Machine learning for classification of postoperative patient status using standardized medical data

Area of Science:

• Medical Informatics
• Health Services Research
• Machine Learning in Healthcare

Background:

• Real-world evidence (RWE) requires standardized data from diverse medical institutions.
• Integrating electronic medical records (EMRs) is crucial for effective RWE analysis.
• Existing EMRs present challenges in data quality and integration for RWE generation.

Purpose of the Study:

• To propose an analysis method for prioritizing clinical interventions using real-world data.
• To develop a system for clustering and visualizing time-series patient data.
• To enhance learning health systems by prioritizing patient outcomes and status during hospitalization.

Main Methods:

• Examined common data items for reimbursement (Diagnosis Procedure Combination [DPC]) and clinical pathway data.
• Utilized a cloud platform to store and analyze long-term hospitalization data from multiple institutions.
• Applied machine learning for classification, visualization, and interpretation of patient data.

Main Results:

• The ePath platform successfully integrated standardized data from multiple institutions.
• Clustering and directed graphs visualized DPC item distribution and temporal trends.
• Constipation identified as a key risk factor for long hospitalization; drainage management as a mitigating factor.

Conclusions:

• Generated high-quality medical process evidence from multi-institutional patient data, overcoming EMR limitations.
• The learning health system will provide feedback to institutions for continuous improvement.
• The system enables re-evaluation and analysis of interventions based on real-world evidence.