The main purpose of the International Classification of Diseases system is to enable different countries or regions to classify their collected disease diagnosis or related health problem data through unified standards, facilitating comparative analysis of health data and providing a reference for decision-making by health authorities in various countries.

Furthermore, accurate disease classification helps with DRGs (Diagnosis-Related Groups) payment applications, thereby affecting a hospital's CMI (Case Mix Index). However, compared to the previous generation disease classification code (ICD-9-CM), the 10th generation diagnosis codes number nearly 70,000, five times more than the previous generation, and the coding composition rules also differ greatly, significantly increasing complexity. Due to the complicated nature of coding work, the typical clinical process involves physicians selecting diagnosis codes similar to the disease classification during consultation, which are then re-verified by disease classification specialists to confirm the final coding.

Given the time-consuming, labor-intensive, and error-prone issues with manual coding, our hospital's Artificial Intelligence Center used AI natural language processing technology to train models on unstructured electronic medical record data (including discharge summaries, hospitalization records, and progress notes) to develop the ICD-10 Intelligent Automatic Coding System.

The current system performance is: AUC (Area Under ROC Curve) 82%, Sensitivity 78%, F1 score 82%, aiming to improve the efficiency, quality, and completeness of ICD-10 coding while saving coding time, optimizing DRGs weight calculation, and enhancing the hospital's overall CMI value.

Clinical evaluations have demonstrated that the AI-ICD10 assisted coding system significantly enhances efficiency. Compared to traditional coding tools, the system reduces the average processing time per case by 10.8 minutes across most medical institutions, resulting in a 35% improvement in overall coding efficiency. This advancement not only streamlines administrative workflows but also alleviates the workload of medical coders. System assessments indicate that AI-ICD10 improves coding accuracy by approximately 4%, leading to substantial cumulative benefits in ensuring the reliability and consistency of diagnostic data. For instance, in the analysis of discharge medical records, professional coders verified that AI-ICD10 assisted in identifying 34 cases where the primary diagnosis could be reclassified into higher-weight Diagnosis-Related Groups (DRGs). This reclassification increased the total weight by 11.0233 and is estimated to raise the severity index by 0.32, thereby enabling more precise allocation of medical resources that better reflect patients' actual conditions.

Awards

2022

• National Healthcare Quality Award (NHQA) for Smart Solutions
• ​Symbol of National Quality (SNQ) in the Smart Healthcare category for medical institutions in Taiwan

Contact window

Project Management Group +886-4-22052121 ext. 12534