Presentation
Examining the Efficacy of Large Language Models for Injury Surveillance
SessionPoster Session 1
DescriptionAccurate and efficient injury coding is critical for effective injury surveillance. Manual injury coding requires considerable time and resources and suffers from inconsistencies in coding quality. In recent years, machine learning models trained on manually-coded historical injury records have been used to predict the injury codes efficiently and with reasonable accuracy but their accuracy has been limited for complex narratives and rare causes of injury. In this study, we examined performance of Large Language Models (LLMs) to parse complicated injury narratives and predict three injury codes: cause-of-injury, product-involved, and nature-of-injury. The evaluation was performed on 100 injury cases randomly selected from the publicly available National Electronic Injury Surveillance System database. The prediction performance of LLM (ChatgGPT-3.5) was compared with a traditional Logistic Regression and a neural network model- Multilayer Perceptron. We observed that LLM was able to effectively (a) extract syntactic relationships, (b) handle misspellings and common acronyms, and (c) decipher semantic information (e.g., contextual meaning of words or use of synonymous words) from the text with reasonable accuracy, even when the narratives were not in a proper grammatical format. The traditional ML models were observed to have limited capability in dealing with these challenges but required less computational resources.
Event Type
Poster
TimeTuesday, October 14th5:30pm - 6:30pm CDT
LocationRiverside East
Similar Presentations