- 學位論文
運用類神經網路建立糖尿病性腎臟病變病患預測模型分析
Designing and analysis an Artificial Neural Network Model for the Prediction of Diabetic Nephropathy
摘要
糖尿病（DM）是在二十世紀的最常見的慢性疾病之一。目前台灣大約有120萬患病人口，發病率與流行率逐年上升，糖尿病腎病佔終末期腎臟病（ESRD）的新病例為35.9％，並於2010年，與終末期糖尿病患者的治療費用超過30億美元。此外，糖尿病腎病與心血管疾病的發病率和死亡率非常高的風險。目標：使用台灣北部某醫學中心資料，以類神經網路技術，建立糖尿病病患腎臟病變的預測模型。方法：收集台灣北部某醫院中心2008至2011年就診診斷為糖尿病病人(ICD-9為250.00)，兩年間新診斷為糖尿病腎臟病變(ICD-9為250.40、250.42)的個案共7873筆，使用隨機抽樣出1000筆，分成訓練組500筆，測試組500筆；經文獻回顧，臨床血液檢驗資料，找出12項與糖尿病腎臟病變有關的項目納入分析，使用類神經網路與邏輯斯迴歸分析，分別以5項(達顯著差異)、7項(包含無顯著差異)的變數為輸入變項，輸入變項有性別、年齡、抽菸史、高血壓病史、BMI、glucose、creatinine、BUN、HDL、LDL、CHOL、HbA1C，得到腎臟病變診斷(ICD-9為250.40、250.42)輸出變項的方式進行測試。結果：以全部12項輸入變項建立的類神經網路預測模型成效最佳(p<0.05)，ROC曲線下的面積 0.925、準確率 0.916、敏感度 0.838、特異度 0.926較邏輯斯迴歸ROC曲線下的面積 0.916、準確率 0.894、敏感度 0.86、特異度 0.838為佳。結論：類神經網路所建立糖尿病腎臟病變的預測模型優於邏輯斯迴歸，可輔助糖尿病病患，在就診時得知兩年內腎臟病變的可能性，及早作臨床醫療上的治療準備，積極回診，也減少個人疾病認知錯誤而不回診的因素，進而延緩疾病進程，降低醫療上的不必要開銷與浪費。
並列摘要
Diabetes Mellitus (DM) is one of the most common chronic diseases in the twentieth century. The incidence and prevalence increased steadily over decades, and there were approximate 120 million cases in the population of Taiwan up to now. Diabetic nephropathy accounts for 35.9% of new cases of end-stage renal disease (ESRD),and in 2010, the cost for treatment of diabetic patients with ESRD was in excess of $30 billion in Taiwan. Moreover, diabetic nephropathy is associated with a very high risk of cardiovascular morbidity and mortality. An artificial neural network( ANN) was established for the prediction of development of diabetic nephropathy in diabetic patients. From Jan. 2008 to Dec. 2011, a total 7873 patients diagnosed as having diabetes mellitus at one hospital in Taiwan were retrospectively reviewed. Patients enrolled in the study included diagnosis Diabetes patients’ within two years of kidney disease. We have chosen a total number of 1000 patients and distinguish between the training sample of 500 patients and test samples of 500 patients. Variables examined were age, BMI, hypertension, blood glucose, BUN, creatinine, HDL, LDL, HbA1C and total cholesterol. An ANN model was developed by using two randomly selected training and testing sets for predicting diabetic nephropathy. We used logistic regression(LR) compared with the ANN model. The ANN model has good performance with overall accuracy of 91.6% and the area under the receiver operating characteristics (ROC) curve is 0.925( p-value < 0.05) better than logistic regression . Diabetic nephropathy can be easily and accurately predicted by the ANN model. By using the model, the clinicians can find out the patients at high risk of developing diabetic nephropathy and delay the occurance.
參考文獻
延伸閱讀
- 陳智堂（2012）。運用類神經網路建構腎結石的預測模型〔碩士論文，臺北醫學大學〕。華藝線上圖書館。https://doi.org/10.6831/TMU.2012.00082
- 洪薇雯（2012）。一位糖尿病合併腎臟病變個案治療之經驗分享。中華民國糖尿病衛教學會會訊，8(1)，9-10。https://doi.org/10.6583/TADE.2012.8(1).5
- 黃玉瑩（2015）。充能介入方案對糖尿病早期腎病變病患之自我效能、自我管理及生理指標之成效〔碩士論文，高雄醫學大學〕。華藝線上圖書館。https://www.airitilibrary.com/Article/Detail?DocID=U0011-0902201502022200
- 蔡富棉（2017）。充能自我管理方案於糖尿病早期腎病變病人之成效〔博士論文，高雄醫學大學〕。華藝線上圖書館。https://www.airitilibrary.com/Article/Detail?DocID=U0011-2208201710365400
- 王志偉（2001）。Neural Network Model Predictive Control Design for Nonlinear Processes with Unmeasured Disturbances〔碩士論文，中原大學〕。華藝線上圖書館。https://doi.org/10.6840/cycu200100246