評估以非侵入方法在家預測高膽固醇血症檢查結果

Using Non- invasive Method for Predicting Hypercholesterolemia Risk at Home

鄧仲華(Chung-Hua Teng)

指導教授 ： 邱泓文

臺北醫學大學/醫學科技學院/醫學資訊研究所/碩士(2015年)

https://doi.org/10.6831/TMU.2015.00060

摘要

高膽固醇血症(hyperlipidemia)為許多心血管與腦血管疾病的重要前驅因子，許多國人的十大死因，都和心血管與腦血管疾病有關，若能早期發現高膽固醇血症並且配合運動飲食及藥物治療，將可避免許多重大疾病的發生及惡化。

此研究以自動神經元(Artificial Neural Network, ANN)，分析非侵入性身體檢查以預測高膽固醇血症的結果，分析資料期間為2012年期間 ，共收集1003名新竹科學園區新人體檢驗數據，以2006年行政院衛生署公布的代謝症候群標準及佛萊明罕危險預估評分表(Framingham Risk Score)以及和膽固醇風險因子有關的文獻。選取非侵入性檢查項目，並且容易在家量測之項目，包括年齡，腰圍身高比WhtR (waist-to-height ratio,腰圍除以身高值)，身體質量指數BMI，抽菸，腰圍，高血壓，吸菸以預測高膽固醇血症發生的機率，軟體採用STATISACA 10.0 分析類神經網路，70%為訓練組(703人)，30%為測試組(300人)。在1003個案例中，平均年齡為35.2歲，最小15歲，最大59歲。有288人(28.71%)為高膽固醇血症的患者6個變數除了抽菸以外在高膽固醇與無高膽固醇血症之間都有顯著差異(p value < 0.05)。測試使用不同神經元數目時，對ANN預測結果的差異性，分別測試了3、4、5、9及10個神經元，以4個神經元對高膽固醇血症預測結果較佳，類神經網路預測模型整體預測結果 ，正確率(accury)89.63%，敏感度(Sensitivity) 為75.00%，特異性(Specificity)95.52%與ROC曲線下面積為0.94。ANN與使用相同參數邏輯斯?A歸(logistic regression)和支持向量機SVM(support vector machine)比較有較佳的結果。

雖然醫療院所在台灣已經非常發達，但民眾定時量測抽血仍需耗時耗力，對健保及民眾的負擔也會增加。而這世界還是有很多地區不易抽血檢查。所以發展能在家中自行推估疾病的風險，還是有其重要性。 以ANN預測五項變數年齡，腰圍身高比WHtR，身體質量指數BMI，腰圍，高血壓可以在家的測量項目的結果，可以用來預測高膽固醇血症的可能性。

WHtR在各項變數中與高膽固醇血症有最高的相關係數，且在相關文獻中對成人高血壓，第二型糖尿病，高血脂，心血管疾病比BMI和腰圍統計顯著性更佳，顯示WHTR為優於WC和BMI檢測心血管代謝危險因素。建議WHtR應列入新的健康指標，良好的WhtR應小於0.5 。

關鍵字

自動神經元 ； 高膽固醇血症 ； 佛萊明罕危險預估評分表 ； 邏輯斯?A規 ； 決策樹型支持向量機 ； 腰圍身高比

並列摘要

High blood cholesterol (Hypercholesterolemia) is precursor of many cardiovascular and cerebrovascular diseases. Cardiovascular and cerebrovascular diseases are also two major diseases of top ten leading cause of death of Taiwanese people. Early detect hypercholesterolemia and combined with exercise, diet and drug treatment could avoid the deterioration of many major diseases. This study use ANN (Artificial Neural Network) to analysis non-invasive medical examination and predict the outcome of hypercholesterolemia.
1003 new employee physical examination data were collected from the Hsinchu
Science Park. According Department of Health in Taiwan published standards of metabolic syndrome 2006, Framingham Risk Score and the literature of cholesterol risk factors, select noninvasive and easy measure at home risk factors of hypercholesterolemia. The risk factor including age, WHtR (waist-to-height ratio), body mass index (BMI), smoking, waist circumference, blood pressure, smoking predicts the probability of occurrence. The analysis software uses STATISACA 10.0 for ANN. 703 physical examination persons (70%) as training group, the rest 300 person (30%) as the test group . In these 1003 case, the average age is 35.2 years old (maxima age is 59 and minima age is 15), 289 case has Hypercholesterolemia. Six risk factor has significant differences (p<0.005) besides smoking. Use 3,4,5,9 and 10 number of neuron for predict hypercholesterolemia, 4 number of neuron has better outcome. Our result demonstrated that our ANN model can predict the occurrence of hypercholesterolemia people, the overall accuracy of this ANN is 89.63%, sensitivity is 75.00%, specificity is 95.52% and ROC area is 0.94 .ANN model compare logistic regression and SVM(support vector machine) which has better result.
Although health care system is consummate in Taiwan .People take blood test still cost money and time. And still has some backwater area difficult to take blood test. It’s important to develop non-invasion method to measure hypercholesterolemia.
Our study highlight the ANN model analysis use 5 risk factor include age, WHtR, BMI waist and hypertension can predict the occurrence of hypercholesterolemia.
WHtR has better correlation coefficient in these variables and in pertinent literature has better statistically significance in adult hypertension, diabetes mellitus, hypercholesterolemia and cardiovascular disease. We suggest WhtR should be as a new health indicator in Taiwan, a favorable WHtR value should be less than 0.5.

並列關鍵字

Artificial Neural Network ； hypercholesterolemia ； Framingham Risk Score ； specific logistic meter loop regulation ； decision tree support vector machine ； WHtR (waist-to-height ratio)

參考文獻

[2] Chou P., Li CL, Tsai ST, 2001. Epidemiology of type 2 diabetes in Taiwan, Diabetes Res Clin Pract; 39(suppl.1): S29-S35.

[3] Luo W, Guo Z, Wu M, Hao C, Zhou Z, Yao X. 2014 Jun 16.Index of central obesity as a parameter to replace waist circumference in definition of metabolic syndrome in predicting cardiovascular disease. J Cardiovasc Med (Hagerstown) 2014 Jun 16.

[5] Cheng-Ding Chang, Chien. Chih Wang, Bernard. Jiang.2010.Using data mining techniques for multi-diseases prediction modeling of hypertension and hyperlipidemia by common risk factors.Expert Systems with Applications 38 (2011) 5507–5513

[6] Chan S, Frochlich J, et al. 2002.Association of skin cholesterol content, measured by a non-invasive method, with markers of inflammation and Framingham risk prediction. Am J Cardiol Mancini J, 89:1313-16.

[9] N.R. Shanker, A. Ezhil, S. Archana .2012.Non-invasive method of detection of cholesterol using image processing. International Journal of Medical Engineering and Informatics, 4 (3): 223 DOI: 10.1504/IJMEI.2012.048384