Heterogeneous Information Network-Based Patient Similarity Search.

Area of Science:

• Medical Informatics
• Artificial Intelligence in Medicine
• Data Science

Background:

• Patient similarity search is crucial for AI-assisted medical services, aiding diagnosis and personalized treatment.
• Current methods using Electronic Health Record (EHR) data often result in high-dimensional, sparse vectors, hindering accurate similarity calculation.
• Existing approaches frequently neglect the temporal dynamics within EHR data, impacting the reliability of patient similarity search.

Purpose of the Study:

• To develop an improved patient similarity search method addressing limitations of existing techniques.
• To enhance the accuracy and relevance of patient similarity measurements by incorporating diverse medical information and temporal factors.

Main Methods:

• A heterogeneous information network approach is proposed, connecting patients, diseases, and drugs.
• Patient similarity is calculated by measuring node similarity within this network, overcoming high-dimensional and sparse vector issues.
• Temporal information from EHRs is encoded into an annotated heterogeneous information network to capture time-dependent relationships.

Main Results:

• The proposed method effectively represents mixed information from patients, diseases, and drugs.
• Incorporating temporal information significantly improves patient similarity search accuracy.
• Experimental results demonstrate superior performance compared to existing baseline methods.

Conclusions:

• The heterogeneous information network method offers a robust solution for patient similarity search.
• Integrating temporal data is vital for accurate patient similarity assessment in medical informatics.
• This approach holds promise for advancing AI-assisted medical diagnosis and treatment planning.