IPV is abusive behavior within relationships when an ex-spouse or current partner abuses violence and aggression. IPV is not confined to one’s location or income. Intimate partner violence is one of today’s biggest public health issues, primarily because it remains hidden. It affects millions of people around the globe.
IPV is not only harmful to the body, but it also leaves scars that last a lifetime. Many survivors carry invisible scars such as chronic pain, infection, menopausal issues, mental health problems like anxiety, depression or PTSD.
It is important to identify IPV as early as possible in order to prevent these issues from worsening. Many victims are afraid to speak up for fear of their own safety, or because they depend on their partners financially.
This makes it difficult for them get help.
Imagine that doctors would be able to recognize early warning signs of violence against intimate partners years before the patient was ready to raise their voice. Researchers at Mass General Brigham have developed new artificial intelligence software tools to achieve this goal. The technology analyzes medical records to identify potential risks before symptoms appear.
Researchers at Mass General Brigham published a study in the npj Women’s Health journal that showed machine-learning models trained using electronic medical records could detect IPV risks up to four year before an individual sought treatment at a domestic abuse treatment center. The researchers developed new AI tools by using data from female patients seeking help at the domestic abuse prevention and intervention center of a large hospital in the United States.
This breakthrough could lead to a world where clinicians can start conversations about sensitive topics earlier and prevent years of suffering.
Dr. Bharti Khurana, founding director and senior author of the Trauma Image Research and Innovation Centre, explains: Our research provides proof-of-concept that AI could support clinicians to flag possible abuse sooner. Early identification of intimate partners violence and future risks may allow clinicians intervene earlier and prevent serious mental and physical consequences.
In a previous study, it was found that when asked by trusted providers in private about their experience with a particular product or service, patients were more willing to talk.
Researchers trained three AI models using records from medical files in collaboration with Dimitris Bertsimas of MIT. The researchers compared data collected from 673 women visiting a domestic abuse prevention center with that of more than 4,500 similar patients.
Study: Domestic violence in developing countries is widespread
A model analyzed structured data, such as diagnoses, medication, and neighbourhood information.
One model analyzed radiology reports and emergency visits. Holistic AI In Medicine (HAIM) is a third approach that combines both methods to give the best picture.
Researchers tested AI with new patients and the results were impressive. The HAIM fusion was most impressive. All three models performed well with the new data. The model correctly predicted risk in 88% cases.
It is also noteworthy that the system could detect warning signs in older medical records with time stamps, even years before they were noticed. The fusion model was able to predict IPV on average more than 3.7 year before the patient sought assistance.
AI uncovered clues that went beyond the obvious. Abuse was more common among people with mental illness, frequent ER visits or chronic pain.
People who received preventive health care like mammograms and vaccines were less likely to be abused.
Researchers note that the models are based on data from patients who have already sought help. Results may be less accurate for people who do not seek assistance. Researchers say that to make even more accurate predictions, they will need to use larger datasets and a wider variety of data.
The AI models may change the way doctors support and identify patients who are at risk for intimate partner violence. Doctors may prevent IPV-related deaths, serious injuries and long-term health issues by providing timely trauma-informed treatment.
Journal Reference
- Gu, J., Carballo, K.V., Ma, Y. et al. Leveraging multimodal learning to accurately identify the risk of violence against intimate partners. Women’s Health, 4, 15 (2026).
DOI: 10.1038/s44294-025-00126-3
