Researchers Develop AI Tool to Predict Patients at Risk of Intimate Partner Violence

Why It Matters

Proactive IPV detection can close the disclosure gap, allowing timely interventions that improve patient safety and reduce long‑term health costs.

Key Takeaways

• Multimodal model reaches 88% accuracy for IPV prediction
• Detects risk on average three years before clinical presentation
• Fusion of structured and unstructured data outperforms single-modality models
• Tool aims to support clinicians, not replace diagnosis
• Planned integration into EHRs for real-time IPV alerts

Pulse Analysis

Intimate partner violence remains one of the most under‑reported health crises, with many victims reluctant to disclose abuse due to fear and stigma. Traditional screening tools capture only a fraction of cases, leaving clinicians without reliable signals. The rise of machine‑learning in healthcare offers a pathway to shift from reactive self‑reporting to proactive risk identification, leveraging data already captured during routine visits.

The Harvard‑affiliated study evaluated three AI approaches, ultimately finding that a multimodal fusion model—combining tabular patient data with narrative radiology and clinical notes—delivered the strongest performance. With an 88% accuracy rate, the system flagged high‑risk individuals up to three years before they would typically present at domestic‑abuse intervention centers. This early window is critical, as it provides clinicians the opportunity to initiate confidential conversations and connect patients with resources before injuries become severe.

Embedding such predictive tools directly into electronic medical‑record platforms promises scalable, real‑time risk alerts across diverse health systems. However, successful deployment hinges on ethical safeguards, clinician training, and patient‑centered communication strategies to avoid unintended bias or coercion. As health networks adopt AI‑enhanced IPV screening, the approach could set a new standard for preventive care, reducing morbidity, lowering downstream costs, and ultimately saving lives.