Addressing diagnostic code variability in intimate partner violence surveillance through natural language processing:...

Area of Science:

• Public Health
• Health Informatics
• Clinical Research

Background:

• Intimate partner violence (IPV) and substance use disorders (SUDs) are critical public health issues.
• Identifying IPV in electronic health records (EHR) is challenging due to inconsistent documentation and billing.

Purpose of the Study:

• To develop and validate a Natural Language Processing (NLP) classifier for detecting IPV in clinical notes of patients with SUDs.
• To compare NLP-based IPV detection with traditional diagnostic codes.
• To examine the association between IPV and fatal overdose risk.

Main Methods:

• Constructed patient cohorts with SUDs or overdoses (stimulants, opioids, or both) from EHR data.
• Utilized the Open Health Natural Language Processing (OHNLP) toolkit for a rule-based IPV classifier.
• Validated the classifier and linked cohorts to a fatal overdose surveillance system.

Main Results:

• Analyzed over 15.5 million clinical notes from nearly 30,000 patients.
• NLP classifier achieved high performance (recall 0.91, precision 0.85, F1-score 0.89).
• NLP identified IPV in 5.6% of patients, significantly more than diagnostic codes (0.2-2.5%), with 90% of NLP-detected cases lacking codes.
• IPV prevalence was highest in polysubstance use (7.5%).
• IPV documentation rates in fatal overdose cases were similar to the general cohort, indicating missed opportunities.

Conclusions:

• NLP analysis of EHR clinical notes substantially increases IPV case identification compared to diagnostic codes alone.
• Individuals with polysubstance use and IPV represent a vulnerable group needing integrated screening and interventions.
• Routine NLP surveillance can enhance IPV case finding in SUD populations.