Skip to main content
Peer-reviewed Authoritative

Are Natural Language Processing Tools Ready for Predicting Violence Toward Self or Others?

PRISMA systematic review of 21 eligible studies applying natural language processing to clinical text for predicting violence toward self or others. Assesses predictive performance and methodological quality across the evidence base.

Publisher

Psychiatric Annals (SLACK Incorporated)

Published

1 Apr 2026

Added

today

Key Findings

  • NLP-enhanced models consistently outperformed structured-data-only approaches, with AUROC values frequently exceeding 0.80 for self-directed outcomes
  • Studies showed methodological inconsistencies and gaps in reporting calibration and fairness metrics
  • Concludes clinical text contains meaningful predictive signal but significant constraints remain before clinical deployment

Methodology Notes

Peer-reviewed, Psychiatric Annals 56(4) (online-first 2026-03-24; print 2026-04-01), DOI 10.3928/00485713-20260324-03. PRISMA systematic review, 21 studies. journals.healio.com bot-blocks fetchers; title/authors and the abstract (source of the neutral-voice fields above) verified via Crossref.

Sources

Authors

Sabrina Grenier, Mattie Fay Arpin St-André, Bao Thy Nguyen, Rolence Pierre, Alexandre Hudon

Tags

nlpviolence-risksystematic-reviewclinical-textprisma

Cite This

APA

Sabrina Grenier et al. (2026). Are Natural Language Processing Tools Ready for Predicting Violence Toward Self or Others?. Psychiatric Annals (SLACK Incorporated). https://doi.org/10.3928/00485713-20260324-03