• 沒有找到結果。

六十項輸入變數預測模型比較

第三章 研究材料與方法

4.4 預測模型比較

4.4.3 六十項輸入變數預測模型比較

表 31 六十項輸入變數預測模型績效比較 敏感度

(sensitivity)

特異度

(specificity)

正確度 (accuracy)

AUC

類神經網路預 測模型

77.7% 75.9% 76.4% 0.791

邏輯斯迴歸預 測模型

87.4% 50.0% 72.67% 0.706

從上表可以看出,類神經網路預測模型,敏感度(sensitivity)77.7%特異度(specificity)

75.9%,正確度(accuracy) 76.4%,AUC 0.791;邏輯斯迴歸預測模型,敏感度(sensitivity)

87.4%,特異度(specificity)50%,正確度(accuracy) 72.67%,AUC 0.706,本研究以AUC 作為評估績效之依據,越大表示效果越好,故類神經網路預測模型的績效高於邏輯斯迴 歸預測模型。

圖 22 六十項變數預測模型 ROC curves 比較

表32 六十項項變數預測模型ROC curves比較

AUC SE a 95% CI b ANN3 0.791 0.0348 0.720 to 0.851 LR3 0.706 0.0407 0.629 to 0.775 Difference between areas 0.0849

Standard Error c 0.0295

95% Confidence Interval 0.0271 to 0.143

z statistic 2.879

Significance level P = 0.004

從上表可以看出,類神經網路預測模型,AUC 0.791,標準差為0.0348 及95%信賴區間 為0.720到0.851:邏輯斯迴歸預測模型AUC 0.706,標準差為0.0407 及95%信賴區間為 0.629 到0.755,P -vaule = 0.004,小於0.05有顯著差異,故類神經網路預測模型的績效 高於邏輯斯迴歸預測模型。

第五章 討論

5.1 研究發現與討論

本研究運用類神經網路及邏輯斯迴歸來建構腦血管動脈粥狀硬化預測模型。研究結果發 現,類神經網路模型(ANN model)有比邏輯斯迴歸模型(LR model)有更好的識別能 力。本研究六十項變數的類神經網路預測模型為本研究最佳模型,以確認組(Validation Set)測試類神經網路模型的曲線下方面積(AUC =0.791)與邏輯斯模型曲線下面積(AUC

=0.706),在預測腦血管動脈粥狀硬化出現異常的表現上,有達到顯著差異(p = 0.004)。

類神經網路模型使用三十九項變數所建構的預測模型,其驗證組的AUC為0.765、正確 率為67%、敏感度為56.3%、特異度為89.7%,為一低敏感度、高特異度的模型;使用四 十七項變數所建構的預測模型,其驗證組的AUC為0.771、正確率為74.5%、敏感度為 75.7%、特異度為69%,為一高敏感度、低特異度的模型;使用六十項變數所建構的預 測模型,其驗證組的AUC為0.791、正確率為76.4%、敏感度為77.7%、特異度為75.9%,

為一高敏感度、高特異度的模型,結果顯示變數增加對於模型的AUC及正確率是有幫助 的。

邏輯斯迴歸模型使用三十九項變數所建構的預測模型,其驗證組的AUC為0.706、正確 率為72.67%、敏感度為79%、特異度為53.4%,為一高敏感度、低特異度的模型;使用 四十七項變數所建構的預測模型,其驗證組的AUC為0.718、正確率為72.67%、敏感度 為91.3%、特異度為44.8%,為一高敏感度、低特異度的模型;使用六十項變數所建構的 預測模型,其驗證組的AUC為0.706、正確率為72.67%、敏感度為87.4%、特異度50.0%,

為一高敏感度、低特異度的模型,結果顯示變數增加對於模型的AUC及正確率是沒有幫 助的,也無法提升特異度。

一個模型好壞除了可以由 AUC 及正確率高低來判斷之外,此模型若能具有一定的解釋 程度與高血壓和糖尿病密切相關(Sun, Lin, Lu, Yip, & Chen, 2002);簡國龍等人發表血脂 蛋白與腦中風發生的相關(Chien, et al., 2002);蘇大成等人以 24 小時血壓監測探討與頸

第六章 結論與建議

6.1 結論

本研究以類神經網路之訓練學習過程尋求最佳預測模型,輸入變數為腦血管動脈粥狀硬 化危險因子,在網路訓練過程中探討預測模型的表現,並與邏輯斯迴歸模型比較其AUC 及正確率,透過實際資料的驗證歸納出以下結論:

運用類神經網路之預測能力,分析民眾健檢資訊之特徵,建構腦血管動脈粥狀硬化預測 模型。由於目前國內外之腦中風相關研究,並未針對個別民眾之腦血管動脈粥狀硬化進 行預測,因此本研究特點是藉由民眾個別之健檢資訊預測腦血管動脈粥狀硬化的可能 性,由於民眾本身存在個別獨特性,所以瞭解民眾在不同狀況下對疾病預測結果之影響。

在預測能力方面,本研究所建構之類神經網路預測模型與邏輯斯迴歸預測模型相比較,

結果發現類神經網路預測模型,有較好的識別能力,推斷當輸入變數呈現非線性組合 時,類神經網路之預測力將優於邏輯斯迴歸預測模型。

本研究所蒐集的資料分成5部份,檢查報告資料、腦中風風險評估分數、個人生活型態、

個人疾病史、個人家族史。透過資料的學習,建構預測模型能夠正確判斷腦血管動脈粥 狀硬化發生的可能性,本研究六十項變數的類神經網路模型,AUC =0.791與邏輯迴歸模 型AUC =0.706),在預測腦血管動脈粥狀硬化有達到顯著差異(p = 0.004)。綜合以上 論述,類神經網路可以得到更準確的預估。因此本研究建議以類神經網路模型預測腦血 管動脈粥狀硬化發生的可能性。

6.2 研究限制

對於本研究,臨床醫師建議對於健康檢查的樣本數若能擴大至全國健檢資料,或許可以 得到一些與國外研究目前未發現的潛在危險性因子。對於醫學領域後續研究上也可採用 世代研究方式,以長時間固定追蹤某族群腦血管動脈粥狀硬化狀況與危險因子之間變化 關係,更能符合臨床所需。

在建立預測模型的過程中,有好幾個階段的處理會影響最後模型的正確率,包含資料收 集前置處理、變數選取、選用的預測方法、預測方法參數設定、推衍組及驗證組與訓練 組及測試組的樣本比重等。本研究的資料前置處理方式是將可判斷錯寫及遺漏的資料欄 位做刪除處理,但對於屬性中資料偏差大的少數資料並未做進一步處理,或許會導致雜 訊發生。

6.3 未來發展與建議

本研究在研究過程中希望能力求嚴謹,但受限於研究限制以及在研究過程中發現某些部 份尚有考慮不足、涵蓋不周地方。因此提供幾項建議供後續研究者注意及參考:

類神經網路的參數設定在過去文獻上並沒有明確的決定方式,在研究過程中發現,訓練 組有最佳的AUC,但測試組並不一定有最好的AUC,會有過度訓練的情形發生。因此,

建議後續研究者可使用決策樹、基因演算法或其他可找最佳化組合的演算法,來彌補本 研究不足之處。

腦血管動脈粥狀硬化的發生與腦血管疾病有很多的關聯,本研究的樣本是取健康檢查裡 的資訊,研究對象大都是健康的族群,異常的數值較少。故建議後續研究者可以針對腦 血管疾病患者的資料來做進一步的探討。

對於醫療領域若能找出因子與因子間或是因子與病症間關係是重要的。除了本研究的方

法外,若能先將資料用關聯性分析或集群分析來找出因子間的相關影響程度,再投入分 析研究,相信可以得到不錯的結果。

中文文獻

相關因子。物理治療物理治療物理治療物理治療,34,1, 53-64。

Alamowitch, S. (2001). Risk, causes, and prevention of ischaemic stroke in elderly patients with symptomatic internal-carotid-artery stenosis. Lancet, 357, 1154~1160.

Andersen, K. K., Olsen, T. S., Dehlendorff, C., & Kammersgaard, L. P. (2009). Hemorrhagic and ischemic strokes compared: stroke severity,mortality, and risk factors. Stroke, 40(6), 2068-2072.

Bamford, J., Sandercock, P., Dennis, M., Burn, J., & Warlow, C. (1991). Classification and natural history of clinically identifiable subtypes of cerebral infarction. Lancet, 337, 1521-1526.

Barnett, H. J., Gunton, R. W., Eliasziw, M., Fleming, L., Sharpe, B., Gates, P., et al. (2000).

Causes and severity of ischemic stroke in patients with internal carotid artery stenosis.

JAMA, 283, 1429-1436.

Bell, T. A., Brown, J. M., Graham, M. J., Lemonidis, K. M., Crooke, R. M., & Rudel, L. L.

(2006). Liver-specific Inhibition of Acyl-coenzyme a: Cholesterol Acyltransferase 2 with Antisense Oligonucleotides Limits Atherosclerosis Development in

Apolipoprotein B100-only Low-density Lipoprotein Receptor-/- mice Arterioscler Thromb Vasc Biol, 26(8), 1814-1820.

Berndt, M. C., Shen, Y., Dopheide, S. M., Gardiner, E. E., & Andrews, R. K. (2001). The vascular biology of the glycoprotein Ib-IX-V complex. Thromb Haemost 86, 178-188.

Bogousslavsky, J., Melle, G. V., & Regli, F. (1988). The Lausanne Stroke Registry: analysis of 1000 consecutive patients with first stroke. Stroke, 19(1083-1092).

Chien, K. L., Sung, F. C., Hsu, H. C., Su, T. C., Lin, R. S., & Lee, Y. T. (2002).

Apolipoprotein A-I and B and Stroke Events in a Community-Based Cohort in Taiwan:

Report of the Chin-Shan Community Cardiovascular Study. Stroke, 33(1), 39-44.

Chu, J. Y., Chu, J. K., & Chung, A. G. (2006). Epidemiology of cerebrovascular disease among. Chinese-Canadians: A ten years retrospective. case-mix study. Neurology Asia, 11, 13-18.

Cooper, G. F., Abraham, V., Aliferis, C. F., Aronis, J. M., Buchanan, B. G., Caruana, R., et al.

(2005). Predicting Dire Outcomes of Patients with Community Acquired Pneumonia.

Biomedical Informatics, 38, 347-366.

Cubrilo-Turek, M. (2004). Stroke Risk Factors : Recent Evidence and New Aspects.

International Congress Series, 1262, 466-469.

Dammerman, M., & Breslow, J. L. (1995). Genetic Basis of Lipoprotein Disorders.

Circulation 91, 505-512.

Do, S. R., Richardson, D., Hung, E., Brorson, J. R., Cruz-Flores, S., Felton III, W. L., et al.

(2006). Predictors of Recurrent Stroke in African Americans. Neurology, 67, 567-571.

Du, X., McNamee, R., & Cruickshank, K. (2000). Stroke Risk from Multiple Risk

FactorsCombined with Hypertension: A Primary Care Based Case-Control Study in a Defined Population of Northwest England Annals of Epidemiology, 10, 380-388.

Ebrahim, S., Sung, J., Song, Y. M., Ferrer, R., Lawlor, D. A., & Smith, G. D. (2006). Serum cholesterol, haemorrhagic stroke, ischaemic stroke,and myocardial infarction: Korean national health system prospective cohort study. British Medical Journal, 333(7557), 22-26.

Fatahzadeh, M., & Glick, M. (2006). Oral Medicine. Oral surgery Oral Medicine Oral Pathology Oral Radiology and Endodontics, 102, 180-191.

Feigin, V. L., Lawes, C. M., Bennett, D. A., & Anderson, C. A. (2003). Stroke

Epidemiology:A Review of Poptlarion-Based Studies of Incidence, Prevalence, and Case-Fatality in the Late 20th Century. Lancet Neurology, 2, 43-53.

Gorelick, P. S., Sacco, R. L., Smith, D. B., Alberts, M., Mustone-Alexander, L., Rader, D., et al. (1999). Prevention of a First Stroke : A Review of Guidelines and a

Multidisciplinary Consensus Statement From the National Stroke Association,.

Journal of the American Medical Association, 281, 1112-1120.

Gresele, P., & Agnelli, G. (2002). Novel approaches to the treatment of thrombosis. Trends

Pharmacol Sci, 23(1), 25-32.

Gulliford, S., Webb, S., Rowbottom , C. G., Corne, D. W., & Dearnaley, D. P. (2004). Use of Artificial Neural Networks to Predict Biological Outcomes for Patient Receiving Radical Radiotherapy of the Prostate. Radiotherapy and Oncology, 71, 3-12.

Hanley, J. A., & McNeil, B. J. (1982). The meaning and use of the area under a receiver operating characteristic (ROC) curve.

Haykin, S. S. (2009). Neural networks and learning machines: Prentice Hall.

Hillen, T., Cosholl, C., Tilling, K., Rude, A. G., McGovern, R., & Wolfe, C. A. ( 2003).

Cause of Stroke Recurrence Is Multifactorial: Patterns, Risk Factors and Outcomes of Stroke Recurrence in the South London Stroke Register. Stroke, 34, 1457-1463.

Howng, S. L., Hung, T. P., Kwan, A. L., & Lin, C. L. (1995). Intracranial aneurysm in Taiwan. J Formos Med Asso, 148(94), S73-S80.

Hu, H. H., Sheng, W. Y., & Chu, F. L. (1992). Incidence of stroke in Taiwan. Stroke, 23, 1237-1241.

Inzitari, D., Eliasziw, M., Gates, P., Sharpe, B. L., Chan, R. K., Meldrum, H. E., et al. (2000).

The causes and risk of stroke in patients with asymptomatic internal carotid artery stenosis. N Engl J Med, 342, 1693-1700.

Iso, H., Baba, S., Mannami, T., Sasaki, S., Okada, K., Konishi, M., et al. (2004). Alcohol Consumption and Risk of Stroke among Middle-aged Men: the JPHC Study Cohort I.

Stroke, 35(5), 124-129.

Iso, H., Moriyama, Y., Sato, S., Kitamura, A., Tanigawa, T., Yamagishi, K., et al. (2004).

SerumTotal Homocysteine Concentrations and Risk of Stroke and Its Subtypes in Japanese. Circulation, 109(22), 2766-2772.

Jood, K., Ladenvall, C., Rosengren, A., Blomstrand, C., & Jern, C. (2005). Family history in ischemic stroke before 70 years of age: the 90 Sahlgrenska Academy Study on Ischemic Stroke. Stroke, 36(7), 1383-1387.

Kaste, M., & Koivisto, P. (1988). Risk of Brain Infarction in Familial Hypercholesterolemia.

Stroke, 19(1097 - 1100).

Kelly, T. N., Gu, D., Chen, J., Huang, J. F., Chen, J. C., & Duan, X. (2008). Cigarette smoking and risk of stroke in the chinese adult. Stroke, 39(6), 1688-1693.

Khan, S. N., & Vohra, E. A. (2007). Risk factors for stroke: A hospital based study. Pakistan Journal of Medical Science 23(1), 17-22.

Kim, H. C., Nam, C. M., Jee, S. H., & Suh, I. (2005). Comparison of Blood

Pressure-Associated Risk of Intracerebral Hemorrhage and Subarachnoid Hemorrhage Korea Medical Insurance Corporation Study. Hypertension, 46(2), 393-397.

Kirshner, H. S. (2003). Medical Prevention of Stroke. Southern Medical Journal, 96, 354-358.

Kowalska, I., Prokop, J., & Bachorzewska-Gajewska, H. (2001). Disturbances of glucose metabolism in men referred for coronary arteriography. Postload glycemia as predictor for coronary atherosclerosis. Diabetes Care, 24, 897-901.

Kuivenhoven, J. A., Jukema, J. W., Zwinderman, A. H., Knijff, P., McPherson, R., Bruschke, A. V., et al. (1998). The Role of a Common Variant of the Cholesteryl EsterTransfer Protein Gene in the Progression of Coronary Atherosclerosis.The Regression Growth Evaluation Statin Study Group. N Engl J Med, 338(2), 86-93.

Lee, R. G., Kelley, K. L., Sawyer, J. K., Farese, R. V., Parks, J. S., & Rudel, L. L. (2004).

Plasma Cholesteryl Esters Provided by Lecithin: Cholesterol Acyltransferase and Acyl-coenzyme a: Cholesterol Acyltransferase 2 Have Opposite Atherosclerotic Potential. Circ Res, 95(10), 998-1004.

Loizou, C. P., Pattichis, C. S., Istepanian, R. S. H., Pantziaris, M., & Nicolaides, A. ( 2004).

“Atherosclerotic carotid plaque segmentation”, Engineering in Medicine and Biology Society, EMBS 2004. Paper presented at the 26th Annual International Conference.

Mark, H. Z., & Gregory, C. (1993). Receiver-Operating Characteristic (ROC) Plots: A Fundamental Evaluation Tool in Clinical Medicine. CLINICAL CHEMISTRY, 39(4), 561-577.

McGill, H. C., McMahan, C. A., Herderick, E. E., Zieske, Z. D., Malcom, G. T., Richard, E., et al. (2002). Obesity Accelerates the Progression of Coronary Atherosclerosis in Young Men. Circulation, 105, 2712-2718.

McMahan, C. A., Gidding, S. S., Malcom, G. T., Tracy, R. E., & Strong, J. P. (2006). McGill HC and Pathobiological Determinants of Atherosclerosis in Youth Research Group.

Pathobiological Determinants of Atherosclerosis in Youth Risk Scores Are Associated with Early and Advanced Atherosclerosis. Pediatrics, 118(4), 1447-1455.

Meigs, J. B., Larson, M. G., & D'Agostino, R. B. (2002). Coronary artery calcification in type 2 diabetes and insulin resistance. The Framingham offspring study , Diabetes Care, 25, 1313-1319.

Milionis, H. J., Liberopoulos, E., Goudevenos, J., Bairaktari, E. T., Seferiadis, K., & Elisaf, M.

S. (2005). Risk factors for first-ever acute ischemic non-embolic stroke in elderly individuals. International journal of cardiology, 99(2), 269-275.

Minelli, C., Lee, F. F., & Minelli, D. P. (2007). Stroke Incidence,Prognosis, 30-Day, and 1-Year Case Fatality Rates in Matão, Brazil: A Population-Based Prospective Study Stroke. Stroke, 38(11), 2906-2911.

Mukamal, K. J., Ascherio, A., Mittleman, M. A., Conigrave, K. M., Camargo, C. A., Kawachi, I., et al. (2005). Alcohol and Risk for Ischemic Stroke in Men: the Role of Drinking Patterns and Usual Beverage. Ann Intern Med, 142(1), 11-19.

Naghavi, M., Libby, P., Falk, E., Casscells, S. W., Litovsky, S., Rumberger J, B. J., Stefanadis C,, et al. (2003). From vulnerable plaque to vulnerable patient: a call for new definitions and risk assessment strategies: part I. Circulation, 108, 1664-1672.

P.K., Y., J.S., J., T.K., L., Y.C., C., & Z.S., H. (1997). Subtype of Ischemic Stroke. A Hospital-Based Stroke Registry in Taiwan (SCAN-IV). Stroke, 28, 2507-2512.

Pan, W. H., Bai, C. H., Chen, J. R., & Chiu, H. C. (1997). Associations between carotid

Pan, W. H., Bai, C. H., Chen, J. R., & Chiu, H. C. (1997). Associations between carotid

相關文件