以兩種資料探勘方式預測透析中低血壓事件:比較支持向量機法與羅吉斯迴歸法

Predict intradialytic hypotension with two data mining methods: comparison between support vector machines and logistic regression

蔡佩芬(Pei-Fen Tsai)

指導教授 ： 陳永福

中臺科技大學/管理學院/健康產業管理研究所/碩士(2012年)

摘要

透析中低血壓 (IDH) 一直都是透析治療過程中棘手且嚴重的問題，臨床上也發展出許多方法介入以預防低血壓事件的發生，我們的研究目的主要是利用人工智慧分級設計一套模組來確立高危險透析患者。
數據來自於160位血液透析病患 (61位是易患低血壓病患 ; 另外99位則是對低血壓具耐受性的病患) ，採用支援向量機 (SVM) 以及羅吉斯回歸分析法 (LRA)。透析中低血壓的定義：低血壓發生時收縮壓<100mmHg 或者是症狀性的發生掉壓事件，（收縮壓掉壓幅度>20mmHg ，合併抽筋、頭暈、噁心、嘔吐、頭痛、冒冷汗、失去意識）， 以這兩種方法進行比較預測發生透析中低血壓的能力。
SVM整體分類的準確率 (81.3%) 優於LRA (76.3%)。另外在靈敏度和特異性部份， 兩者在SVM (70.5% & 87.9%) 的表現上亦是高於LRA (62.3% & 84.8%)。
SVM在用於預測透析中低血壓風險的表現上確實優於LRA。臨床決策支持系統設計與支持向量機的臨床應用可能有助於醫師鑑別透析治療的高危險患者。

關鍵字

羅吉斯迴歸法 ； 支持向量機法 ； 資料探勘 ； 血液透析 ； 低血壓

並列摘要

Intradialytic hypotension (IDH) remains an important issue during hemodialysis treatment. Many tools were developed to investigate the mechanisms and predict the hypotensive events when receiving dialysis. Our aim is to design intelligent classifies to identify high risk hemodialysis patients.
Data from 160 hemodialysis patients (61 IDH-prone, 99 IDH-resistant) were analyzed with support vector machines (SVM) and logistic regression analysis (LRA). The IDH event was defined as a systolic pressure <100mmHg or a fall in systolic pressure >20mmHg associated with symptoms of hypotension (cramps, dizziness, nausea, vomiting, headache, sweating, loss of consciousness). The different abilities to predict the occurrences of IDH were compared between these two methods.
SVM demonstrated better overall classification accuracy (81.3%) compared to LRA (76.3%). It was also shown that SVM had higher sensitivity (70.5%) and specificity (87.9%) than LRA (sensitivity: 62.3% and specificity: 84.8%).
SVM is superior to logistic regression method in predicting the risk of hypotension during hemodialysis. Clinical application of the clinical decision support system designed with SVM may be helpful for nephrologist in discriminating the high risk patients when receiving hemodialysis treatments.

並列關鍵字

Logistic regression analysis ； Support vector machines ； Data mining ； Hypotension ； Hemodialysis

參考文獻

Richard A. Sherman, J. T. D., and Todd S. Ing. (2007). Complications During Hemodialysis. Handbook of dialysis, 236, p170-191.

台灣腎臟醫學會. (2010). 慢性腎臟病防治手冊. In 行政院衛生署國民健康局 (Eds.) (Vol. 第四章, pp. p9-12).

Rubinger, D., Sapoznikov, D., Pollak, A., Popovtzer, M. M., & Luria, M. H. (1999). Heart rate variability during chronic hemodialysis and after renal transplantation: studies in patients without and with systemic amyloidosis. Journal of the American Society of Nephrology : JASN, 10(9), 1972-1981.

Kuo, T. B., Lin, T., Yang, C. C., Li, C. L., Chen, C. F., & Chou, P. (1999). Effect of aging on gender differences in neural control of heart rate. [Research Support, Non-U.S. Gov''t]. The American journal of physiology, 277(6 Pt 2), H2233-2239.

Aronszajn, N. (1950). Theory of reproducing kernels. Transactions of the American mathematical society, 68, 337-404.