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.

Background: In the emergency department (ED), adequate pain control is essential for managing patients; however, children with pain are known to receive less analgesia than adults with pain. We introduce the Pain Passport to improve pain management in paediatric patients with suspected fractures in the ED. Methods: This was a before-and-after study. We reviewed the medical records of paediatric patients who were primarily diagnosed with fractures from May to August 2015. After the introduction of the Pain Passport, eligible children were enrolled from May to August 2016. Demographics, analgesic administration rates, time intervals between ED arrival and analgesic administration, and satisfaction scores were obtained. We compared the analgesic prescription rate between the two periods using multiple logistic regression. Results: A total of 58 patients were analysed. The baseline characteristics of subjects during the two periods were not significantly different. Before the introduction of the Pain Passport, 9 children (31.0%) were given analgesics, while after the introduction of the Pain Passport, a significantly higher percentage of patients (24/29, 82.8%) were treated with analgesics (P < 0.001). The median administration times were 112 (interquartile range [IQR], 64-150) minutes in the pre-intervention period and 24 (IQR, 20-74) minutes in the post-intervention period. The median satisfaction score for the post-intervention period was 4 (IQR, 3-5). The adjusted odds ratio for providing analgesics in the post-intervention period was 25.91 (95% confidence interval, 4.36-154.02). Conclusions Patient-centred pain scoring with the Pain Passport improved pain management in patients with suspected fractures in the paediatric ED.

Microbial colonization of the infant gut is unstable and shows a wide range of diversity between individuals. Gut microbiota play an important role in the development of the immune system, and an imbalance in these organisms can affect health, including an increased risk of allergic diseases. Microbial colonization of young infants is affected by the delivery mode at birth and the consequent alterations of gut microbiota in early life affect the development of allergic diseases. We investigated the effects of the delivery mode on the temporal dynamics of gut microbiota in healthy Korean infants. Fecal samples were collected at 1-3 days, 1 month, and 6 months after birth in six healthy infants. Microbiota were characterized by 16S rRNA shotgun sequencing. At the first and third days of life, infants born by vaginal delivery showed a higher richness and diversity of gut microbiota compared with those born by cesarean section. However, these differences disappeared with age. The Bacteroides genus and Bacteroidetes phylum were abundant in infants born by vaginal delivery, whereas Bacilli and Clostridium g4 were increased in infants born by cesarean section. The Firmicutes phylum and Bacteroides genus showed convergent dynamics with age. This study demonstrated the effect of delivery mode on the dynamics of gut microbiota profiles in healthy Korean infants.
Bacteroides ; Bacteroidetes ; Cesarean Section ; Clostridium ; Colon ; Female ; Firmicutes ; Gastrointestinal Microbiome* ; Humans ; Immune System ; Infant* ; Microbiota ; Parturition ; Pregnancy