Data Science and Bayesian Decision in Extubation Prediction of Surgical Intensive Care Unit

• Main Authors: Tsung-LunTsai, 蔡宗倫
• Other Authors: Chia-Yen Lee
• Format: Others
• Language: en_US
• Published: 2017
• Online Access: http://ndltd.ncl.edu.tw/handle/3359db

碩士 === 國立成功大學 === 製造資訊與系統研究所 === 105 === Nowadays, with the rapid development in technology, we can collect the data more easily and efficiently than before. However, the typical statistical methodologies applied in medicine are not comprehensive when comparing to those applied in manufacturing. In the past, endotracheal extubation in the hospital simply conducted the experiments by simple statistics or clinical experience and then it generates lots of “rule of thumbs” which cannot address large dataset effectively, identify the interactions among the sequential processes, and thus might lead to misclassification in the medical treatment. This study focuses on the intubate/extubate treatments in the surgical intensive care unit (ICU). We apply the machine learning methodologies to the real setting regarding the decision of extubation. Since each patient shows different nature, status, constitution, and the length of staying in the ICU, the above measures are extracted by each individual and lead to the case-by-case decision of the intubate/extubate treatments. With the difficulty mentioned above, the data collected from the hospital is usually incomplete and inconsistent and this issue makes the application unstable as well as difficult. In this study, we apply the data science framework and establish the extubation prediction model, involving definition of the problem, data preprocessing, variable selection, support vector machine, boosting logistic regression, cross validation etc. Emphasizing on how to improve the success rate of extubation to assist the clinical extubation decision, and the accuracy rate is up to 81.5%. On the other hand, this paper proposes a Bayesian decision framework to correct the probability of the prediction result, inference the posterior probability, and quantify the value of information provided by our proposed model, so as to enhance the decision-making process of “Precision Medicine”.