Overview
- A peer‑reviewed study in npj Women's Health reports that machine‑learning tools trained on electronic medical records can identify elevated intimate partner violence risk up to roughly four years in advance.
- Among three approaches, a multimodal fusion model that combines structured data with clinical and radiology notes achieved the strongest performance at about 88% accuracy.
- The models were trained on hospital cohorts and then validated on independent patient groups, with the fusion model detecting more cases in advance while a tabular model recognized risk slightly earlier on average.
- Signals linked to higher risk included mental health disorders, chronic pain, and frequent emergency department visits, while consistent uptake of preventive services correlated with lower risk.
- The team, funded in part by NIH, frames the tools as clinician decision support rather than diagnostics and plans EMR‑embedded prototypes, noting limits from training on disclosed cases, potential false negatives, and the need for broader, privacy‑conscious, trauma‑informed deployment.