Measuring Patient Similarity on Multiple Diseases by Joint Learning via a Convolutional Neural Network

Area of Science:

• Healthcare Informatics
• Medical Data Analysis
• Computational Medicine

Background:

• Patient similarity is crucial for clinical decision-making, cohort analysis, and personalized treatments.
• Medical data's complexity, irregularity, and sequential nature pose significant challenges to accurate similarity measurement.
• Existing methods often focus on single diseases, which is unrealistic given comorbidities.

Purpose of the Study:

• To develop and validate a novel model for measuring patient similarity in the context of multiple diseases.
• To address the limitations of existing patient similarity measurement techniques.
• To improve the accuracy and applicability of patient similarity analysis in clinical practice.

Main Methods:

• A convolution neural network-based model was proposed, integrating feature learning and similarity learning.
• The model was trained and evaluated using cohort data from the National Health Insurance Sharing Service of Korea.
• Comparative analysis was performed against existing patient similarity measurement models.

Main Results:

• The proposed model demonstrated outstanding performance in measuring similarity among patients with multiple diseases.
• The integrated approach of feature and similarity learning proved effective for complex medical data.
• Experimental results confirmed the model's superiority over existing methods.

Conclusions:

• The developed convolution neural network model offers a robust solution for multi-disease patient similarity.
• This approach enhances the potential for more accurate clinical decision support and personalized medicine.
• The findings highlight the importance of considering multiple diseases in patient similarity research.