Skip to main content
Peer-reviewed Authoritative

Risk prediction using natural language processing of electronic mental health records in an inpatient forensic psychiatry setting

Applies seven machine-learning algorithms with four word-list dictionaries (UMLS mental-health terms, DSM-IV diagnoses, a sentiment lexicon, and corpus frequencies) to de-identified forensic-inpatient clinical notes, predicting clinician-assigned risk ratings on three structured instruments: the HCR-20, START, and DASA. Reports best accuracy on the DASA dataset and flags that predicting actual endpoints (self-harm, harm-to-others, victimisation) needs further work.

Publisher

Journal of Biomedical Informatics (Elsevier)

Published

1 Oct 2018

Added

today

Key Findings

  • Structured violence-risk instrument ratings (HCR-20/START/DASA) can be partially predicted from free-text clinical notes via NLP
  • A sentiment dictionary with LMT/SVM classifiers gave the strongest performance, on the DASA dataset
  • Predicting downstream endpoints (self-harm, harm-to-others, victimisation) remained substantially harder than predicting the clinician rating

Methodology Notes

Peer-reviewed, Journal of Biomedical Informatics vol. 86, pp. 49-58 (issue October 2018; exact day not stated), DOI 10.1016/j.jbi.2018.08.007. Retrospective NLP on de-identified forensic inpatient notes; small single-site corpus. sciencedirect.com bot-blocks fetchers; metadata verified via PubMed (PMID 30118855) and Crossref.

Sources

Authors

Duy Van Le, James Montgomery, Kenneth C. Kirkby, Joel Scanlan

Tags

hcr-20forensicnlpviolence-riskclinical-notes

Cite This

APA

Duy Van Le et al. (2018). Risk prediction using natural language processing of electronic mental health records in an inpatient forensic psychiatry setting. Journal of Biomedical Informatics (Elsevier). https://pubmed.ncbi.nlm.nih.gov/30118855/