臺灣博碩士論文加值系統

研究背景：
臺灣地區的口腔癌發病率和死亡率隨著時間有逐年增加的趨勢，為了防制口腔癌，臺灣地區於2004年開始進行以醫師視診為主的大規模族群篩檢，隨著早期癌症比例的增加，我們也對口腔癌存活狀況感到興趣。然而，早期口腔癌的存活分析易受競爭死亡風險的影響。儘管競爭風險模型已被廣泛應用，但這些統計模型極少應用於以族群為基礎的癌症篩檢計畫，也很少應用於多階段疾病-死亡競爭風險模式。如何將累積危險性(Cumulative risk)和考慮競爭死因之次分布瞬間風險(Subdistribution hazard, SDH)概念融入多階段疾病-死亡競爭風險模式亦鮮被探討。
目的：
本論文的主要目的旨在發展一個多階段病病-死亡競爭風險模型，並應用於臺灣口腔癌篩檢資料庫，據此估計算口腔癌發生及死亡的累積危險性及嚼食檳榔所貢獻的死因別瞬間風險比（Cause-specific hazard ratio, CSH ratio）與次分布瞬間風險比（SDH ratio）。
資料：本論文分析臺灣地區2004-2009年參加口腔癌篩檢計畫之資料，共計2,332,430位18歲以上具有吸菸或嚼食檳榔的民眾曾經參與篩檢，其中有8009個口腔癌個案發現、2223位民眾死於口腔癌，另外，在非口腔癌及口腔癌的族中，各有75582及667人死於競爭死因。
統計模型建構：
本論文發展一個四階段疾病-死亡模式，該模式的四個階段分別為沒有口腔癌（Free of cancer, FOC）（階段1）、口腔癌（Oral cancer, OC）（階段2），口腔癌死亡（Oral cancer death, OCD）（階段3），競爭風險死亡（Competing risks of death, CRD）（階段四）。利用均質性（指數分佈）馬可夫模型 (Homogeneous Markov model)和非均質性韋伯分佈為基礎之隨機過程(Non-homogeneous Weibull-based stochastic process)估計不同狀態之間的瞬間轉移率，再將之轉換成累積轉移風險。本論文亦利用Gray和Fine概念探討檳榔嚼食對於口腔癌發生的死因別瞬間風險比（Cause-specific hazard ratio, CSH ratio）與次分布瞬間風險比（SDH ratio）。
結果：
調整競爭風險後之口腔癌累積死亡風險較未考慮競爭風險下之口腔癌累積死亡風險低。本論文利用多階段疾病-死亡競爭風險模型得到在考慮及未考慮競爭死亡風險的十年預測機率分別為0.20％及0.27％。口腔癌患者及一般族群在考慮及未考慮競爭風險下之十年口腔癌死亡預測機率分別為67.4％和81.98％，及0.33％和0.39。
次分布瞬間風險比（SDH ratio）相對於死因別瞬間風險比（CSH ratio）而言，其估計值會隨著追蹤時間而異。
結論：
本論文發展多階段病病-死亡競爭風險模型，並將之應用於以族群為主之口腔癌篩檢計畫，根據此模型的估計結果，我們可以在考慮競爭風險之下預測不同狀態(如發病或死亡)之長期累積風險，此結果亦可以被用在評估癌症篩檢過度診斷(Overdiagnosis)之議題。

Background
An increasing trend of incidence of and mortality from oral cancer called for a nationwide secondary prevention through oral cancer screening with dental inspection in Taiwan. It is therefore of great interest to examine the survival of oral cancer while more proportion of early-detected oral cancer was noted. However, statistical analysis of early stage of oral cancer is subject to competing risks of death. In spite of wide applications of statistical competing risks model, very few studies were conducted to apply these statistical models to population-based cancer screening data. Moreover, it is also very rare to develop a multi-state illness-death model with the incorporation of competing risks of death as one of absorbing states. How to integrate the concept of cumulative incidence and subdistribution hazard into the illness-death competing risks model has been barely addressed.
Aims
The main purposes of this thesis were to develop a multi-state illness-death competing risks model so as to apply the proposed model to estimate cumulative incidence for occurrence of and death from oral cancer and also to estimate case-specific hazard (CSH) ratio and subdistribution hazard (SDH) ratio for the effect of betel quids on oral cancer death taking competing risks into account.
Data
Data on 2,332,430 Taiwanese residents aged 18 years or older attending the population-based screening for oral cancer with dental inspection from which 8009 oral cancer, and 2223 oral cancer deaths, together with 75582 and 667 deaths from competing causes among subjects free of oral cancer and patients of oral cancer, respectively, were ascertained. These data were exploited to estimate cumulative risk of occurrence of oral cancer and death from oral cancer. Information on betel quids chewing and smoking was also collected for assessing the CSH and SDH ratios for the effect of betel quids chewing.
Model Specification
A four-state illness-death model was proposed, including free of oral cancer (FOC) (State 1), oral cancer (State 2), oral cancer death (State 3), and competing risks of death (State 4). Both homogeneous (exponential) Markov model and non-homogeneous Weibull-based stochastic process were applied to estimating the parameters corresponding to each transition from state i to state j (i&;#8804;j, i, j=1, 2, 3, 4). Cumulative risks for each transition were estimated by using the corresponding transition probabilities.
The effect of betel quids on occurrence of oral cancer was assessed by cause-specific hazard (CSH) ratio and subdistribution hazard (SDH) ratio based on Gray and Fine idea.
Results
Cumulative risks with adjustment for competing risks of death for oral cancer and oral cancer death were slightly lower than those without considering competing risks of death. By using the proposed multi-state illness-death competing risks model, we predicted 10-year cumulative risks for occurrence of oral cancer were 0.20% and 0.27% with and without adjustment for competing risks of death, respectively. Both 10-year cumulative risk figures for oral cancer death were 67.4% and 81.98% for oral cancer patients, and 0.33% and 0.39% for the underlying screened population, respectively.
Using SDH ratio as opposed to cause-specific hazard (CSH) shows the effect of betel quids chewing varied with time of follow-up.
Conclusion
The proposed multi-state illness-death competing risks model permits one to predict long-term risk of multi-state outcome subject to presence of competing risks of death. Information provided from the results after the application of this model make contribution to the concern about overdiagnosis in population-based screen.

口試委員會審定書 i
致謝 ii
中文摘要 iii
Abstract v
目錄 viii
1. Introduction 1
1.1 Impact and Risk Factors of Oral Cancer 1
1.2 Intervention of Incidence and Death from Oral Cancer 2
1.3 Statistical Analysis of Competing Risk Data 3
1.4 Study aims 5
2. Literature review 6
2.1 Conceptual Model of Competing Risk Analysis 6
2.2 Controversy in Statistical Analysis of Competing Survival Analysis 7
2.3 Statistical Methods for Competing Risk Analysis 10
3. Data Sources 14
3.1 Oral Cancer Screening 14
3.2 Descriptive Results 15
4. Multi-state Illness-Death Model with Competing Risks 20
4.1 Model Specification 20
4.2 Homogeneous Multi-state Illness-death Model with Competing Risk 21
4.3 Non-homogeneous Multi-state Illness-death Model with Competing Risk 23
4.4 Incorporating Covariates into Model 24
4.5 Likelihood Function and Estimation 25
4.6 Cause-specific Hazard Function and Sub Distribution Hazard (SDH) for Multi-state Illness-death Model 25
5. Results 29
5.1 Non-parametric Method for Cumulative Risk Curve 29
5.2 Competing risks model and AFT Model 30
5.3 Cumulative Risk by Betel Quids Chewing 31
5.4 Competing risks model 32
5.5 The Four-state Illness-death Weibull (Non-homogeneous)-based Stochastic Model 34

6. Discussion 75
6.1 Application of Competing Risks Analysis to Disease Progression of Oral Cancer 75
6.2 Overestimation of Disease Progression of Oral Cancer without Considering Competing Risk 76
6.3 Effect of Covariates (Betel Quids Chewing) with Sub distribution Hazard 77
6.4 Implications for Oral Cancer Screening 78
6.5 Methodological Thoughts 79
6.6 Limitation 80
7. Conclusion 82
References 83

Beyersmann J, Latouche A, Buchholz A, Schumacher M. (2009) Simulating competing risks data in survival analysis. Stat Med 28(6):956-971.
Blot WJ, McLaughlin JK, Winn DM. (1988) Smoking and drinking in relation to oral and pharyngeal cancer. Cancer Res 48:3282-3287.
Chen THH, Kuo HS, Yen MF, Lai MS, Tabar L, Duffy SW. (2000) Estimation of sojourn time in chronic disease screening without data on interval cases. Biometrics 56(1):167-172.
Clark TG, Bradburn MJ, Love SB, Altman DG. (2003a) Survival analysis part I: basic concepts and first analyses. Br J Cancer 2003;89(2):232-238.
Clark TG, Bradburn MJ, Love SB, Altman DG (2003b). Survival analysis part IV: further concepts and methods in survival analysis. Br J Cancer 2003;89(5):781-786.
Cox DR, Miller HD. (1965) The theory of stochastic processes. London: Methuen.
Dignam JJ, Kocherginsky MN. (2008) Choice and interpretation of statistical tests used when competing risks are present. J Clin Oncol 26(24):4027-4034. doi: 10.1200/JCO.2007.12.9866.
Dignam JJ, Zhang Q, Kocherginsky M. (2012) The use and interpretation of competing risks regression models. Clin Cancer Res 18:2301-2308.
Fine JP, Gray RJ. (1999) A Proportional hazards model for the subdistribution of a competing risk. Journal of the American Statistical Association, 94:446, 496-509
Gail M. (1975) A review and critique of some models used in competing risk analysis. Biometrics 31(1):209-222.
Gray RJ. (1988) A class of k-sample tests for comparing the cumulative incidence of a competing risk. Annals of Statistics 16:1141-1154.
IARC Working Group on the Evaluation of Carcinogenic Risks to Humans (2004) Betel-quid and areca-nut chewing and some areca-nut-derived nitrosamines. IARC Monogr Eval Carcinog Risks Hum 85:1-334.
Klein JP, Andersen PK. (2005) Regression modeling of competing risks data based on pseudovalues of the cumulative incidence function. Biometrics 61:223-229.
Pace LE, Keating NL. (2014) A systematic assessment of benefits and risks to guide breast cancer screening decisions. JAMA 311(13):1327-1335. doi: 10.1001/jama.2014.1398.
Prentice RL, Kalbfleisch JD, Peterson AV, Flournoy N, Farewell VT, Breslow NE. (1978) The analysis of failure times in the presence of competing risks. Biometrics 34:541-554.
Putter H, Fiocco M, Geskus RB. (2007) Tutorial in biostatistics: competing risks and multi-state models. Stat Med 26(11):2389-2430.
Satagopan JM, Ben-Porat L, Berwick M, Robson M, Kutler D, Auerbach AD. (2004) A note on competing risks in survival data analysis. Br J Cancer 91(7): 1229-1235.
Shiu MN, Chen THH, Chang SH, Hahn LJ. (2000) Risk factors for leukoplakia and malignant transformation to oral carcinoma: a leukoplakia cohort in Taiwan. Br J Cancer 82:1871-1874.
Tai BC, Machin D, White I, Gebski V; EOI (The European Osteosarcoma Intergroup). (2001) Competing risks analysis of patients with osteosarcoma: a comparison of four different approaches. Stat Med 20(5):661-684.
Wu GH, Auvinen A, Yen AMF, Hakama M, Walter SD, Chen HH. (2012a) A Stochastic Model for Survival of Early Prostate Cancer with Adjustments for Leadtime, Length Bias, and Over-detection. Biom J 2012a;54(1):20-44.
Wu GH, Auvinen A, M&;auml;&;auml;tt&;auml;nen L, Tammela TL, Stenman UH, Hakama M, Yen AM, Chen HH. (2012b) Number of screens for overdetection as an indicator of absolute risk of overdiagnosis in prostate cancer screening. Int J Cancer 131(6):1367-1375. doi: 10.1002/ijc.27340.
Yen AMF, Chen SC, Chen THH. (2007) Dose-response relationships of oral habits associated with the risk of oral pre-malignant lesions among men who chew betel quid. Oral Oncol 43:634-638.