Related Experiment Video
Updated: Aug 8, 2026
Augmenting Large Language Models via Vector Embeddings to Improve Domain-Specific Responsiveness
Published on: December 6, 2024
Mortality Research With Artificial Intelligence Tools: Applying Language Models to the National Violent Death
John R Blosnich1, Jaspreet Ranjit1, Swabha Swayamdipta1
1John R. Blosnich is with the Suzanne Dworak-Peck School of Social Work at the University of Southern California, Los Angeles. Jaspreet Ranjit and Swabha Swayamdipta are with the Thomas Lord Department of Computer Science in the Viterbi School of Engineering at the University of Southern California.
Abstract:
Suicide deaths in the United States have increased for a generation with no current evidence of meaningful decline. With advances in data collection and accumulation, suicide prevention research has started utilizing artificial intelligence (AI) tools, predominantly machine learning, predictive analytics, and language models, to help mainly in prediction of suicide by utilizing large electronic health record data sets. The National Violent Death Reporting System (NVDRS) compiles significant amounts of structured and unstructured data about suicide decedents, including narrative information about the death (i.e., death narrative) that summarizes details such as the scene, decedent characteristics, and witness and next-of-kin details from interviews. Over the last 20 years, the NVDRS has accumulated hundreds of thousands of suicide death narratives-more than any human can (or would care to) read because of their grim, and sometimes graphic, nature. This analytic essay briefly reviews how AI tools have been applied to suicide prevention research. It also summarizes an example of how AI tools (e.g., language models) can expedite analysis using 756 645 narratives from 483 303 suicide deaths contained in the 2003-2023 NVDRS. (Am J Public Health. Published online ahead of print August 6, 2026:e1-e7. https://doi.org/10.2105/AJPH.2026.308618).