Development and Validation of an AI-Based Emergency Triage Model for Predicting Critical Outcomes in Emergency Department

Development and Validation of an AI-Based Emergency Triage Model for Predicting Critical Outcomes in Emergency Department | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Article Development and Validation of an AI-Based Emergency Triage Model for Predicting Critical Outcomes in Emergency Department Patipan Sitthiprawiat, Borwon Wittayachamnankul, Wachiranun Sirikul, and 1 more This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-6229836/v1 This work is licensed under a CC BY 4.0 License Status: Published Journal Publication published 25 Aug, 2025 Read the published version in Scientific Reports → Version 1 posted 10 You are reading this latest preprint version Abstract Emergency department (ED) overcrowding contributes to delayed patient care and worse clinical outcomes. Traditional triage systems face accuracy and consistency limitations. This study developed and internally validated a machine learning model predicting intensive care unit (ICU) admissions and resource utilization in ED patients. A retrospective analysis of 163,452 ED visits (2018–2022) from Maharaj Nakhon Chiang Mai Hospital evaluated logistic regression, random forest, and XGBoost models against the Canadian Triage and Acuity Scale (CTAS). The XGBoost model achieved superior predictive performance (AUROC 0.917 vs. 0.882, AUPRC 0.629 vs. 0.333). Key predictors included mode of arrival, patient age, and free-text chief complaints analyzed with multilingual sentence embeddings. Results demonstrate machine learning’s potential to enhance triage accuracy and resource allocation, effectively identifying critically ill patients compared to traditional triage methods. Biological sciences/Computational biology and bioinformatics Health sciences/Risk factors Health sciences/Signs and symptoms Health sciences/Health care Health sciences/Health care/Prognosis Figures Figure 1 Figure 2 Figure 3 Figure 4 Introduction Over the past decade, visits to emergency departments (ED) have steadily increased globally, resulting in overcrowding that significantly reduces the quality of patient care and satisfaction. 1 – 5 Triage, the initial process of identifying life-threatening conditions and prioritizing care, is vital for efficient resource allocation and optimal patient care in the ED. Standardized triage systems, such as the Emergency Severity Index (ESI) 6 and the Canadian Triage and Acuity Scale (CTAS), 7 exhibit variable accuracy, ranging from 59.2–82.9%. These systems rely heavily on clinical judgment, which can lead to significant variability and suboptimal outcomes 8 , 9 . Inaccurate triage contributes to ED overcrowding, delays in care, and increased mortality risks. 10 – 12 Over-triage puts a strain on resources, while under-triage impedes critical care. 2 Nurse-based triage is influenced by cognitive biases, including overconfidence, anchoring (the excessive reliance on initial information), and risk aversion, along with factors such as patient presentation, nurse experience, and the clinical environment. Disparities in outcomes related to ethnicity and sociodemographic factors raise concerns about fairness in ED care. 13 – 17 To address these limitations, several data-driven predictive models have been developed to improve triage accuracy and predict multiple clinical outcomes. These models incorporate predictors that are readily available at the time of triage, including vital signs, coded chief complaint, ambulance use, and medical history. The outcomes these models aim to predict include ESI levels, hospital admission, ICU admission, in-hospital mortality, critical conditions like sepsis and cardiovascular events, length of stay, readmission, and emergency department resource utilization (such as radiographs, medications, and laboratory tests) in adults. 18 – 24 Electronic health records (EHRs) contain a significant amount of unstructured data, including clinical notes, representing approximately 80% of all health information for a patient 25 , 26 , such as free-text chief complaints found in triage notes. Natural language processing enables machine learning models to utilize these free texts by transforming them into various numerical features through semantic embedding. 19 – 21 , 24 , 27 However, relatively few studies have adopted this method. This underscores a considerable gap in the current research. These free-text narratives frequently contain subtle details about patient presentations that structured data might not be able to capture. This study aims to develop a machine learning model that integrates structured data with free-text chief complaints documented by triage nurses, facilitating the triage process based on the need for critical care prediction. The goal is to reduce variability, improve predictive accuracy, and mitigate health disparities in triage outcomes through a real-time, clinically applicable model. Methods Study design and participants This retrospective cohort study utilized data from the EHRs of Maharaj Nakhon Chiang Mai Hospital, a tertiary university hospital in northern Thailand with approximately 60,000 annual ED visits. Data were collected from January 1, 2018, to December 31, 2022. This study was approved by the Institutional Review Board (IRB) - Research Ethics Committee Panel 5 (Research ID: 0068 / Study code: EME-2566-0068), with a waiver of informed consent. All patient identifiers were removed before analysis. All methods were performed in accordance with relevant guidelines and regulations. We included consecutive adult patient visits (≥ 18 years) presenting to the ED with either trauma or non-trauma complaints, regardless of mode of arrival (walk-in, emergency medical services, or referral), and with recorded CTAS triage levels from 1 to 5 were assigned by a qualified emergency nurse. We excluded duplicated, missing triage labels, dead on arrival, left without being seen, transferred to another hospital, or missing ED disposition targets. The study included patient demographic and clinical data available during triage. Demographic variables included patient age and biological sex. Arrival characteristics included the mode of arrival (walk-in, emergency medical services, or referral) and the case type (trauma or non-trauma). Clinical data included the chief complaint, which was documented as free text, vital signs (heart rate, respiratory rate, blood pressure, oxygen saturation, and temperature), and the level of consciousness, measured using the Glasgow Coma Scale (GCS). The primary outcome was Intensive care unit (ICU) admission directly from the ED regardless of boarding time. Data analysis Sample size calculations were conducted using the pmsampsize module. 28 Based on a prior research study’s ICU admission prevalence of 0.8% 19 , a c-statistic of 0.85, a shrinkage factor of 0.9, and 15 predictive parameters, the required sample size was determined to be 15,713 resulting in 4.19 events per predictor variable. This study developed a machine learning-based model using logistic regression with a lasso penalty, random forest, and XGBoost (eXtreme Gradient Boosting). These models are commonly used in clinical predictive model development sorted by their complexity compared to a reference model based on the CTAS triage label. The included visits were stratified and randomized, split into a training set (80%) and a test set (20%) to preserve a balanced distribution of labels across groups. We tuned hyperparameters with a random search and a 5-fold cross-validation strategy. The missing input features were imputed using K-nearest neighbors. Other structured input features were handled by labeling, scaling, and standardization. The pre-trained Multilingual Universal Sentence Encoder converted chief complaints recorded as free text into semantic vector representations. This approach captures contextual and semantic meanings, enabling the model to understand free text. Then, Principal Component Analysis was used to reduce the complexity of the embeddings. We used SHapley Additive exPlanations (SHAP) to determine the importance of the feature of the predictor to the model. Model performance was evaluated in a test set using the area under the receiver operating characteristic curve (AUROC) and the area under the precision-recall curve (AUPRC). While AUROC assesses discriminative performance, it can be less reliable with imbalanced datasets, whereas AUPRC emphasizes positive predictive value, offering a better measure in such cases. 29 The model performance was reported as mean and 95% confidence intervals generated using 1,000 bootstrapped samples from the test set. We used 500 bootstrapped samples from the test set for the prediction instability plot, mean absolute prediction error, and calibration plot to confirm the stability and reliability of predictions across datasets. 30 , 31 Analyses were performed using Python version 3.10.15. Results Participants This study reviewed 172,791 patient visits during its duration. After these exclusions, the final cohort consisted of 163,452 visits, as shown in Figure 1. Table 1 describes the characteristics of patients in this study. Overall, 13,406 visits experienced ICU admission (8.2%). A total of 2,016 visits (1.2%) were triaged as CTAS level 3-5 and below and eventually admitted to the ICU. Chief complaint free text is not classified. Model performance The model performance metrics are shown in Table 2 and Figure 2. The logistic regression model with lasso penalty exhibited comparable discrimination in terms of sensitivity-specificity (AUROC 0.879 [0.872 – 0.886] vs. 0.882 [0.877 – 0.887]). It statistically outperformed the CTAS system in precision-recall, achieving an AUPRC of 0.506 (0.484 – 0.528) compared to 0.333 (0.319 – 0.347). The random forest and XGBoost models achieved higher AUROC and AUPRC values than logistic regression and CTAS, underscoring their superior performance. XGBoost demonstrated the highest discrimination (AUROC: 0.917 [0.911 – 0.922]) and precision-recall (AUPRC: 0.629 [0.608 – 0.649]). SHAP of the models revealed that the top important features contributing to prediction were the mode of arrival, age, vital signs, and chief complaint, as illustrated in Figure 3. The waterfall plot in Figure 4 shows how individual sampled features influence the likelihood of ICU admission for a specific prediction. The model's average output or base value (E[f(x)]) adjusts incrementally based on the contributions of these features. The cumulative contributions from all features lead to a final log-odds of 2.812, which corresponds to a predicted probability of approximately 94.3% for ICU admission for the patient. Left: Feature importance summary plot.(y-axis: predictors, x-axis: mean of absolute SHAP value) Right: Waterfall plot illustrating how individual features affect the prediction. (y-axis: predictors, x-axis: log-odds of ICU admission) Discussion This study shows machine learning models' potential in predicting ICU admissions. Compared to traditional CTAS methods, these models achieved higher discriminative performance. This advantage is crucial in resource-limited settings where ICU beds are limited. The study found XGBoost and Random Forest to be the top-performing models, surpassing both the CTAS system and logistic regression due to their capacity to manage complex interactions and nonlinear relationships among predictors. One of this study's key findings is the importance of factors like mode of arrival, age, and chief complaint text in enhancing the model’s predictive performance. These factors are often overlooked in conventional triage systems and were potentially beneficial in improving the identification of patients at risk for ICU admission. However, using a mode of arrival also introduces potential bias when referral cases often have critical conditions requiring ICU care. Previous studies have shown that machine learning can improve triage outcomes using both structured and unstructured data, primarily in developed countries. 18 , 19 , 21 , 27 , 32 This study builds on those findings and demonstrates that such improvements are also achievable in resource-limited settings. Differences include excluding pain scores as they are subjective and unreliable for determining patient acuity. 33 Furthermore, the integration of NLP (Natural Language Processing) to analyze chief complaints in free text enabled the model to interpret textual data. Conventional triage systems, including ESI and CTAS, rely on chief complaints for categorization. The NLP approach can capture the subtle variations in clinical presentations, allowing for a broader categorization of chief complaints. Additionally, the use of multilingual embeddings effectively manages the linguistic diversity of clinical documentation in the local context, allowing the model to interpret text written in Thai with occasional English medical terms. However, reliance on free-text chief complaints introduces variability that could affect model prediction reliability. This study shows that data-driven tools can make emergency department decisions more effective. The model predicts ICU admission risks in real-time and can flag high-risk cases for triage nurses to quickly identify and focus on patients who might otherwise be under-triaged. The model also helps with managing resources. Its predictions can guide decisions on ICU bed allocation and assist hospitals in anticipating patient demand more accurately. Limitations However, several limitations should be acknowledged. First, the single-center nature of the study, focused on an Asian population, may limit its generalizability to other ethnic or geographic groups. Second, the outcome was limited to ICU admissions. Certain conditions, such as anaphylaxis or reactive airway disease, require immediate attention but may not result in ICU admission. In contrast, conditions associated with high mortality, such as unconsciousness, may lead to death in the ED rather than admission to the ICU. Outcomes such as emergency procedures, early mortality, or ED resource utilization could provide a more comprehensive evaluation of patient acuity. Third, the absence of detailed patient history as a predictor may have constrained the model’s discrimination performance. Prior medical information could significantly enhance prediction accuracy. Further research To enhance its generalizability, Future studies should validate this model with temporal and multi-center data, especially from resource-limited facilities. Expanding the dataset to include additional demographic variables, such as ethnicity and socioeconomic status, as well as incorporating patient history, could significantly improve model performance and address potential biases. Furthermore, extending the model’s capabilities to predict other outcomes, such as short-term in-hospital mortality, emergent procedure needs, and ED resource utilization (e.g., radiographs, blood tests, and medications), would broaden its clinical applicability. Also, improving the quality of free-text inputs should be prioritized. Encoding chief complaints into standardized categories using systems like the International Statistical Classification of Diseases and Related Health Problems 10th (ICD-10) or Systematized Nomenclature of Medicine Clinical Terms (SNOMED CT) could reduce variability, enhance data consistency, and boost overall model stability. Conclusion This study demonstrates that machine learning models leveraging structured and unstructured data can effectively predict ICU admission needs, achieving strong performance in AUROC and AUPRC metrics. The incorporation of free-text chief complaints and multilingual embeddings enhanced prediction accuracy, offering a practical and efficient alternative to traditional triage systems. Declarations Acknowledgment We thank Rudklao Sairai and the Research Unit of the Department of Emergency Medicine, Faculty of Medicine, Chiang Mai University, for their valuable support. We also sincerely appreciate Chiraphat Boonnag for expert guidance and Piyapong Khumrin for inspiring our enthusiasm in data science. This research was funded by the Faculty of Medicine, Chiang Mai University Research Fund (Grant No. INV08/2567). Author Contributions Statement P.S., B.W., W.S., and K.L. contributed to the study's conceptualization and methodology. P.S. and W.S. developed software, performed validation, and carried out formal analysis, with additional input from K.L. P.S. and W.S. conducted data curation. All authors contributed to the investigation. K.L. provided resources, supervised the project, and was the corresponding author. P.S. prepared all figures and tables. P.S., B.W., and W.S. wrote the original manuscript draft, and all authors reviewed and edited the manuscript. Conflict of interest The authors declare that there is no conflict of interest. Funding This work was supported by the Faculty of Medicine, Chiang Mai University [grant number: INV08/2567, 2024]. Data Availability The datasets generated and/or analyzed during this study are not publicly available due to hospital confidentiality policies and research grant restrictions but are available from the corresponding author upon reasonable request. References Higginson, I. Emergency department crowding. Emerg. Med. J. 29 (6), 437–443. 10.1136/emermed-2011-200532 (2012). Chen, W. et al. The effects of emergency department crowding on triage and hospital admission decisions. Am. J. Emerg. Med. 38 (4), 774–779. 10.1016/j.ajem.2019.06.039 (2020). Hoot, N. R. et al. Does crowding influence emergency department treatment time and disposition? J. Am. Coll. Emerg. Physicians Open. 2 (1), e12324. 10.1002/emp2.12324 (2021). Morley, C., Unwin, M., Peterson, G. M., Stankovich, J. & Kinsman, L. Emergency department crowding: A systematic review of causes, consequences and solutions. PLoS One . 13 (8), e0203316. 10.1371/journal.pone.0203316 (2018). Ruxin, T., Feldmeier, M., Addo, N. & Hsia, R. Y. Trends by Acuity for Emergency Department Visits and Hospital Admissions in California, 2012 to 2022. JAMA Netw. Open. 6 (12), e2348053. 10.1001/jamanetworkopen.2023.48053 (2023). Gilboy, N., Tanabe, P., Travers, D. A., Rosenau, A. M. & Eitel, D. R. Emergency severity index, version 4: implementation handbook. Rockv. MD: Agency Healthc. Res. Qual. ; 1 . (2005). Bullard, M. J. et al. Revisions to the Canadian Emergency Department Triage and Acuity Scale (CTAS) Guidelines 2016. CJEM 19 (S2), S18–S27. 10.1017/cem.2017.365 (2017). Zachariasse, J. M. et al. Performance of triage systems in emergency care: a systematic review and meta-analysis. BMJ Open. 9 (5), e026471. 10.1136/bmjopen-2018-026471 (2019). Tam, H. L., Chung, S. F. & Lou, C. K. A review of triage accuracy and future direction. BMC Emerg. Med. 18 (1), 58. 10.1186/s12873-018-0215-0 (2018). Chiu, I. M. et al. The influence of crowding on clinical practice in the emergency department. Am. J. Emerg. Med. 36 (1), 56–60. 10.1016/j.ajem.2017.07.011 (2018). Pines, J. M. & Hollander, J. E. Emergency department crowding is associated with poor care for patients with severe pain. Ann. Emerg. Med. 51 (1), 1–5. 10.1016/j.annemergmed.2007.07.008 (2008). Sun, B. C. et al. Effect of Emergency Department Crowding on Outcomes of Admitted Patients. Ann. Emerg. Med. 61 (6), 605–611e6. 10.1016/j.annemergmed.2012.10.026 (2013). Thirsk, L. M., Panchuk, J. T., Stahlke, S. & Hagtvedt, R. Cognitive and implicit biases in nurses’ judgment and decision-making: A scoping review. Int. J. Nurs. Stud. 133 , 104284. 10.1016/j.ijnurstu.2022.104284 (2022). Essa, C. D., Victor, G., Khan, S. F., Ally, H. & Khan, A. S. Cognitive biases regarding utilization of emergency severity index among emergency nurses. Am. J. Emerg. Med. 73 , 63–68. 10.1016/j.ajem.2023.08.021 (2023). Arslanian-Engoren, C. Gender and age bias in triage decisions. J. Emerg. Nurs. 26 (2), 117–124. 10.1016/S0099-1767(00)90053-9 (2000). Schrader, C. D. & Lewis, L. M. Racial Disparity in Emergency Department Triage. J. Emerg. Med. 44 (2), 511–518. 10.1016/j.jemermed.2012.05.010 (2013). Vigil, J. M. et al. Ethnic Disparities in Emergency Severity Index Scores among U.S. Veteran’s Affairs Emergency Department Patients. PLOS ONE . 10 (5), e0126792. 10.1371/journal.pone.0126792 (2015). Ivanov, O. et al. Improving ED Emergency Severity Index Acuity Assignment Using Machine Learning and Clinical Natural Language Processing. J. Emerg. Nurs. 47 (2), 265–278e7. 10.1016/j.jen.2020.11.001 (2021). Fernandes, M. et al. Predicting Intensive Care Unit admission among patients presenting to the emergency department using machine learning and natural language processing. PLoS One . 15 (3), e0229331. 10.1371/journal.pone.0229331 (2020). Sterling, N. W., Patzer, R. E., Di, M. & Schrager, J. D. Prediction of emergency department patient disposition based on natural language processing of triage notes. Int. J. Med. Informatics . 129 , 184–188. 10.1016/j.ijmedinf.2019.06.008 (2019). Zhang, X. et al. Prediction of Emergency Department Hospital Admission Based on Natural Language Processing and Neural Networks. Methods Inf. Med. 56 (05), 377–389. 10.3414/ME17-01-0024 (2017). Hong, W. S., Haimovich, A. D. & Taylor, R. A. Predicting hospital admission at emergency department triage using machine learning. PLoS One . 13 (7), e0201016. 10.1371/journal.pone.0201016 (2018). Raita, Y. et al. Emergency department triage prediction of clinical outcomes using machine learning models. Crit. Care . 23 (1), 64. 10.1186/s13054-019-2351-7 (2019). Joseph, J. W. et al. Deep-learning approaches to identify critically Ill patients at emergency department triage using limited information. J. Am. Coll. Emerg. Physicians Open. 1 (5), 773–781. 10.1002/emp2.12218 (2020). Consultant, H. Why unstructured data holds the key to intelligent healthcare systems [Internet]. Atlanta (GA): HIT Consultant . Published online 2015. Kong, H. J. Managing Unstructured Big Data in Healthcare System. Healthc. Inf. Res. 25 (1), 1–2. 10.4258/hir.2019.25.1.1 (2019). Tahayori, B., Chini-Foroush, N. & Akhlaghi, H. Advanced natural language processing technique to predict patient disposition based on emergency triage notes. Emerg. Med. Australas . 33 (3), 480–484. 10.1111/1742-6723.13656 (2021). Riley, R. D. et al. Minimum sample size for developing a multivariable prediction model: PART II - binary and time-to-event outcomes. Stat. Med. 38 (7), 1276–1296. 10.1002/sim.7992 (2019). Davis, J. & Goadrich, M. The relationship between Precision-Recall and ROC curves. In: Proceedings of the 23rd International Conference on Machine Learning - ICML ’06 . ACM Press; :233–240. (2006). 10.1145/1143844.1143874 Riley, R. D. & Collins, G. S. Stability of clinical prediction models developed using statistical or machine learning methods. Biom. J. 65 (8), 2200302. 10.1002/bimj.202200302 (2023). Riley, R. D. et al. Clinical prediction models and the multiverse of madness. BMC Med. 21 (1), 502. 10.1186/s12916-023-03212-y (2023). Grant, L. et al. Machine learning outperforms the Canadian Triage and Acuity Scale (CTAS) in predicting need for early critical care. Can. J. Emerg. Med. Published online November . 19 10.1007/s43678-024-00807-z (2024). Ueareekul, S., Changratanakorn, C., Tianwibool, P., Meelarp, N. & Wongtanasarasin, W. Accuracy of Pain Scales in Predicting Critical Diagnoses in Non-Traumatic Abdominal Pain Cases; a Cross-sectional Study. Arch. Acad. Emerg. Med. 11 (1), e68. 10.22037/aaem.v11i1.2131 (2023). Tables Table 1. Demographic data of the patient across different triage levels Variable Missing, n (%) Overall (n = 163,452) ICU Admission (n = 13,406) Non-ICU Admission (n = 150,046) Demographics Age, med (IQR) 0 (0.0) 48.0 (26.0-66.0) 61.0 (45.0 - 72.0) 46.0 (26.0 - 65.0) Sex - male, n (%) 0 (0.0) 79,558 (48.7) 8,493 (63.4) 71,065 (47.4) Mode of arrival 12403 (7.6) Walk-in, n (%) 116, 229 (77.0) 3,612 (28.4) 112,617 (81.4) EMS, n (%) 22,661 (15.0) 2,602 (20.5) 20,059 (14.5) Referral, n (%) 12,159 (8.1) 6,506 (51.2) 5,653 (4.1) Case type 0 (0.0) Trauma, n (%) 73,534 (45.0) 3,858 (28.8) 69,676 (46.4) Non-trauma, n (%) 89,918 (55.0) 9,548 (71.2) 80,370 (53.6) Vital signs Heart rate, med (IQR) 5647 (3.5) 88.0 (76.0 – 101.0) 91.0 (77.0 - 109.0) 88.0 (76.0 - 101.0) Respiratory rate, med (IQR) 7900 (4.8) 18.0 (18.0 - 20.0) 20.0 (18.0 - 24.0) 18.0 (18.0 - 20.0) SBP, med (IQR) 6106 (3.7) 132.0 (117.0 - 150.0) 132.0 (111.0 - 154.0) 132.0 (117.0 - 150.0) DBP, med (IQR) 6975 (4.3) 79.0 (69.0 - 90.0) 79.0 (65.0 - 93.0) 80.0 (69.0 - 90.0) O 2 saturation, med (IQR) 5600 (3.4) 98.0 (96.0 - 99.0) 98.0 (95.0 - 99.0) 98.0 (96.0 - 99.0) Temperature, med (IQR) 40409 (24.7) 36.6 (36.3 - 37.0) 36.6 (36.2 - 37.2) 36.6 (36.3 - 37.0) Level of consciousness GCS E < 4, n (%) 6660 (4.1) 5,799 (3.7) 2,699 (22.0) 3,100 (2.2) GCS V < 5, n (%) 7059 (4.3) 7,065 (4.5) 3,369 (27.8) 3,696 (2.6) GCS M < 6, n (%) 6770 (4.1) 5,171 (3.3) 2,531 (20.6) 2,640 (1.8) CTAS Triage 0 (0.0) Level 1, n (%) 9,285 (5.7) 4,639 (34.6) 4,646 (3.1) Level 2, n (%) 30,548 (18.7) 6,751 (50.4) 23,797 (15.9) Level 3, n (%) 52,677 (32.2) 1,867 (13.9) 50,810 (33.9) Level 4, n (%) 55,813 (34.2) 126 (0.9) 55,687 (37.1) Level 5, n (%) 15,129 (9.3) 23 (0.2) 15,106 (10.1) Abbreviations: CTAS, Canadian Triage and Acuity Scale; EMS, Emergency Medical Services; GCS, Glasgow Coma Scale; ICU, Intensive Care Unit; IQR, Interquartile Range; SBP, Systolic Blood Pressure; DBP, Diastolic Blood Pressure. Table 2. Comparison of predictive performance between CTAS triage and predicting models Metric XGBoost Random forest Logistic regression CTAS AUROC (95% CI) 0.917 (0.911 – 0.922) 0.904 (0.898– 0.911) 0.879 (0.872 – 0.885) 0.882 (0.877 - 0.887) AUPRC (95% CI) 0.629 (0.608 – 0.649) 0.609 (0.589 – 0.628) 0.506 (0.484 – 0.528) 0.333 (0.319 - 0.347) Abbreviations: AUPRC, Area under precision recall curve; AUROC, Area Under the Receiver Operating Characteristic Curve; CTAS, Canadian Triage and Acuity Scale; XGBoost, eXtreme Gradient Boosting Additional Declarations No competing interests reported. Cite Share Download PDF Status: Published Journal Publication published 25 Aug, 2025 Read the published version in Scientific Reports → Version 1 posted Editorial decision: Revision requested 25 Jun, 2025 Reviews received at journal 22 Jun, 2025 Reviewers agreed at journal 10 Jun, 2025 Reviews received at journal 05 May, 2025 Reviewers agreed at journal 25 Apr, 2025 Reviewers invited by journal 27 Mar, 2025 Editor assigned by journal 27 Mar, 2025 Editor invited by journal 19 Mar, 2025 Submission checks completed at journal 18 Mar, 2025 First submitted to journal 14 Mar, 2025 You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-6229836","acceptedTermsAndConditions":true,"allowDirectSubmit":false,"archivedVersions":[],"articleType":"Article","associatedPublications":[],"authors":[{"id":439168700,"identity":"def54ab2-05ff-47a4-a63d-bc12f8dba4f6","order_by":0,"name":"Patipan Sitthiprawiat","email":"","orcid":"","institution":"Chiang Mai University","correspondingAuthor":false,"prefix":"","firstName":"Patipan","middleName":"","lastName":"Sitthiprawiat","suffix":""},{"id":439168701,"identity":"26a4bc24-e3b2-4199-9b77-caae8b94edb1","order_by":1,"name":"Borwon Wittayachamnankul","email":"","orcid":"","institution":"Chiang Mai University","correspondingAuthor":false,"prefix":"","firstName":"Borwon","middleName":"","lastName":"Wittayachamnankul","suffix":""},{"id":439168702,"identity":"0a4252c6-bbfe-4371-a5f3-3d458b119679","order_by":2,"name":"Wachiranun Sirikul","email":"","orcid":"","institution":"Chiang Mai University","correspondingAuthor":false,"prefix":"","firstName":"Wachiranun","middleName":"","lastName":"Sirikul","suffix":""},{"id":439168703,"identity":"a0a6c4a9-a3ff-49e9-82e5-8b558d342a39","order_by":3,"name":"Korsin Laohavisudhi","email":"data:image/png;base64,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","orcid":"","institution":"Chiang Mai University","correspondingAuthor":true,"prefix":"","firstName":"Korsin","middleName":"","lastName":"Laohavisudhi","suffix":""}],"badges":[],"createdAt":"2025-03-15 02:08:07","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-6229836/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-6229836/v1","draftVersion":[],"editorialEvents":[{"content":"https://doi.org/10.1038/s41598-025-17180-1","type":"published","date":"2025-08-25T15:58:11+00:00"}],"editorialNote":"","failedWorkflow":false,"files":[{"id":80709474,"identity":"5010b211-ea5e-4e6d-b134-de941a950ade","added_by":"auto","created_at":"2025-04-16 08:55:30","extension":"jpg","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":25091,"visible":true,"origin":"","legend":"\u003cp\u003eInclusion \u0026amp; exclusion flowchart\u003c/p\u003e","description":"","filename":"Picture1.jpg","url":"https://assets-eu.researchsquare.com/files/rs-6229836/v1/f1f7759f6adacb37d286477a.jpg"},{"id":80709479,"identity":"0e8076af-1595-40bc-a900-dbce3ec7b45f","added_by":"auto","created_at":"2025-04-16 08:55:30","extension":"jpg","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":247325,"visible":true,"origin":"","legend":"\u003cp\u003eAUROC and AUPRC comparison between CTAS triage and the models\u003c/p\u003e","description":"","filename":"Picture2.jpg","url":"https://assets-eu.researchsquare.com/files/rs-6229836/v1/03111a135c42d19cf1b9608d.jpg"},{"id":80710958,"identity":"afc76bf2-2adf-46df-ae8b-6866c6ddc318","added_by":"auto","created_at":"2025-04-16 09:03:30","extension":"jpg","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":102255,"visible":true,"origin":"","legend":"\u003cp\u003eSHAP feature importance summary plot\u003c/p\u003e","description":"","filename":"Picture3.jpg","url":"https://assets-eu.researchsquare.com/files/rs-6229836/v1/931a89d80199effb6fe44309.jpg"},{"id":80712760,"identity":"b4a605ff-7463-4538-a87a-1ee20ab5e06b","added_by":"auto","created_at":"2025-04-16 09:19:30","extension":"jpg","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":95300,"visible":true,"origin":"","legend":"\u003cp\u003eSHAP waterfall plot on a sampled individual\u003c/p\u003e","description":"","filename":"Picture4.jpg","url":"https://assets-eu.researchsquare.com/files/rs-6229836/v1/5f5a80a26658f53501d18004.jpg"},{"id":90345561,"identity":"7d93a481-503b-4680-89f3-8e8b3117f0d0","added_by":"auto","created_at":"2025-09-01 16:10:33","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":1234264,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-6229836/v1/002151d6-4c77-470b-909f-a37f3db627dd.pdf"}],"financialInterests":"No competing interests reported.","formattedTitle":"Development and Validation of an AI-Based Emergency Triage Model for Predicting Critical Outcomes in Emergency Department","fulltext":[{"header":"Introduction","content":"\u003cp\u003eOver the past decade, visits to emergency departments (ED) have steadily increased globally, resulting in overcrowding that significantly reduces the quality of patient care and satisfaction.\u003csup\u003e\u003cspan additionalcitationids=\"CR2 CR3 CR4\" citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e\u003c/sup\u003e Triage, the initial process of identifying life-threatening conditions and prioritizing care, is vital for efficient resource allocation and optimal patient care in the ED. Standardized triage systems, such as the Emergency Severity Index (ESI)\u003csup\u003e\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e\u003c/sup\u003e and the Canadian Triage and Acuity Scale (CTAS), \u003csup\u003e\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e\u003c/sup\u003e exhibit variable accuracy, ranging from 59.2\u0026ndash;82.9%. These systems rely heavily on clinical judgment, which can lead to significant variability and suboptimal outcomes\u003csup\u003e\u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e,\u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e\u003c/sup\u003e. Inaccurate triage contributes to ED overcrowding, delays in care, and increased mortality risks.\u003csup\u003e\u003cspan additionalcitationids=\"CR11\" citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e\u003c/sup\u003e Over-triage puts a strain on resources, while under-triage impedes critical care.\u003csup\u003e\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e\u003c/sup\u003e Nurse-based triage is influenced by cognitive biases, including overconfidence, anchoring (the excessive reliance on initial information), and risk aversion, along with factors such as patient presentation, nurse experience, and the clinical environment. Disparities in outcomes related to ethnicity and sociodemographic factors raise concerns about fairness in ED care.\u003csup\u003e\u003cspan additionalcitationids=\"CR14 CR15 CR16\" citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e17\u003c/span\u003e\u003c/sup\u003e\u003c/p\u003e \u003cp\u003eTo address these limitations, several data-driven predictive models have been developed to improve triage accuracy and predict multiple clinical outcomes. These models incorporate predictors that are readily available at the time of triage, including vital signs, coded chief complaint, ambulance use, and medical history. The outcomes these models aim to predict include ESI levels, hospital admission, ICU admission, in-hospital mortality, critical conditions like sepsis and cardiovascular events, length of stay, readmission, and emergency department resource utilization (such as radiographs, medications, and laboratory tests) in adults.\u003csup\u003e\u003cspan additionalcitationids=\"CR19 CR20 CR21 CR22 CR23\" citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e24\u003c/span\u003e\u003c/sup\u003e\u003c/p\u003e \u003cp\u003eElectronic health records (EHRs) contain a significant amount of unstructured data, including clinical notes, representing approximately 80% of all health information for a patient \u003csup\u003e\u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e25\u003c/span\u003e,\u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e26\u003c/span\u003e\u003c/sup\u003e, such as free-text chief complaints found in triage notes. Natural language processing enables machine learning models to utilize these free texts by transforming them into various numerical features through semantic embedding.\u003csup\u003e\u003cspan additionalcitationids=\"CR20\" citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e,\u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e24\u003c/span\u003e,\u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e27\u003c/span\u003e\u003c/sup\u003e However, relatively few studies have adopted this method. This underscores a considerable gap in the current research. These free-text narratives frequently contain subtle details about patient presentations that structured data might not be able to capture. This study aims to develop a machine learning model that integrates structured data with free-text chief complaints documented by triage nurses, facilitating the triage process based on the need for critical care prediction. The goal is to reduce variability, improve predictive accuracy, and mitigate health disparities in triage outcomes through a real-time, clinically applicable model.\u003c/p\u003e"},{"header":"Methods","content":"\u003cdiv id=\"Sec3\" class=\"Section2\"\u003e \u003ch2\u003eStudy design and participants\u003c/h2\u003e \u003cp\u003eThis retrospective cohort study utilized data from the EHRs of Maharaj Nakhon Chiang Mai Hospital, a tertiary university hospital in northern Thailand with approximately 60,000 annual ED visits. Data were collected from January 1, 2018, to December 31, 2022. This study was approved by the Institutional Review Board (IRB) - Research Ethics Committee Panel 5 (Research ID: 0068 / Study code: EME-2566-0068), with a waiver of informed consent. All patient identifiers were removed before analysis. All methods were performed in accordance with relevant guidelines and regulations. We included consecutive adult patient visits (\u0026ge;\u0026thinsp;18 years) presenting to the ED with either trauma or non-trauma complaints, regardless of mode of arrival (walk-in, emergency medical services, or referral), and with recorded CTAS triage levels from 1 to 5 were assigned by a qualified emergency nurse. We excluded duplicated, missing triage labels, dead on arrival, left without being seen, transferred to another hospital, or missing ED disposition targets.\u003c/p\u003e \u003cp\u003eThe study included patient demographic and clinical data available during triage. Demographic variables included patient age and biological sex. Arrival characteristics included the mode of arrival (walk-in, emergency medical services, or referral) and the case type (trauma or non-trauma). Clinical data included the chief complaint, which was documented as free text, vital signs (heart rate, respiratory rate, blood pressure, oxygen saturation, and temperature), and the level of consciousness, measured using the Glasgow Coma Scale (GCS). The primary outcome was Intensive care unit (ICU) admission directly from the ED regardless of boarding time.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec4\" class=\"Section2\"\u003e \u003ch2\u003eData analysis\u003c/h2\u003e \u003cp\u003eSample size calculations were conducted using the \u003cem\u003epmsampsize\u003c/em\u003e module.\u003csup\u003e\u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e28\u003c/span\u003e\u003c/sup\u003e Based on a prior research study\u0026rsquo;s ICU admission prevalence of 0.8%\u003csup\u003e19\u003c/sup\u003e, a c-statistic of 0.85, a shrinkage factor of 0.9, and 15 predictive parameters, the required sample size was determined to be 15,713 resulting in 4.19 events per predictor variable.\u003c/p\u003e \u003cp\u003eThis study developed a machine learning-based model using logistic regression with a lasso penalty, random forest, and XGBoost (eXtreme Gradient Boosting). These models are commonly used in clinical predictive model development sorted by their complexity compared to a reference model based on the CTAS triage label. The included visits were stratified and randomized, split into a training set (80%) and a test set (20%) to preserve a balanced distribution of labels across groups. We tuned hyperparameters with a random search and a 5-fold cross-validation strategy.\u003c/p\u003e \u003cp\u003eThe missing input features were imputed using K-nearest neighbors. Other structured input features were handled by labeling, scaling, and standardization. The pre-trained Multilingual Universal Sentence Encoder converted chief complaints recorded as free text into semantic vector representations. This approach captures contextual and semantic meanings, enabling the model to understand free text. Then, Principal Component Analysis was used to reduce the complexity of the embeddings. We used SHapley Additive exPlanations (SHAP) to determine the importance of the feature of the predictor to the model.\u003c/p\u003e \u003cp\u003eModel performance was evaluated in a test set using the area under the receiver operating characteristic curve (AUROC) and the area under the precision-recall curve (AUPRC). While AUROC assesses discriminative performance, it can be less reliable with imbalanced datasets, whereas AUPRC emphasizes positive predictive value, offering a better measure in such cases.\u003csup\u003e\u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e29\u003c/span\u003e\u003c/sup\u003e The model performance was reported as mean and 95% confidence intervals generated using 1,000 bootstrapped samples from the test set. We used 500 bootstrapped samples from the test set for the prediction instability plot, mean absolute prediction error, and calibration plot to confirm the stability and reliability of predictions across datasets.\u003csup\u003e\u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e30\u003c/span\u003e,\u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e31\u003c/span\u003e\u003c/sup\u003e Analyses were performed using Python version 3.10.15.\u003c/p\u003e \u003c/div\u003e"},{"header":"Results","content":"\u003cp\u003e\u003cstrong\u003eParticipants\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThis study reviewed 172,791 patient visits during its duration. After these exclusions, the final cohort consisted of 163,452 visits, as shown in Figure 1.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eTable 1 describes the characteristics of patients in this study. Overall, 13,406 visits experienced ICU admission (8.2%). A total of 2,016 visits (1.2%) were triaged as CTAS level 3-5 and below and eventually admitted to the ICU. Chief complaint free text is not classified.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eModel performance\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe model performance metrics are shown in Table 2 and Figure 2. The logistic regression model with lasso penalty exhibited comparable discrimination in terms of sensitivity-specificity (AUROC 0.879 [0.872 \u0026ndash; 0.886] vs. 0.882 [0.877 \u0026ndash; 0.887]). It statistically outperformed the CTAS system in precision-recall, achieving an AUPRC of 0.506 (0.484 \u0026ndash; 0.528) compared to 0.333 (0.319 \u0026ndash; 0.347).\u003c/p\u003e\n\u003cp\u003eThe random forest and XGBoost models achieved higher AUROC and AUPRC values than logistic regression and CTAS, underscoring their superior performance. XGBoost demonstrated the highest discrimination (AUROC: 0.917 [0.911 \u0026ndash; 0.922]) and precision-recall (AUPRC: 0.629 [0.608 \u0026ndash;\u0026nbsp;\u0026nbsp;0.649]).\u003c/p\u003e\n\u003cp\u003eSHAP of the models revealed that the top important features contributing to prediction were the mode of arrival, age, vital signs, and chief complaint, as illustrated in Figure 3.\u0026nbsp;The waterfall plot in Figure 4 shows how individual sampled features influence the likelihood of ICU admission for a specific prediction. The model\u0026apos;s average output or base value (E[f(x)]) adjusts incrementally based on the contributions of these features. The cumulative contributions from all features lead to a final log-odds of\u0026nbsp;2.812, which corresponds to a predicted probability of approximately\u0026nbsp;94.3%\u0026nbsp;for ICU admission for the patient.\u003c/p\u003e\n\u003cp\u003eLeft: Feature importance summary plot.(y-axis: predictors, x-axis: mean of absolute SHAP value) Right: Waterfall plot illustrating how individual features affect the prediction. (y-axis: predictors, x-axis: log-odds of ICU admission)\u003c/p\u003e"},{"header":"Discussion","content":"\u003cp\u003eThis study shows machine learning models' potential in predicting ICU admissions. Compared to traditional CTAS methods, these models achieved higher discriminative performance. This advantage is crucial in resource-limited settings where ICU beds are limited. The study found XGBoost and Random Forest to be the top-performing models, surpassing both the CTAS system and logistic regression due to their capacity to manage complex interactions and nonlinear relationships among predictors.\u003c/p\u003e \u003cp\u003eOne of this study's key findings is the importance of factors like mode of arrival, age, and chief complaint text in enhancing the model\u0026rsquo;s predictive performance. These factors are often overlooked in conventional triage systems and were potentially beneficial in improving the identification of patients at risk for ICU admission. However, using a mode of arrival also introduces potential bias when referral cases often have critical conditions requiring ICU care.\u003c/p\u003e \u003cp\u003ePrevious studies have shown that machine learning can improve triage outcomes using both structured and unstructured data, primarily in developed countries.\u003csup\u003e\u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e,\u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e,\u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e,\u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e27\u003c/span\u003e,\u003cspan citationid=\"CR32\" class=\"CitationRef\"\u003e32\u003c/span\u003e\u003c/sup\u003e This study builds on those findings and demonstrates that such improvements are also achievable in resource-limited settings. Differences include excluding pain scores as they are subjective and unreliable for determining patient acuity.\u003csup\u003e\u003cspan citationid=\"CR33\" class=\"CitationRef\"\u003e33\u003c/span\u003e\u003c/sup\u003e Furthermore, the integration of NLP (Natural Language Processing) to analyze chief complaints in free text enabled the model to interpret textual data. Conventional triage systems, including ESI and CTAS, rely on chief complaints for categorization. The NLP approach can capture the subtle variations in clinical presentations, allowing for a broader categorization of chief complaints. Additionally, the use of multilingual embeddings effectively manages the linguistic diversity of clinical documentation in the local context, allowing the model to interpret text written in Thai with occasional English medical terms. However, reliance on free-text chief complaints introduces variability that could affect model prediction reliability.\u003c/p\u003e \u003cp\u003eThis study shows that data-driven tools can make emergency department decisions more effective. The model predicts ICU admission risks in real-time and can flag high-risk cases for triage nurses to quickly identify and focus on patients who might otherwise be under-triaged. The model also helps with managing resources. Its predictions can guide decisions on ICU bed allocation and assist hospitals in anticipating patient demand more accurately.\u003c/p\u003e\n\u003ch3\u003eLimitations\u003c/h3\u003e\n\u003cp\u003eHowever, several limitations should be acknowledged. First, the single-center nature of the study, focused on an Asian population, may limit its generalizability to other ethnic or geographic groups. Second, the outcome was limited to ICU admissions. Certain conditions, such as anaphylaxis or reactive airway disease, require immediate attention but may not result in ICU admission. In contrast, conditions associated with high mortality, such as unconsciousness, may lead to death in the ED rather than admission to the ICU. Outcomes such as emergency procedures, early mortality, or ED resource utilization could provide a more comprehensive evaluation of patient acuity. Third, the absence of detailed patient history as a predictor may have constrained the model\u0026rsquo;s discrimination performance. Prior medical information could significantly enhance prediction accuracy.\u003c/p\u003e\n\u003ch3\u003eFurther research\u003c/h3\u003e\n\u003cp\u003eTo enhance its generalizability, Future studies should validate this model with temporal and multi-center data, especially from resource-limited facilities. Expanding the dataset to include additional demographic variables, such as ethnicity and socioeconomic status, as well as incorporating patient history, could significantly improve model performance and address potential biases. Furthermore, extending the model\u0026rsquo;s capabilities to predict other outcomes, such as short-term in-hospital mortality, emergent procedure needs, and ED resource utilization (e.g., radiographs, blood tests, and medications), would broaden its clinical applicability. Also, improving the quality of free-text inputs should be prioritized. Encoding chief complaints into standardized categories using systems like the International Statistical Classification of Diseases and Related Health Problems 10th (ICD-10) or Systematized Nomenclature of Medicine Clinical Terms (SNOMED CT) could reduce variability, enhance data consistency, and boost overall model stability.\u003c/p\u003e"},{"header":"Conclusion","content":"\u003cp\u003eThis study demonstrates that machine learning models leveraging structured and unstructured data can effectively predict ICU admission needs, achieving strong performance in AUROC and AUPRC metrics. The incorporation of free-text chief complaints and multilingual embeddings enhanced prediction accuracy, offering a practical and efficient alternative to traditional triage systems.\u003c/p\u003e"},{"header":"Declarations","content":"\u003cp\u003e\u003cstrong\u003eAcknowledgment\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e\u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp;\u0026nbsp;We thank Rudklao Sairai and the Research Unit of the Department of Emergency Medicine, Faculty of Medicine, Chiang Mai University, for their valuable support. We also sincerely appreciate Chiraphat Boonnag for expert guidance and Piyapong Khumrin for inspiring our enthusiasm in data science. This research was funded by the Faculty of Medicine, Chiang Mai University Research Fund (Grant No. INV08/2567).\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAuthor Contributions Statement\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eP.S., B.W., W.S., and K.L. contributed to the study\u0026apos;s conceptualization and methodology. P.S. and W.S. developed software, performed validation, and carried out formal analysis, with additional input from K.L. P.S. and W.S. conducted data curation. All authors contributed to the investigation. K.L. provided resources, supervised the project, and was the corresponding author. P.S. prepared all figures and tables. P.S., B.W., and W.S. wrote the original manuscript draft, and all authors reviewed and edited the manuscript.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eConflict of interest\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e\u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp;\u0026nbsp;The authors declare that there is no conflict of interest.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eFunding\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThis work was supported by the Faculty of Medicine, Chiang Mai University [grant number: INV08/2567, 2024].\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eData Availability\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe datasets generated and/or analyzed during this study are not publicly available due to hospital confidentiality policies and research grant restrictions but are available from the corresponding author upon reasonable request.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\u003cli\u003e\u003cspan\u003eHigginson, I. Emergency department crowding. \u003cem\u003eEmerg. Med. J.\u003c/em\u003e \u003cb\u003e29\u003c/b\u003e (6), 437\u0026ndash;443. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1136/emermed-2011-200532\u003c/span\u003e\u003cspan address=\"10.1136/emermed-2011-200532\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2012).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eChen, W. et al. The effects of emergency department crowding on triage and hospital admission decisions. \u003cem\u003eAm. J. Emerg. Med.\u003c/em\u003e \u003cb\u003e38\u003c/b\u003e (4), 774\u0026ndash;779. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1016/j.ajem.2019.06.039\u003c/span\u003e\u003cspan address=\"10.1016/j.ajem.2019.06.039\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2020).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHoot, N. R. et al. Does crowding influence emergency department treatment time and disposition? \u003cem\u003eJ. Am. Coll. Emerg. Physicians Open.\u003c/em\u003e \u003cb\u003e2\u003c/b\u003e (1), e12324. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1002/emp2.12324\u003c/span\u003e\u003cspan address=\"10.1002/emp2.12324\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2021).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMorley, C., Unwin, M., Peterson, G. M., Stankovich, J. \u0026amp; Kinsman, L. Emergency department crowding: A systematic review of causes, consequences and solutions. \u003cem\u003ePLoS One\u003c/em\u003e. \u003cb\u003e13\u003c/b\u003e (8), e0203316. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1371/journal.pone.0203316\u003c/span\u003e\u003cspan address=\"10.1371/journal.pone.0203316\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2018).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eRuxin, T., Feldmeier, M., Addo, N. \u0026amp; Hsia, R. Y. Trends by Acuity for Emergency Department Visits and Hospital Admissions in California, 2012 to 2022. \u003cem\u003eJAMA Netw. Open.\u003c/em\u003e \u003cb\u003e6\u003c/b\u003e (12), e2348053. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1001/jamanetworkopen.2023.48053\u003c/span\u003e\u003cspan address=\"10.1001/jamanetworkopen.2023.48053\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2023).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eGilboy, N., Tanabe, P., Travers, D. A., Rosenau, A. M. \u0026amp; Eitel, D. R. Emergency severity index, version 4: implementation handbook. \u003cem\u003eRockv. MD: Agency Healthc. Res. Qual.\u003c/em\u003e ;\u003cb\u003e1\u003c/b\u003e. (2005).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBullard, M. J. et al. Revisions to the Canadian Emergency Department Triage and Acuity Scale (CTAS) Guidelines 2016. \u003cem\u003eCJEM\u003c/em\u003e \u003cb\u003e19\u003c/b\u003e (S2), S18\u0026ndash;S27. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1017/cem.2017.365\u003c/span\u003e\u003cspan address=\"10.1017/cem.2017.365\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2017).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eZachariasse, J. M. et al. Performance of triage systems in emergency care: a systematic review and meta-analysis. \u003cem\u003eBMJ Open.\u003c/em\u003e \u003cb\u003e9\u003c/b\u003e (5), e026471. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1136/bmjopen-2018-026471\u003c/span\u003e\u003cspan address=\"10.1136/bmjopen-2018-026471\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2019).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eTam, H. L., Chung, S. F. \u0026amp; Lou, C. K. A review of triage accuracy and future direction. \u003cem\u003eBMC Emerg. Med.\u003c/em\u003e \u003cb\u003e18\u003c/b\u003e (1), 58. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1186/s12873-018-0215-0\u003c/span\u003e\u003cspan address=\"10.1186/s12873-018-0215-0\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2018).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eChiu, I. M. et al. The influence of crowding on clinical practice in the emergency department. \u003cem\u003eAm. J. Emerg. Med.\u003c/em\u003e \u003cb\u003e36\u003c/b\u003e (1), 56\u0026ndash;60. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1016/j.ajem.2017.07.011\u003c/span\u003e\u003cspan address=\"10.1016/j.ajem.2017.07.011\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2018).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003ePines, J. M. \u0026amp; Hollander, J. E. Emergency department crowding is associated with poor care for patients with severe pain. \u003cem\u003eAnn. Emerg. Med.\u003c/em\u003e \u003cb\u003e51\u003c/b\u003e (1), 1\u0026ndash;5. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1016/j.annemergmed.2007.07.008\u003c/span\u003e\u003cspan address=\"10.1016/j.annemergmed.2007.07.008\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2008).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSun, B. C. et al. Effect of Emergency Department Crowding on Outcomes of Admitted Patients. \u003cem\u003eAnn. Emerg. Med.\u003c/em\u003e \u003cb\u003e61\u003c/b\u003e (6), 605\u0026ndash;611e6. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1016/j.annemergmed.2012.10.026\u003c/span\u003e\u003cspan address=\"10.1016/j.annemergmed.2012.10.026\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2013).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eThirsk, L. M., Panchuk, J. T., Stahlke, S. \u0026amp; Hagtvedt, R. Cognitive and implicit biases in nurses\u0026rsquo; judgment and decision-making: A scoping review. \u003cem\u003eInt. J. Nurs. Stud.\u003c/em\u003e \u003cb\u003e133\u003c/b\u003e, 104284. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1016/j.ijnurstu.2022.104284\u003c/span\u003e\u003cspan address=\"10.1016/j.ijnurstu.2022.104284\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2022).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eEssa, C. D., Victor, G., Khan, S. F., Ally, H. \u0026amp; Khan, A. S. Cognitive biases regarding utilization of emergency severity index among emergency nurses. \u003cem\u003eAm. J. Emerg. Med.\u003c/em\u003e \u003cb\u003e73\u003c/b\u003e, 63\u0026ndash;68. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1016/j.ajem.2023.08.021\u003c/span\u003e\u003cspan address=\"10.1016/j.ajem.2023.08.021\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2023).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eArslanian-Engoren, C. Gender and age bias in triage decisions. \u003cem\u003eJ. Emerg. Nurs.\u003c/em\u003e \u003cb\u003e26\u003c/b\u003e (2), 117\u0026ndash;124. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1016/S0099-1767(00)90053-9\u003c/span\u003e\u003cspan address=\"10.1016/S0099-1767(00)90053-9\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2000).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSchrader, C. D. \u0026amp; Lewis, L. M. Racial Disparity in Emergency Department Triage. \u003cem\u003eJ. Emerg. Med.\u003c/em\u003e \u003cb\u003e44\u003c/b\u003e (2), 511\u0026ndash;518. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1016/j.jemermed.2012.05.010\u003c/span\u003e\u003cspan address=\"10.1016/j.jemermed.2012.05.010\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2013).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eVigil, J. M. et al. Ethnic Disparities in Emergency Severity Index Scores among U.S. Veteran\u0026rsquo;s Affairs Emergency Department Patients. \u003cem\u003ePLOS ONE\u003c/em\u003e. \u003cb\u003e10\u003c/b\u003e (5), e0126792. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1371/journal.pone.0126792\u003c/span\u003e\u003cspan address=\"10.1371/journal.pone.0126792\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2015).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eIvanov, O. et al. Improving ED Emergency Severity Index Acuity Assignment Using Machine Learning and Clinical Natural Language Processing. \u003cem\u003eJ. Emerg. Nurs.\u003c/em\u003e \u003cb\u003e47\u003c/b\u003e (2), 265\u0026ndash;278e7. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1016/j.jen.2020.11.001\u003c/span\u003e\u003cspan address=\"10.1016/j.jen.2020.11.001\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2021).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eFernandes, M. et al. Predicting Intensive Care Unit admission among patients presenting to the emergency department using machine learning and natural language processing. \u003cem\u003ePLoS One\u003c/em\u003e. \u003cb\u003e15\u003c/b\u003e (3), e0229331. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1371/journal.pone.0229331\u003c/span\u003e\u003cspan address=\"10.1371/journal.pone.0229331\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2020).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSterling, N. W., Patzer, R. E., Di, M. \u0026amp; Schrager, J. D. Prediction of emergency department patient disposition based on natural language processing of triage notes. \u003cem\u003eInt. J. Med. Informatics\u003c/em\u003e. \u003cb\u003e129\u003c/b\u003e, 184\u0026ndash;188. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1016/j.ijmedinf.2019.06.008\u003c/span\u003e\u003cspan address=\"10.1016/j.ijmedinf.2019.06.008\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2019).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eZhang, X. et al. Prediction of Emergency Department Hospital Admission Based on Natural Language Processing and Neural Networks. \u003cem\u003eMethods Inf. Med.\u003c/em\u003e \u003cb\u003e56\u003c/b\u003e (05), 377\u0026ndash;389. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.3414/ME17-01-0024\u003c/span\u003e\u003cspan address=\"10.3414/ME17-01-0024\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2017).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHong, W. S., Haimovich, A. D. \u0026amp; Taylor, R. A. Predicting hospital admission at emergency department triage using machine learning. \u003cem\u003ePLoS One\u003c/em\u003e. \u003cb\u003e13\u003c/b\u003e (7), e0201016. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1371/journal.pone.0201016\u003c/span\u003e\u003cspan address=\"10.1371/journal.pone.0201016\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2018).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eRaita, Y. et al. Emergency department triage prediction of clinical outcomes using machine learning models. \u003cem\u003eCrit. Care\u003c/em\u003e. \u003cb\u003e23\u003c/b\u003e (1), 64. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1186/s13054-019-2351-7\u003c/span\u003e\u003cspan address=\"10.1186/s13054-019-2351-7\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2019).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eJoseph, J. W. et al. Deep-learning approaches to identify critically Ill patients at emergency department triage using limited information. \u003cem\u003eJ. Am. Coll. Emerg. Physicians Open.\u003c/em\u003e \u003cb\u003e1\u003c/b\u003e (5), 773\u0026ndash;781. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1002/emp2.12218\u003c/span\u003e\u003cspan address=\"10.1002/emp2.12218\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2020).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eConsultant, H. Why unstructured data holds the key to intelligent healthcare systems [Internet]. \u003cem\u003eAtlanta (GA): HIT Consultant\u003c/em\u003e. Published online 2015.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKong, H. J. Managing Unstructured Big Data in Healthcare System. \u003cem\u003eHealthc. Inf. Res.\u003c/em\u003e \u003cb\u003e25\u003c/b\u003e (1), 1\u0026ndash;2. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.4258/hir.2019.25.1.1\u003c/span\u003e\u003cspan address=\"10.4258/hir.2019.25.1.1\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2019).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eTahayori, B., Chini-Foroush, N. \u0026amp; Akhlaghi, H. Advanced natural language processing technique to predict patient disposition based on emergency triage notes. \u003cem\u003eEmerg. Med. Australas\u003c/em\u003e. \u003cb\u003e33\u003c/b\u003e (3), 480\u0026ndash;484. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1111/1742-6723.13656\u003c/span\u003e\u003cspan address=\"10.1111/1742-6723.13656\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2021).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eRiley, R. D. et al. Minimum sample size for developing a multivariable prediction model: PART II - binary and time-to-event outcomes. \u003cem\u003eStat. Med.\u003c/em\u003e \u003cb\u003e38\u003c/b\u003e (7), 1276\u0026ndash;1296. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1002/sim.7992\u003c/span\u003e\u003cspan address=\"10.1002/sim.7992\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2019).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eDavis, J. \u0026amp; Goadrich, M. The relationship between Precision-Recall and ROC curves. In: \u003cem\u003eProceedings of the 23rd International Conference on Machine Learning - ICML \u0026rsquo;06\u003c/em\u003e. ACM Press; :233\u0026ndash;240. (2006). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1145/1143844.1143874\u003c/span\u003e\u003cspan address=\"10.1145/1143844.1143874\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eRiley, R. D. \u0026amp; Collins, G. S. Stability of clinical prediction models developed using statistical or machine learning methods. \u003cem\u003eBiom. J.\u003c/em\u003e \u003cb\u003e65\u003c/b\u003e (8), 2200302. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1002/bimj.202200302\u003c/span\u003e\u003cspan address=\"10.1002/bimj.202200302\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2023).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eRiley, R. D. et al. Clinical prediction models and the multiverse of madness. \u003cem\u003eBMC Med.\u003c/em\u003e \u003cb\u003e21\u003c/b\u003e (1), 502. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1186/s12916-023-03212-y\u003c/span\u003e\u003cspan address=\"10.1186/s12916-023-03212-y\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2023).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eGrant, L. et al. Machine learning outperforms the Canadian Triage and Acuity Scale (CTAS) in predicting need for early critical care. \u003cem\u003eCan. J. Emerg. Med. Published online November\u003c/em\u003e. \u003cb\u003e19\u003c/b\u003e \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1007/s43678-024-00807-z\u003c/span\u003e\u003cspan address=\"10.1007/s43678-024-00807-z\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2024).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eUeareekul, S., Changratanakorn, C., Tianwibool, P., Meelarp, N. \u0026amp; Wongtanasarasin, W. Accuracy of Pain Scales in Predicting Critical Diagnoses in Non-Traumatic Abdominal Pain Cases; a Cross-sectional Study. \u003cem\u003eArch. Acad. Emerg. Med.\u003c/em\u003e \u003cb\u003e11\u003c/b\u003e (1), e68. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.22037/aaem.v11i1.2131\u003c/span\u003e\u003cspan address=\"10.22037/aaem.v11i1.2131\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2023).\u003c/span\u003e\u003c/li\u003e\u003c/ol\u003e"},{"header":"Tables","content":"\u003ctable border=\"0\" cellspacing=\"0\" cellpadding=\"0\" width=\"665\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd colspan=\"5\" valign=\"top\" style=\"width: 97.2932%;\"\u003e\n \u003cp\u003e\u003cstrong\u003eTable 1. Demographic data of the patient across different triage levels\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 2.70677%;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 25.5639%;\"\u003e\n \u003cp\u003e\u003cstrong\u003eVariable\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 14.2857%;\"\u003e\n \u003cp\u003e\u003cstrong\u003eMissing,\u003c/strong\u003e\u003c/p\u003e\n \u003cp\u003e\u003cstrong\u003en (%)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 19.3985%;\"\u003e\n \u003cp\u003e\u003cstrong\u003eOverall\u003c/strong\u003e\u003c/p\u003e\n \u003cp\u003e\u003cstrong\u003e(n = 163,452)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 20.9023%;\"\u003e\n \u003cp\u003e\u003cstrong\u003eICU Admission\u003c/strong\u003e\u003c/p\u003e\n \u003cp\u003e\u003cstrong\u003e(n = 13,406)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" valign=\"top\" style=\"width: 19.8496%;\"\u003e\n \u003cp\u003e\u003cstrong\u003eNon-ICU Admission\u003c/strong\u003e\u003c/p\u003e\n \u003cp\u003e\u003cstrong\u003e(n = 150,046)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 25.5639%;\"\u003e\n \u003cp\u003e\u003cstrong\u003eDemographics\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 14.2857%;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 19.3985%;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 20.9023%;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" valign=\"top\" style=\"width: 19.8496%;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 25.5639%;\"\u003e\n \u003cp\u003eAge, med (IQR)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 14.2857%;\"\u003e\n \u003cp\u003e0 (0.0)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 19.3985%;\"\u003e\n \u003cp\u003e48.0 (26.0-66.0)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 20.9023%;\"\u003e\n \u003cp\u003e61.0 (45.0 - 72.0)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" valign=\"top\" style=\"width: 19.8496%;\"\u003e\n \u003cp\u003e46.0 (26.0 - 65.0)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 25.5639%;\"\u003e\n \u003cp\u003eSex - male, n (%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 14.2857%;\"\u003e\n \u003cp\u003e0 (0.0)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 19.3985%;\"\u003e\n \u003cp\u003e79,558 (48.7)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 20.9023%;\"\u003e\n \u003cp\u003e8,493 (63.4)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" valign=\"top\" style=\"width: 19.8496%;\"\u003e\n \u003cp\u003e71,065 (47.4)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 25.5639%;\"\u003e\n \u003cp\u003e\u003cstrong\u003eMode of arrival\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 14.2857%;\"\u003e\n \u003cp\u003e12403 (7.6)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 19.3985%;\"\u003e\n \u003cp\u003e\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 20.9023%;\"\u003e\n \u003cp\u003e\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" valign=\"top\" style=\"width: 19.8496%;\"\u003e\n \u003cp\u003e\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 25.5639%;\"\u003e\n \u003cp\u003eWalk-in, n (%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 14.2857%;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 19.3985%;\"\u003e\n \u003cp\u003e116, 229 (77.0)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 20.9023%;\"\u003e\n \u003cp\u003e3,612 (28.4)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" valign=\"top\" style=\"width: 19.8496%;\"\u003e\n \u003cp\u003e112,617 (81.4)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 25.5639%;\"\u003e\n \u003cp\u003eEMS, n (%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 14.2857%;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 19.3985%;\"\u003e\n \u003cp\u003e22,661 (15.0)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 20.9023%;\"\u003e\n \u003cp\u003e2,602 (20.5)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" valign=\"top\" style=\"width: 19.8496%;\"\u003e\n \u003cp\u003e20,059 (14.5)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 25.5639%;\"\u003e\n \u003cp\u003eReferral, n (%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 14.2857%;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 19.3985%;\"\u003e\n \u003cp\u003e12,159 (8.1)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 20.9023%;\"\u003e\n \u003cp\u003e6,506 (51.2)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" valign=\"top\" style=\"width: 19.8496%;\"\u003e\n \u003cp\u003e5,653 (4.1)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 25.5639%;\"\u003e\n \u003cp\u003e\u003cstrong\u003eCase type\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 14.2857%;\"\u003e\n \u003cp\u003e0 (0.0)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 19.3985%;\"\u003e\n \u003cp\u003e\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 20.9023%;\"\u003e\n \u003cp\u003e\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" valign=\"top\" style=\"width: 19.8496%;\"\u003e\n \u003cp\u003e\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 25.5639%;\"\u003e\n \u003cp\u003eTrauma, n (%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 14.2857%;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 19.3985%;\"\u003e\n \u003cp\u003e73,534 (45.0)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 20.9023%;\"\u003e\n \u003cp\u003e3,858 (28.8)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" valign=\"top\" style=\"width: 19.8496%;\"\u003e\n \u003cp\u003e69,676 (46.4)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 25.5639%;\"\u003e\n \u003cp\u003eNon-trauma, n (%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 14.2857%;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 19.3985%;\"\u003e\n \u003cp\u003e89,918 (55.0)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 20.9023%;\"\u003e\n \u003cp\u003e9,548 (71.2)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" valign=\"top\" style=\"width: 19.8496%;\"\u003e\n \u003cp\u003e80,370 (53.6)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 25.5639%;\"\u003e\n \u003cp\u003e\u003cstrong\u003eVital signs\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 14.2857%;\"\u003e\n \u003cp\u003e\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 19.3985%;\"\u003e\n \u003cp\u003e\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 20.9023%;\"\u003e\n \u003cp\u003e\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" valign=\"top\" style=\"width: 19.8496%;\"\u003e\n \u003cp\u003e\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 25.5639%;\"\u003e\n \u003cp\u003eHeart rate, med (IQR)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 14.2857%;\"\u003e\n \u003cp\u003e5647 (3.5)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 19.3985%;\"\u003e\n \u003cp\u003e88.0 (76.0 \u0026ndash; 101.0)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 20.9023%;\"\u003e\n \u003cp\u003e91.0 (77.0 - 109.0)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" valign=\"top\" style=\"width: 19.8496%;\"\u003e\n \u003cp\u003e88.0 (76.0 - 101.0)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 25.5639%;\"\u003e\n \u003cp\u003eRespiratory rate, med (IQR)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 14.2857%;\"\u003e\n \u003cp\u003e7900 (4.8)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 19.3985%;\"\u003e\n \u003cp\u003e18.0 (18.0 - 20.0)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 20.9023%;\"\u003e\n \u003cp\u003e20.0 (18.0 - 24.0)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" valign=\"top\" style=\"width: 19.8496%;\"\u003e\n \u003cp\u003e18.0 (18.0 - 20.0)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 25.5639%;\"\u003e\n \u003cp\u003eSBP, med (IQR)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 14.2857%;\"\u003e\n \u003cp\u003e6106 (3.7)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 19.3985%;\"\u003e\n \u003cp\u003e132.0 (117.0 - 150.0)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 20.9023%;\"\u003e\n \u003cp\u003e132.0 (111.0 - 154.0)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" valign=\"top\" style=\"width: 19.8496%;\"\u003e\n \u003cp\u003e132.0 (117.0 - 150.0)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 25.5639%;\"\u003e\n \u003cp\u003eDBP, med (IQR)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 14.2857%;\"\u003e\n \u003cp\u003e6975 (4.3)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 19.3985%;\"\u003e\n \u003cp\u003e79.0 (69.0 - 90.0)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 20.9023%;\"\u003e\n \u003cp\u003e79.0 (65.0 - 93.0)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" valign=\"top\" style=\"width: 19.8496%;\"\u003e\n \u003cp\u003e80.0 (69.0 - 90.0)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 25.5639%;\"\u003e\n \u003cp\u003eO\u003csub\u003e2\u003c/sub\u003e saturation, med (IQR)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 14.2857%;\"\u003e\n \u003cp\u003e5600 (3.4)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 19.3985%;\"\u003e\n \u003cp\u003e98.0 (96.0 - 99.0)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 20.9023%;\"\u003e\n \u003cp\u003e98.0 (95.0 - 99.0)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" valign=\"top\" style=\"width: 19.8496%;\"\u003e\n \u003cp\u003e98.0 (96.0 - 99.0)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 25.5639%;\"\u003e\n \u003cp\u003eTemperature, med (IQR)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 14.2857%;\"\u003e\n \u003cp\u003e40409 (24.7)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 19.3985%;\"\u003e\n \u003cp\u003e36.6 (36.3 - 37.0)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 20.9023%;\"\u003e\n \u003cp\u003e36.6 (36.2 - 37.2)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" valign=\"top\" style=\"width: 19.8496%;\"\u003e\n \u003cp\u003e36.6 (36.3 - 37.0)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 25.5639%;\"\u003e\n \u003cp\u003e\u003cstrong\u003eLevel of consciousness\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 14.2857%;\"\u003e\n \u003cp\u003e\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 19.3985%;\"\u003e\n \u003cp\u003e\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 20.9023%;\"\u003e\n \u003cp\u003e\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" valign=\"top\" style=\"width: 19.8496%;\"\u003e\n \u003cp\u003e\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 25.5639%;\"\u003e\n \u003cp\u003eGCS E \u0026lt; 4, n (%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 14.2857%;\"\u003e\n \u003cp\u003e6660 (4.1)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 19.3985%;\"\u003e\n \u003cp\u003e5,799 (3.7)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 20.9023%;\"\u003e\n \u003cp\u003e2,699 (22.0)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" valign=\"top\" style=\"width: 19.8496%;\"\u003e\n \u003cp\u003e3,100 (2.2)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 25.5639%;\"\u003e\n \u003cp\u003eGCS V \u0026lt; 5, n (%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 14.2857%;\"\u003e\n \u003cp\u003e7059 (4.3)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 19.3985%;\"\u003e\n \u003cp\u003e7,065 (4.5)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 20.9023%;\"\u003e\n \u003cp\u003e3,369 (27.8)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" valign=\"top\" style=\"width: 19.8496%;\"\u003e\n \u003cp\u003e3,696 (2.6)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 25.5639%;\"\u003e\n \u003cp\u003eGCS M \u0026lt; 6, n (%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 14.2857%;\"\u003e\n \u003cp\u003e6770 (4.1)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 19.3985%;\"\u003e\n \u003cp\u003e5,171 (3.3)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 20.9023%;\"\u003e\n \u003cp\u003e2,531 (20.6)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" valign=\"top\" style=\"width: 19.8496%;\"\u003e\n \u003cp\u003e2,640 (1.8)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 25.5639%;\"\u003e\n \u003cp\u003e\u003cstrong\u003eCTAS Triage\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 14.2857%;\"\u003e\n \u003cp\u003e0 (0.0)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 19.3985%;\"\u003e\n \u003cp\u003e\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 20.9023%;\"\u003e\n \u003cp\u003e\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" valign=\"top\" style=\"width: 19.8496%;\"\u003e\n \u003cp\u003e\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 25.5639%;\"\u003e\n \u003cp\u003eLevel 1, n (%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 14.2857%;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 19.3985%;\"\u003e\n \u003cp\u003e9,285 (5.7)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 20.9023%;\"\u003e\n \u003cp\u003e4,639 (34.6)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" valign=\"top\" style=\"width: 19.8496%;\"\u003e\n \u003cp\u003e4,646 (3.1)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 25.5639%;\"\u003e\n \u003cp\u003eLevel 2, n (%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 14.2857%;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 19.3985%;\"\u003e\n \u003cp\u003e30,548 (18.7)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 20.9023%;\"\u003e\n \u003cp\u003e6,751 (50.4)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" valign=\"top\" style=\"width: 19.8496%;\"\u003e\n \u003cp\u003e23,797 (15.9)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 25.5639%;\"\u003e\n \u003cp\u003eLevel 3, n (%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 14.2857%;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 19.3985%;\"\u003e\n \u003cp\u003e52,677 (32.2)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 20.9023%;\"\u003e\n \u003cp\u003e1,867 (13.9)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" valign=\"top\" style=\"width: 19.8496%;\"\u003e\n \u003cp\u003e50,810 (33.9)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 25.5639%;\"\u003e\n \u003cp\u003eLevel 4, n (%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 14.2857%;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 19.3985%;\"\u003e\n \u003cp\u003e55,813 (34.2)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 20.9023%;\"\u003e\n \u003cp\u003e126 (0.9)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" valign=\"top\" style=\"width: 19.8496%;\"\u003e\n \u003cp\u003e55,687 (37.1)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 25.5639%;\"\u003e\n \u003cp\u003eLevel 5, n (%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 14.2857%;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 19.3985%;\"\u003e\n \u003cp\u003e15,129 (9.3)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 20.9023%;\"\u003e\n \u003cp\u003e23 (0.2)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" valign=\"top\" style=\"width: 19.8496%;\"\u003e\n \u003cp\u003e15,106 (10.1)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd colspan=\"5\" valign=\"top\" style=\"width: 97.2932%;\"\u003e\n \u003cp\u003e\u003cstrong\u003eAbbreviations:\u003c/strong\u003e CTAS, Canadian Triage and Acuity Scale; EMS, Emergency Medical Services; GCS, Glasgow Coma Scale; ICU, Intensive Care Unit; IQR, Interquartile Range; SBP, Systolic Blood Pressure; DBP, Diastolic Blood Pressure.\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 2.70677%;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e\n\u003cp\u003e\u003cbr\u003e\u003c/p\u003e\n\u003ctable border=\"1\" cellspacing=\"0\" cellpadding=\"0\" width=\"662\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd colspan=\"5\" valign=\"top\" style=\"width: 662px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eTable 2. Comparison of predictive performance between CTAS triage and predicting models\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 123px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eMetric\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 132px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eXGBoost\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 132px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eRandom forest\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 132px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eLogistic regression\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 142px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eCTAS\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 123px;\"\u003e\n \u003cp\u003eAUROC (95% CI)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 132px;\"\u003e\n \u003cp\u003e0.917 (0.911 \u0026ndash; 0.922)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 132px;\"\u003e\n \u003cp\u003e0.904 (0.898\u0026ndash; 0.911)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 132px;\"\u003e\n \u003cp\u003e0.879 (0.872 \u0026ndash; 0.885)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 142px;\"\u003e\n \u003cp\u003e0.882 (0.877 - 0.887)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 123px;\"\u003e\n \u003cp\u003eAUPRC (95% CI)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 132px;\"\u003e\n \u003cp\u003e0.629 (0.608 \u0026ndash; 0.649)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 132px;\"\u003e\n \u003cp\u003e0.609 (0.589 \u0026ndash; 0.628)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 132px;\"\u003e\n \u003cp\u003e0.506 (0.484 \u0026ndash; 0.528)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 142px;\"\u003e\n \u003cp\u003e0.333 (0.319 - 0.347)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd colspan=\"5\" valign=\"top\" style=\"width: 662px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eAbbreviations:\u003c/strong\u003e AUPRC, Area under precision recall curve; AUROC, Area Under the Receiver Operating Characteristic Curve; CTAS, Canadian Triage and Acuity Scale; XGBoost, eXtreme Gradient Boosting\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":false,"highlight":"","institution":"","isAcceptedByJournal":true,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"scientific-reports","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"scirep","sideBox":"Learn more about [Scientific Reports](http://www.nature.com/srep/)","snPcode":"","submissionUrl":"","title":"Scientific Reports","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"stoa","reportingPortfolio":"Scientific Reports","inReviewEnabled":true,"inReviewRevisionsEnabled":true},"keywords":"","lastPublishedDoi":"10.21203/rs.3.rs-6229836/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-6229836/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003eEmergency department (ED) overcrowding contributes to delayed patient care and worse clinical outcomes. Traditional triage systems face accuracy and consistency limitations. This study developed and internally validated a machine learning model predicting intensive care unit (ICU) admissions and resource utilization in ED patients. A retrospective analysis of 163,452 ED visits (2018\u0026ndash;2022) from Maharaj Nakhon Chiang Mai Hospital evaluated logistic regression, random forest, and XGBoost models against the Canadian Triage and Acuity Scale (CTAS). The XGBoost model achieved superior predictive performance (AUROC 0.917 vs. 0.882, AUPRC 0.629 vs. 0.333). Key predictors included mode of arrival, patient age, and free-text chief complaints analyzed with multilingual sentence embeddings. Results demonstrate machine learning\u0026rsquo;s potential to enhance triage accuracy and resource allocation, effectively identifying critically ill patients compared to traditional triage methods.\u003c/p\u003e","manuscriptTitle":"Development and Validation of an AI-Based Emergency Triage Model for Predicting Critical Outcomes in Emergency Department","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2025-04-16 08:55:25","doi":"10.21203/rs.3.rs-6229836/v1","editorialEvents":[{"type":"communityComments","content":0},{"type":"decision","content":"Revision requested","date":"2025-06-26T03:38:17+00:00","index":"","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2025-06-23T03:50:22+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"90742084949470884844229017326610710122","date":"2025-06-11T02:36:23+00:00","index":"hide","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2025-05-05T13:29:09+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"123991654888415594743885095963820234303","date":"2025-04-25T11:52:58+00:00","index":"hide","fulltext":""},{"type":"reviewersInvited","content":"","date":"2025-03-27T08:00:41+00:00","index":"","fulltext":""},{"type":"editorAssigned","content":"","date":"2025-03-27T07:56:24+00:00","index":"","fulltext":""},{"type":"editorInvited","content":"","date":"2025-03-19T04:12:52+00:00","index":"","fulltext":""},{"type":"checksComplete","content":"","date":"2025-03-18T06:59:34+00:00","index":"","fulltext":""},{"type":"submitted","content":"Scientific Reports","date":"2025-03-15T01:57:57+00:00","index":"","fulltext":""}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"scientific-reports","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"scirep","sideBox":"Learn more about [Scientific Reports](http://www.nature.com/srep/)","snPcode":"","submissionUrl":"","title":"Scientific Reports","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"stoa","reportingPortfolio":"Scientific Reports","inReviewEnabled":true,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"a17912ba-cc4e-4deb-a7aa-26057cfda6f7","owner":[],"postedDate":"April 16th, 2025","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"published-in-journal","subjectAreas":[{"id":46751644,"name":"Biological sciences/Computational biology and bioinformatics"},{"id":46751645,"name":"Health sciences/Risk factors"},{"id":46751646,"name":"Health sciences/Signs and symptoms"},{"id":46751647,"name":"Health sciences/Health care"},{"id":46751648,"name":"Health sciences/Health care/Prognosis"}],"tags":[],"updatedAt":"2025-09-01T16:08:09+00:00","versionOfRecord":{"articleIdentity":"rs-6229836","link":"https://doi.org/10.1038/s41598-025-17180-1","journal":{"identity":"scientific-reports","isVorOnly":false,"title":"Scientific Reports"},"publishedOn":"2025-08-25 15:58:11","publishedOnDateReadable":"August 25th, 2025"},"versionCreatedAt":"2025-04-16 08:55:25","video":"","vorDoi":"10.1038/s41598-025-17180-1","vorDoiUrl":"https://doi.org/10.1038/s41598-025-17180-1","workflowStages":[]},"version":"v1","identity":"rs-6229836","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-6229836","identity":"rs-6229836","version":["v1"]},"buildId":"8U1c8b4HqxoKbykW_rLl7","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}