Researchers at Mass General Brigham have developed AI models that can identify patients at risk of intimate partner violence up to four years before they reach a domestic violence treatment center. The most accurate model, which combines medical records, clinical notes, vital signs, and demographics, correctly predicted abuse in 88% of cases. Subtle indicators like chest pain, painkiller use, and increased arm radiology scans were found to correlate with higher likelihood of abuse.
The study, published Friday in NPJ Women's Health, was trained on data from nearly 850 women enrolled at Brigham's domestic abuse intervention center between 2017–2019 and 2021–2022, along with 5,200 control patients. Three model types were tested — one analyzing structured medical data, one evaluating clinical notes, and a combined model that achieved the highest accuracy.
The researchers say the tool could help healthcare providers identify at-risk patients earlier and offer intervention resources before violence escalates. However, they stress that AI predictions alone should never replace clinical judgment and that patient privacy protections must be carefully maintained.
