子宮頸癌是常見的女性生殖器惡性腫瘤之一，每年死亡人數為女性癌症中的前十名。本文針對子宮頸癌復發和死亡的危險因子做討論與分析，共分成兩個部分。第一部分利用病人的病理相關資訊，建立 Cox 比例風險模型 (Cox proportional hazard model)，找出對復發和死亡有顯著影響的危險因子，但有些危險因子的影響會隨著時間變化，因此再進一步利用廣義 Cox 模型探討這些危險因子是如何隨時間影響病情。第二部分是分析子宮頸癌變量跟變量間的關係，在臨床上用連續變數作為變量時，時常需要決定一個切割點 (cutoff point) 將病人分群，在此提供一個選取切割點的方法。首先要決定切分點的個數和資料型態，可以選擇不同判別準則決定最佳的切分點，提供臨床上評估和決策參考。

Cervical cancer is one of the most common genital tumors and ranks among the top ten leading cause of death from gynecologic cancers. We discuss how risk factors influence recurrence and death of cervical cancer.
In the first part, we used Cox proportional hazards model to find out which clinicopathologic factors posed a significant risk for recurrence or death. The effect of some of the risk factors changes with time. So, we built a flexible Cox model that allowed time-varying effects, and variables with time-varying effects and those without could thus be analyzed side by side for their impact on recurrence and death.
In The second part, we analyzed the correlation among risk factors. It is often necessary to determine a cutoff point to stratify patients into groups when using continuous variables in clinical studies. We provided a method of how to choose cut-off points. Given the number of cutoff points and the data type, our method offered a criterion to decide cut-off point, which may be helpful to clinical assessment and decision-making.

論文審定書 i
摘要 ii
Abstract iii
1.研究動機與目的 1
2.資料描述 3
2.1 資料處理 3
2.2 變數介紹 3
2.2.1 連續型變數 3
2.2.2 離散型變數 4
3.研究方法
3.1 Kaplan-Meier 存活函數 9
3.2 對數等級檢定 9
3.3 Cox 比例風險模型 10
3.4 Schoenfeld 殘差 11
3.5 廣義 Cox 風險模型 11
3.6 Additive Cox 比例風險模型 13
3.7 羅吉斯迴歸 13
3.8 決定切分點 14
4.研究結果
4.1 子宮頸癌復發與死亡的危險因子 19
4.1.1 單變量分析 19
4.1.2 多變量分析 23
4.2 變數間的關係與決定切分點 26
5. 結論與建議 32
參考文獻 34
A 附錄 35
A.1 附錄一 35

Araujo, A. and Meira-Machado, L. (2015). Smooth Hazard Ratio Curves Taking a Reference
Value. R package version 1.0.2.
Budczies, J., Klauschen, F., Sinn, B. V., Gyory, B., Schmitt, W. D., DarbEsfahani, S., and
Denkert, C. (2012). Cuto nder: A comprehensive and straightforward web application
enabling rapid biomarker cuto optimization. PLOS ONE, 7:1{7.
Carmen, C. S., Luis, M. M., Thomas, K., and Francisco, G. (2010). Flexible hazard ratio
curves for continuous predictors in multi-state models: an application to breast cancer
data. Statistical Modelling, 10:291{314.
R Core Team (2016). R: A Language and Environment for Statistical Computing. R
Foundation for Statistical Computing, Vienna, Austria.
Wang, H. C. (2015). Analyzing data in ovarian cancer study using extended cox propor-
tional hazards model (including time-varying coecients) and nonlinear mixed-eects
model.