AI tools can detect early signs of intimate partner violence

AI tools flagged intimate partner violence risk in medical records up to four years before patients sought care, research suggests.

The models were built to identify patterns in health records that could point to abuse before someone enrolled in care at a domestic violence treatment centre.

Researchers said the findings suggest AI could support earlier screening and help healthcare providers start conversations about intimate partner violence sooner, though the tools still need broader testing.

The study was led by researchers at Mass General Brigham, working with collaborators at the Massachusetts Institute of Technology in the US.

Bharti Khurana is principal investigator, corresponding and senior author, founding director of the Trauma Imaging Research and Innovation Center and an emergency radiologist in the Mass General Brigham department of radiology.

Khurana said: “Our research offers proof of concept that AI can support clinicians in flagging possible abuse earlier.

“Earlier identification of intimate partner violence and future risk may enable clinicians to intervene sooner and help prevent significant mental and physical health consequences.”

The researchers trained three machine-learning models using electronic medical record data from 673 women who visited a domestic abuse intervention and prevention centre at a US academic health centre between 2017 and 2022, as well as 4,169 demographically matched controls who did not report intimate partner violence.

Electronic medical records are digital versions of a patient’s health history, while machine learning is a type of AI that identifies patterns in data to make predictions.

The three models included a tabular model using structured electronic medical record data such as diagnoses, medications and a social deprivation index based on zip code, a notes model using unstructured clinical notes and radiology and emergency department reports, and a fusion model combining both data types called Holistic AI in Medicine.

When tested on a separate group of 168 patients who visited the intimate partner violence intervention and prevention centre in the same timeframe and 1,043 controls, all three models showed high accuracy. The fusion model performed best at 88 per cent.

Using archived, time-stamped medical records, that fusion model identified 80.5 per cent of cases in advance, on average more than 3.7 years before patients sought care.

The researchers then validated the models using data from two additional patient groups that were not included in the training or testing data, as well as controls, and found similarly high accuracy.

Previous research led by Bharti Khurana found that women who frequently undergo imaging studies in the emergency department and have specific types of injuries are more likely to later report intimate partner violence.

This new AI research identified additional risk factors for intimate partner violence.

People with mental health disorders, chronic pain and frequent emergency department visits were more likely to experience intimate partner violence, whereas patients who regularly accessed preventive services such as mammograms and immunisations had a lower risk.

More than one-third of women and one in 10 men will experience intimate partner violence in their lifetimes, yet people rarely disclose it to health providers because of fear, stigma, or financial or psychosocial dependence on the person abusing them.

The authors said the models were developed and validated in patients who had sought care for or disclosed intimate partner violence, which may limit accuracy in predicting intimate partner violence in people who are less likely to seek care or disclose it to providers.

They also said the control group in the training data may have included false negatives, or patients who were experiencing intimate partner violence but did not report it, which could reduce model accuracy.

Khurana said future training with larger, more diverse patient datasets over longer time periods will improve the model’s accuracy.