肝癌手術病人二年内焦慮症狀與憂鬱症狀預測模式之探討

摘要研究背景與目的肝癌手術病人因為擔心復發，伴隨著焦慮症狀與憂鬱症狀，因此本研究將利用類神經網路(Artificial Neural Network, ANN)、最近鄰居演算法(K-Nearest Neighbor, KNN)與支持向量機(Support Vector Machine, SVM)預測模式評估肝癌病患術後二年內焦慮與憂鬱之準確性，並利用全域敏感度分析(Global Sensitivity Analysis)評估影響肝癌手術病患術後二年內焦慮與憂鬱重要預測因子之權重。研究方法本研究乃前瞻性縱貫性研究設計，以問卷方式進行資料蒐集，收取2014年5月1日至2016年12月31日，選取台灣南部二間教學醫院共計409位接受肝癌手術病人為研究樣本。分別以結構式問卷、貝克焦慮量表(Beck Anxiety Inventory, BAI)及貝克憂鬱量表第二版(The Beck Depression Inventory-Second Edition, BDI-II)收集病人人口學特性、臨床特性、醫療照護品質特性及焦慮症狀與憂鬱症狀評估分數，追蹤時間點分別為術前術後二年。本研究利用均方根誤差（Root-Mean-Square Error, RMSE）、均方誤差(Mean-Square Error, MSE)及平均絕對百分比誤差(Mean Absolute Percentage Error, MAPE)作為上述三種預測模式評估其績效之指標。研究結果研究結果發現，肝癌手術病人人口學特性、臨床特性、醫療照護品質特性及術前焦慮症狀與憂鬱症狀分數，皆與其術後二年焦慮症狀與憂鬱症狀分數有顯著性相關(p<0.001)，再者，類神經網路預測模式其績效指標之數值均較其他兩種預測模式為佳，最後利用全域敏感度分析發現，麻醉風險分類是影響病人術後二年焦慮症狀之最重要影響因子，接著為術前焦慮症狀分數及術後住院天數，術前憂鬱症狀分數是影響病人術後二年憂鬱症狀之最重要影響因子，接著為合併症指數及B型肝炎。

關鍵字

肝癌；類神經網路；焦慮；憂鬱

並列摘要

Abstract Background and Purposes Patients with hepatocellular carcinoma (HCC) surgery are often accompanied by anxiety symptoms and depression symptoms because of fear of cancer recurrence. Therefore, this study uses artificial neural network (ANN), k-nearest neighbor (KNN), and support vector machine (SVM) to compare and assess the accuracy of anxiety and depression within two years after the end of liver cancer surgery. Besides, this study would use global sensitivity analysis to evaluate the weight of important predictors of anxiety and depression. Research Methods This prospective study uses Beck anxiety inventory (BAI) and the Beck depression inventory second edition (BDI-II) to evaluate anxiety and depression from May 1, 2014 to December 31, 2016. A total of 409 patients from two teaching hospitals in the south of Taiwan were included into the study. The patient attributes, clinical attributes, quality of care, pre-operative anxiety symptom and depression symptom were collected by using the structured questionnaire (SQ). The tracking time of this study would be two years before and after HCC surgery. In this study, root mean square error (RMSW), mean square error (MSE), and mean absolute percentage error (MAPE) would be used as performance indicators for evaluating the accuracy of the three prediction models. Research Results It showed that the patient attributes, clinical attributes, quality of care, and pre-operative anxiety symptom and depression symptom were statistically significant with post-operative two-year anxiety symptom and depression symptom (p < 0.001). Besides, the prediction model of artificial neural network (ANN) has better performance than the other two prediction models. Finally, it is found the most important predictor to post-operative 2-year anxiety symptoms is the classification of anesthesia by using global sensitivity analysis (GSA). Following predictors are the pre-operative anxiety symptom and the number of days in hospitals. On the other hand, the most important predictor to post-operative 2-year depression symptom is the pre-operative depression symptom and following predictors are Charlson comorbidity index (CCI) and hepatitis B. Conclusions and Implications It showed that the artificial neural network (ANN) was the best prediction model for the post-operative 2-year anxiety symptoms and depression symptoms among HCC patients. This study also found that the significant predictors of post-operative 2-year anxiety symptom and depression symptom could be used to educate HCC patients undergoing surgeries in the expected rehabilitation process and other surgical outcomes. Keywords: hepatocellular carcinoma、artificial neural network、anxiety、depression