Information quality measurement of medical encoding support based on usability

Area of Science:

• Medical Informatics
• Health Information Management
• Computational Linguistics

Background:

• Medical encoding systems are crucial for billing, reimbursement, and health policy.
• Current quality assessment relies on information retrieval metrics, which are inadequate for practical medical encoding.
• A need exists for a comprehensive measure to evaluate the quality of automatically generated medical code lists.

Purpose of the Study:

• To define and validate a new information quality measure for medical codes lists.
• To address the limitations of existing measures in assessing the practical value of codes for computer-assisted medical encoding.
• To provide a method for globally comparing the quality of multiple generated codes lists.

Main Methods:

• Developed a novel encoding information quality measure named ADN (Accuracy, Dispersion, Noise).
• Conducted a usability study involving expert coders and physicians in computer-assisted medical encoding.
• Validated the ADN measure through simulations and application to real-world data from 502 inpatient stays.

Main Results:

• The ADN measure provides a single-point representation of codes list quality within a feature space.
• The approach reliably measures and compares the information quality of numerous codes lists, demonstrating practical value.
• ADN measure revealed variability in encoding-support system performance across different clinic departments.

Conclusions:

• The ADN measure offers a robust and adaptable method for assessing medical codes list quality.
• This new measure overcomes limitations of previous metrics, providing a global comparison of code list quality.
• ADN facilitates a comprehensive evaluation of encoding-support systems, aiding in the improvement of medical billing and health policy decisions.