Mass General Brigham researchers have developed a set of artificial intelligence (AI) tools that use machine learning to use information from electronic medical records (EMRs) to identify individuals who may be at risk for intimate partner violence (IPV). In a study published in npj women’s health, Researchers reported that the tool was able to detect IPV up to four years before an individual sought care at a domestic violence treatment center. This finding highlights proactive screening and its potential to support health care providers in initiating early conversations with patients about IPV.
Our study provides proof of concept that AI can support clinicians to detect potential abuse early. Early identification of intimate partner violence and future risk may allow clinicians to intervene sooner and prevent serious consequences for mental and physical health. ”
Bharti Khurana, MD, MBA, principal investigator, correspondent and senior author, founding director of the Center for Trauma Imaging Research and Innovation, emergency radiologist in the Department of Radiology, Brigham General Hospital, Massachusetts
More than one-third of women and one in 10 men will experience IPV in their lifetime. However, despite its high prevalence, people rarely disclose IPV to health care providers for reasons such as fear, stigma, and economic and psychosocial dependence on the abuser. Previous research has shown that people experiencing IPV are more likely to disclose abuse if asked privately in a trauma-informed manner by a trusted health care provider.
To facilitate early detection and intervention by health care providers, Khurana’s research team, in collaboration with collaborators at the Massachusetts Institute of Technology (MIT) led by Dr. Dimitris Bartsimas, trained three machine learning models using EMR data from 673 women who visited domestic violence intervention and prevention centers at academic health centers in the United States between 2017 and 2022, as well as 4,169 demographically matched controls who did not report. IPV.
The three AI models tested included a tabular model that uses structured EMR data, including postcode-based diagnoses, medications, and social deprivation index. A note model using unstructured clinical notes, radiology and emergency department reports. The other is a fusion model that combines both data types, called Holistic AI in Medicine (HAIM).
When tested on 168 patients and 1,043 controls who visited an IPV intervention and prevention center during the same time period, all three models showed high accuracy, with the fusion model achieving the highest (88%). When tested using time-stamped archived medical records, the fusion model was able to predict 80.5% of cases on average more than 3.7 years in advance before patients sought treatment.
The model was then validated using data from two additional patient groups and controls that were not included in the training or test data and yielded similarly high accuracy.
Previous research led by Khurana found that women who underwent frequent imaging tests in the emergency department and sustained certain types of injuries were more likely to later report IPV. This new AI study identified additional risk factors for IPV. They found that people with mental health disorders, chronic pain, and frequent emergency department visits were more likely to experience IPV, whereas patients who regularly received preventive services such as mammograms and immunizations were at lower risk.
The authors note that the AI tool was developed and validated with patients who sought care or disclosed IPV, which may limit its accuracy in predicting IPV in individuals who are less likely to seek care or disclose IPV to health care providers. Additionally, the control group in the training data may include false negatives or patients who have experienced IPV but did not report it, potentially reducing the accuracy of the model. In the future, accuracy will improve by training on larger, more diverse patient datasets over longer periods of time, Khurana said.
sauce:
Reference magazines:
Goo, Jay. Others. (2026) Leverage multimodal machine learning to accurately identify risk of intimate partner violence. npj women’s health. DOI: 10.1038/s44294-025-00126-3. https://www.nature.com/articles/s44294-025-00126-3
