Harnessing the Power of Machine Learning and Electronic Health Records to Support Child Abuse and Neglect Identification in Emergency Department Settings

Abstract

Emergency departments (EDs) are pivotal in detecting child abuse and neglect, but this task is often complex. Our study developed a machine learning model using structured and unstructured electronic health record (EHR) data to predict when children in EDs might need intervention from child protective services. We used a case-control study design, analyzing data from a pediatric ED. Clinical notes were processed with natural language processing (NLP) techniques to identify suspected cases and matched in a 1:9 ratio to ensure dataset balance. The features from these notes were combined with structured EHR data to construct a model using the XGBoost algorithm. The model achieved a precision of 0.95, recall of 0.88, and F1-score of 0.92, with improvements seen from integrating NLP-derived data. Key indicators for abuse included hospital admissions, extended ED stays, and specific clinical orders. The model's accuracy and the utility of NLP suggest the potential for EDs to better identify at-risk children. Future work should validate the model further and explore additional features while considering ethical implications to aid healthcare providers in safeguarding children.