), for enhanced complex disease risk prediction.
• To integrate diverse data sources including genetic variants, lifestyle factors, physical attributes, and medical history.
• To leverage a causal Transformer architecture for nuanced medical history pattern extraction.

Main Methods:

• Proposed a Multi-source Integration for Disease Risk Prediction (MIDRP) framework.
• Utilized a causal Transformer architecture to analyze medical history records.
• Evaluated MIDRP against multiple baseline models using UK Biobank data for Coronary Artery Disease, Type 2 Diabetes, and Breast Cancer.

Main Results:

• MIDRP achieved state-of-the-art performance across three complex diseases.
• Achieved Area Under the Receiver Operating Characteristic Curve (AUROC) scores of 0.783 for Coronary Artery Disease, 0.841 for Type 2 Diabetes, and 0.784 for Breast Cancer.
• Demonstrated superior predictive accuracy compared to established methods like LDPred2, random forest, and Med-Bert.

Conclusions:

• The MIDRP framework offers a robust and accurate approach to complex disease risk prediction.
• Integrating multi-source data, particularly medical history, significantly enhances predictive capabilities.
• The causal Transformer architecture effectively extracts valuable patterns from electronic health records for improved risk assessment.