Back

Assessing an ICD-10 Code Approach for Tracking Xylazine-Involved Overdose Deaths in the United States

Friedman, J. R.

2024-12-01 addiction medicine
10.1101/2024.11.27.24318111 medRxiv
Show abstract

IntroductionThe national prevalence of the veterinary sedative xylazine in US overdose deaths rose between 2018 and 2021. More updated estimates are limited, partially due to the lack of a dedicated ICD-10 code--a primary mechanism used to specify drugs implicated in overdose deaths in the US, including in the CDC WONDER system, which provides public data requests with a 6-month lag. For other emerging substances lacking dedicated codes, over time umbrella codes have come to de facto represent them, yet this has not yet been demonstrated for xylazine. MethodsOverdose deaths in CDC WONDER involving T42.7 ("Antiepileptic and sedative-hypnotic drugs, unspecified") or T46.5 ("Other antihypertensive drugs, not elsewhere classified") were compared to two more specific, albeit delayed, sources: NVSS describing national trends in 2018-2021 and SUDORS describing state-level trends in 2020-2022. The CDC WONDER approach was also used to visualize trends in xylazine-involved deaths through Q1 2024 by geography, race/ethnicity, substance co-involvement, and demographic categories. ResultsAt the national level, concordance between CDC WONDER records and previous NVSS estimates improved after 2019 and became highly similar in 2021 (3,480 vs 3,468 deaths). Concordance was also high for estimates stratified by race, age, and region. At the state-level, across 49 state-year pairs, correlation between CDC WONDER and SUDORS was 0.97. Xylazine-involved deaths doubled between 2021 and 2024 Q1, and racial inequalities widened. DiscussionT42.7 or T46.5, together, may have become the de-facto coding scheme representing xylazine-involved deaths. This approach provides more up-to-date results, showing increasing prevalence and worsening racial inequalities in xylazine-involved deaths into 2024.

Matching journals

The top 5 journals account for 50% of the predicted probability mass.

50% of probability mass above

"Similar papers" are the closest papers from that journal in the model's embedding space. They show what the match is built on, but the ranking comes mostly from a classifier over the whole training set, not from these examples alone.