慢性腎臟病伴隨心血管疾病之評估研究

A Study on the Assessment of the Chronic Kidney Disease Associated with Cardiovascular Disease

鄭柏鑫(Bo-sin Jheng)

指導教授 ： 張俊郎

國立虎尾科技大學/管理學院/工業管理系工業工程與管理碩士班/碩士(2016年)

摘要

隨著醫療水準提升與衛生環境改善，台灣社會人口結構呈現高齡化社會且逐漸邁向超高齡化社會。慢性病是導致高齡人口死亡的主要原因，其中慢性腎臟病是一種初期症狀不明顯卻嚴重威脅國人健康的慢性病。慢性腎臟病與心血管疾病為台灣十大死因中的兩種疾病，皆是影響國人健康的重大疾病。
本研究以近年某醫療機構資料庫中慢性腎臟病患者為研究對象，透過文獻探討與醫師訪談，篩選出增加罹患心血管疾病風險之重要因子後，運用演算法計算因子權重，並結合預測模型與案例式評估系統，用以評估慢性腎臟病患者是否需要進一步檢測心血管疾病伴隨與否。研究結果顯示，雖以基因邏輯斯演算法結合案例式推理評估系統有較高之準確度，但傅利曼檢定結果顯示三種演算法之權重值對於計算相似度不存在顯著差異，且準確率與ROC曲線下面積皆達88%與0.813以上，皆適合做為評估系統之權重值計算；預測模型部分，雖然基因邏輯斯迴歸演算法結合支援向量機有較高之準確度， K疊驗證後平均準確率為88.24%，ROC曲線下面積為0.818，但六項預測模型經統計檢定後，並不存顯著差異。本研究結果能提供醫療機構及臨床工作者，做為輔助診斷之參考依據，進而達成早期治療減輕病患疾病負擔。

關鍵字

慢性腎臟病 ； 心血管疾病 ； 基因邏輯斯迴歸演算法 ； 粒子群最佳化演算法 ； 交叉熵演算法 ； 倒傳遞類神經網路 ； 支援向量機 ； 案例式推理技術

並列摘要

With the improved healthcare standards and health environment, Taiwan’s socio-demographic structure presents an aging society gradually moving towards and ultra-aging society. Chronic diseases are the main causes of death among older persons. Chronic kidney diseases, CKDs that have nonspecific early symptoms are chronic diseases that pose serious health threats. Chronic disease and cardiovascular disease are two diseases listed among the leading ten causes of death, which are major diseases affecting people’s health.
CKD patients from the recent database of an anonymous medical institution were adopted as the research participants in this study. Through literature reviews and interviews with physicians and the selection of important factors contributing to increased cardiovascular disease risks, algorithms were applied to calculate the factor weights. In addition, a predictive model and case-based assessment system were used to assess whether chronic disease patients require further examination of accompanying cardiovascular disease. Research results show that although the genetic algorithm with logistic regression combined with the case-based reasoning assessment system possess higher accuracy, the Friedman’s test results show that the weights of the three algorithms and calculation proximity show no significant difference. Additionally, the accuracy rate and the area under the ROC curve reached over 88% and 0.813, suitable for calculating the weights of the assessment systems. For the predictive model part, although the genetic algorithm with logistic regression coupled with Support Vector Machine, SVM possesses higher accuracy, the average accuracy after k-fold verification is 88.24%, and the area under the ROC curve is 0.818.The six predictive models after statistical analysis show no significant difference. The research results shall serve as a reference for medical institutions and clinical workers during aided diagnosis, thereby achieving early treatment to reduce the load of patients with the disease.

並列關鍵字

Chronic Kidney Disease ； Cardiovascular Disease ； Particle Swarm Optimization ； Genetic Algorithms Logistic Regression ； Cross Entropy ； Back Propagation Neural Network ； Case Based Reasoning

參考文獻

1. 王春和、蔡勝旺(2014)，「同時最佳化元件維修周期和維修活動之多元件系統預防維修模式」，品質期刊，21卷，4期，頁247-267。

2. 宋裕祺、王俊穎(2015)，「結合粒子群演算法與遺傳演算法於斜張橋鋼索預力之最佳化設計」，中國土木水利工程學刊，27卷，1期，頁1-10。

3. 李俊賢、陳玟? (2012)「適應式影像還原-以計算智慧為方法」，資訊管理學報，19卷，3期，頁475-507。

4. 李建邦、俞星辰(2014)，「以粒子群最佳化演算法結合分類器預測疾病及搜尋變數之研究」，數據分析，9卷，1期，頁57-74。

6. 李維平、張加憲 (2011)，「改良粒子群演算法求解旅行銷售員問題」，先進工程學刊，6卷，1期，頁21-30。