Leveraging Natural Language Processing to Identify Veterans Who Inject Drugs to Assess Preexposure Prophylaxis and Sexually Transmitted Infection Testing Services at the Veterans Health Administration.

BACKGROUND

People who inject drugs (PWID) face disproportionate risks for infectious diseases yet remain difficult to identify within health care systems. Natural language processing (NLP) offers potential solutions for identifying PWID to improve access to harm reduction services.

METHODS

We evaluated an NLP dashboard designed to identify Veterans with evidence of injection drug use across 6 Veterans Health Administration facilities between August and October 2024. Four independent reviewers assessed electronic health records to confirm recent injection drug use and evaluated preventive care delivery, including HIV/hepatitis screening, sexually transmitted infection testing, preexposure prophylaxis usage, and harm reduction services.

RESULTS

Among 502 075 veterans, the dashboard identified 507 potential PWID, with 78 (15%) confirmed through chart review. Of confirmed PWID, 49% injected opiates, 41% cocaine, and 37% methamphetamines. HIV prevalence was 6%, hepatitis C antibody positivity 45% (28% viremic), and hepatitis B exposure 13%. Despite 94% engaging with mental health services and 82% with social work, only 29% saw infectious disease specialists. Most PWID (88%) had not received syringes, 74% lacked recent gonorrhea/chlamydia screening, and only 1 received HIV preexposure prophylaxis. Independent reviewers completed most chart reviews within 1 to 2 minutes.

CONCLUSIONS

The NLP dashboard efficiently identified PWID within an extensive health care system, revealing significant gaps in preventive care delivery despite high engagement with mental health services. Findings suggest opportunities to leverage existing therapeutic relationships while enhancing collaboration among mental health, social work, and infectious disease services to improve care for this vulnerable population.