Temporal topic model for clinical pathway mining from electronic medical records

Area of Science:

• Medical Informatics
• Artificial Intelligence in Healthcare
• Data Mining

Background:

• Discovering clinical pathways (CPs) from electronic medical records (EMRs) is crucial for intelligent healthcare.
• Existing topic models for CPs lack temporal considerations, limiting practical medical applications.
• Current methods often yield similar treatment patterns, failing to capture patient journey nuances.

Purpose of the Study:

• To propose a novel method for mining clinical pathways from EMRs.
• To address the limitations of existing topic models by incorporating temporal dynamics.
• To enhance the development of intelligent healthcare services through improved CPs discovery.

Main Methods:

• Developed an extended temporal topic model (TTM) based on traditional topic models.
• TTM is designed to derive CPs from EMR data, capturing treatment topics and timestamps.
• The model assumes CPs can be extracted from the sequential patient treatment data recorded in EMRs.

Main Results:

• Experimental validation on a real-world patient hospitalization dataset confirmed the method's applicability.
• The proposed TTM demonstrated significant improvements and robustness compared to existing benchmarks.
• The results highlight the model's effectiveness in mining clinically relevant pathways.

Conclusions:

• The TTM offers a more competitive approach for mining potential CPs by considering temporal EMR data features.
• This method provides a prospective tool to support clinical diagnostic decisions.
• The temporal aspect enhances the utility of discovered pathways for practical healthcare.