Intimate partner violence (IPV) is abuse within relationships, when a current or former spouse or partner uses violence or aggression. It doesn’t matter where someone lives or how much money they make; IPV cuts across all boundaries. Because it often stays hidden, intimate partner violence has become one of the biggest health problems of our time, quietly affecting millions of people across the world.
IPV doesn’t just cause immediate harm; it leaves lasting scars on health. Survivors often carry invisible wounds, chronic pain, infections, menopausal problems, and mental health struggles like depression, anxiety, PTSD, or even self‑harm.
Identifying IPV early is vital to stop these problems from getting worse. But many victims remain silent, held back by fears for their safety or financial dependence on their partner, which makes it harder for them to get timely help.
Imagine if doctors could recognize the early signs of intimate partner violence years before a patient felt ready to speak up. That’s the goal of new artificial intelligence tools developed by researchers at Mass General Brigham. This technology is designed to analyze medical records and identify risks long before they become apparent.
In a study published in npj Women’s Health, researchers at Mass General Brigham showed that machine learning models trained on electronic medical records (EMRs) could detect IPV risk up to four years before individuals sought care at a domestic violence treatment center. They developed the new AI tools using a dataset of female patients who sought help at a domestic abuse intervention and prevention center of a major hospital in the United States.
The breakthrough suggests a future where clinicians might be able to start sensitive conversations earlier, potentially preventing years of hidden suffering.
Dr. Bharti Khurana, senior author and founding director of the Trauma Imaging Research and Innovation Center, explained: “Our research offers proof of concept that AI can support clinicians in flagging possible abuse earlier. Earlier identification of intimate partner violence and future risk may enable clinicians to intervene sooner and help prevent significant mental and physical health consequences.”
A past study found that patients are more likely to share their experiences when asked privately by a trusted provider.
In collaboration with MIT’s Dimitris Bertsimas, researchers trained three different AI models using medical records. They studied data from 673 women who had visited a domestic violence prevention center and compared it with more than 4,000 similar patients who had not reported abuse.
Domestic violence is widely accepted in most developing countries, study
One model examined structured information, including diagnoses, medications, and neighborhood data. Another analyzed doctors’ notes, radiology reports, and emergency visits. The third, called Holistic AI in Medicine (HAIM), combined both approaches, giving the most complete picture.
When researchers tested the AI on new patients, the results were striking. All three models show strong performance with new data, but the HAIM fusion model was the most impressive. It correctly identified risk in 88% of cases.
Even more notably, when the system examined older, time-stamped medical records, it could identify warning signs years in advance. On average, the fusion model predicted IPV more than 3.7 years before patients sought help.
The AI found clues beyond injuries. People with frequent ER visits, chronic pain, or mental health issues were more likely to face abuse. Those who kept up with preventive care, like mammograms or vaccines, were less likely.
But the researchers note the models were trained on patients who had already reported abuse, so results may not be as accurate for those who never seek help. To make predictions even stronger, researchers say larger and more diverse datasets will be needed.
These AI models could change how doctors identify and support patients at risk of intimate partner violence. With timely, trauma‑informed care, doctors may be able to prevent abuse from escalating into serious injuries, long‑term health problems, or even deaths linked to IPV.
Journal Reference:
- Gu, J., Carballo, K.V., Ma, Y. et al. Leveraging multimodal machine learning for accurate risk identification of intimate partner violence. npj Women’s Health 4, 15 (2026). DOI: 10.1038/s44294-025-00126-3
