Knowledge-aware patient representation learning for multiple disease subtypes.

Area of Science:

• Medical Informatics
• Computational Biology
• Machine Learning

Background:

• Effective patient representation is crucial for downstream applications like clinical endpoint prediction.
• Existing methods often fail to adequately address disease subtypes, impacting representation learning performance.
• Patient data heterogeneity across disease subtypes necessitates integrated learning approaches.

Purpose of the Study:

• To develop a method for effectively integrating data from all disease subtypes to enhance patient representation learning.
• To improve the performance of downstream tasks, such as clinical endpoint prediction, by learning more robust patient representations.
• To address the limitations of existing models that either ignore subtype distinctions or neglect inter-subtype correlations.

Main Methods:

• Proposed a knowledge-aware shared-private neural network model.
• Explicitly incorporated disease-oriented knowledge into the model.
• Learned shared and specific representations from both disease and subtype perspectives.
• Evaluated the model on a clinical endpoint prediction task using real-world clinical datasets.

Main Results:

• The proposed model significantly improved patient representation learning compared to state-of-the-art methods.
• Integrated learning across disease subtypes yielded superior performance in the downstream task.
• The knowledge-aware approach effectively captured both shared and subtype-specific patient information.

Conclusions:

• The developed knowledge-aware shared-private neural network model offers a superior approach to patient representation learning.
• Integrating subtype information and disease-level knowledge enhances predictive accuracy for clinical endpoints.
• This method provides a promising direction for leveraging complex patient data in healthcare applications.