MGB AI Predicts Domestic Abuse

What happened

Mass General Brigham scientists, led by Dr. Bharti Khurana, developed AI models to predict intimate partner violence (IPV) risk. Models analyse electronic medical records, combining structured data (diagnoses, medications) with unstructured clinical and radiology notes. Trained on nearly 850 women from a domestic abuse intervention centre and about 5,200 control patients, the fusion model achieved 88% accuracy, predicting IPV up to 3.7 years before patients sought care. Researchers plan to integrate this as a decision support tool within electronic medical record systems.

Why it matters

This AI capability offers clinicians a mechanism for proactive screening, identifying patients at risk of intimate partner violence years earlier than traditional disclosure methods. For healthcare providers and public health officials, this shifts the metric from reactive intervention to early detection, potentially preventing escalating harm. However, the models were trained on patients who had already disclosed IPV, which may constrain accuracy for individuals less likely to seek help, and implementation in non-research settings with privacy considerations remains to be addressed. This follows other recent AI advancements in early disease detection, such as WPI AI predicting Alzheimer's.