TraumaICDBERT Excels in Predicting Injury ICD-10 Codes

A natural language processing algorithm, traumaicdbert, demonstrated superior performance in predicting injury ICD-10 diagnosis codes compared to existing tools. Analyzing over 3000 tertiary survey notes, traumaicdbert outperformed Amazon Web Services Comprehend Medical, showing potential for improved real-time availability of diagnosis codes in clinical settings. Each survey note, on average, had 3.8 extracted injury ICD-10 codes, highlighting the algorithm’s accuracy and potential impact on downstream prediction tools.

Journal Article by Choi J, Chen Y (…) Spain DA et 3 al. in Ann Surg

Copyright © 2023 Wolters Kluwer Health, Inc. All rights reserved.

read the whole article in Ann Surg

open it in PubMed