Examining the Use of Machine Learning Algorithms to Enhance the Pediatric Triaging Approach

Abstract

Hussain J Aljubran,1 Maitham J Aljubran,2 Ahmed M AlAwami,1 Mohammad J Aljubran,3 Mohammed A Alkhalifah,4 Moayd M Alkhalifah,5 Ahmed S Alkhalifah,6 Tawfik S Alabdullah7 1College of Medicine, Imam Abdulrahman Bin Faisal University, Dammam, Saudi Arabia; 2Pediatric Department, King Faisal Specialist Hospital & Research Centre, Riyadh, Saudi Arabia; 3Department of Energy Science & Engineering, Stanford University, Stanford, CA, USA; 4Emergency Medicine Department, Johns Hopkins Aramco Healthcare, Al-Hasa, Saudi Arabia; 5Neurology Unit, Neurosciences Institute, Johns Hopkins Aramco Healthcare, Dhahran, Saudi Arabia; 6Pediatric Department, Qatif Central Hospital, Al Qatif, Saudi Arabia; 7Pediatric Emergency Medicine, King Faisal Specialist Hospital & Research Centre, Riyadh, Saudi ArabiaCorrespondence: Hussain J Aljubran, College of Medicine, Imam Abdulrahman Bin Faisal University, Dammam, 34212, Saudi Arabia, Tel [phone], Email [email protected] Maitham J Aljubran, Pediatric Department, King Faisal Specialist Hospital & Research Centre, Riyadh, 12212, Saudi Arabia, Tel [phone], Email [email protected]: Triage systems play a vital role in effectively prioritizing patients according to the seriousness of their condition. However, conventional emergency triage systems in pediatric care predominantly rely on subjective evaluations. Machine learning technologies have shown significant potential in various medical fields, including pediatric emergency medicine. Therefore, this study seeks to employ pediatric emergency department records to train machine learning algorithms and evaluate their effectiveness and outcomes in the triaging system. This model will improve accuracy in pediatric emergency triage by categorizing cases into three urgency levels (nonurgent, urgent, and emergency).Patients and Methods: This is a retrospective observational cohort study that used emergency patient records obtained from the Emergency Department at King Faisal Specialist Hospital & Research Centre. Using the emergency severity index (a scale of 1 to 5), various machine learning techniques were employed to build different machine learning models, such as regression, instance-based, regularization, tree-based, Bayesian, dimensionality reduction, and ensemble algorithms. The accuracy of these models was compared to reach the most accurate and precise model.Results: A total of 38,891 pediatric emergency patient records were collected. However, due to numerous outliers and incorrectly labeled data, clinical knowledge and a confident learning algorithm were employed to preprocess the dataset, leaving 18,237 patient records. Notably, ensemble algorithms surpassed other models in all evaluation metrics, with CatBoost achieving an F-1 score of 90%. Importantly, the model never misclassified an urgent patient as nonurgent or vice versa.Conclusion: The study successfully created a machine learning model to classify pediatric emergency department patients into three urgency levels. The model, tailored to the specific needs of pediatric patients, shows promise in improving triage accuracy and patient care in pediatric emergency departments. The implication of this model in the real-life sitting will increase the accuracy of the pediatric emergency triage and will reduce the possibilities of over or under triaging.Keywords: machine learning, artificial intelligence, emergency medicine, pediatric, triage

Keywords

• machine learning
• artificial intelligence
• emergency medicine
• pediatric
• triage