近來大醫院急診持續嚴重壅塞，日益增加的病患人數造成急診醫療資源供不應求，長期以來急診壅塞的問題也導致延誤病患的就診或住院的時間。本研究以合作醫院「台北馬偕紀念醫院」之2011年至2018年之八年度急診室病歷資料，合計共1,065,480筆急診病患於檢傷階段之就醫紀錄，以預測住院之可能性。研究首先採用自然語言處理之BERT預訓練模型進行微調訓練，透過急診室病歷資料之主訴語進行住院預測。研究結果發現經過不平衡處理的BERT模型，期住院預測結果之AUC指標可達0.950、Accuracy指標可達0.891；此外，透過特定檢傷資料（檢傷一級與檢傷五級）進行預測結果AUC指標可達0.954、Accuracy指標可達0.960。然而單獨只考慮結構化變數，如：檢傷等級、年齡、體溫、到院時間與到院方式，並採用弱分類器XGBoost模型之預測效力，其不如以主訴透過BERT模型之預測結果。故研究進一步比較XGBoost透過篩選之病患五項結構化重要特徵所生成之擴充主訴，納入BERT模型預測之效力，透過BERT訓練後，其AUC指標可達0.958、Accuracy指標可達0.904，遠高於過去的相關研究。研究方法與發現提供急診住院預測參考，並期盼降低急診室病床等候時間，進而改善急診壅塞問題。

In recent years, emergency departments (EDs) of hospitals are crowded constantly. The issue of EDs congestion is due to the prolonged situation of public demands exceeding the supply of EDs medical resources, and the increasing numbers of patients. As the result, this issue may delay a patient’s time for receiving treatments or patient admissions. To predict the probability of patient admissions, the study was based on 2011-2018 EDs medical records of the collaborated hospital MacKay Memorial Hospital in Taipei (Taiwan). This 8-year dataset included a total of 1,065,480 triage medical records of EDs.
The study adopted Bidirectional Encoder Representations from Transformers(BERT), a pre-training model in natural language processing(NLP) for fine-tuning. By training the chief complaints (CCs) on EDs medical records, the study aimed to predict the possibility of patient admissions. The result carried out by a BERT model that underwent imbalanced processing was that the AUC index reached 0.950, and the Accuracy index reached 0.891 for the possibility of patient admissions. On top of that, by analyzing specific levels of triage (level 1st and level 5th), the prediction result reached 0.954 on AUC index and 0.960 on Accuracy index. Nevertheless, when only considering structured variables, for instance, the levels of triage, age, body temperature, arrival time, and mode of arrival, and adopting weak classifier eXtreme Gradient Boosting (XGBoost). The predictive validity was lower than the result of CC data undergone BERT. Therefore, the study took one step forward. First, this study further screened the five important variables by comparing with the structured characteristics of the EDs medical records through XGBoost. Later, inserted the variables into the CC-expanded of BERT. And after BERT Training, the AUC index reached 0.958 and the Accuracy index reached 0.904, which was far higher than the related studies in the past. The research methods and findings are for references to predict EDs presentations and hospital admissions. Furthermore, the study aimed to decrease the waiting time for EDs beds and reduce EDs congestion.

陳子立、吳怡瑾、馮嚴毅、楊昌倫 (Oct., 2016)。基於急診滯留時間探勘醫療行為之實證研究，管理與系統，第二十三卷‧第四期：527-561頁。
陳芃諭(2020).探討使用多面向方法在文字不平衡資料集之分類問題影響. https://hdl.handle.net/11296/h24e95
陳彥伯(2021). 基於Word2vec與XGBoost方法之急診住院預測研究. https://hdl.handle.net/11296/3zje5u
Cameron A., Ireland A.J., Mckay G.A., Stark A., Lowe D.J. Predicting Admission at Triage: Are Nurses Better Than a Simple Objective Score? Emerg Med J 2017;34(1):2–7.
Chang D., Hong W.S., Taylor R.A., Generating contextual embeddings for emergency department chief complaints, JAMIA Open, Volume 3, Issue 2, July 2020, Pages 160–166, https://doi.org/10.1093/jamiaopen/ooaa022.
Devlin J., Chang M.W., Lee K., Toutanova K. 2019. BERT: Pre-training of deep bidirectional transformers for language understanding. In NAACL, pages 4171–4186.
Ebker-White, A., Bein, K.J., Berendsen Russell, S. et al. The Sydney triage to admission risk tool (START) to improve patient flow in an emergency department: a model of care implementation pilot study. BMC Emerg Med 19, 79 (2019). https://doi.org/10.1186/s12873-019-0290-x.
Fernandes M., Vieira S.M., Leite F., Palos C., Finkelstein S., João M.C. Sousa, Clinical Decision Support Systems for Triage in the Emergency Department using Intelligent Systems: a Review, Artificial Intelligence in Medicine, Volume 102, 2020, 101762, ISSN 0933-3657, https://doi.org/10.1016/j.artmed.2019.101762.
Friedman, J. H. (2001). Greedy function approximation: A gradient boosting machine. Ann. Statist. 29(5) 1189-1232. https://doi.org/10.1214/aos/1013203451
Gligorijevic D., Stojanovic J., Satz W., Stojkovic I., Schreyer K., Del Portal D., Obradovic Z. Deep attention model for triage of emergency department patients. Proceedings of the 2018 SIAM International Conference on Data Mining 2018:297–305, https://doi.org/10.1137/1.9781611975321.34.
Handly N., Thompson D.A., Li J., Chuirazzi D.M., Venkat A. Evaluation of a hospital admission prediction model adding coded chief complaint data using neural network methodology. Eur J Emerg Med. 2015 Apr;22(2):87-91. doi: 10.1097/MEJ.0000000000000126. PMID: 24509606.
Hong W.S., Haimovich A.D., Taylor R.A.(2018) Predicting hospital admission at emergency department triage using machine learning. PLoS ONE 13(7): e0201016. https://doi.org/10.1371/journal.pone.0201016
Lee, S.J., Weinberg, B.D., Gore, A. et al. A Scalable Natural Language Processing for Inferring BT-RADS Categorization from Unstructured Brain Magnetic Resonance Reports. J Digit Imaging 33, 1393–1400 (2020). https://doi.org/10.1007/s10278-020-00350-0.
Leslie N. Smith. (2015) Cyclical Learning Rates for Training Neural Networks. https://arxiv.org/abs/1506.01186
Levin S., Toerper M., Hamrock E., Hinson J.S., Barnes S., Gardner H., Dugas A., Linton B., Kirsch T., Kelen G. Machine-Learning-Based Electronic Triage More Accurately Differentiates Patients With Respect to Clinical Outcomes Compared With the Emergency Severity Index. Ann Emerg Med. 2018 May;71(5):565-574.e2. doi: 10.1016/j.annemergmed.2017.08.005. Epub 2017 Sep 6. PMID: 28888332.
Mohammed R., Rawashdeh J. and Abdullah M., "Machine Learning with Oversampling and Undersampling Techniques: Overview Study and Experimental Results," 2020 11th International Conference on Information and Communication Systems (ICICS), 2020, pp. 243-248, doi: 10.1109/ICICS49469.2020.239556.
Parker C.A., Liu N., Wu S.X., Shen Y., Lam S.S.W., Ong M.E.H. Predicting hospital admission at the emergency department triage: A novel prediction model. Am. J. Emerg. Med. 2019;37:1498–1504. doi: 10.1016/j.ajem.2018.10.060.
Patel S.J., Chamberlain D.B., Chamberlain J.M. A Machine Learning Approach to Predicting Need for Hospitalization for Pediatric Asthma Exacerbation at the Time of Emergency Department Triage. Acad Emerg Med. 2018 Dec;25(12):1463-1470. doi: 10.1111/acem.13655. Epub 2018 Nov 29. PMID: 30382605.
Saha B., Lisboa S., Ghosh S.Understanding patient complaint characteristics using contextual clinical BERT embeddings. 2020. https://arxiv.org/abs/2002.05902.
Sun B.C., Hsia R.Y., Weiss R.E., Zingmond D., Liang L.J., Han W., McCreath H., Asch S.M. Effect of emergency department crowding on outcomes of admitted patients. Ann Emerg Med. 2013 Jun;61(6):605-611.e6. doi: 10.1016/j.annemergmed.2012.10.026. Epub 2012 Dec 6. PMID: 23218508; PMCID: PMC3690784.
Vaswani A., Shazeer N., Parmar N., Uszkoreit J., Jones L., Gomez A.N., Kaiser L., Polosukhin I. Attention is all you need. Advances in Neural Information Processing Systems. 2017. https://arxiv.org/abs/1706.03762.
Wu, IC., Chen, CE., Lin, ZR., Chen, TL., Feng, YY. (2022). Predicting Hospital Admission by Adding Chief Complaints Using Machine Learning Approach. In: Fui-Hoon Nah, F., Siau, K. (eds) HCI in Business, Government and Organizations. HCII 2022. Lecture Notes in Computer Science, vol 13327. Springer, Cham. https://doi.org/10.1007/978-3-031-05544-7_18.
Y.-Y. Feng、I-C. Wu, & T.-L. Chen, &W.H. Chang (2019). A Hybrid Data Mining Approach for Generalizing Characteristics of Emergency Department Visits Causing Overcrowding, Journal of Library and Information Studies (JLIS), Vol. 17, No. 1, pp.1-35
Zhai Q., Lin Z., Ge H., Liang Y., Li N., Ma Q., Ye C. Using machine learning tools to predict outcomes for emergency department intensive care unit patients. Sci Rep. 2020 Dec 1;10(1):20919 https://pubmed.ncbi.nlm.nih.gov/33262471/.
Zhang X., Kim J., Patzer R.E., Pitts S.R., Patzer A., Schrager J.D. Prediction of emergency department hospital admission based on natural language processing and neural networks. Methods Inf Med. 2017;56(05):377–389. doi: 10.3414/ME17-01-0024.