Clinical and Medical Coding: A New Pathway for Automation-An Updated Review | ||||
| Journal of Medical and Life Science | ||||
| Volume 6, Issue 4, December 2024, Page 633-651 PDF (1.14 MB) | ||||
| Document Type: Original Article | ||||
| DOI: 10.21608/jmals.2024.413890 | ||||
 View on SCiNiTO | ||||
| Authors | ||||
Naif Fahad Almarshadi  1; MOHAMMED MUBARAK AWAD ALHARBI2; Mohammed Saleh Alharbi3; Waked Ahmed Alwaked3; Hatem Mazyad Alsayer3; Saber Maqbel Alhussain3; Oawed Sultain Aeed Almotary4
| ||||
| 1Hospital management specialist, Northern Area Armed Forces Hospital, Saudi Arabia | ||||
| 2HEALTH INFORMATICS, Northern Area Armed Forces Hospital, Saudi Arabia | ||||
| 3Health Information Technician, Northern Area Armed Forces Hospital, Saudi Arabia | ||||
| 4Pharmacy Technician, Northern Area Armed Forces Hospital, Hafar Albaten, Saudi Arabia | ||||
| Abstract | ||||
| Background: Clinical coding is a critical process in healthcare, involving the transformation of free-text medical records into structured codes using classification systems like ICD-10. This process ensures consistent and comparable clinical data, supporting healthcare planning, policy-making, and epidemiological research. Aim: This review aims to explore the evolution of automated clinical coding, evaluate the performance of state-of-the-art deep learning models, and identify key challenges and future directions for improving automated coding systems. Methods: The review synthesizes findings from 113 studies on automated clinical coding, focusing on the transition from rule-based symbolic AI to neural AI, particularly deep learning. It examines the performance of multi-label classification models, the integration of knowledge-based approaches, and the challenges of handling long documents, imbalanced data, and terminology changes. The review also highlights the importance of human-in-the-loop learning and explainability in automated systems. Results: Deep learning models, particularly transformer-based architectures like BERT, have achieved Micro-F1 scores of 58-60% on benchmark datasets like MIMIC-III. However, challenges such as handling infrequent codes, processing long documents, and incorporating symbolic reasoning persist. Hybrid approaches combining symbolic and neural AI show promise, as do knowledge-augmented deep learning methods. Studies also emphasize the need for high-quality datasets, explainability, and adaptability to new coding systems like ICD-11. Conclusion: Automated clinical coding has made significant progress but remains a complex task requiring further research. Future directions include integrating symbolic reasoning, improving explainability, and developing more representative datasets. Collaboration between AI researchers and clinical coding experts is essential to advance the field. | ||||
| Keywords | ||||
| Clinical coding; automated coding; ICD-10; knowledge graphs; explainability; hybrid AI | ||||
|
Statistics Article View: 409 PDF Download: 237 |
||||
