利用臨床數據預測冠狀動脈血管繞道手術術後心房顫動發生

The prediction of atrial fibrillation after coronary artery bypass surgery using artificial neural network with clinical data

沈玫君(Mei-Chun Shen)

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(2012年)

摘要

冠狀動脈心臟病（Coronary Artery Disease），是全球現代人主要健康殺手之一，也台灣名列十大死因第二名。治療方式開始大多以藥物控制為主，較為嚴重的患者則採冠狀動脈血管繞道術（Coronary Artery Bypass Graft Surgery）進行治療。而手術後心律不整（Postoperative Arrhythmia），是冠狀動脈血管繞道手術後常見合併症之一。術後心律不整發生的種類以心房顫動（Atrial fibrillation）最為常見。這類心律不整屬於持續性心律不整疾病，且發生此類心律不整之病人的預後較差，往往會造成住院天數、醫療費用、中風及死亡率等風險增加。由於心房顫動生理機轉尚未明瞭，因此本研究利用臨床可得的數據，建立類神經網路預測模型（Artificial Neural Network），預測手術後心房顫動發生，並將此預測模型與邏輯斯迴歸模型（Logistic Regression）相比。

由於冠狀動脈繞道手術後心房顫動發生機轉複雜，本實驗利用年齡、身體質量指數、血液中C-反應蛋白濃度、血液鈉離子濃度、心臟超音波檢查心容積射出率、左心房腔室是否擴張、左心室收縮功能是否良好、是否為急性心肌梗塞之患者和手術過程是否行體外循環，共9個參數，建立類神經網路預測模型，預測冠狀動脈血管繞道手術後心房顫動的發生率。於STATISTICA 7.0輸入資料陣列，輸入250筆資料，80％為訓練組（202人），20％為測試組（48人），選擇多層感知器MLP（Multiple Layer Perception）網路，輸入層13個神經元，隱藏層6個神經元，輸出層2個神經元，代表心房顫動是否發生。類神經網路預測模型整體預測結果，正確率75.6％、敏感度69.28％、特異度81.2％、ROC曲線下面積0.80。使用相同參數建立邏輯斯迴歸模型進行比較，邏輯回歸模型整體預測結果，正確率73.2％、敏感度68.4％、特異度77.4％、ROC曲線下面積0.72。

由於類神經網路模型預測結果表現不佳，檢視參數在組距間的關係，共去除12筆資料。以238筆資料輸入陣列中，80％為訓練組（191人），20％為測試組（47人），選擇多層感知器，輸入相同的9個參數，重新建立預測模型。新建立的類神經網路預測模型，輸入層13個神經元，隱藏層12個神經元，輸出層2個神經元，分別代表心房顫動是否發生。整體的預測結果，正確率86.97％、敏感度86.36％、特異度87.50％、ROC曲線下面積0.92。使用相同的參數再次建立邏輯斯迴歸模型，邏輯斯迴歸模型整體的預測結果，正確率75.2％、敏感度69.1％、特異度80.5％、ROC曲線下面積0.75。兩模型相較結果，類神經網路預測模型整體的表現比邏輯斯迴歸模型有較好的預測結果。

此系統期能建立，可協助臨床醫療人員提早預防併發症發生，提升醫療照護品質，間接的可減少病患手術後住院天數、醫療費用、中風及死亡率。

關鍵字

冠狀動脈血管繞道術 ； 心房顫動 ； 類神經網路 ； 邏輯斯迴歸

並列摘要

Heart disease is the second leading cause of death in Taiwan. Coronary artery disease(CAD) continues to be a leading cause of morbidity and mortality in the world. The treatments of CAD include medical treatment, percutaneous coronary intervention (PCI) or coronary artery bypass graft surgery (CABG). CABG is an optimal treatment when medical therapy and PCI failed. CABG can improve the survival rate of severe CAD patients. However, postoperative atrial fibrillation (Af) is a common complication of CABG. It not only associated with an increase incidence of other complications (e.g. postoperative stroke), and hospitalization days, but also increased the overall costs of admission in hospital.

In this study, we developed an ANN（Artificial Neural Network）and logistic regression to early detection of the patient with Af. We collected the clinical data from 250 patients. These patients were divided into training group (80%) and test group (20%). Our results demonstrated that our ANN model can predict the occurrence of Af in CABG patients, the overall accuracy of this ANN is 75.6％, sensitivity is 69.23％, specificity is 81.20％, and ROC area 0.80. The overall accuracy of this logistic regression is 73.2％, sensitivity is 68.4％, specificity is 77.44％, and ROC area 0.72. I found that the accuracy of prediction in our study can be get better prediction rate when 12 outliers were excluded. The modified model can reach the better accuracy (86.97％ vs. 75.2％), sensitivity (86.36％ vs. 69.1％), specificity (87.50％ vs. 80.5％), and ROC area (0.92 vs. 0.75). In conclusion, we could predict the development pathogenesis of Af effectively in CABG patients by using my ANN analysis model.

Our study highlight the ANN model analysis can be a good tool to predict the occurrence of Af when patents received CABG intervention.

並列關鍵字

Artificial Neural Network ； logistic regression ； coronary artery bypass graft surgery ； postoperative atrial fibrillation

參考文獻

Akbarzadeh Najar, R., Ghaderian, S. M., & Tabatabaei Panah, A. S. (2012). C-reactive protein (CRP) gene polymorphisms: implication in CRP plasma levels and susceptibility to acute myocardial infarction. Mol Biol Rep, 39(4), 3705-3712.

Alam, M., Siddiqui, S., Lee, V. V., Elayda, M. A., Nambi, V., Yang, E. Y., et al. (2011). Isolated coronary artery bypass grafting in obese individuals: a propensity matched analysis of outcomes. Circ J, 75(6), 1378-1385.

Ascione, R., Caputo, M., Calori, G., Lloyd, C. T., Underwood, M. J., & Angelini, G. D. (2000). Predictors of atrial fibrillation after conventional and beating heart coronary surgery: A prospective, randomized study. Circulation, 102(13), 1530-1535.

Aviles, R. J., Martin, D. O., Apperson-Hansen, C., Houghtaling, P. L., Rautaharju, P., Kronmal, R. A., et al. (2003). Inflammation as a risk factor for atrial fibrillation. Circulation, 108(24), 3006-3010.

Bardaji, A., Barrabes, J. A., Sanchis, J., & Sanchez, P. L. (2010). Ischemic heart disease: 2009 update. Rev Esp Cardiol, 63 Suppl 1, 49-60.