The main purpose of the International Classification of Diseases is to allow different countries or regions to classify the data on disease diagnosis or related health problems collected by them through a unified standard, so as to conduct comparative analysis of health data and provide reference for decision-making by health authorities in various countries.

In addition, accurate disease classification can assist in DRGs (Diagnosis Related Groups hospital payment system) benefit application, and then affect the hospital CMI (Case Mix Index, case mix index) value. However, compared with the previous generation of disease classification codes (ICD-9 CM), the number of diagnostic codes of the 10th generation is nearly 70,000, which is 5 times as the previous generation, and the composition rules of the codes are also greatly different, whereas the complexity is greatly improved. Since the coding operation is quite complicated, most of the clinical procedures are as follows: when a patient sees a doctor, first select a diagnostic code with a similar disease classification, and then the disease classifier will check again to confirm the final code.

In view of the time-consuming, labor-intensive, error-prone problems of manual coding, this case uses natural language processing (NLP) technology for model training by unstructured electronic medical record data, including discharge summary, hospitalization records, and progress notes, for developing ICD Auto-Coding System.

Current system performance: F1 score > 80%, expecting to improve the efficiency, quality, integrity of ICD-10 coding, save encoding time, optimize DRG weight calculation, and upgrade the overall hospital CMI (Case Mix Index). Currently, it has been introducing into the medical institutions as follows: CMUH, CMUBH, TMANH, CMUHCH, WGMH, AUH, CMUHTB.

（Figure description）User interface of ICD Auto-Coding System