• Jeff Choi, Yifu Chen, Alexander Sivura, Edward B Vendrow, Jenny Wang, and David A Spain.
• Department of Surgery, Stanford University, Stanford, CA.
• Ann. Surg. 2024 Jul 1; 280 (1): 150155150-155.

ObjectiveTo develop and validate TraumaICDBERT, a natural language processing algorithm to predict injury International Classification of Diseases, 10th edition (ICD-10) diagnosis codes from trauma tertiary survey notes.BackgroundThe adoption of ICD-10 diagnosis codes in clinical settings for injury prediction is hindered by the lack of real-time availability. Existing natural language processing algorithms have limitations in accurately predicting injury ICD-10 diagnosis codes.MethodsTrauma tertiary survey notes from hospital encounters of adults between January 2016 and June 2021 were used to develop and validate TraumaICD Bidirectional Encoder Representation from Transformers (TraumaICDBERT), an algorithm based on BioLinkBERT. The performance of TraumaICDBERT was compared with Amazon Web Services Comprehend Medical, an existing natural language processing tool.ResultsA data set of 3478 tertiary survey notes with 15,762 4-character injury ICD-10 diagnosis codes was analyzed. TraumaICDBERT outperformed Amazon Web Services Comprehend Medical across all evaluated metrics. On average, each tertiary survey note was associated with 3.8 (SD: 2.9) trauma registrar-extracted 4-character injury ICD-10 diagnosis codes.ConclusionsTraumaICDBERT demonstrates promising initial performance in predicting injury ICD-10 diagnosis codes from trauma tertiary survey notes, potentially facilitating the adoption of downstream prediction tools in clinical settings.Copyright © 2023 Wolters Kluwer Health, Inc. All rights reserved.

12 keywords...

hide…