以資料探勘建立之糖尿病小血管併發症關聯模式

Investigation of Diabetic Microvascular Complications Using Data Mining Techniques

劉又甄(Yu-Chen Liu)

指導教授 ： 詹前隆

元智大學/資訊學院/資訊管理學系/碩士(2007年)

https://doi.org/10.6838/YZU.2007.00329

摘要

本論文主要是探討糖尿病病患生理資料、實驗室資料以及歷史疾病與糖尿病小血管併發症間的關聯性，研究中主要探討的小血管併發症包含: 視網膜病變、腎病變、神經病變(足部問題)。資料分析來自北區健保局糖尿病共同照護資料庫，總共有8,736糖尿病人進行分析，研究期間為2003年到2006年，採用的分析方法是資料探勘的C5.0與Neural network。研究結果發現肌酸酐(creatinine)對於視網膜病變預測有相當高的重要性，一但肌酸酐控制不良，即使其他控制變項良好罹患視網膜病變的比例卻相當的高。腎病變的發生通常是有多項控制變項不良，其中女性糖尿病患具有糖尿病家族病史者罹患腎病變的比例也較高。在神經病變方面發現，女性患者通常都有BMI, HbA1c以及 AC sugar的控制不良; 而男性通常是血壓控制不良。另外，抽菸者在神經病變上也是一個重要的預測因子。年齡與糖尿病發病時間對於三種併發症而言都是相當重要的預測因子，但是這兩個變項是人力無法改變的，因此自我管理以及生活型態的改變對於糖尿病患而言是相當重要，本研究利用C5.0 以及Neural network清楚的探討出小血管併發症的關係，此模式可應用於潛在的小血管併發症病患預測。

關鍵字

糖尿病 ； 視網膜病變 ； 腎病變 ； 神經病變 ； C5.0, Neural Network

並列摘要

Purpose: This thesis theoretically analyzes and numerically explores the relationship between the physiological data and three diabetic microvascular complications: diabetic retinopathy, diabetic nephropathy, and diabetic neuropathy (foot problem).
Method: The analysis results of 8,736 diabetic patients in northern Taiwan by using two data mining models: C5.0 and neural network were presented and compared.
Results: It is found that creatinine is the most important predictor for diabetic retinopathy. If creatinine is out of control, diabetics will easily suffer from diabetic retinopathy in spite of many other laboratory evaluations are qualified. The sensitivity and specificity for diabetic retinopathy prediction using C5.0 are 58.62 and 74.73, and those using neural network are 59.48 and 99.86, respectively. In addition, diabetic nephropathy will happen when several laboratory evaluation values are worse than target valus. Female diabetics with diabetic family history are easier to undergo this complication. The sensitivity and specificity for diabetic nephropathy prediction using C5.0 are 69.44 and 81.36, and those using neural network are 74.44 and 98.55, respectively. For diabetic neuropathy, female diabetics feature unqualified BMI, HbA1c and AC sugar, while male diabetics mostly have uncontrolled blood pressure. Besides, smoking diabetics are more difficult to avoid this complication. The sensitivity and specificity for diabetic foot problem prediction using C5.0 are 64.71 and 83.48, and those using neural network are 67.63 and 99.70, respectively.
Conclusions: The work presented in this paper demonstrates that age and the duration of diabetes are highly important but uncontrollable variables to predict the three complications, which imply that one can take advantages of other controllable variables through lifestyle changes and self-management to reduce the onset possibility of diabetic complications. Using C5.0 and neural network, we clearly identify and categorize the features of diabetics and relative importance of predictors for complications, respectively. Such prediction mechanism, which can diagnose diabetic complications in the early phases, has great potential for future treatment and prevention of diabetic complications.

並列關鍵字

Diabetes Mellitus ； Retinopathy ； Nephropathy ； Neuropathy ； C5.0 ； Neural Network.

參考文獻

[29] Taiwan Bureau of National Health Insurance, “Insurance analyses”, National Health Insurance Annual Statistical Report, Nov 2005.

[22] S.J. Russell, P. Norvig, Artificial intelligence: a modern approach, Prentice-Hall, Inc. Upper Saddle River, NJ, USA, 1995.

[1] H.M. Ye, F.T. Dai, Y.T. Yan, ” Elder patients with diabetes mellitus”, Medical Professionals Journal in Taiwan, 44(3) p27-28, 2001.

[3] Bureau of National Health Insurance, “National Health Insurance Annual Statistical Report”, 2003.

[4]. American Diabetes Association (ADA), “Diagnosis and classification of diabetes mellitus”, Diabetes Care, 28:S37-S42, 2005.