A new AI system can identify patients exposed to domestic violence years before they seek help, improving early interventions.

Artificial intelligence is rapidly expanding its role in the field of health, having promising applications in the early detection of intimate partner violence (IPV). A recent study published in Nature presents an AI system capable of identifying patients at risk of domestic abuse years before they seek help, facilitating life-saving interventions. The AI system analyzes routine medical data, looking for subtle and cumulative signals, instead of waiting for the disclosure of abuse. Three automated learning models have been developed: one based on structured data, one on unstructured clinical notes, and a hybrid model. The hybrid model had the highest accuracy, correctly identifying risk in 88% of cases, signaling abuse up to three years before formal intervention. This approach relies on timely pattern recognition, allowing the AI to detect recurring patterns of harm that are often too subtle for clinicians to observe. It is important to mention that the system is a decision support tool, not a diagnostic, providing risk signals without forcing disclosure.