Note: Single-source report; awaiting corroboration.
Researchers supported by the National Institutes of Health have developed an artificial intelligence (AI) tool to predict patients at risk of intimate partner violence (IPV) long before they seek assistance. IPV involves abuse from current or former partners and can lead to serious physical and mental health consequences. Current screening tools rely mostly on patient self-reporting and detect only a fraction of IPV cases.
The AI tool was created using machine learning, with three models built from electronic medical records of 841 patients in a domestic abuse intervention center and 5,212 non-IPV patients matched by age and background. The models utilized structured data, unstructured medical notes, or both. After training on 80% of the dataset, they were tested on the remaining 20%, achieving over 80% accuracy. The model combining both data types reached 88% accuracy.
This combined model could identify IPV risk more than three years before patients sought help. Its accuracy, ranging from 82% to 88%, was validated in three additional patient groups. Factors such as mental health conditions, chest pain, painkiller use, high social deprivation, and frequent radiology tests were associated with increased IPV risk in the model's predictions. In contrast, patients regularly obtaining preventive services like mammograms and cervical cancer screenings were less likely to be identified as at risk, which researchers suggested may indicate better access to care and comfort in seeking medical services.
The tool may enable healthcare providers to intervene earlier by connecting patients with resources and support, potentially reducing the long-term health impacts of IPV.