運用資料探勘技術建構脂肪肝預測模式

A Research of Data Mining Applied to the Predictive Model of Fatty Liver

姚志成(Norman Yau)

指導教授 ： 李維平博士

中原大學/商學院/資訊管理研究所/碩士(2005年)

https://doi.org/10.6840/cycu200500705

摘要

脂肪肝病症不容易察覺，需由專業醫師進行腹部超音波檢查始可發現。然目前全民健保未提供腹部超音波脂肪肝檢查，部分機關律定由50歲以上實施檢查，取決標準較無科學依據。經查閱相關文獻，罹患脂肪肝風險因素甚多，目前尚無一完整確認，且過去研究所探討關鍵性指標較不豐富。其中僅利用臨床基本資料簡易統計、運用線性邏輯迴歸分析抽樣統計，樣本及相關變數並不周延完整，影響準確度。
台北松山醫院體檢資料庫，完整記錄歷年人員體檢資料，其中2000至2004年資料計2230筆，體檢29項參數，足以運用資料探勘技術，進行脂肪肝預測模式研究，以突破以往研究樣本數較少、相關變數不完整，所獲預測不夠精確之限制。
研究方法以一般物理檢查、血脂肪異常指數、肝功能異常指數、腎功能異常指數、尿液常規指數及血液常規指數等六個構面，運用決策樹探勘技術以建構脂肪肝預測模型，藉以完整觀察並找尋項目中關鍵性指標，以提高預測準度。並與線性邏輯迴歸分析、類神經網路預測法進行實驗比較，以證明決策樹分類方法運用在脂肪肝預測上較其它方法佳，可建立一個容易閱讀、高準度且有脈絡可循之脂肪肝預測模式。
研究實驗重要發現：1.決策樹演算法預測準度78%優於線性邅k方法，另預測中、重度脂肪肝患者達93%。2.BMI(體重/身高2)>24.18、三酸甘油脂>92.5、ALT(GPT) >26.4、腎功能尿酸UA >5.15、血液紅血球指數>4.675 為本研究方法獲得重要預測因子及指數。3.一般物理檢查BMI、年齡、血壓、脈博便可初步預測脂肪肝，預測度76%。4.實驗獲得BMI>24.57且ALT(GPT) >40，則100%幾乎可能是中重度患者，說明肝炎為脂肪肝重要指標。

關鍵字

脂肪肝 ； 預測 ； 線性邏輯迴歸分析 ； 決策樹 ； 類神經網路

並列摘要

Fatty liver disease is a subtle disease, and it usually can be diagnosed by ultrasonography.Many risk factors are attributable to fatty liver disease. However, the accuracy of these risk factors and the model of fatty liver disease screening are not well-established. It is essential to create a reliable model for fatty liver disease screening.
The database of health examination in Taipei Shong-Shan Hospital was analyzed. It enrolled 2,230 persons from 2000 to 2004, and 29 routine tests were arranged for each person in health examination. Six parameters, including general physical examination, blood lipid profile, liver function, renal function, urinalysis, and complete blood count (CBC), were assessed by decision-tree mining algorithm, logistic regression, and neural network. Their accuracy and sensitivity were calculated and compared.
In this study, we conclude that: 1. The decision-tree algorithm for fatty liver screening has an accuracy of 78%, and it is better than logistic regression; 2. The accuracy of decision-tree algorithm for moderate to severe fatty liver disease is 93%; 3. The cut points of six parameters in decision-tree algorithm are: body-mass index (BMI) > 24.18kg/m2; triglyceride > 92.5mg/dL; alanine aminotransferase (ALT) > 26.4 U/L; uric acid (UA) > 5.15mg/dL; RBC count > 4.675 x 1012/L; 4. BMI, age, blood pressure, and pulse rate are reliable parameters for predicting fatty liver disease. The accuracy is 76%; 5. BMI > 24.57 kg/m2 and ALT > 40 U/L had an approximately 100% accuracy rate for predicting moderate to severe fatty liver disease.

並列關鍵字

decision tree ； fatty liver disease ； prediction

參考文獻

2. 吳國禎，2000，資料探索在醫學資料庫之應用，中原大學醫學工程學系碩士論文。

4. 李博智，2002，資料探勘在慢性病預測模式之建構，元智大學資訊管理學系碩士論文。

6. 曹德弘，1997，腰圍和腰臀圍比與內臟脂肪體積的相關研究，國立師範大學體育學系碩士論文。

1. Akahoshi M, Amasoki Y, Soda M, et al. ,2001,Correlation between fatty liver and coronary risk factors: A population study of elderly men and women in Nagasaki, Japan. Hypertens Res(24),pp:337–343.

2. Angulo P, Keach JC, Batts KP, Lindor KD. ,1999,Independent predictors of liver fibrosis in patients with nonalcoholic steatohepatitis., Hepatology(30) , pp:1356–1362.