A method for modeling co-occurrence propensity of clinical codes with application to ICD-10-PCS auto-coding

Area of Science:

• Medical Informatics
• Natural Language Processing
• Computational Linguistics

Background:

• Current medical auto-coding methods often assign billing codes independently, leading to inaccuracies.
• This can result in the assignment of unrelated or improbable co-occurring codes from clinical narratives.

Purpose of the Study:

• To develop and evaluate a novel method for medical auto-coding that incorporates code co-occurrence probabilities.
• To improve the accuracy of automatic generation of medical billing codes from electronic health records.

Main Methods:

• A model was trained to estimate the conditional probability of code co-occurrence.
• An iterative algorithm was developed to adjust confidence scores of existing auto-coders using co-occurrence data.
• The method was integrated with a primary auto-coder for International Statistical Classification of Diseases-10 procedure codes.

Main Results:

• A 12% relative improvement in F-score was achieved for International Statistical Classification of Diseases-10 procedure code auto-coding.
• The method demonstrated enhanced accuracy compared to the baseline auto-coder.

Conclusions:

• The proposed method effectively improves medical auto-coding accuracy by leveraging code co-occurrence.
• This approach shows potential for broad application across various auto-coding systems and code sets.