EHR2CCAS: A framework for mapping EHR to disease knowledge presenting causal chain of disorders - chronic kidney...

Area of Science:

• Medical Informatics
• Computational Biology
• Health Data Science

Background:

• Understanding disease progression requires detailed patient state tracking.
• Electronic Health Records (EHR) contain rich patient data but are complex to analyze.
• Existing methods struggle to capture the nuanced, temporal nature of disease evolution.

Purpose of the Study:

• To develop a framework (EHR2CCAS) for mapping electronic health record (EHR) data to a novel disease representation called Causal Chains of Abnormal States (CCAS).
• To enable fine-grained analysis of disease progression by identifying transitions between medical conditions.
• To improve disease phenotyping and patient management through comprehensive EHR data utilization.

Main Methods:

• Developed the EHR2CCAS framework integrating rule-based and machine learning approaches.
• Processed heterogeneous EHR data, including structured and unstructured clinical texts.
• Employed data-driven methods for text analysis and imputation of abnormal states based on temporal EHR properties and CCAS structure.
• Constructed and evaluated a CCAS mapping system for chronic kidney disease using EHR data from the University of Tokyo Hospital.

Main Results:

• The EHR2CCAS system demonstrated high prediction performance in identifying abnormal patient states.
• Strong agreement was observed among annotators, validating the system's accuracy.
• Improved prediction performance was achieved through effective handling of narrative text variations and state imputation.
• The system successfully presented temporal patient data on abnormal states within the CCAS framework.

Conclusions:

• EHR2CCAS represents a significant advancement in integrating disease knowledge with EHR data for abnormality extraction.
• The framework facilitates disease progression management and deep phenotyping by detailing abnormal states and their transitions.
• Further analysis of state transitions can contribute to the identification of distinct disease subtypes.