Correction: Automatic Extraction of Mental Health Disorders From Domestic Violence Police Narratives: Text Mining Study

George Karystianis, Armita Adily, Peter W. Schofield, Lee G. Knight, Clara Galdon, David M. Greenberg, Louisa Jorm, Goran Nenadić, Tony ButlerJournal of Medical Internet Research 2018 · DOI · OpenAlex

Introduction

This document serves as a formal corrigendum for the study "Automatic Extraction of Mental Health Disorders From Domestic Violence Police Narratives: Text Mining Study." While the original research successfully demonstrated the utility of Natural Language Processing (NLP) in identifying mental health trends within police reports, this update provides critical corrections to the quantitative data and the hierarchical mapping logic used to categorize these disorders. These adjustments ensure the accuracy of the prevalence rates reported for victims and persons of interest (POIs) in domestic violence (DV) cases.

Methodology and Taxonomy Refinement

The study's core methodology involves using text mining to identify mentions of mental health disorders in unstructured police narratives and mapping them to the International Classification of Diseases, Tenth Revision (ICD-10) framework. The correction clarifies the specific structure of this mapping hierarchy, which was previously misreported.

The refined taxonomy includes:

  • Level 1 (Generic): 26 categories (corrected from 24).
  • Level 2: 62 categories (corrected from 71).
  • Level 3: 98 categories (corrected from 163).
  • Level 4: 27 categories (corrected from 25).

Furthermore, nomenclature for certain categories was updated to better align with clinical data. For instance, "Diseases of the nervous system" was more accurately renamed to "Other degenerate diseases of the nervous system." These changes provide a more granular and clinically relevant framework for analyzing the population.

Data Scale and Mapping Efficiency

The scale of the NLP task is significant, involving a massive corpus of domestic violence narratives. The corrected data highlights the volume of information processed:

  • Total Mentions: 103,232 primary-level mental health disorder mentions were identified.
  • Mapping Success: Approximately 60% (62,526) of these mentions provided sufficient context for mapping to more specific secondary ICD-10 levels.

Key Findings and Corrected Prevalence

The updated statistics offer a more precise view of the mental health landscape in DV contexts. While the numerical values have shifted, the relative prevalence of certain conditions remains a key insight.

Prevalence Rates

  • Depression: This was identified in 3,269 victim events (22.25%) and 8,944 POI events (18.70%). The correction shows that depression is slightly more prevalent among victims than initially reported.
  • Alcohol Abuse: This remains a dominant factor for POIs, appearing in 5,829 events (12.19%).
  • Anxiety Disorders: These were corrected to 11.66% (1,714 events) for victims.
  • Mood (Affective) Disorders: These were the most frequent mentions overall, with 15,330 mentions for POIs and 4,946 for victims.

The "Unspecified" Challenge

A significant finding is the high volume of "unspecified mental health disorders" (32.46% for POIs and 22.79% for victims). This indicates a gap in police reporting where a mental health issue is recognized but not explicitly diagnosed in the narrative. This highlights a limitation of using police data as a proxy for clinical records.

Conclusion

Despite these numerical and taxonomic corrections, the authors emphasize that the core conclusions of the original study remain unchanged. Text mining is a powerful and viable tool for large-scale mental health surveillance within criminal justice data. These corrections refine the accuracy of the data, providing a more reliable foundation for public health researchers and policy-makers working at the intersection of mental health and domestic violence. Future work will focus on improving the NLP's ability to distinguish between historical mentions and acute crises, as well as validating these findings against clinical health records.

Summary updated

Abstract

[This corrects the article DOI: 10.2196/11548.].
BibTeX
@article{karystianis2018correction,
  author = {George Karystianis and Armita Adily and Peter W. Schofield and Lee G. Knight and Clara Galdon and David M. Greenberg and Louisa Jorm and Goran Nenadić and Tony Butler},
  title = {{Correction: Automatic Extraction of Mental Health Disorders From Domestic Violence Police Narratives: Text Mining Study}},
  journal = {Journal of Medical Internet Research},
  year = {2018},
  doi = {10.2196/13007},
}
Figure