Complaint Driven Clinical Guidance and ICD Chapter Suggestion: An AI-Based Text Classification Tool.

Authors

Anday Duru, Eftal Sehirli, Muhammed Kamil Turan, Emrullah Demiral, Mehmet Tankul, Mehmet Coskun

Published in

Journal of evaluation in clinical practice. Volume 32. Issue 8.

Abstract

Accurate routing of patients to appropriate clinical departments is essential for timely diagnosis, effective treatment, and efficient use of hospital resources. However, hospital statistics and feedback from patients and healthcare staff indicate that many patients select clinics that are not appropriate for their presenting complaints. This mismatch unnecessarily occupies appointment slots, limits access to appropriate care, and increases the operational burden on healthcare systems. To address this problem, a complaint-driven clinical guidance framework was developed to map patient complaints to suitable clinical departments and corresponding International Classification of Diseases (ICD) chapters.
A 7-year electronic health record dataset comprising 1,795,645 patient complaints was used to train and evaluate conventional machine learning algorithms and BERT-based language models. The final classification tasks included 15 clinical departments for department recommendation and 12 ICD chapters for suggestion.
For clinical department recommendation, Turkish BERT achieved the best performance, with a mean accuracy of 92.02%, precision of 92.91%, and F1-score of 92.08%. McNemar's test confirmed that its improvement over multilingual BERT-based alternatives was statistically significant. Error analysis indicated that most department-level misclassifications occurred between clinically related or symptomatically overlapping departments, including Internal Medicine-Cardiology, Neurosurgery and Physical Therapy, and Cardiovascular Surgery-Cardiology. For ICD chapter suggestion, Turkish BERT achieved the best performance among the evaluated models, with a top-1 accuracy of 66.7%, top-3 accuracy of 82.9%, top-5 accuracy of 85.4%.
Although exact ICD chapter prediction remained challenging, the substantially higher top-3 and top-5 accuracies indicate that the model may be useful for generating ranked candidate ICD chapters. Confusion analysis showed that errors were mainly concentrated between clinically overlapping ICD chapters, particularly neoplasms/blood and immune disorders, endocrine/metabolic diseases, circulatory system diseases, and symptom-based chapters. These findings suggest that artificial intelligence-based systems can effectively support clinical department recommendation to improve hospital workflow efficiency, whereas ICD chapter suggestion may serve as a preliminary ranked decision-support tool.

PMID:
42976144
Bibliographic data and abstract were imported from PubMed on 04 Nov 2026.

Read full publication at:
Please sign in to see all details.

Stats

1-terrible, 9-excellent. How would you rate this publication? Sign in in to submit your rating.

Post a comment

You need to be signed in to post comments. You can sign in here.

Comments

There are no comments yet.