Options
預測肝細胞癌病患復發風險及其預後
Date Issued
2022
Author(s)
賴玟潔
Wen-Chieh Lai
Advisor
張啟昌
Abstract
背景: 肝細胞癌倖存者面臨最大的臨床問題之一是復發,過去研究多以存活為優先目標。近年來由於治療技術的進步,如何以實證醫學觀點進行整合性的分析,是目前肝細胞癌倖存者防治復發的重要課題。方法: 本研究以實證醫學回溯五家醫院癌症登記庫資料共5,665筆。依據文獻與臨床專家意見共選取14個候選危險因子:診斷年齡、性別、腫瘤大小、腫瘤顆數、巴塞隆納肝細胞癌分期、根治性治療、BMI值、吸菸行為、嚼檳榔行為、喝酒行為、甲型胎兒蛋白、總膽紅素、B型肝炎、C型肝炎。特徵擷取採取統計與機器學習分類器進行差異比較。預測模式採用CART (Classification And Regression Tree) 、C4.5 (C4.5 Decision tree) 、LDA (Linear Discriminant Analysis) 、RF (Random Forest) 與LGR (Logistic Regression),用以分析整體復發及其不同復發型式與治療。預測模式的績效評估使用:敏感度、特異度、準確性、F度量評分、Kappa、Matthews相關係數和曲線以下面積(AUC)。結果: 整體復發前三名危險因子:腫瘤顆數、BCLC 期別、根治性治療;局部區域復發前三名危險因子:手術方式、腫瘤顆數、BCLC期別;遠端復發前三名危險因子:腫瘤大小、甲型胎兒蛋白、腫瘤顆數;切除性手術復發前三名危險因子:病理期別、腫瘤大小、腫瘤顆數;射頻消融復發前三名危險因子:腫瘤顆數、腫瘤大小、BCLC期別;經導管動脈化學栓塞復發前三名危險因子:C型肝炎、腫瘤顆數、嚼檳榔行為為顯著危險因子;在所有預測模式中,C4.5的AUC值最高,在0.7573-0.9675之間。 討論/結論: 本研究從肝細胞癌復發三個層面的六組資料分析結果顯示:共同危險因子是腫瘤顆數。整體復發的危險因子:腫瘤顆數、甲型胎兒蛋白、腫瘤大小與先前文獻結果一致;而性別與BCLC期別則是不同。復發型式的危險因子:腫瘤大小及甲型胎兒蛋白與先前文獻結果一致;而腫瘤顆數與BCLC 期別則是不同。手術切除治療後復發的危險因子:腫瘤大小、腫瘤顆數、甲型胎兒蛋白、分級分化及手術邊緣與先前文獻結果一致。射頻消融治療後復發的危險因子:腫瘤顆數、腫瘤大小及甲型胎兒蛋白與先前文獻結果一致;而年齡與性別則是不同。經導管動脈化學栓塞治療後復發的危險因子:腫瘤顆數及BCLC期別,與先前文獻結果一致;而腫瘤大小及甲型胎兒蛋白則是不同。本研究結合機器學習分類器與統計方法分析台灣癌症登記資料庫。此外,篩選的重要危險因子作為預測模型補助臨床決策,這將使醫師能夠識別疾病風險,並對臨床調整提供最佳的監測與追蹤。
Background: Recurrence is one of the major clinical problems for survivors of hepatocellular carcinoma (HCC). A comprehensive analysis of evidence-based medicine is currently necessary to prevent the recurrence of HCC.Methods: A retrospective analysis of 5,665 records from five cancer registries was conducted, and 14 candidate risk factors were selected based on literature and clinical experts, which are age at diagnosis, sex, tumor size, tumor number, Barcelona Clinic Liver Cancer (BCLC) stage, curative treatment, body mass index (BMI), smoking behavior, betel nut chewing behavior, alcohol use, alpha-fetoprotein (AFP), total bilirubin, hepatitis B virus (HBV), hepatitis C virus (HCV). Statistical and machine learning classifiers were used to select the important features. In addition, five predictive models (CART (Classification And Regression Tree), C4.5 (C4.5 Decision tree), LDA (Linear Discriminant Analysis), RF (Random Forest) and LGR (Logistic Regression)) were used to analyze different recurrence types and treatment modalities. We evaluated the predictive models based on their sensitivity, specificity, accuracy, F-measure score, Kappa, Matthews correlation coefficient, and area under the curve (AUC).Results: The three most important risk factors for overall recurrence were tumor number, BCLC stage and curative treatment. For local regional recurrence, surgical procedure, tumor number and BCLC stage were the three most common risk factors; relative to local regional recurrence, tumor size, AFP and tumor number were the three most common risk factors of distant recurrence. According to the results of the feature selection for different treatment modalities, the most common risk factors for surgical resection were pathologic stage, tumor size, and tumor number. For radiofrequency ablation (RFA), the three main risks were tumor number, tumor size, and BCLC stage; for transcatheter arterial chemoembolization (TACE), the three main risks were HCV, tumor number, and betel nut chewing behavior. In all prediction modes, C4.5 had the highest AUC value, between 0.7573 and 0.9675. Discussions and Conclusion: The results of this study showed that tumor number was the common risk factor. The risk factors for overall recurrence, tumor number, AFP, and tumor size were consistent with previous findings, but sex and BCLC Stage were different. The risk factors for the different types of recurrence, tumor size and AFP were consistent with previous findings, while tumor number and BCLC stage were not. The risk factors for the surgical resection treatment were consistent with previous findings, including tumor size, tumor number, AFP, grade and surgical margins. The RFA results showed that tumor number, tumor size, and AFP were consistent with previous findings; however, age and sex were different. In TACE, tumor number and BCLC stage were consistent with previous findings; however, tumor size and AFP were not. From an empirical medical perspective, this study uses machine learning classifiers and statistical methods to analyze the Taiwan cancer registry database. Furthermore, the selected risk factors can be used as predictive models for clinical decision-making. This will allow physicians to identify disease risks and provide the best monitoring and tracking of clinical adjustment.
Subjects
肝細胞癌
復發
機器學習技術
危險因子
實證醫學
hepatocellular carcinoma
recurrence
machine learning
risk factors
evidence-based medicine
Type
thesis
File(s)
No Thumbnail Available
Name
110CSMU5585001-001.pdf
Size
3.03 MB
Format
Adobe PDF
Checksum
(MD5):914060d4a79b2322fe6d84016d4733d6