Integrating predictive coding and a user-centric interface for enhanced auditing and quality in cancer registry data.

Area of Science:

• Medical Informatics
• Artificial Intelligence in Healthcare
• Cancer Registry Data Management

Background:

• Manual data abstraction from electronic health records (EHRs) is labor-intensive for cancer registrars.
• Accurate cancer registry data is crucial for research, quality improvement, and patient care.

Purpose of the Study:

• To develop and evaluate a hybrid natural language processing (NLP) and expert system to streamline cancer registry data abstraction.
• To improve the efficiency and accuracy of identifying lung cancer registry-related concepts and generating codes.

Main Methods:

• Developed a hybrid system combining deep learning and rule-based NLP for concept identification.
• Integrated a symbolic expert system for weighted rule-based registry coding.
• Implemented a patient journey visualization platform within the hospital information system.

Main Results:

• The system achieved high F1-scores (0.85-1.00) across 30 coding items on a lung cancer dataset (1428 patients).
• Registrar feedback confirmed the system's reliability for assisting and auditing data abstraction.
• Demonstrated significant reduction in labor and time for data abstraction.

Conclusions:

• The proposed hybrid neural-symbolic system is effective and efficient for cancer registry coding.
• The system enhances the quality of registrar outcomes and supports clinical decision-making.
• Advancements in AI can optimize cancer registry workflows and contribute to better clinical outcomes.