SP 2.05 Standardization of Emergency Department Triage for Rib Fractures Using a Machine Learning–Based Clinical Decision Support System Integrating Physiological Parameters and Natural Language Processing
Yasser Mohamed, Fatima Zulfiqar, Raghav Anirudh, Antonios Anthanasiou, Alex Bull, Muhammad S SajidAbstract
Background
Triage decisions for patients with rib fractures are often inconsistent, particularly regarding the need for hospital admission. This study evaluated a machine learning–based clinical decision support system designed to standardize triage decisions by integrating physiological variables with unstructured clinical and radiological data.
Methods
A retrospective cohort study was conducted including adult patients presenting with rib fractures to a tertiary hospital emergency department. Structured clinical features incorporated into the model included Glasgow Coma Scale score, temperature, and pain severity. Natural Language Processing (NLP) techniques were applied to free-text triage documentation and radiology reports to extract clinically relevant red flags and complications. Feature engineering and rule-based safety constraints were guided by established clinical guidelines and published literature. A supervised logistic regression classifier was trained to categorize patients as requiring hospital admission or suitable for safe discharge.
Results
The model was evaluated on 27 patients and demonstrated high sensitivity for identifying those requiring admission. Twenty patients (74%) were classified as needing hospital admission, while seven (26%) were identified as suitable for outpatient management. Rule-based safeguards correctly flagged all patients with severe pain, reduced Glasgow Coma Scale scores, or pyrexia >38.0 °C. Incorporating NLP-derived features enhanced model performance by capturing clinically significant findings within radiology reports not reflected in structured data fields.
Conclusion
Machine learning–based clinical decision support systems that integrate physiological parameters with NLP of clinical documentation can improve consistency and safety of rib fracture triage. These tools may reduce unnecessary admissions and support ambulatory management for optimum resource utilization.