Artificial Intelligence to Predict Billing Code Levels of Emergency Department Encounters

Area of Science:

• Medical Informatics
• Artificial Intelligence in Healthcare
• Clinical Documentation

Background:

• Accurate medical billing is crucial for healthcare reimbursement.
• Emergency Department (ED) coding is complex and time-consuming.
• AI offers potential solutions for automating administrative tasks in healthcare.

Purpose of the Study:

• To develop and evaluate an AI model for predicting ED billing code levels.
• To assess the model's performance using clinical notes, characteristics, and orders.
• To identify key features influencing billing code prediction.

Main Methods:

• Utilized an ensemble model combining natural language processing and machine learning.
• Trained the model on 321,893 adult ED encounters from January to September 2023.
• Employed explainable AI (Shapley Additive Explanations) to identify important predictive features.

Main Results:

• The AI model achieved high performance in predicting billing code levels 4 and 5 (AUC 0.94-0.95, accuracy 0.80-0.92).
• Key predictors included critical care notes, number of orders, and discharge disposition.
• At a 95% threshold, level 5 prediction showed 0.99 precision and 0.57 recall.

Conclusions:

• AI models can accurately predict ED billing code levels from clinical data.
• This technology has the potential to automate ED coding processes.
• Automation can lead to substantial savings in administrative costs and time.