AI tool predicts patients at risk of intimate partner violence.

NIH-funded researchers developed machine-learning models that predict patients at risk of intimate partner violence; the multimodal model showed about 88% accuracy and detected risk on average more than three years earlier. The team plans to embed the decision-support tool into electronic health records for real-time risk evaluations.

A team funded by the National Institutes of Health developed an AI clinical decision support tool to identify patients at risk of intimate partner violence using data routinely collected in healthcare settings. The researchers trained and compared three machine-learning models that used structured records and unstructured clinical notes, including radiology reports. The multimodal fusion model performed best in the study, and the authors say the tool is intended to assist clinicians rather than provide definitive diagnoses. The team published methods and guidance on a project website and described plans to integrate the tool into electronic health records. Key findings: - The study evaluated three AI models trained on structured and unstructured hospital data from affected and matched control patients. - The multimodal fusion model achieved about 88% accuracy in the study and detected risk on average more than three years before patients later enrolled at hospital-based domestic abuse intervention centers. - The models are described as decision-support tools to help clinicians have earlier, supportive conversations and are not diagnostic on their own. - The research team plans to develop a real-time decision-support implementation within electronic health records and has shared guidance for clinician communication on the project website. Summary: The reported research describes an AI approach that identifies patterns in existing clinical data to recognize patients at elevated risk of intimate partner violence earlier than current detection paths. The authors describe plans to embed the decision-support tool into electronic health records for real-time evaluations and further clinical implementation work.