Minha Kim

This study aimed to develop an interpretable machine learning-based scoring system for predicting sepsis and septic shock among febrile patients at emergency department (ED) triage using longitudinal data. This retrospective, single-center study included adult patients, presented to ED of tertiary academic hospital with fever from January 2016 to December 2021. Using the AutoScore framework, we developed a novel scoring system for predicting sepsis and septic shock at the triage stage, incorporating nine variables and a maximum score of 29. The predictive performance of our score was assessed by calculating the area under the receiver operating characteristic curve (AUROC), and its performance was compared with that of two existing scoring systems: the quick Sequential Organ Failure Assessment (qSOFA) and the Modified Early Warning Score (MEWS). Our model incorporated nine variables including initial vital signs, age, baseline platelet count, total bilirubin, and creatinine levels. Among these, initial systolic blood pressure was identified as the most important predictor. AUROC of our model was 0.844 (95% confidence interval [CI], 0.812–0.875) in predicting septic shock and 0.703 (95% CI, 0.687–0.720) for sepsis. Compared to qSOFA ≥ 2 (AUROC: 0.605) and MEWS ≥ 5 (AUROC: 0.678), our scoring system demonstrated superior predictive performance for septic shock. For comparable specificity levels (ranging from 0.50 to 0.95), our scoring system achieved higher sensitivity than MEWS. Our scoring system is an interpretable and practical scoring tool for predicting sepsis and septic shock among patients with documented fever at ED triage using patient’s longitudinal data.

BACKGROUND Antibiotics are among the most frequently prescribed drugs globally, yet their use carries substantial risks of adverse drug reactions (ADRs), including nephrotoxicity, hepatotoxicity, and pharmacokinetic variability driven by patient body composition. Up to 63.6% of antibiotic prescriptions are issued empirically without confirmed pathogen identification, placing the responsibility for balancing efficacy and safety on real-time clinical judgment, particularly in the emergency setting. Existing clinical decision support systems (CDSSs) for antibiotic prescribing are constrained either by the rigidity of rule-based decision trees or by the lack of explainability in machine learning approaches, and none simultaneously offers transparent, safety-focused scoring with minimal clinical input. OBJECTIVE This study aimed to develop and validate a quantitative, transparent CDSS that pre-filters empirical antibiotic candidates by patient-specific organ safety risk, supporting safer prescription decisions. METHODS We developed ASSET (Antibiotic Safety Scoring and Selection Evaluation Tool), a CDSS that quantitatively computes a composite organ safety risk score for the seven most prescribed antibiotics using a Euclidean distance-based method across renal, hepatic, and body mass index (BMI) axes. Through its mathematical scoring framework, ASSET maintains high transparency while minimizing the clinical input required for recommendations. RESULTS Among 17 in-scope test cases, ASSET achieved 100% (17/17) recommendation concordance, with no clinician-selected antibiotic classified as carrying an unacceptable safety burden. Prospective validation is warranted before clinical application. CONCLUSIONS ASSET provides a transparent, quantitative safety pre-filter for empirical antibiotic selection, supporting more consistent safety-oriented prescribing in the emergency setting.

Objective: This study aimed to develop and validate MEDIVAL (Medical Documentation Validation), a progressive chain-of-thought (CoT) evaluation framework for automated assessment of large language model (LLM)-generated emergency department documentation, designed to align with expert clinical judgment in acute care settings.Methods: We designed a three-tier evaluation framework incorporating persona-based, error-enhanced, and insight-integrated strategies. The framework was tested across four LLMs (GPT-4o, GPT-4.1, Claude-3.5, Claude-3.7) on 33 emergency department records reviewed by four expert emergency physicians. Each model applied the three CoT strategies across five criteria: appropriateness, accuracy, structure/format, conciseness, and clinical validity. Model outputs were compared with expert ratings using Spearman correlation coefficients. Differences were analyzed with the Friedman test and Wilcoxon signed rank test with Bonferroni correction. Reproducibility was assessed through intraclass correlation coefficient (ICC) analysis.Results: All models demonstrated stronger alignment with expert ratings as CoT complexity increased, with Claude-3.7 (r=0.712, P<0.001) and GPT-4o (r=0.702, P<0.001) showing the highest correlations under the insight-integrated strategy. GPT-4.1 showed the greatest relative improvement (43.3% increase, r=0.457 to r=0.655, P<0.001). Significant overall differences were observed across strategies (χ2(2)=48.39, P0.919), with Claude-3.5 achieving the most consistent results (ICC, 0.997–0.998).Conclusion: MEDIVAL demonstrates that progressive CoT strategies systematically improve automated evaluation of emergency department documentation while maintaining excellent reproducibility. This framework offers a viable prescreening tool to reduce expert workload and support reliable artificial intelligence integration into emergency medicine workflows.

Objective: Trauma is one of the leading causes of obstetric morbidity and mortality, and incorrect seat belt use is a major determinant of adverse outcomes. This study evaluated the impact of structured education on proper seat belt use during pregnancy and examined preferences for future interventions.Methods: This prospective pilot study was conducted at a single center in Korea, over an 8-week period. Pregnant women between 20 and 37 weeks of gestation who visited the outpatient obstetrics and gynecology department were enrolled. A pre-education survey assessed baseline seat belt practices. Participants then received structured education, and a post-education survey 1 month later evaluated changes in awareness and behavior.Results: Sixty participants were included in the final analysis. Most reported obtaining seat belt information from unverified sources, particularly online communities. Many indicated a preference for structured education delivered by obstetricians, especially early in pregnancy. After the intervention, knowledge of proper seat belt use increased from 21.7% to 85.0% (P<0.001), and correct seat belt usage rose from 16.7% to 88.3% (P<0.001). Seat belt use while driving also improved markedly, increasing from 19.5% to 78.0% (P<0.001), demonstrating the program’s effectiveness in enhancing awareness and behavior.Conclusion Education on seat belt use during pregnancy significantly improved both knowledge and correct usage among pregnant women. Routine, formal educational programs are recommended to promote maternal and fetal safety.

Cho H, et al. Yonsei Med J. 2026 ;67:e34. https://doi.org/10.3349/ymj.2025.0125

Background: Ultrasound education is transitioning from in-person training to remote methods using mixed reality (MR) and 5G networks. Previous studies are mainly experimental, lacking randomized controlled trials in direct training scenarios. Objective: This study aimed to compare an MR-based telesupervised ultrasound education platform on private 5G networks with traditional in-person training for novice doctors. Methods: Conducted at a tertiary academic hospital from November to December 2023, the prospective unblinded randomized controlled pilot study assigned doctors without prior abdominal ultrasound education experience to either the telesupervision group (TG; n=20) or direct supervision group (DG; n=20). Participants received a 15-minute video lecture, conducted ultrasound on a phantom, and had 18 images scored by 2 blinded experts. Additionally, the TG received 5 minutes of training on the basic operation of a head-mounted display. Communication between doctors in the TG and supervisors was facilitated through a head-mounted display, whereas those in the DG interacted directly with supervisors. Primary outcomes were image quality scores, while secondary outcomes included procedure time, number of supervisor interventions, user experience using National Aeronautics and Space Administration-Task Load Index (NASA-TLX), System Usability Scale (SUS), and self-confidence through pre- and postsurveys. Results: Image quality scores and procedure times showed no significant differences between the groups (TG: 66.8 [SD 10.3] vs DG: 66.8 [SD 10.4], P=.84; TG: 23.8 [SD 8.0] min vs DG: 24.0 [SD 8.1] min, P=.95, respectively). However, the TG engaged in more educational interventions (TG: 4.0 [SD 2.5] vs DG: 0.8 [SD 1.1], P<.001), reflecting a more interactive training environment. TG participants reported lower NASA-TLX scores for mental demand (43.8 [SD 24.8] vs 60.6 [SD 22.4], P=.03), effort (43.1 [SD 22.9] vs 67.9 [SD 17], P<.001), and frustration (26.9 [SD 20.3] vs 45.2 [SD 27.8], P=.02), indicating a reduced cognitive load compared to the DG. The mean SUS score was also higher in the TG (66.6 [SD 9.1] vs 60.2 [SD 10.4], P=.046), suggesting better usability. Both groups showed significant improvements in confidence, with the TG showing notably greater improvement in abdominal ultrasound proficiency (pre-education TG: 1.6 [SD 0.9] vs DG: 1.7 [SD 0.9], P=.73; post-education TG: 3.8 [SD 0.9] vs DG: 2.8 [SD 1.0], P=.006). Conclusions: Although no significant differences in image quality scores were observed between groups, considerable differences in positive educational interactions, workload, and usability were evident. These findings emphasize the platform’s potential to enhance the ultrasound training experience, suggesting more interactive and efficient learning.

Background: The advancement of large language models (LLMs) offers significant opportunities for health care, particularly in the generation of medical documentation. However, challenges related to ensuring the accuracy and reliability of LLM outputs, coupled with the absence of established quality standards, have raised concerns about their clinical application. Objective: This study aimed to develop and validate an evaluation framework for assessing the accuracy and clinical applicability of LLM-generated emergency department (ED) records, aiming to enhance artificial intelligence integration in health care documentation. Methods: We organized the Healthcare Prompt-a-thon, a competitive event designed to explore the capabilities of LLMs in generating accurate medical records. The event involved 52 participants who generated 33 initial ED records using HyperCLOVA X, a Korean-specialized LLM. We applied a dual evaluation approach. First, clinical evaluation: 4 medical professionals evaluated the records using a 5-point Likert scale across 5 criteria—appropriateness, accuracy, structure/format, conciseness, and clinical validity. Second, quantitative evaluation: We developed a framework to categorize and count errors in the LLM outputs, identifying 7 key error types. Statistical methods, including Pearson correlation and intraclass correlation coefficients (ICC), were used to assess consistency and agreement among evaluators. Results: The clinical evaluation demonstrated strong interrater reliability, with ICC values ranging from 0.653 to 0.887 (P<.001), and a test-retest reliability Pearson correlation coefficient of 0.776 (P<.001). Quantitative analysis revealed that invalid generation errors were the most common, constituting 35.38% of total errors, while structural malformation errors had the most significant negative impact on the clinical evaluation score (Pearson r=–0.654; P<.001). A strong negative correlation was found between the number of quantitative errors and clinical evaluation scores (Pearson r=–0.633; P<.001), indicating that higher error rates corresponded to lower clinical acceptability. Conclusions: Our research provides robust support for the reliability and clinical acceptability of the proposed evaluation framework. It underscores the framework’s potential to mitigate clinical burdens and foster the responsible integration of artificial intelligence technologies in health care, suggesting a promising direction for future research and practical applications in the field.