Scientists have developed an AI tool that detects intimate partner violence (IPV) risk in patients years before they seek help. Trained on hospital records from nearly 850 women with IPV and over 5,200 controls, the system analyzed medical histories and clinical notes.
The combined model-using both structured data and physician notes-correctly identified abuse risk in 88% of cases. It flagged potential abuse more than three years ahead of formal intervention programs.
The tool does not diagnose or force disclosures. Instead, it alerts clinicians to subtle patterns of trauma, enabling earlier, compassionate screening.
Researchers at the National Institutes of Health and Mass General Brigham plan to integrate the system into electronic health records for real-time use during routine care.