Researchers at Mass General Brigham (MGB) have made significant strides in developing artificial intelligence models aimed at predicting the risk of domestic abuse in patients. Their findings suggest that these AI tools can identify individuals at risk of experiencing physical, sexual, or psychological violence in intimate relationships up to four years before they seek help from a domestic violence treatment center. This research was published on October 6, 2023, in the journal Nature Partner Journals Women’s Health.
The AI models, which analyze various data points including medical records, vital signs, and demographic information, achieved an impressive accuracy rate of 88 percent in predicting intimate partner violence. Notably, certain medical indicators, such as chest pain, painkiller usage, and an increased number of radiology tests of the arms, were found to correlate with a higher likelihood of abuse. According to Dr. Bharti Khurana, an emergency radiologist at MGB and one of the study’s authors, these tools can help clinicians identify potential abuse cases proactively rather than waiting for victims to disclose their situations.
The Centers for Disease Control and Prevention estimates that one in three women and one in six men will experience intimate partner violence in their lifetimes. Despite its prevalence, many victims do not disclose their experiences to healthcare providers due to fears of judgment, partner retaliation, or financial dependency on their abuser. Dr. Khurana observed subtle patterns in imaging results that hinted at potential intimate partner violence but acknowledged that radiologists typically lack the time to analyze comprehensive medical histories for signs of abuse.
By harnessing the capabilities of AI, clinicians can leverage electronic medical records to identify at-risk patients more effectively. This approach represents a promising development in enhancing the limited attention and time available to healthcare professionals, enabling them to flag conditions that might otherwise go unnoticed.
Development and Testing of AI Models
The study involved training the AI models using data from nearly 850 women enrolled at the Brigham’s domestic abuse intervention and prevention center between 2017 and 2019 and 2021 and 2022. Data from 2020 was excluded due to the impact of the COVID-19 pandemic. The researchers also utilized a control group of approximately 5,200 patients who had not experienced intimate partner violence but shared similar demographics with those who had.
Three distinct AI models were created for this study. One analyzed medication, vital signs, and demographic data; another focused on clinical and radiology notes; and the third combined both approaches. The combination model showed the highest accuracy in predicting instances of violence.
Navigating the complexities of intimate partner violence discussions is inherently challenging, and missteps can have serious consequences. While the AI models signify a notable advancement, Dr. Brigid McCaw, former medical director of the Kaiser Permanente Family Violence Prevention Program, emphasized the need for caution. She remarked, “We need to be very, very cautious about how [AI] information is used for clinicians so that they don’t become over-reliant on algorithms without understanding what the data are that drive the algorithms.”
Future Directions and Considerations
Dr. Khurana noted that the models were adjusted to ensure they could accurately identify victims without mistakenly flagging individuals who were not at risk. She stated, “If there are too many false positives, then you lose trust and nobody’s using it.” The research team is committed to refining the models through 2025 and is engaging with international researchers to enhance the tool’s effectiveness.
“My hope is to bring more institutions in so that we can learn from different ZIP codes, different areas, not only in the US,” Dr. Khurana added. The ongoing development of these AI models holds the potential to revolutionize how healthcare professionals approach the sensitive issue of domestic violence, ultimately fostering earlier intervention and support for vulnerable individuals.
For future applications, researchers must ensure any domestic violence screening tools undergo rigorous testing and incorporate the perspectives of survivors. The integration of these AI models into clinical practice could significantly improve early detection of domestic abuse, providing a vital lifeline to those in need.
