Predicting Diagnosis Code from Medication List of an Electronic Medical Record Using Convolutional Neural Network

Area of Science:

• Medical Informatics
• Artificial Intelligence in Healthcare
• Clinical Data Analysis

Background:

• Automated coding systems are crucial for enhancing healthcare quality of care.
• Accurate diagnosis coding from electronic medical records (EMR) is essential for clinical documentation and billing.
• Predicting diagnosis codes from medication data presents a novel approach to automated coding.

Purpose of the Study:

• To develop and evaluate a convolutional neural network (CNN) model for predicting International Statistical Classification of Diseases 10 (ICD 10) codes.
• To utilize medication lists from EMR as input for diagnosis code prediction.
• To assess the performance of the CNN model across different physicians and departments.

Main Methods:

• Collected clinical notes from an outpatient department (OPD) at Wanfang Hospital, Taiwan (2016).
• Utilized medication lists as input and ICD 10 codes as output for the CNN model.
• Employed word2vector CNN architecture after data preprocessing, with a 90% training and 10% testing data split.

Main Results:

• The CNN model demonstrated improved performance across all three physicians from different departments.
• The best performance was achieved by a model associated with a cardiology physician, yielding a precision of 69%, recall of 89%, and F-score of 78%.
• These results indicate the feasibility of predicting diagnosis codes from medication data.

Conclusions:

• Convolutional neural networks can effectively predict ICD 10 codes from EMR medication lists.
• The model shows promising results, particularly in specific clinical departments like cardiology.
• Future enhancements should incorporate additional data sources such as clinical text and lab reports for improved evaluation.