Intimate partner violence (IPV)—abuse or aggression in a romantic relationship—is a significant yet under-discussed public health issue worldwide. The term ‘intimate partner’ includes both current and former spouses as well as dating partners. IPV can range from a single episode of violence to chronic and severe abuse that lasts over several years. It includes physical violence, sexual violence, stalking, and psychological aggression intended to harm a partner mentally or emotionally or to exert control over them.
Intimate partner violence often lead to injuries and can even be fatal. In India, IPV remains widely under-reported. Nevertheless, data from the National Family Health Survey-5 indicates that one in three married women aged 18–49 has experienced spousal violence at least once. Reports from around the world suggest that a significant share of homicide victims are killed by an intimate partner.
A team of researchers supported by the National Institutes of Health in the US has developed an AI tool that provides decision support to clinicians by predicting whether patients are at risk of IPV. Using data routinely collected during medical visits, the team trained a machine-learning model that showed high accuracy in detecting IPV among patients in a study.
As part of the study, a research team led by Harvard Medical School developed three AI models for IPV detection in health care settings and compared how well they predict risk. The study notes that many IPV cases go unnoticed, limiting chances for early intervention. Existing screening tools capture only a small share of cases, while clinical records and imaging data offer important clues. The AI works by identifying factors such as the location of injuries and signs of older trauma, and then assessing whether a patient has a high or a low risk of IPV. Radiologists, in particular, are often better at spotting patterns of injury linked to IPV.
An Indian-origin emergency radiologist, Bharti Khurana, is the senior author of this NIH-supported study. Khurana, who was born and raised in New Delhi, began working on IPV about a decade ago, initially aiming to publish a single paper exploring whether radiological scans could help diagnose IPV and raise awareness among radiologists.
As part of the study, researchers analysed several years of hospital data from about 850 affected women and 5,200 matched controls. They built two separate models—one using structured data (such as tables) and another using unstructured data from medical notes, including radiology reports. They also created a third, multimodal model that combines both data types.
All three models performed well, but the multimodal model was the most accurate, correctly identifying IPV risk 88 per cent of the time. Both the tabular and multimodal models could flag risk more than three years before patients entered hospital-based intervention programmes. While the tabular model detected risk slightly earlier, the multimodal model identified more cases overall.
According to the researchers, this AI approach aims to shift IPV detection from reactive disclosure to proactive risk recognition within routine clinical care. By analysing patterns already present in health care data, it can help clinicians initiate earlier, safer, and more informed conversations with patients.
The research team now plans to develop a decision-support tool based on these AI models, which can be integrated into electronic medical record systems to provide real-time IPV risk assessments in clinical settings.