Information fusion for diagnosis coding support.

Area of Science:

• Medical Informatics
• Health Information Management
• Clinical Documentation Improvement

Background:

• Standardized medical coding is mandatory for hospital discharge documents in France.
• Physicians find the diagnostic coding process non-intuitive, leading to frequent errors, inaccuracies, and missing data.
• These coding deficiencies have significant economic consequences.

Purpose of the Study:

• To develop and evaluate an information fusion system designed to enhance medical coding accuracy.
• To compare the effectiveness of different information fusion techniques (voting and possibilistic) against individual data extraction methods.

Main Methods:

• A coding support system was developed, integrating outputs from various Hospital Information System applications.
• Three distinct information processing methodologies were employed to generate partial, heterogeneous data.
• Information fusion techniques, specifically voting and possibilistic approaches, were utilized to synthesize diverse data inputs.
• The system was tested on a database comprising 1,000 patient discharge summaries.

Main Results:

• Information fusion methods demonstrated superior performance in improving medical coding accuracy compared to partial information extraction.
• Both voting and possibilistic fusion approaches showed benefits in synthesizing heterogeneous information for coding support.
• The synthesized information was presented to physicians as an ordered list of suggested diagnostic codes.

Conclusions:

• Information fusion is a valuable approach for overcoming challenges in medical coding.
• The developed coding support system effectively enhances the accuracy and completeness of diagnostic coding.
• Implementing such systems can mitigate economic losses associated with coding errors and improve data quality.