Background: Injuries pose a significant global health challenge due to their high incidence and mortality rates. Although injury surveillance is essential for prevention, it is resource-intensive.This study aimed to develop and validate locally deployable large language models (LLMs) to extract core injury-related information from Emergency Department (ED) clinical notes. Methods: We conducted a diagnostic study using retrospectively collected data from January 2014 to December 2020 from two urban academic tertiary hospitals. One served as the derivation cohort and the other as the external test cohort. Adult patients presenting to the ED with injury-related complaints were included. Primary outcomes included classification accuracies for information extraction tasks related to injury mechanism, place of occurrence, activity, intent, and severity. We fine-tuned a single generalizable Llama-2 model and five distinct Bidirectional Encoder Representations from Transformers (BERT) models for each task to extract information from initial ED physician notes. The Llama-2 model was able to perform different tasks by modifying the instruction prompt. Data recorded in injury registries provided the gold standard labels. Model performance was assessed using accuracy and macro-average F1 scores. Results: The derivation and external test cohorts comprised 36,346 and 32,232 patients, respectively. In the derivation cohort’s test set, the Llama-2 model achieved accuracies (95% confidence intervals) of 0.899 (0.889–0.909) for injury mechanism, 0.774 (0.760–0.789) for place of occurrence, 0.679 (0.665–0.694) for activity, 0.972 (0.967–0.977) for intent, and 0.935 (0.926–0.943) for severity. The Llama-2 model outperformed the BERT models in accuracy and macro-average F1 scores across all tasks in both cohorts. Imposing constraints on the Llama-2 model to avoid uncertain predictions further improved its accuracy. Conclusion Locally deployable LLMs, trained to extract core injury-related information from free-text ED clinical notes, demonstrated good performance. Generative LLMs can serve as versatile solutions for various injury-related information extraction tasks.

Background: Injuries pose a significant global health challenge due to their high incidence and mortality rates. Although injury surveillance is essential for prevention, it is resource-intensive.This study aimed to develop and validate locally deployable large language models (LLMs) to extract core injury-related information from Emergency Department (ED) clinical notes. Methods: We conducted a diagnostic study using retrospectively collected data from January 2014 to December 2020 from two urban academic tertiary hospitals. One served as the derivation cohort and the other as the external test cohort. Adult patients presenting to the ED with injury-related complaints were included. Primary outcomes included classification accuracies for information extraction tasks related to injury mechanism, place of occurrence, activity, intent, and severity. We fine-tuned a single generalizable Llama-2 model and five distinct Bidirectional Encoder Representations from Transformers (BERT) models for each task to extract information from initial ED physician notes. The Llama-2 model was able to perform different tasks by modifying the instruction prompt. Data recorded in injury registries provided the gold standard labels. Model performance was assessed using accuracy and macro-average F1 scores. Results: The derivation and external test cohorts comprised 36,346 and 32,232 patients, respectively. In the derivation cohort’s test set, the Llama-2 model achieved accuracies (95% confidence intervals) of 0.899 (0.889–0.909) for injury mechanism, 0.774 (0.760–0.789) for place of occurrence, 0.679 (0.665–0.694) for activity, 0.972 (0.967–0.977) for intent, and 0.935 (0.926–0.943) for severity. The Llama-2 model outperformed the BERT models in accuracy and macro-average F1 scores across all tasks in both cohorts. Imposing constraints on the Llama-2 model to avoid uncertain predictions further improved its accuracy. Conclusion Locally deployable LLMs, trained to extract core injury-related information from free-text ED clinical notes, demonstrated good performance. Generative LLMs can serve as versatile solutions for various injury-related information extraction tasks.

Background: Injuries pose a significant global health challenge due to their high incidence and mortality rates. Although injury surveillance is essential for prevention, it is resource-intensive.This study aimed to develop and validate locally deployable large language models (LLMs) to extract core injury-related information from Emergency Department (ED) clinical notes. Methods: We conducted a diagnostic study using retrospectively collected data from January 2014 to December 2020 from two urban academic tertiary hospitals. One served as the derivation cohort and the other as the external test cohort. Adult patients presenting to the ED with injury-related complaints were included. Primary outcomes included classification accuracies for information extraction tasks related to injury mechanism, place of occurrence, activity, intent, and severity. We fine-tuned a single generalizable Llama-2 model and five distinct Bidirectional Encoder Representations from Transformers (BERT) models for each task to extract information from initial ED physician notes. The Llama-2 model was able to perform different tasks by modifying the instruction prompt. Data recorded in injury registries provided the gold standard labels. Model performance was assessed using accuracy and macro-average F1 scores. Results: The derivation and external test cohorts comprised 36,346 and 32,232 patients, respectively. In the derivation cohort’s test set, the Llama-2 model achieved accuracies (95% confidence intervals) of 0.899 (0.889–0.909) for injury mechanism, 0.774 (0.760–0.789) for place of occurrence, 0.679 (0.665–0.694) for activity, 0.972 (0.967–0.977) for intent, and 0.935 (0.926–0.943) for severity. The Llama-2 model outperformed the BERT models in accuracy and macro-average F1 scores across all tasks in both cohorts. Imposing constraints on the Llama-2 model to avoid uncertain predictions further improved its accuracy. Conclusion Locally deployable LLMs, trained to extract core injury-related information from free-text ED clinical notes, demonstrated good performance. Generative LLMs can serve as versatile solutions for various injury-related information extraction tasks.

Background: Injuries pose a significant global health challenge due to their high incidence and mortality rates. Although injury surveillance is essential for prevention, it is resource-intensive.This study aimed to develop and validate locally deployable large language models (LLMs) to extract core injury-related information from Emergency Department (ED) clinical notes. Methods: We conducted a diagnostic study using retrospectively collected data from January 2014 to December 2020 from two urban academic tertiary hospitals. One served as the derivation cohort and the other as the external test cohort. Adult patients presenting to the ED with injury-related complaints were included. Primary outcomes included classification accuracies for information extraction tasks related to injury mechanism, place of occurrence, activity, intent, and severity. We fine-tuned a single generalizable Llama-2 model and five distinct Bidirectional Encoder Representations from Transformers (BERT) models for each task to extract information from initial ED physician notes. The Llama-2 model was able to perform different tasks by modifying the instruction prompt. Data recorded in injury registries provided the gold standard labels. Model performance was assessed using accuracy and macro-average F1 scores. Results: The derivation and external test cohorts comprised 36,346 and 32,232 patients, respectively. In the derivation cohort’s test set, the Llama-2 model achieved accuracies (95% confidence intervals) of 0.899 (0.889–0.909) for injury mechanism, 0.774 (0.760–0.789) for place of occurrence, 0.679 (0.665–0.694) for activity, 0.972 (0.967–0.977) for intent, and 0.935 (0.926–0.943) for severity. The Llama-2 model outperformed the BERT models in accuracy and macro-average F1 scores across all tasks in both cohorts. Imposing constraints on the Llama-2 model to avoid uncertain predictions further improved its accuracy. Conclusion Locally deployable LLMs, trained to extract core injury-related information from free-text ED clinical notes, demonstrated good performance. Generative LLMs can serve as versatile solutions for various injury-related information extraction tasks.

PURPOSE: Pedestrian injury is one of the most frequent injury mechanism in pediatrics. This study aimed to measure the association between time of pedestrian injury and injury severity among pediatric patients. METHODS: We used the Emergency Department based Injury In-depth Surveillance (EDIIS) database from 23 emergency departments between 2013 and 2016. All pediatric (≤15 years old) patients with pedestrian injury were eligible, excluding cases with unknown outcomes. Primary and secondary endpoints was severe injury. We calculated adjusted odds ratios (AORs) of time of injury (8 am to 2 pm, 2 pm to 8 pm, 8 pm to 8 am) to investigate out-comes while adjusting for potential confounders. RESULTS: Among 6,748 eligible patients, 4,184 (62.0%) suffered pedestrian injury at 2 pm to 8 pm, 1,566 (23.2%) at 8 am to 2 pm, and 998 (14.8%) at 8 pm to 8 am. Among them, 52 (0.8%) had case-fatalities, 572 (8.5%) had severe injuries, and 1,246 (18.5%) were admitted to hospital. In terms of severe injury, the 8 am to 2 pm group (10.5%) had higher proportions of severe injury compared to the 2 pm to 8 pm (8.0%; AOR {95% confidence interval [CI]}, 0.73 [0.60 to 0.89]) and 8 pm to 8 am (7.2%; AOR [95% CI], 0.65 [0.49 to 0.88]) groups. CONCLUSION: Pediatric pedestrian injury was frequent at 2 pm to 8 pm and was more severe at 8 am to 2 pm. Public health efforts to decrease pediatric pedestrian injury are needed to reduce health burden.
Emergency Service, Hospital ; Humans ; Odds Ratio ; Pedestrians ; Pediatrics ; Public Health ; Wounds and Injuries

BACKGROUND: Injury is a major public health problem and accounts for 10% of the global burden of disease. This study intends to present the temporal trend in the injury burden in Korea and to compare the burden size by injury mechanism and age group. METHODS: This study was a nationwide population-based observational study. We used two data sets, the death certificates statistics and the Korean National Hospital Discharge Survey data (2004–2012). We calculated age-standardized disability-adjusted life year (DALY) from years of life lost (YLL) and years lived with disability (YLD) and trend analysis. RESULTS: The DALYs of road injury decreased (P = 0.002), falls did not exhibit a trend (P = 0.108), and self-harm increased overall (P = 0.045). In the road injury, the YLLs decreased across all 4 age groups (0–14, 15–49, 50–79, ≥ 80) and the YLDs decreased in the 0–14-year-old group. In total, the DALYs of road injuries decreased in the 0–14-year-old group. In the fall injury, although the YLLs decreased in the over 80-year-old group, the YLDs increased in the 50–79-year-old group and the over 80-year-old group. The burden of self-harm injury was high in the age group 15 years and over, especially in the 15–49-year-old group. CONCLUSION: The leading causes of the injury burden were road injuries, falls, and self-harm. The burden of road injury and self-harm have recently shown a gradual decreasing tendency. On the other hands, that of fall injuries are continually high in the age group over 50 years of age.
Accidental Falls ; Accidents, Traffic ; Aged, 80 and over ; Dataset ; Death Certificates ; Hand ; Health Care Surveys ; Humans ; Korea* ; Observational Study ; Public Health ; Suicide