探討糖尿病的高危險因子

Abstract

本文為對中國醫藥大學2001至2003年期間參加糖尿病檢測的民眾，對檢測的資料做分析與了解，其中欲探討的是在多元的檢測項目下，可能會導致糖尿病的高危險因子。醫學證實糖尿病的發生與血糖變化有密切相關，因此控制血糖是預防糖尿病的不二法則。而空腹血漿血糖 (又稱飯前血糖值)是判斷罹患糖尿病與否的重要準則，本文將會探討舒張壓、糖化血色素、低密度脂蛋白......等八項測量值與飯前血糖值之間的關係，並採取五種迴歸重要變數的篩選方式，「向前選取法」、「向後消去法」、「逐步選取法」、「Cp選取法」、「校正後的複判斷係數法」，以此挑選出最終的模型。並且嚴格地審查最終模型，須檢查有無異常點以及影響點、是否符合殘差的四項基本假設，「殘差平均數是否為零」、「殘差是否來自常態」、「殘差是否獨立」、「殘差變異數是否為常數」，並對異常點、影響點以及違反殘差假設的部分另做修正，以確保最終模型能對可能導致糖尿病的高危險因子這宗旨提供有效並且可信的解釋力，並以此作為糖尿病預防以及防範的重要典範與參考。

Abstract This report is to investigate 500 patients which were collected by China Medical University from 2001 to 2003. These patients participated in the detection of diabetes. We extract partial data set which may relate in a high risk factor for diabetes mellitus. The response variable is fasting blood glucose. The explanatory variables include family medical history, smoking, drinking, diastolic blood pressure, glycated hemoglobin, low-density lipoprotein, total cholesterol, and triglyceride. Forward Selection, Backward Elimination, Stepwise Selection, Cp selection method, and all subset selection are used to choose the best subset model. We conclude that diastolic blood pressure, glycated hemoglobin and triglyceride are the most important explanatory variables to fasting blood glucose. Diagnostic checking and residual analysis are performed to check whether the final model is adequate.