Dr. Chiou-Hwa Chuang's team at National Taiwan University Hospital (NTUH) has successfully developed an intelligent disease classification coding support system. This system utilizes natural language processing (NLP) technology to analyze the content of medical records, offering suggested International Classification of Diseases, 10th Edition (ICD-10) diagnostic and procedure codes. By assisting physicians and coders in classifying diseases, the system automates the complex coding process and significantly enhances coding accuracy through the integration of deep learning models and expert knowledge.

Current Complex and Time-Consuming Disease Classification Process Relies on Professional Expertise

The current ICD-10 system includes approximately 69,000 diagnostic codes and 72,000 procedure codes, with a complex structure and coding rules. This process heavily relies on experienced professional coders, who must spend significant time reading medical records to accurately identify the correct ICD-10 codes.

Integrating Expert Knowledge with AI Models for Accurate Disease Classification Predictions

The intelligent coding system developed by Dr. Chuang’s team employs NLP technology, which integrates both structured and unstructured medical record information. It incorporates external knowledge from disease classification experts and utilizes a deep learning model to automatically analyze medical records. The system then provides suggested ICD-10 diagnostic and procedure codes, assisting physicians and coders in the coding process.

The system has five key innovative features:

1. Systematic Expert Knowledge Feedback Process: Periodically updates the model with expert feedback.
2. Automated Principal Diagnosis Selection and Inpatient Coding: Enhances efficiency and reduces human intervention.
3. DRG (Diagnostic Related Groups) Placement Assistance: Provides hints for selecting the best DRG, along with cost reminders for healthcare providers.
4. Incorporating Continuous Patient Information: Integrates parameters like diagnostic and procedure codes to improve prediction accuracy.
5. Performance Analysis for Different Disease Chapters:  Continuously optimizes the model based on performance in various disease categories.

The system’s diagnostic code prediction achieves an F-score of 80.33%, while the procedure code prediction reaches an F-score of 75.09%. It has garnered a high satisfaction rate of 91% from users and has already been implemented across NTUH's network of hospitals.

Enhancing Coding Efficiency and Optimizing Healthcare Quality

Dr. Chuang highlighted that this patented system not only significantly improves coding efficiency and quality but also assists hospitals in obtaining appropriate health insurance reimbursements and enhancing the quality of patient care. Looking ahead, the team plans to continue refining the system, aiming to develop it into a commercial application that further contributes to the advancement of smart healthcare.

Resource (mandarin): 無須耗時猛讀！智慧編碼系統自動分析病歷歸類疾病！