COIOSDeep risk insights through multi-disciplinary research across the globe
a view of a city at night from the top of a building
Item new Peer-reviewed Added

Language models identified overdose deaths from New York investigators' written reports

Reading the reports New York City death investigators write on each case, the better of two language models identified suspected overdose deaths with 92% sensitivity and 88% positive predictive value.

The New York City medical examiner's office trained two language models on 9,263 reports written by investigators after each death in 2022 and 2023. The stronger model identified suspected overdose deaths with 92% sensitivity and 88% positive predictive value. Specificity was 76%, and both models did better than the office's existing surveillance tool and logistic regression on the same reports.

Why it is interesting: Both models read free text better than the office's existing surveillance tool, though specificity was the weaker measure.

Source
American journal of public health, 1 October 2026
Paper
Human-Centered Artificial Intelligence Framework for Overdose Mortality Surveillance, New York City, 2022‒2023
Authors
Hochstatter KR, Nadel T, Gupta R, Jangra A, D'Anna N, Gryczynski J et al.
DOI
10.2105/ajph.2026.308698
Type
Peer-reviewed article
Design
Model development and internal evaluation on 9,263 narrative death investigation reports, New York City OCME, 2022–2023, two LLM families against the incumbent tool and logistic regression
Verdict
New finding
Driver
AI in medicine
Driver
Alcohol and drug use

Other drivers in Data, models and systems