A research team funded by the National Institutes of Health (NIH) has developed an advanced Artificial Intelligence (AI) tool capable of predicting the risk of Intimate Partner Violence (IPV), offering a new approach to early detection and prevention within healthcare settings.

The AI-powered clinical decision support tool uses routinely collected patient data during medical visits to identify individuals at potential risk of IPV. The machine learning models demonstrated high accuracy in detecting cases, marking a significant step toward proactive intervention in a major public health concern.

IPV, which includes abuse by current or former partners, can lead to severe health consequences such as life-threatening injuries, chronic pain and mental health disorders. Despite affecting millions of individuals globally, many cases remain undiagnosed due to underreporting driven by stigma, fear and safety concerns.

The study, led by researchers from Harvard Medical School, introduced three distinct AI models designed to detect IPV risk. These included models trained on structured clinical data, unstructured medical notes such as radiology reports and a multimodal fusion model combining both data types.

Among the three, the multimodal AI model delivered the highest performance, achieving approximately 88 percent accuracy. The models were trained using several years of hospital data, including records from nearly 850 affected patients and over 5,200 matched control individuals.

Importantly, the AI models demonstrated the ability to identify IPV risk more than three years before patients sought help at hospital-based domestic abuse intervention centres. While the structured data model detected risks slightly earlier, the fusion model identified a greater number of cases overall.

The research also highlighted the critical role of clinical and imaging data in identifying patterns associated with IPV. Radiology reports, in particular, can reveal recurring trauma patterns that may indicate abuse, providing valuable insights for early intervention.

According to Qi Duan, the AI-based tool has the potential to transform how healthcare providers approach IPV detection and prevention. He noted that the system could become a valuable asset in addressing the widespread yet often hidden issue.

Similarly, Bhati Khurana from Mass General Brigham emphasised that the technology represents a shift from reliance on patient self-disclosure to proactive risk identification using existing healthcare data.

Researchers clarified that the tool is intended to support, not replace, clinical judgement. It is designed to help healthcare professionals initiate sensitive, patient-centred conversations and connect individuals with appropriate support services.

The team is now working to integrate the AI models into electronic medical record systems, enabling real-time IPV risk assessments in clinical environments. This integration could enhance early detection efforts and improve long-term health outcomes for at-risk individuals.

The study was co-funded by the National Institute of Biomedical Imaging and Bioengineering (NIBIB) and the NIH Office of the Director, reinforcing the growing role of AI in advancing public health and clinical care.