自然語言處理在繁體中文護理紀錄預測呼吸衰竭患者拔管失敗之應用

Application of Natural Language Processing in Traditional Chinese Nursing Records to predict extubation failure in patients with respiratory failure

吳中凱(Chung-Kai Wu)

指導教授 ： 張家瑋

國立臺中科技大學/資訊與流通學院/資訊工程系碩士班/碩士(2021年)

摘要

藉由重症加護病房呼吸衰竭患者之拔管前和拔管後的繁體中文護理記錄以及呼吸照護紀錄之淺快呼吸指數，應用深度學習之人工神經網路、長短期記憶神經網路以及機器學習的支援向量機、邏輯迴歸、決策樹和集成學習的隨機森林和極限梯度提升等模型，做為呼吸衰竭患者重新插管成功和失敗的預測。

實驗所用的數據集是由評鑑為「醫學中心暨甲類教學醫院」所提供的急重症加護病房患者的護理紀錄和呼吸照護紀錄。收治的患者大多有呼吸衰竭的問題，除了要密切的監控和治療之外，若患者無法自主呼吸或氧氣不足以供應生理所需時，就必須要插管及使用呼吸器給予通氣治療方能維持生命。

插管與呼吸器為加護病房重症病患重要器官的支持，屬於高度不適之侵入性醫療，故病情稍有改善的情況時，臨床即會朝向引導患者做呼吸訓練，以利拔管和脫離呼吸器。惟仍有部分拔管的病人仍會因為呼吸功能不足須要重新插管，故如何能夠準確預測拔管的結果進而儘快提前重新插管，乃是目前重症醫療急待解決之問題。

關鍵字

護理紀錄 ； 自然語言 ； 深度學習 ； 機器學習 ； 淺快呼吸指數

並列摘要

Using the traditional Chinese nursing records before and after the extubation of the respiratory failure patients in the intensive care unit and the shallow rapid breathing index of the respiratory care records, the application of deep learning artificial neural networks, long short-term memory neural networks and machine learning support vector machines, logistic regression, decision trees, random forests with ensemble learning, and extreme gradient boosting models are used to predict the success and failure of reintubation in patients with respiratory failure.

The data set used in the experiment is the nursing records and respiratory care records of patients in the acute intensive care unit provided by the "medical center and class A teaching hospital". Most of the admitted patients have respiratory failure. In addition to close monitoring and treatment, if the patient cannot breathe spontaneously or the oxygen is insufficient to supply the physiological needs, they must be intubated and used a ventilator to give ventilation treatment to maintain it. life.

Intubation and respirator support the vital organs of critically ill patients in the intensive care unit. They are highly uncomfortable and invasive medical treatment. Therefore, when the condition improves slightly, the clinic will guide the patient to do breathing exercises to facilitate extubation and escape from breathing Device. However, some patients with extubation still need to be re-intubated due to insufficient respiratory function. Therefore, how to accurately predict the result of extubation and re-intubate as soon as possible is a problem that urgently needs to be solved in intensive care.

並列關鍵字

Nursing records ； Natural language processing ； Deep learning ； Machine learning ； RSBI

參考文獻

1. Epstein, S. K., Ciubotaru, R. L. (1998). Independent effects of etiology of failure and time to reintubation on outcome for patients failing extubation. American journal of respiratory and critical care medicine, 158(2), 489-493.

2. Roussos, C., Koutsoukou, A. (2003). Respiratory failure. European Respiratory Journal, 22(47 suppl), 3s-14s.

3. Dasta, J. F., McLaughlin, T. P., Mody, S. H., Piech, C. T. (2005). Daily cost of an intensive care unit day: the contribution of mechanical ventilation. Critical care medicine, 33(6), 1266-1271.

4. Weissman, G. E., Hubbard, R. A., Ungar, L. H., Harhay, M. O., Greene, C. S., Himes, B. E., Halpern, S. D. (2018). Inclusion of unstructured clinical text improves early prediction of death or prolonged ICU stay. Critical care medicine, 46(7), 1125.

5. Knaus, W. A., Draper, E. A., Wagner, D. P., Zimmerman, J. E. (1985). APACHE II: a severity of disease classification system. Critical care medicine, 13(10), 818-829.