Deep-ADCA: Development and Validation of Deep Learning Model for Automated Diagnosis Code Assignment Using Clinical...

Area of Science:

• * Medical Informatics
• * Artificial Intelligence
• * Natural Language Processing

Background:

• * International Classification of Diseases (ICD) codes are crucial for clinical, financial, and administrative functions.
• * Inaccurate ICD coding negatively impacts care quality and reimbursement.
• * Manual ICD code selection is time-consuming and requires specialized knowledge.

Purpose of the Study:

• * To develop and evaluate a deep learning-based natural language processing (NLP) model for predicting ICD-10 codes.
• * To assess the model's potential to aid healthcare providers in clinical decision-making and service improvement.

Main Methods:

• * Retrospective collection of clinical notes from five outpatient departments (January 2016 - December 2016).
• * Application of NLP techniques (GloVe, Word2Vec, embeddings) for data processing.
• * Development of a Convolutional Neural Network (CNN) model trained on 90% of the data and tested on 10%.

Main Results:

• * The CNN model demonstrated clinically satisfactory performance across five departments (Precision: 0.50–0.69, Recall: 0.78–0.91).
• * The cardiology department showed the highest performance: Precision 69%, Recall 89%, F-score 78%.
• * A total of 21,953 medical records from 5016 patients were analyzed.

Conclusions:

• * The CNN model effectively predicts ICD-10 codes, offering a significant advancement in healthcare data management.
• * Implementation in clinical settings can reduce manual coding burdens and enhance efficiency.
• * This predictive model supports physicians in making informed clinical decisions, ultimately improving patient care quality.