April 28, 2027
At a Glance
- A groundbreaking artificial intelligence tool has been developed to predict which patients may face intimate partner violence, potentially years before they seek assistance.
- This tool could enable healthcare providers to identify at-risk patients and implement early interventions.
The tool could assist clinicians in promptly connecting patients with necessary resources and support.
Fizkes / Shutterstock
Intimate partner violence (IPV) affects millions of individuals annually in the United States. This form of abuse, stemming from current or former spouses and partners, can result in severe injuries, chronic pain, and lasting mental health issues. Many victims choose not to disclose their experiences to healthcare providers due to fears for their safety, stigma, and anxiety.
Existing screening tools often miss a significant number of IPV cases, as they typically depend on patients to self-report their situations. Early recognition of IPV allows for timely interventions, which can mitigate long-term health repercussions.
To address this issue, a research team supported by the NIH, led by Dr. Bharti Khurana from Mass General Brigham, created an innovative artificial intelligence (AI) tool aimed at predicting patients who may be vulnerable to IPV. The findings concerning this tool were published on March 13, 2026, in npj Women’s Health.
The research team utilized machine learning—a form of AI—to develop three distinct computer models designed to forecast IPV risks. They drew on electronic medical records from 841 patients involved in a domestic abuse intervention and prevention center, and compared them with records from an additional 5,212 non-IPV patients of similar demographics. One model utilized structured patient data, while another analyzed unstructured data extracted from medical notes. A third model combined both data types.
During development, the researchers used 80% of the patient data to train the models and then assessed their accuracy against the remaining 20%. All three models demonstrated over 80% accuracy, with the combined model achieving the highest accuracy at 88%. Notably, both the table and combined models could identify IPV risk more than three years prior to a patient seeking help.
The model’s reliability was further confirmed across three additional patient cohorts, maintaining an accuracy rate between 82-88% for the combined model.
According to the models, higher risks for IPV were associated with mental health disorders, chest pain, and the use of painkillers, as well as socio-economic factors such as social deprivation and frequent radiology tests. Conversely, patients who engaged in routine preventive services, like mammograms and cervical cancer screenings, exhibited a lower risk of IPV, likely due to improved access to healthcare and enhanced comfort when seeking medical attention.
The researchers stressed the importance of evaluating the model in broader populations before its application in clinical environments. They noted that the model is not intended for diagnosing IPV; instead, it aims to assist healthcare providers in identifying patients who may benefit from discussions regarding IPV and available support resources.
“By analyzing existing patterns within healthcare data, this approach empowers clinicians to have earlier, safer, and more informed discussions with patients,” said Khurana. “The goal isn’t to force disclosure, but to help clinicians communicate in a supportive manner and connect patients with necessary resources and support.”
Related Links
References
Leveraging multimodal machine learning for accurate risk identification of intimate partner violence. Gu J, Carballo KV, Ma Y, Bertsimas D, Khurana B. NPJ Womens Health. 2026;4(1):15. doi: 10.1038/s44294-025-00126-3. Epub 2026 Mar 13. PMID: 41836047.
Funding
NIH’s National Institute of Biomedical Imaging and Bioengineering (NIBIB), Office of the Director (OD).
