Automated MKB-10 Triage and Clinical Documentation in Thyroid Care Study

High-volume specialized endocrine centers face operational bottlenecks in manual ICD-10 (MKB-10) coding, clinical documentation, and diagnostic triage. This study evaluates an automated machine learning decision support architecture deployed across a retrospective dataset of 30,000 patient encounters at the Čigota Institute. Methods: We validated a multi-output classification and natural language processing pipeline trained on structured clinical parameters and diagnostic records. Performance was evaluated using standard metrics, including macro-averaged AUROC, subset accuracy, weighted F1-score, and targeted recall for critical diagnostic codes. Results: The system achieved a Macro AUROC of 0.9597, a Subset Accuracy of 0.7820, and a Weighted F1-score of 0.7948 across all triage categories. For high-risk malignant thyroid pathology (MKB-10 code C73), the model demonstrated a Recall of 0.9805 (98.05%) and a Precision of 0.9542. Conclusion: The validation results confirm that machine-learning-assisted triage and documentation systems can provide highly accurate, scalable decision support in specialized clinical environments while maintaining exceptionally high sensitivity for critical oncological diagnoses.