利用資料探勘技術預測呼吸器患者脫離呼吸器的最佳時機

鄭儒勵(Cheng Ju Li)

指導教授 ： 胡雅涵

國立中正大學/管理學院/資訊管理系醫療資訊管理研究所/碩士(2016年)

摘要

摘要
影響呼吸器脫離的因素很多，且不是單一因素所導致，因此至今學者們仍不斷地在尋找脫離因素以掌握適合脫離呼吸器的患者。為了避免患者脫離呼吸器失敗而造成病患的傷害和增加醫療的費用，評估脫離呼吸器成功的時機對於臨床人員而言是相當重要的。
過去研究大都採用線性迴歸的方式找尋影響因子，卻發現結果差異很大造成臨床應用之困難。本研究將採用資料探勘之監督式學習技術來確認影響脫離呼吸器的因素，並建立病患可被脫離成功之時機預測模式，收集1067筆脫離呼吸器患者的資料，為避免實驗結果的偏誤，將其817筆拔管成功和250筆拔管失敗，以各隨機抽樣方式使數量一致，將兩者合併為一組資料，共隨機抽樣30次後，使用分類技術進行資料分析。
實驗結果，以隨機林(Random forest)分類技術所建構的預測模式效能最佳，但Random forest是同時用多個決策樹一起做預測的分類器，所以無法解釋其規則。因此最後以也具有高正確率(0.787)和高AUC值(0.8142)的分類迴歸樹(Classification and Regression Tree, CART) 的預測模式作規則說明。實驗結果發現，RSBI數值大於105脫離失敗的機會較大之外，也要考慮病患的整體情形，使用呼吸器天數增加、加護病房住院天數增加和營養狀況不佳易導致感染的產生。受到感染的患者再加上代謝能力變差，脫離呼吸器失敗的機會則會增加。

關鍵字

拔管 ； 機械式通氣 ； 呼吸器脫離 ； 資料探勘

並列摘要

Abstract
There are many factors that affect the ventilator weaning, and is not caused by a single factor. To date, scholars are still looking for factors for the patients wean from the ventilator. To avoid from any weaning failures that lead to patients’ harm and increase in health care costs; timing of success in clinical trials is very important.
Most of the previous studies use linear regression method to find impact factors and found significant difference in the results that cause difficulties in clinical trials. In this study, using supervised learning techniques of date mining to identify factors influence ventilator weaning. In addition, we established prediction mode of success in weaning time. We collected 1067 samples from the patients’ ventilator information. In order to avoid deviation of the results, using consistent quantity in random sampling from 817 successful extubation and 250 failed extubation. Merge the random samples into a set of data and after 30 times of random sampling; use classification techniques to perform data analysis.
Test results showed Random forest’s classification technique has the best prediction method, but Random forest classifier uses a number of decision trees at the same time, therefore cannot explain the rules of the method. For this reason, the study use high accuracy(0.787) and high AUC (0.8142) CART(Classification and Regression Tree) to rule analysis. From the results, data showed RSBI greater than 105 has bigger chance of failure. General status of the patients should be considered, concomitant poor nutritional status, increase of ventilator day and ICU length of stay can lead to the possibility of infection. Infected patients coupled with metabolism deterioration, weaning failure will increase.

並列關鍵字

Extubation ； Mechanical ventilation ； Weaning ； Data mining

參考文獻

Chen, H. C., Lin, M. C., Lin, A. S., Wang, C. C., & Wang, Y. H. (2012). Risk Factors and Outcomes of Patients with Prolonged Mechanical Ventilation after Successful Weaning. Thoracic Medicine, 27(4), 209-216.

Lee, C. H., Wu, C. L., & Chen, P. J. (2005). Effect of Respiratory Muscle Performance on Ventilatory Weaning. Thoracic Medicine, 20(5), 371-379.

Lin, Y. T., Cho, C. M., Yang, H. C., Chan, M. C., & Wu, C. L. (2009). Usefulness of Rapid Shallow Respiratory Breathing Index Measured through Mechanical Ventilator Monitoring. Thoracic Medicine, 24(5), 243-251.

Agarwal, R., Aggarwal, A. N., Gupta, D., & Jindal, S. K. (2007). Role of Noninvasive Positive-Pressure Ventilation in Postextubation Respiratory Failure: A Meta-Analysis. Respiratory care, 52(11), 1472-1479.

Alía, I., & Esteban, A. (2000). Weaning from mechanical ventilation. Critical Care, 4(2), 72-81.