Related Experiment Videos
Associations Between Natural Language Processing‒Enriched Social Determinants of Health and Fatal Opioid Overdose
Avijit Mitra1, Richeek Pradhan1, Kirsha S Gordon1
1Avijit Mitra was with the Manning College of Information and Computer Sciences, University of Massachusetts Amherst, at the time of this study. Richeek Pradhan is with the Centre for Medicine Use and Safety, Monash University, Parkville, Australia. Kirsha S. Gordon is with the Veterans Affairs Connecticut Healthcare System, West Haven. Haijuan Yan is with the Center for Healthcare Optimization and Implementation Research, Veterans Affairs Bedford Healthcare, Bedford, MA. Robert D. Kerns is with the Department of Psychiatry, Yale School of Medicine, New Haven. William C. Becker is with the Department of Internal Medicine, Yale School of Medicine. Hong Yu is with Miner School of Computer and Information Sciences, University of Massachusetts Lowell.
Abstract:
Objectives. To examine associations between opioid overdose (OOD) deaths among veterans and natural language processing (NLP)‒enriched social determinants of health (SDOH). Methods. We conducted a nested case‒control study using a cohort (n = 6 854 031) from the US Veterans Health Administration database with any record of service between October 2017 and September 2021. Participants were followed for up to 2 years from cohort entry, ending in September 2021. An NLP system extracted 8 SDOHs from unstructured notes, while structured data captured 6, yielding 9 distinct SDOHs with 5 common to both. Results. A total of 5411 veterans experienced fatal OOD over 21 397 462 person-years of follow-up and were matched with 21 623 controls. NLP-extracted SDOHs captured 90.08% of structured and 97.28% of all SDOHs. All measured SDOHs were significantly associated with increased risk of fatal OOD. The strongest association was for housing instability (adjusted odds ratio [AOR] = 3.36; 95% confidence interval [CI] = 2.96, 3.81), followed by legal problems (AOR = 3.07; 95% CI = 2.55, 3.70) and financial problems (AOR = 2.63; 95% CI = 2.32, 2.98). Conclusions. NLP-extracted SDOHs were significantly associated with fatal OOD risks among veterans and exhibited comparable direction and magnitude of association to structured SDOHs. (Am J Public Health. 2026;116(10):1575-1585. https://doi.org/10.2105/AJPH.2026.308697).