Chu-Lin Tsai

Background: Early prediction of hospital admission at the emergency department (ED) triage can improve patient flow and resource allocation. Most existing models rely solely on structured data. Incorporating multimodal physiologic information, such as cardiac and respiratory signals, may better capture subtle clinical signs. This study aimed to develop a deep learning fusion model that integrates electrocardiogram (ECG) images and respiration videos to predict hospital admission at ED triage. Methods: We prospectively collected ECG images and respiration videos from adult ED patients in a tertiary medical center. The proposed model consisted of modality-specific branches: an ECG branch using transfer learning, a respiration-signal-estimation pipeline, and a spatial attention module to enhance cardiopulmonary feature extraction. Class-balanced loss was applied to address dataset imbalance. Model performance was compared with the baseline Taiwan Triage Acuity Score (TTAS) and single-modality variants. Attention heatmaps using score-class activation mapping (score-CAM) were analyzed for interpretability. Results: The study included 118 patients, with a 27% admission rate. The ECG-respiration fusion model achieved an accuracy of 0.814, a recall of 0.749, a precision of 0.635, an F1-score of 0.680, an area under the receiver operating characteristic curve (AUROC) of 0.863, and an area under the precision-recall curve (AUPRC) of 0.675, outperforming both TTAS and single-modality models. Ablation analyses confirmed the additive value of spatial attention and class-balanced loss. Score-CAM visualizations indicated the model focused on physiologically relevant features, such as respiratory waveform inflection points. Conclusions: A multimodal deep learning approach combining ECG and respiration video signals can predict hospital admission at ED triage. Integrating such models into triage systems may enhance earlier risk identification and resource prioritization. Clinical trial number: Not applicable.

Background: Diabetic retinopathy (DR) and age-related macular degeneration (AMD) are 2 of the leading causes of vision loss worldwide. As population aging and diabetes prevalence increase, timely detection of these conditions has become essential. However, limited professionalism and insufficient training in ophthalmic screening among general medicine physicians may lead to delayed diagnosis and treatment. Artificial intelligence (AI)–assisted diagnostic tools may help to improve the screening of DR and AMD in routine clinical practice. Objective: This study aims to evaluate the clinical effectiveness and cost-effectiveness of AI-assisted fundus imaging for DR and AMD screening in adults with diabetes and older adults at risk of macular degeneration. Methods: This multicenter, 2-arm, parallel-group, open-label, individual-level randomized controlled trial and patient recruitment are performed at the settings of Family Medicine and Geriatric and Gerontology Care over 4 medical centers in Taiwan. Eligibility includes (1) diabetic individuals aged ≥20 years for DR screening, and (2) individuals aged ≥50 years for AMD screening. The study protocol has been approved by the ethics committees of all participating hospitals, and all participants will provide written informed consent. Results: The study was funded in September 2024, began on October 2, 2025, and is expected to be completed in December 2027. After the pilot implementation phase without randomization, participants will be randomized 1:1 into two groups: (1) AI-assisted screening, and (2) usual physician-only screening. The primary outcomes will include the detection rates (defined as participants with confirmed DR or AMD among all screened participants) and the positive predictive values (defined as participants with confirmed DR or AMD among those who tested positive). Cost-effectiveness analyses will be performed using data derived from the trial results. Conclusions: This study will provide robust evidence on the effectiveness of AI-assisted ophthalmic screening in improving patient eye health outcomes through timely screening and accurate early detection. This strategy may be cost-effective.

Background: Older adults account for a disproportionate share of emergency department (ED) visits and are often present with complex needs. Frailty is a key determinant of adverse outcomes in this population. We previously developed the Emergency Department Frailty Scale (ED‐FraS), a novel clinician‐judgment‐based instrument. This study prospectively validated the ED‐FraS when used by triage nurses and evaluated its predictive performance for hospital admission and ED length of stay (EDLOS). Methods: This prospective observational study was conducted at a tertiary academic medical center in Taiwan between February and July 2025. Triage nurses assessed patients aged ≥ 65 years using the ED‐FraS during routine triage encounters. We evaluated the association between ED‐FraS levels (1–5) and hospital admission and EDLOS. We compared the predictive performance of ED‐FraS, the standard Taiwan Triage and Acuity Scale (TTAS), and a modified TTAS (mTTAS), which integrated frailty scores. Discriminative ability was measured using the area under the receiver operating characteristic curve (AUROC). Results: A total of 550 older adults were enrolled (mean age 77.2 years). Higher ED‐FraS levels were significantly associated with increased admission rates (25.0% in level 1 vs. 59.4% in level 5) and prolonged median EDLOS (2.6 vs. 27.3 h). The mTTAS demonstrated superior discriminatory ability for hospital admission (AUROC 0.720) compared to TTAS (0.657) or ED‐FraS alone (0.638). Nurses reported the tool was feasible, taking < 30 s to complete. Conclusions: The ED‐FraS is a feasible and valid tool for identifying older adults at risk of adverse outcomes during triage. Integrating frailty assessment into standard triage systems enhances risk stratification and may improve resource allocation for vulnerable older adults.

BACKGROUND Diabetic retinopathy (DR) and age-related macular degeneration (AMD) are two of the leading causes of vision loss worldwide. As population aging and diabetes prevalence increase, timely detection of these conditions has become essential. However, limited professionalism and insufficient training of ophthalmic screening in general medicine physicians may lead to delayed diagnosis and treatment. Artificial intelligence (AI)-assisted diagnostic tools may help to improve the screening of DR and AMD in the routine clinical practice. OBJECTIVE To evaluate the clinical effectiveness and cost-effectiveness of AI-assisted fundus imaging for DR and AMD screening in adults with diabetes and older adults at risk of macular degeneration. METHODS This multicenter, two-arm, parallel-group, open-label, individual-level randomized controlled trial and patient recruitment is performed at the settings of Family Medicine and Geriatric and Gerontology Care over four medical centers in Taiwan. Eligibility includes (1) diabetic individuals aged 20 years or more for DR screening, and (2) individuals aged 50 years or more for AMD screening. The study protocol has been approved by the ethics committees of all participating hospitals, and all participants will provide written informed consent. RESULTS The study was funded in September 2024, began on October 2, 2025, and is expected to be completed in December 2027. After the pilot implementation phase without randomization, participants will be randomized 1:1 into two groups: (1) AI-assisted screening and (2) usual physician-only screening. The primary outcomes included the detection rates (defined as participants with confirmed DR or AMD among all screened participants) and the positive predictive values (defined as participants with confirmed DR or AMD among those who tested positive). Cost-effectiveness analyses will be performed using data derived from the trial results. CONCLUSIONS This study will provide robust evidence on the effectiveness of AI-assisted ophthalmic screening in improving patient eye health outcomes through timely screening and accurate early detection. This strategy may be cost-effective. CLINICALTRIAL ClinicalTrials.gov: NCT07069647

Background: Emergency Department (ED) revisits represent a critical issue in emergency medicine. Identifying high-risk revisit cases (revisits with intensive care unit admissions, cardiac arrest, or requiring emergency surgery) is particularly important. While prior studies have explored machine learning models for ED revisit prediction, few deep learning approaches exist, and dynamic features remain underutilized. Methods: We used data from National Taiwan University Hospital (NTUH), incorporating both static (e.g., age, sex, triage) and dynamic (vital signs) features. A preprocessing strategy was developed to handle temporal irregularities. We proposed a hybrid deep learning model combining Temporal Convolutional Network (TCN) and feature tokenizer (FT)-Transformer to integrate static and short-term dynamic information. Results: We evaluated our model on NTUH 2016–2019 data, achieving the area under the receiver operating characteristic curve (AUROC) of 0.8453 and the area under precision recall curve (AUPRC) of 0.0935 for high-risk revisits (base rate = 0.01), and AUROC of 0.7250 and AUPRC of 0.2005 for general revisits (base rate = 0.042). The model maintained robust performance when validated on 2020–2022 data. Compared to the static-only logistic regression baseline, our model improved AUPRC from 0.0288 to 0.0935 and precision from 0.0281 to 0.0428. Conclusion: Our model significantly outperformed the static-only baseline. It demonstrates the effectiveness of multimodal clinical data fusion in improving ED revisit prediction and supporting clinical decision-making.

Introduction: While research has begun to understand emergency department-based cardiac arrest (EDCA), consensus on what exactly constitutes EDCA remains unknown. In this study we aimed to explore the grouping of EDCA by using an unsupervised machine-learning algorithm and to investigate how these underlying clusters related to patient outcomes. Methods: We retrieved electronic health record data from an ED in a tertiary medical center. The EDCAs were identified via the cardiopulmonary resuscitation log. We used k-means cluster analysis to group EDCAs and t-distributed stochastic neighbor embedding (t-SNE) for visualization. Primary outcomes were ED mortality and ED length of stay (LOS). The analyses were repeated using an independent ED data set, the Medical Information Mart for Intensive Care IV Emergency Department (MIMIC-IV-ED) dataset. Results: From 2019 to 2022, there were 366 EDCA events. Cluster analysis identified three distinct clusters (Cluster 1 or immediate risk, n=54 [15%]; Cluster 2 or early risk, n=274 [75%]; Cluster 3 or late risk, n=38 [10%]). Cluster 1 patients had the shortest median time to EDCA (< 1 hour), followed by Cluster 2 (3 hours) and Cluster 3 (81 hours). Near cardiac arrest at triage was the most common cause of EDCA in Cluster 1, while respiratory illnesses and sepsis were more common in Cluster 3. The causes of EDCA in Cluster 2 were diverse, with predominantly cardiovascular and neurologic emergencies. The t-SNE revealed farther distances from Cluster 1 to the other two clusters, suggesting its most critical nature. Cluster 3 had the highest mortality (58%), followed by Clusters 1 (48%) and 2 (35%) (P = .01). Cluster 1 had the shortest median LOS (median, 4 hours), while Cluster 3 had the longest LOS (81 hours) (P < .001). In the independent data set, Cluster 1 remained, but Clusters 2 and 3 appeared to merge due to a shorter ED LOS overall. Conclusion: We identified three novel clusters (immediate, early, and late risk) with distinct patterns in clinical presentation, putative causes of ED-based cardiac arrest, and ED outcomes. Understanding these clinical phenotypes may help develop cluster-specific interventions to prevent EDCA or intervene most appropriately. Cluster 1 patients may benefit from resuscitation efforts, and Clusters 2 or 3 patients can benefit from timely interventions for cardiac, respiratory, and neurologic emergencies. In addition, for patients with prolonged ED boarding, periodic monitoring with an early warning system may prevent a cardiac arrest event.

Background: This study aimed to develop a machine learning‐based model to predict the risk of major adverse cardiac events (MACE) in patients presenting to the emergency department (ED) with chest pain, for whom acute myocardial infarction was excluded after serial high‐sensitivity cardiac troponin testing. Methods: This retrospective analysis included adult patients presenting with chest pain at 5 study hospitals between 2021 and 2024 in Texas. Patients diagnosed with acute myocardial infarction during the index visit were excluded. The primary outcome was the occurrence of 30‐day MACE, defined as myocardial infarction or all‐cause mortality within 30 days of the index ED visit. A long short‐term memory algorithm was used to develop the prediction model. Results: The analysis included 14 177 patients with a median age of 49.7 years, 41.2% of whom were men. A total of 535 patients (3.8%) had at least 1 high‐sensitivity cardiac troponin level above the 99th percentile. Thirty‐nine patients (0.3%) experienced 30‐day MACE, including 15 (0.1%) with myocardial infarction and 24 (0.2%) with all‐cause mortality. The long short‐term memory model demonstrated excellent performance in predicting 30‐day MACE (area under the receiver operating characteristic curve [AUC], 0.884 [95% CI, 0.815–0.941]), myocardial infarction (AUC, 0.963 [95% CI, 0.926–0.993]), and all‐cause mortality (AUC, 0.849 [95% CI, 0.698–0.948]). Conclusions: The long short‐term memory model accurately predicted 30‐day MACE in ED patients presenting with chest pain and no acute myocardial infarction, using patient demographics, vital signs at ED presentation, electrocardiographic findings, and serial high‐sensitivity cardiac troponin levels measured at flexible time points within 24 hours of ED arrival.

Background: Advancements in artificial intelligence (AI) have driven substantial breakthroughs in computer-aided detection (CAD) for chest x-ray (CXR) imaging. The National Taiwan University Hospital research team previously developed an AI-based emergency CXR system (Capstone project), which led to the creation of a CXR module. This CXR module has an established model supported by extensive research and is ready for application in clinical trials without requiring additional model training. This study will use 3 submodules of the system: detection of misplaced endotracheal tubes, detection of misplaced nasogastric tubes, and identification of pneumothorax. Objective: This study aims to apply a real-time CXR CAD system in emergency and critical care settings to evaluate its clinical and economic benefits without requiring additional CXR examinations or altering standard care and procedures. The study will evaluate the impact of CAD system on mortality reduction, postintubation complications, hospital stay duration, workload, and interpretation time, as wells as conduct a cost-effectiveness comparison with standard care. Methods: This study adopts a pilot trial and cluster randomized controlled trial design, with random assignment conducted at the ward level. In the intervention group, units are granted access to AI diagnostic results, while the control group continues standard care practices. Consent will be obtained from attending physicians, residents, and advanced practice nurses in each participating ward. Once consent is secured, these health care providers in the intervention group will be authorized to use the CAD system. Intervention units will have access to AI-generated interpretations, whereas control units will maintain routine medical procedures without access to the AI diagnostic outputs. Results: The study was funded in September 2024. Data collection is expected to last from January 2026 to December 2027. Conclusions: This study anticipates that the real-time CXR CAD system will automate the identification and detection of misplaced endotracheal and nasogastric tubes on CXRs, as well as assist clinicians in diagnosing pneumothorax. By reducing the workload of physicians, the system is expected to shorten the time required to detect tube misplacement and pneumothorax, decrease patient mortality and hospital stays, and ultimately lower health care costs. International Registered Report Identifier (IRRID): PRR1-10.2196/72928

Objectives: Little is known about the effect of pain on the relationship between triage and patient outcomes in United States emergency departments (ED). In this study we aimed to describe pain-associated ED visits and to explore how pain modifies the ability of ED triage to predict patient outcomes (hospitalization and ED length of stay [EDLOS)]. Methods: We obtained data from the National Hospital Ambulatory Medical Care Survey (NHAMCS), 2010-2021. Adult ED visits without missing data on pain score or triage level were included. We assessed pain scores at triage using a numeric rating scale (NRS) of 0-10. We further categorized the NRS scores into no (0), mild (1-3), moderate (4-6), and severe (7-10) pain. The five-level Emergency Severity Index was used for ED triage. The primary outcomes were hospital admission during the ED visit and EDLOS. For the analyses we used descriptive statistics and multivariable regression accounting for NHAMCS’s complex survey design. Results: Over the 12-year study period, there were 132,308 adult ED visits (representing 773,000,000 ED visits nationwide). Approximately 50% were triaged to level 3, followed by 30% to level 4. Approximately 45% reported severe pain, 21% moderate pain, 9% mild pain, and 25% no pain. Triage level 1 was associated with the highest rate of hospitalization (35%), with a gradual decrease in hospitalization rate from levels 2 to 4. Triage level 2 was associated with the longest mean EDLOS (5.6 hours), with a gradual decrease in EDLOS from levels 3 to 5. When stratified by pain intensity, the pattern of hospitalization altered in the mild and moderate pain groups. In these two pain-intensity groups, triage level 1 was associated with lower-than-expected odds of hospitalization, a 31% reduction suggested by the interaction term (adjusted odds ratio 0.69; 95% confidence interval .51-.92, P = .01). By contrast, the pattern of EDLOS persisted across all pain-intensity groups. Conclusion: Mild and moderate levels of pain intensity appear to negatively impact the ability of triage to predict hospitalization, resulting in overtriage among patients in these two pain-intensity groups. Pain intensity in the ED should be carefully evaluated to avoid overtriage and ensure the appropriate allocation of resources. [West J Emerg Med. 2025;XX(X)XXX–XXX.]