Junsang Yoo

Background: Emergency department (ED) crowding compromises patient safety and increases clinical workload. Existing crowding metrics capture structural indicators but not how clinicians subjectively experience crowding in real time. Clinician perception may provide a complementary measure. Objective: This study aimed to examine the concurrent and temporal associations between clinicians’ real-time ED perceived crowding (EDPC), captured via a smartwatch-based ecological momentary assessment (EMA) system, and 3 objective crowding metrics: the ED occupancy ratio (EDOR), the National Emergency Department Overcrowding Score (NEDOCS), and the ED Work Index (EDWIN). Methods: We conducted a prospective observational study over a continuous 449-hour period (November 25, 2024, 7 AM to December 13, 2024, 11:59 PM) in a single tertiary ED operating under reduced bed capacity and attending-only staffing due to a national physician shortage. We enrolled 28 clinicians (5 attending physicians and 23 charge nurses; mean age 36.3, SD 5.9 years; 21 female and 7 male) by voluntary nonprobability sampling. Clinicians recorded EDPC hourly during active shifts on a continuous 0-200 scale anchored to the NEDOCS crowding categories, from “Not Busy” to “Dangerously Overcrowded.” All valid ratings within each hourly period were aggregated into a single median EDPC value. Interrater agreement was assessed with the Gwet weighted agreement coefficient (AC2) at hourly time points with at least 2 ratings. Main analyses used Pearson correlations between hourly median EDPC and median EDOR, NEDOCS, and EDWIN. Linear mixed-effects models additionally accounted for repeated ratings within clinicians and clustering at the same hourly time point. Time-lagged correlations were exploratory. Statistical significance was set at α=.05. Results: Across 1124 scheduled on-duty assessment opportunities, clinicians completed 1100 individual EDPC ratings (response rate 1100/1124, 97.86%). These ratings yielded 449 hourly aggregated EDPC observations, with a median of 2 (IQR 2-3; range 1-4) clinicians contributing ratings per hour. Interrater reliability was substantial (AC2=0.742, 95% CI 0.712-0.772). Objective metrics were derived from 1846 ED visits. Hourly median EDPC correlated with median EDOR (r=0.542, 95% CI 0.473-0.604; P<.001) and NEDOCS (r=0.483, 95% CI 0.409-0.551; P<.001), but not EDWIN (r=0.060, 95% CI –0.033 to 0.152; P=.20). Conclusions: We deployed smartwatch-based EMA to capture high-frequency, in situ clinician perceptions during active shifts, potentially reducing susceptibility to retrospective recall bias while minimizing the spatial constraints of traditional survey methods. Hourly median EDPC was moderately associated with EDOR and NEDOCS in this single tertiary ED during an exceptional staffing crisis. The association with EDWIN was less consistent and became significant only after adjustment for time-of-day patterns. The high response rate (1100/1124, 97.9%) supports the short-term feasibility of repeated wearable-based data collection during routine clinical shifts. These findings support further investigation of EDPC as a complementary operational measure but do not establish that it directly measures cognitive workload or predicts subsequent crowding.

BACKGROUND ICU discharge decisions rely largely on subjective clinician judgment and on severity scores such as the Sequential Organ Failure Assessment (SOFA), which were designed to quantify organ dysfunction rather than to assess discharge readiness. No validated tool systematically evaluates discharge readiness from the physiological parameters routinely monitored at the ICU bedside. OBJECTIVE We developed and internally validated the ICU Discharge Readiness Score (iDRS), a two-tier screening tool for ICU discharge readiness using bedside physiological parameters and minimal laboratory data. METHODS Using the MIMIC-IV v3.1 database (2008–2022), we constructed daily 08:00 assessment grids for 76,801 ICU stays, generating 266,070 assessment points. iDRS employs a two-tier structure: Tier 1 automatically assigns a score of zero to patients receiving active life-sustaining treatments; Tier 2 scores the remaining patients across four physiological domains (respiratory, hemodynamic, neurological, renal/perfusion) on a 0–8 scale. Discriminative performance was evaluated using AUROC, compared against the SOFA score via the Paired patient-level cluster bootstrap, and supplemented by decision curve analysis and discordant case characterization. Temporal validation across early (2008-2016) and recent (2017-2022) cohorts was performed. RESULTS iDRS achieved an AUROC of 0.750 (95% CI 0.747–0.753), significantly outperforming SOFA (0.650; P<.001). At the optimal cutoff of ≥5, sensitivity was 0.895 and NPV was 0.949. Decision curve analysis demonstrated superior net benefit across clinically relevant thresholds. Discordant case analysis revealed that discharge failure was associated with chronic comorbidity burden (Charlson index r = −0.169) and hospitalization complexity (hospital LOS r = −0.367) rather than acute organ dysfunction. Temporal validation confirmed stable performance across both periods (AUROC 0.756 vs. 0.738). CONCLUSIONS iDRS demonstrated superior discrimination over SOFA with high sensitivity and NPV, supporting its role as a bedside screening tool that complements clinician judgment in ICU discharge decision-making.

BACKGROUND Emergency department (ED) crowding compromises patient safety and increases clinical workload. Existing crowding metrics capture structural indicators but not how clinicians subjectively experience crowding in real time. Clinician perception may provide a complementary measure. OBJECTIVE This study aimed to examine the concurrent and temporal associations between clinicians’ real-time ED perceived crowding (EDPC), captured via a smartwatch-based ecological momentary assessment (EMA) system, and 3 objective crowding metrics: the ED occupancy ratio (EDOR), the National Emergency Department Overcrowding Score (NEDOCS), and the ED Work Index (EDWIN). METHODS We conducted a prospective observational study over a continuous 449-hour period (November 25, 2024, 7 AM to December 13, 2024, 11:59 PM) in a single tertiary ED operating under reduced bed capacity and attending-only staffing due to a national physician shortage. We enrolled 28 clinicians (5 attending physicians and 23 charge nurses; mean age 36.3, SD 5.9 years; 21 female and 7 male) by voluntary nonprobability sampling. Clinicians recorded EDPC hourly during active shifts on a continuous 0-200 scale anchored to the NEDOCS crowding categories, from “Not Busy” to “Dangerously Overcrowded.” All valid ratings within each hourly period were aggregated into a single median EDPC value. Interrater agreement was assessed with the Gwet weighted agreement coefficient (AC2) at hourly time points with at least 2 ratings. Main analyses used Pearson correlations between hourly median EDPC and median EDOR, NEDOCS, and EDWIN. Linear mixed-effects models additionally accounted for repeated ratings within clinicians and clustering at the same hourly time point. Time-lagged correlations were exploratory. Statistical significance was set at α=.05. RESULTS Across 1124 scheduled on-duty assessment opportunities, clinicians completed 1100 individual EDPC ratings (response rate 1100/1124, 97.86%). These ratings yielded 449 hourly aggregated EDPC observations, with a median of 2 (IQR 2-3; range 1-4) clinicians contributing ratings per hour. Interrater reliability was substantial (AC2=0.742, 95% CI 0.712-0.772). Objective metrics were derived from 1846 ED visits. Hourly median EDPC correlated with median EDOR (r=0.542, 95% CI 0.473-0.604; P<.001) and NEDOCS (r=0.483, 95% CI 0.409-0.551; P<.001), but not EDWIN (r=0.060, 95% CI –0.033 to 0.152; P=.20). CONCLUSIONS We deployed smartwatch-based EMA to capture high-frequency, in situ clinician perceptions during active shifts, potentially reducing susceptibility to retrospective recall bias while minimizing the spatial constraints of traditional survey methods. Hourly median EDPC was moderately associated with EDOR and NEDOCS in this single tertiary ED during an exceptional staffing crisis. The association with EDWIN was less consistent and became significant only after adjustment for time-of-day patterns. The high response rate (1100/1124, 97.9%) supports the short-term feasibility of repeated wearable-based data collection during routine clinical shifts. These findings support further investigation of EDPC as a complementary operational measure but do not establish that it directly measures cognitive workload or predicts subsequent crowding. CLINICALTRIAL

Emergency Departments (ED) require rapid differentiation of time-sensitive critical conditions. We present a dynamic Clinical Decision Support System (CDSS) utilizing a Time-Aware Retrieval-Augmented Generation (RAG) framework. In a retrospective pilot study of 70 cases (Aortic Dissection, MI, Stroke, and Trauma Controls), the system’s Top-1 diagnostic accuracy significantly improved from 69.5% (triage) to 86.7% (discharge). This validates the system’s capability to integrate evolving FHIR data with clinical guidelines to reduce the diagnostic uncertainty.

The online version contains supplementary material available at 10.1038/s41598-026-42378-2.

Background: Artificial intelligence (AI)–enabled clinical decision support systems (CDSSs) are increasingly embedded within electronic health record (EHR) environments; however, their introduction can disrupt existing workflows and raise patient safety concerns, particularly in high-stakes settings such as surgical transfusion. Limited qualitative evidence exists regarding how frontline professionals anticipate the clinical, organizational, and workflow implications of such systems before wider deployment. Objective: This study aims to qualitatively examine the anticipated clinical, organizational, and workflow-level implications of implementing personalized Maximum Surgical Blood Order Schedule—Thoracic Surgery (pMSBOS-TS), an AI-enabled CDSS for personalized surgical blood ordering, before large-scale deployment. Methods: We conducted a consensual qualitative study with 14 multidisciplinary health care professionals involved in transfusion-related tasks at a large tertiary hospital. Following 1 pilot focus group to refine the interview guide and workflow diagram, 2 semistructured focus group discussions were held with 14 participants (5 physicians, 6 nurses, and 3 blood bank staff). Transcripts were analyzed using the Systems Engineering Initiative for Patient Safety (SEIPS) 101 framework, focusing on People, Environment, Tools, and Tasks, and were supported by task- and workflow-based analyses of transfusion processes. Member checking was conducted with participants and external clinicians to enhance validity. Results: A total of 189 semantic units and 61 core ideas were identified across 18 subdomains and 7 overarching domains. Participants anticipated that pMSBOS-TS could reduce unwarranted variation in blood ordering and planning, provided that algorithmic performance is reliable and the interface is tightly integrated into existing EHR workflows. At the same time, they expressed concerns regarding increased verification burden, system limitations in unexpected clinical scenarios, and potential communication bottlenecks between clinical units and the blood bank. Organizational culture, governance structures, and local transfusion logistics were viewed as critical determinants of whether the system would reduce or inadvertently increase workload and blood product waste. Conclusions: This preimplementation, SEIPS-based qualitative evaluation suggests that the successful adoption of an AI-enabled transfusion CDSS depends not only on predictive performance but also on sociotechnical readiness, including user trust, workflow fit, and organizational support. These findings provide practice-based insights to inform staged implementation, training, and governance strategies aimed at safely integrating predictive transfusion CDSSs into EHR-supported surgical workflows.

Medication errors pose a significant threat to patient safety. Although Bar-Code Medication Administration (BCMA) has reduced error rates, it is constrained by handheld devices, workflow interruptions, and incomplete safeguards against wrong patients, wrong doses, or drug incompatibility. In this study, we developed and evaluated a next-generation BCMA system by integrating artificial intelligence and mixed reality technologies for real-time safety checks: Optical Character Recognition verifies medication–label concordance, facial recognition confirms patient identity, and a rules engine evaluates drug–diluent compatibility. Computer vision models achieved high recognition accuracy for drug vials (100%), medication labels (90%), QR codes (90%), and patient faces (90%), with slightly lower performance for intravenous fluids (80%). A mixed-methods evaluation was conducted in a simulated environment using the System Usability Scale (SUS), Reduced Instructional Materials Motivation Survey (RIMMS), Virtual Reality Sickness Questionnaire (VRSQ), and NASA Task Load Index (NASA-TLX). The results indicated excellent usability (median SUS = 82.5/100), strong user motivation (RIMMS = 3.7/5), minimal cybersickness (VRSQ = 0.4/6), and manageable cognitive workload (NASA-TLX = 31.7/100). Qualitative analysis highlighted the system’s potential to streamline workflow and serve as a digital “second verifier.” These findings suggest strong potential for clinical integration, enhancing medication safety at the point of care.

BACKGROUND Artificial intelligence (AI)–enabled clinical decision support systems (CDSSs) are increasingly embedded within electronic health record (EHR) environments; however, their introduction can disrupt existing workflows and raise patient safety concerns, particularly in high-stakes settings such as surgical transfusion. Limited qualitative evidence exists regarding how frontline professionals anticipate the clinical, organizational, and workflow implications of such systems before wider deployment. OBJECTIVE This study aims to qualitatively examine the anticipated clinical, organizational, and workflow-level implications of implementing personalized Maximum Surgical Blood Order Schedule—Thoracic Surgery (pMSBOS-TS), an AI-enabled CDSS for personalized surgical blood ordering, before large-scale deployment. METHODS We conducted a consensual qualitative study with 14 multidisciplinary health care professionals involved in transfusion-related tasks at a large tertiary hospital. Following 1 pilot focus group to refine the interview guide and workflow diagram, 2 semistructured focus group discussions were held with 14 participants (5 physicians, 6 nurses, and 3 blood bank staff). Transcripts were analyzed using the Systems Engineering Initiative for Patient Safety (SEIPS) 101 framework, focusing on People, Environment, Tools, and Tasks, and were supported by task- and workflow-based analyses of transfusion processes. Member checking was conducted with participants and external clinicians to enhance validity. RESULTS A total of 189 semantic units and 61 core ideas were identified across 18 subdomains and 7 overarching domains. Participants anticipated that pMSBOS-TS could reduce unwarranted variation in blood ordering and planning, provided that algorithmic performance is reliable and the interface is tightly integrated into existing EHR workflows. At the same time, they expressed concerns regarding increased verification burden, system limitations in unexpected clinical scenarios, and potential communication bottlenecks between clinical units and the blood bank. Organizational culture, governance structures, and local transfusion logistics were viewed as critical determinants of whether the system would reduce or inadvertently increase workload and blood product waste. CONCLUSIONS This preimplementation, SEIPS-based qualitative evaluation suggests that the successful adoption of an AI-enabled transfusion CDSS depends not only on predictive performance but also on sociotechnical readiness, including user trust, workflow fit, and organizational support. These findings provide practice-based insights to inform staged implementation, training, and governance strategies aimed at safely integrating predictive transfusion CDSSs into EHR-supported surgical workflows.

Clinical decision-making substantially impacts patients’ lives and their quality of life. However, the black-box nature of AI-powered clinical decision support systems (CDSSs) complicates the interpretation of how decisions are derived. Explainable AI (XAI) improves acceptance and trust with explanations, but the effectiveness of different methods remains uncertain. We compared the acceptance, trust, satisfaction and usability of various explanatory methods among clinicians. We also explored the factors associated with acceptance levels for each item using trust, satisfaction and usability score questionnaires. Surgeons and physicians (N = 63), who had prescribed blood products before surgery, made decisions before and after receiving one of three CDSS explanation methods, each comprising six vignettes, in a counterbalanced design. We found empirical evidence, which indicates that providing a clinical explanation enhances clinicians’ acceptance than presenting ‘results only’ or ‘results with SHapley Additive exPlanations (SHAP)’. Additionally, trust, satisfaction and usability were correlated with acceptance. This study suggests best practices for the strategic application of the XAI–CDSS in the medical field.

Objective: This study used a nationwide database to identify and analyze factors that influence emergency department (ED) length of stay (LOS) and improve the efficiency of emergency care.Methods: This retrospective study analyzed data from the National Emergency Department Information System (NEDIS) database in Korea: 25,578,263 ED visits from 2018 to 2022. Patient demographics, clinical characteristics, and ED operational variables were examined. Univariate and multivariate logistic regression analyses were used to assess the associations between the variables and prolonged ED LOS, defined as 6 hours or more.Results: Among the 25,578,263 patients, the median ED LOS was 2.1 hours (interquartile range, 1.050–3.830 hours), with 12.6% experiencing a prolonged ED LOS. Elderly patients (aged ≥65 years) were significantly more likely than younger patients to experience prolonged ED LOS (adjusted odds ratio [aOR], 1.415; 95% confidence interval [CI]: 1.411–1.419). Patients transferred from other hospitals (aOR, 1.469; 95% CI, 1.463–1.474) and those arriving by emergency medical services (aOR, 1.093; 95% CI, 1.077–1.108) also had high odds of prolonged LOS. Conversely, pediatric patients had a low likelihood of extended stay (aOR, 0.682; 95% CI, 0.678–0.686). Severe illness, including sepsis (aOR, 1.324; 95% CI, 1.311–1.340) and COVID-19 infection (aOR, 1.413; 95% CI, 1.399–1.427), was strongly associated with prolonged LOS.Conclusion: Prolonged ED LOS is influenced by a combination of patient demographics, clinical severity, and systemic factors. Targeted interventions for older adults, severe illness, and operational inefficiencies such as hospital transfers are essential for reducing ED LOS and improving overall emergency care delivery.