Real-time automated billing for tobacco treatment: developing and validating a scalable machine learning approach

Area of Science:

• Medical Informatics
• Artificial Intelligence in Healthcare

Background:

• Manual medical billing for tobacco cessation counseling is administratively burdensome.
• Inaccuracies in billing can hinder the provision of essential cessation services.

Purpose of the Study:

• To develop CigStopper, a prototype for automated identification of eligible tobacco cessation care codes.
• To reduce administrative workload and enhance billing accuracy in healthcare.

Main Methods:

• Generated synthetic clinical notes using ChatGPT prompt engineering for CPT codes 99406/99407.
• Trained machine learning models, including decision trees and random forests, on clinician-annotated data.
• Evaluated model performance using PRC AUC and F1 scores, with generalizability testing on deidentified notes.

Main Results:

• Tree-based models, specifically decision trees and random forests, demonstrated superior performance.
• The models achieved a mean PRC AUC of 0.857 and an F1 score of 0.835.
• Generalizability testing confirmed the effectiveness of tree-based models on real-world data.

Conclusions:

• CigStopper shows potential for optimizing billing processes and supporting tobacco cessation care.
• Machine learning approaches provide a foundation for automating billing tasks, improving efficiency.
• Automating these tasks can simplify healthcare practice operations and enhance financial sustainability.