AI Models Developed to Detect Domestic Abuse Risks in Patients

Researchers at Mass General Brigham have made significant strides in using artificial intelligence to predict the risk of domestic abuse among patients. Their study, published in NPJ Women’s Health on March 15, 2024, reveals that AI models can identify individuals at risk of experiencing physical, sexual, or psychological violence in intimate relationships up to four years before they seek help at domestic violence treatment centers.

The AI tools developed by the research team demonstrated an impressive accuracy rate, predicting intimate partner violence in approximately 88 percent of cases. These predictions are based on a variety of factors, including patients’ medical records, vital signs, and demographic data. Notably, symptoms such as chest pain, painkiller usage, and a higher frequency of radiology tests on the arms were found to correlate with an increased likelihood of abuse.

Dr. Bharti Khurana, an emergency radiologist and one of the study’s authors, emphasized the potential of AI to assist healthcare professionals in recognizing signs of abuse earlier. She described this approach as “proactive screening,” allowing clinicians to intervene before patients disclose their experiences of violence. “The idea is to share resources sooner rather than later,” Khurana stated.

According to the Centers for Disease Control and Prevention, about one in three women and one in six men will encounter intimate partner violence at some point in their lives. Despite the prevalence of this issue, many victims do not disclose their situations to medical professionals due to fear of judgment, concerns about their partner discovering their intentions, or financial and psychological dependencies.

Khurana noted that she began to observe subtle indicators in medical scans of patients who were victims of intimate partner violence. However, since radiologists typically spend only a few minutes reviewing imaging results, they often lack the time to thoroughly examine past medical records for further signs of abuse. AI has the capability to analyze vast amounts of electronic medical records, identifying patterns that may suggest someone is in danger at home.

The study involved training AI models on data from nearly 850 women enrolled at the Brigham’s domestic abuse intervention and prevention center between 2017 and 2019 and again from 2021 to 2022. Patients from 2020 were excluded due to the pandemic’s unique dynamics. The models were also trained on data from around 5,200 control patients who had not experienced intimate partner violence but shared similar demographic characteristics with the affected patients.

To enhance their predictions, the researchers employed three distinct AI models: one focused on medications, vital signs, and demographics; another analyzed clinical and radiology notes; and a third combined these elements. The integrated model achieved the highest accuracy in predicting intimate partner violence.

While the advancements in AI present promising tools for healthcare providers, experts urge caution in their application. Dr. Brigid McCaw, former medical director of the Kaiser Permanente Family Violence Prevention Program, highlighted the importance of ensuring that these tools do not lead clinicians to rely solely on algorithms without understanding the underlying data. “We need to be very, very cautious about how AI information is used for clinicians,” McCaw stressed. She also noted the necessity of rigorous testing for any domestic violence screening tools and called for the inclusion of survivors’ perspectives in the development process.

Khurana and her team are committed to refining these AI models to maximize their effectiveness in identifying victims while minimizing false positives. “If there are too many false positives, then you lose trust and nobody’s using it,” she explained. The team continues to train the models with data through 2025 and is collaborating with researchers worldwide to enhance the tool’s capabilities.

“My hope is to bring more institutions in so that we can learn from different ZIP codes, different areas, not only in the US,” Khurana added, reflecting the broader ambition of the project to address intimate partner violence through innovative technology.

The ongoing research underscores the potential of AI to transform how healthcare providers approach domestic abuse, offering a proactive means to identify and support vulnerable individuals before they seek help.