Knowledge-aware multi-center clinical dataset adaptation: Problem, method, and application.

Area of Science:

• Computational biology
• Medical informatics
• Artificial intelligence in healthcare

Background:

• Multi-center clinical data presents distribution shifts, hindering unified analysis.
• Leveraging diverse datasets is crucial for advancing clinical applications.
• Existing methods struggle to effectively integrate heterogeneous clinical data from multiple sources.

Purpose of the Study:

• To propose a novel approach for multi-center clinical data mining using external knowledge graphs.
• To enhance patient feature representation by capturing shared characteristics across datasets.
• To improve the semantic enrichment of clinical data through KG-infused learning.

Main Methods:

• Developed an adversarial learning model for shared patient feature representation.
• Employed an external knowledge graph (KG) to enrich data semantics.
• Utilized a graph convolutional autoencoder for knowledge feature training.
• Evaluated the model on Chinese hospital cardiology data and the MIMIC III dataset.

Main Results:

• The proposed model effectively captures shared patient features from heterogeneous multi-center data.
• Knowledge graph infusion significantly enriches clinical data semantics.
• The model demonstrates efficacy in predicting acute kidney injury in heart failure patients.
• Experimental results validate the superiority of the KG-infused approach over traditional methods.

Conclusions:

• Integrating external knowledge graphs with adversarial learning offers a powerful solution for multi-center clinical data mining.
• The proposed method enhances the interpretability and predictive power of clinical data analysis.
• This approach holds significant potential for improving patient outcomes in various clinical settings.