請用此 Handle URI 來引用此文件:
http://tdr.lib.ntu.edu.tw/jspui/handle/123456789/707
標題: | 糞便免疫化學法大腸癌族群篩檢之間隔癌統計方法 Statistical Methods for Interval Cancers from Fecal Immunochemical Test in Population-based Screening for Colorectal Cancer |
作者: | Wen-Feng Hsu 許文峰 |
指導教授: | 陳秀熙(Hsiu-Hsi Chen) |
關鍵字: | 大腸直腸癌,大腸癌篩檢,糞便潛血免疫法,大腸鏡,間隔癌症, Colorectal cancer,Colorectal cancer screening,Fecal Immunochemical Test,Colonoscopy,Interval cancer, |
出版年 : | 2019 |
學位: | 博士 |
摘要: | 研究背景
當糞便潛血檢查被廣泛使用於以族群為主之大腸癌篩檢時,對於利用潛血檢測之兩階段篩檢下之篩檢間隔癌相關品質確保議題,無論如何強調都不為過。篩檢間隔癌有許多類型,包含腸鏡後的篩檢間隔癌及因潛血檢測所產生的篩檢間隔癌,雖然兩者已有清楚定義,但對於兩類篩檢間隔個案之相關量性研究迄今仍十分缺乏,包含兩類篩檢間隔個案在大腸癌自無症狀至有症狀之進展表現上,對第一階段糞便潛血檢查及第二階段腸鏡檢查之效度影響、以及在存活分析上,兩類篩檢間隔個案與其他篩檢偵測癌症在考量截切資料及前導期校正後之比較,以及與未轉介、未參加篩檢之大腸癌個案存活互相比較。此外,在區分篩檢間隔個案為新發個案或篩檢工具造成偽陰性個案之方法學發展,也是相當有興趣的議題,而這些也可能伴隨因第二階段腸鏡檢查之工具效度、不同篩檢間隔、不同篩檢率、與不同個人特質而形成更為複雜之議題。 研究目的 本論文旨在 1. 進行系統性文獻回顧估計不同類型糞便潛血檢查(化學法及免疫法)之敏感度,並考慮篩檢參與率及大腸鏡檢轉介率, 2. 發展一新穎統計模式估計與大腸直腸癌疾病進展相關的參數以及因應此兩階段篩檢模式在免疫法糞便潛血與大腸鏡檢階段所對應的敏感度,進一步估算篩檢間隔癌來自於新發生癌症及前次篩檢偽陰性兩種可能所佔的比例,及這兩個比例在篩檢間隔變換或應用於不同人口學特質之差異, 3. 發展一進階統計模式估計估算在免疫法糞便潛血後之間隔癌(FIT Interval Cancer)與大腸鏡檢後之間隔癌(Colonoscopy Interval Cancer)其分別來自新發癌症及前次偽陰性之比例,並模擬不同篩檢間隔之下之差異, 4. 比較免疫法糞便潛血後之間隔癌(FIT Interval Cancer)、大腸鏡檢後之間隔癌(Colonoscopy Interval Cancer)及篩檢偵測個案之長期存活狀況,並與調整前導期之篩檢偵測個案及進一步調整截切之首次篩檢偵測個案、大腸鏡檢未順從個案、及未參與篩檢癌症個案之長期追蹤狀況進行比較, 5. 估算篩檢間隔由2年改為3年後因免疫法糞便潛血後之間隔癌(FIT Interval Cancer)與大腸鏡檢後之間隔癌(Colonoscopy Interval Cancer)增加後所造成死亡人數增加後的潛在人命年損失(Potential Years of Life Lost, PYLL)。 研究方法 1. 貝氏Beta-binominal統合分析 利用糞便潛血化學法為主之篩檢系統性文獻回顧資料及現行台灣糞便潛血免疫法篩檢之資料進行三種敏感度包含工具敏感度、整體敏感度及計畫敏感度之貝氏Beta-binominal統合分析。 2. 廣義非線性測量誤差廻顧模式 本論文發展廣義非線性測量誤差廻歸模式以對臨床症前期癌症發生率及自無症狀轉移至臨床症狀之速率以三階段馬可夫模式分別以貝氏及非貝氏方法進行估計,模式並考量二階段篩檢之敏感度。為評估篩檢間隔癌在不同篩檢間隔及篩檢工具之影響,就不同年齢層、性別之個別速率及敏感度之估計結果應用於不同假設設計之試驗設計上。 更進一步將進階廣義非線性測量誤差廻歸模式,對臨床症前期癌症發生率及自無症狀轉移至臨床症狀之速率進行估計外,並分別估計腸鏡敏感度及潛血檢測敏感度,用以反映腸鏡後的篩檢間隔癌及因潛血檢測所產生的篩檢間隔癌。 3. 比較不同大腸癌偵測型態存活,包含篩檢偵測大腸癌、因潛血檢測所產生的篩檢間隔癌、腸鏡後的篩檢間隔癌、以及臨床偵測大腸癌包含未轉介鏡檢大腸癌以及未參加篩檢之大腸癌個案。此部份除利用寇斯比例風險廻歸模型進行多變量分析,亦對篩檢測偵個案進行截切資料及前導期校正。 資料來源 本論文資料來源因應上述三個部份描述如下: 1. 利用文獻搜尋找到以糞便檢測為主之族群大腸直腸癌篩檢之文獻,共取得5篇以化學法為工具及2篇以免疫法為工具的研究, 2. 利用臺灣大規模族群大腸直腸癌篩檢計畫估計廣義非性迴歸測量誤差模式相關參數,該篩檢計畫的目標族群為50-69歲民眾,篩檢間隔為兩年,篩檢資料區間介於2004年至2014年間, 3. 利用參與上述篩檢計畫目標族群在2004年至2012年間發生的大腸直癌個案,計8992名個案,依照不同偵測模式追蹤其後續存活狀況至2016年止。 結果 第一部份:不同類型糞便潛血檢查工具之敏感度 統合分析結果顯示化學法糞便潛血篩檢工具敏感度為52.5% (95%信賴區間: 51.2%-53.8%),整體敏感度為49.6% (95信賴區間: 48.3%-50.9%),計畫敏感度則降至33.3% (95%信賴區間: 32.0%-34.5%)。而在糞便潛血免疫法為篩檢工具的分析結果中,工具敏感度為80.9% (95%信賴區間: 80.0%-81.7%) 整體敏感度為65.7% (95%信賴區間: 64.6%-67.3%)。其中工具及整體的敏感度皆高於化學法篩檢工具(gFOBT),但因為較低的篩檢參與率,免疫法的計畫敏感度卻較化學法低,為34.7% (95%信賴區間: 33.7%-35.7%)。 第二部份:以免疫糞便潛血法進行篩檢之篩檢間隔個案 應用馬可夫二階段方法,以台灣兩年一次大腸直腸癌篩檢資料估計結果得到基礎發生率為0.00151 (95%信賴區間: 0.00147-0.00155),疾病進展速率為0.36 (95%信賴區間: 0.34-0.38)。若將測量誤差也納入考量之模型,工具敏感度估計為75.47% (95%信賴區間: 72.99%-77.80%)。性別及年齡在癌症發生中扮演的角色,男性的影響程度為女性的1.75倍(95%信賴區間: 1.68-1.82),老年為年輕的1.79倍 (95%信賴區間: 1.73-1.86)。性別及年齡在癌症疾病進展中扮演的角色而言,男性的影響程度為女性的0.82倍(95%信賴區間: 0.77-0.87),老年為年輕的0.91倍 (95%信賴區間:0.86-0.97)。 在兩年一篩的間隔個案中大多數為新發生的大腸癌個案(約佔68.8%),若對應至篩檢政策為一年一篩時會降低至61.1%,若為三年一篩時則增加至74.7%。對兩階段糞便潛血篩檢,使用廣義非線性錯誤分類廻歸模式所估計之糞便潛血敏感度為71.9% (95%信賴區間: 71.0-72.8%),大腸鏡檢敏感度為93.6% (95%信賴區間: 91.9-94.9%)。依據此估計結果,就糞便潛血篩檢而言,分別有61%的篩檢間隔為新發篩檢間隔癌,而有39%為偽陰性篩檢間隔癌。就大腸鏡檢間隔癌而言,分別有88%的篩檢間隔為新發篩檢間隔癌,而有12%為偽陰性篩檢間隔癌。 第三部份:不同偵測模式大腸直腸存活分析 在調整年齡、性別、病灶位置及治療後,相較於未參與篩檢個案,大腸鏡檢未順從個案、免疫法糞便潛血後之間隔癌、大腸鏡檢後之間隔癌、第一次及後續篩檢偵測個案死於大腸癌的調整危險對比值分別為0.56(0.50-0.64)、0.57 (0.52-0.64)、0.42 (0.32–0.54)、0.30 (0.27-0.33)及0.22 (0.19-0.26)。對篩檢測偵個案進行截切資料及前導期校正後,後續及第一次篩檢偵測個案死於大腸癌的調整危險對比值分別為0.25 (0.22-0.30)及0.63 (0.58–0.69)。 利用廣義非線性聯合測量誤差模式的估計結果及不同偵測模式個案之存活率,可以得到兩年一次篩檢改為三年一次篩檢將會導致2337年的潛在人年損失,其2314人年損失是來自糞便潛血檢查、23人年損失來自大腸鏡。 結論 本論文發展一新穎廣義非線性測量誤差迴歸模式,用以評估大規模大腸直腸癌篩檢計畫因糞便潛血及大腸鏡在間隔癌個案的分別貢獻程度,在考量前導期及截切性質下的存活分析後,估算篩檢間隔改變對大腸直腸癌死亡減少的影響。 Background. While fecal immunochemical test (FIT) has widespread use in population-based colorectal cancer (CRC) screening quality assurance of interval cancers (ICs) ensuing from two-stage screening with FIT test cannot be over emphasized. Various types of ICs (including colonoscopy ICs and FIT ICs) and sensitivity have been defined but there are no formal quantitative studies yet addressing how both ICs in relation to the disease progression of CRC from asymptomatic phase to symptomatic phase, the performance of screening with FIT at first stage and colonoscopic examination at second stage, and their survival rates in comparison with screen-detected CRCs with adjustment for truncation and lead-time, and CRCs from non-referral, and non-participants. One of intractable methodological issues in modelling ICs is that newly developed CRCs after screening cannot be directly separated from false negative CRCs missed at the previous screen on the basis of empirical data. This issue is further complicated by the performance of the second-stage confirmatory method such as colonoscopy, inter-screening interval, and screening rate, and demographic features. Aims. The objectives of this thesis are to 1. conduct a systematic review and meta-analysis of estimating different types of sensitivity for g-FOBT and FIT making allowance for attendance rate of the uptake of screening and non-referral of colonoscopy; 2. develop a new statistical model for estimating the transition parameters pertaining to the disease natural history of CRC and the sensitivity of the overall two-stage screening resulting from FIT and colonoscopy in order to quantify the respective contributions of false-negative and newly developed CRCs to total ICs and to estimate the corresponding figures varying with different inter-screening interval and demographic features; 3. develop an advanced statistical model to quantifying the respective contributions of false-negative and newly developed CRCs to FIT ICs and colonoscopy ICs, respectively, to quantify the respective contributions of false-negative and newly developed CRCs to total ICs and to estimate the corresponding figures varying with different inter-screening intervals; 4. elucidate and compare the long-term survival of FIT ICs and colonoscopy ICs in comparison with the lead-time and truncation adjusted survival of prevalent screen-detected CRC and subsequent screen-detected CRC, non-referral (not complying with colonoscopy) CRC, and CRCs from non-participants. 5. estimate potential years of life lost (PYLL) when excess deaths attributed to FIT ICs and colonoscopy ICs were averted by changing inter-screening interval. Methods. Part I: Bayesian Beta-binominal Meta-analysis Three degree of sensitivity: test, episode and program sensitivity in gFOBT-based and FIT-based screening programs of previous studies and the current Taiwanese study were systematically reviewed and a meta-analysis was conducted by Bayesian beta binomial meta-analysis to account for differences of these three sensitivities between gFOBT screening program and FIT screening program. Part II: Generalized non-linear measurement error regression model A generalized non-linear measurement error regression model was developed to estimate the pre-clinical incidence rate and the transition rate from asymptomatic phase to symptomatic phase underpinning a three-state Markov model and the sensitivity of two-stage screening program with Bayesian and non-Bayesian method. To assess the impact of inter-screening interval on interval CRC, the estimated results on age and sex specific rates and sensitivities were further applied to the proposed multi-arm trials. Another advanced generalized non-linear measurement error regression model was further developed to estimate the pre-clinical incidence rate and the transition rate from asymptomatic phase to symptomatic phase underpinning a three-state Markov model and the joint estimates of FIT-based and of colonoscopy-based sensitivity to reflect FIT ICs and colonoscopy ICs. To assess the impact of inter-screening interval on two separate types of interval CRC were further applied to the proposed multi-arm trials. Part III: Lead-time- and Truncation-adjusted Survival The survival status of CRCs was compared stratifying with different detection modes: screen-detected CRC, FIT IC, colonoscopy IC, and clinically diagnosed CRC (colonoscopy noncompliers, and screening non). Multivariable analyses were conducted with Cox proportional hazards regression models. We also further adjusted the lead-time and truncation adjustment for screen-detected CRCs. Data Sources. There are three parts of retrieving data for this thesis. The first part consists of studies on evaluating the sensitivity of tests for colorectal cancer were derived from 5 gFOBT-based screening programs, and 2 FIT-based screening programs. The second part is to use the empirical data on Taiwanese Nationwide Colorectal Cancer Screening Program recruiting residents aged 50 to 69 years to have the uptake of a biennial FIT during 2004-2014 for estimating the parameters encoded in generalized non-linear regression measurement error model. The third part is to use the data of CRCs with various detection modes resulting from Taiwanese population-based FIT screening program. Totally 8,992 CRCs were identified from the cohort who were considered as eligible for screening during 2004-2012 in Taiwanese CRC Screening Program and were followed up until 2016. Results. Part I: Meta-analysis of Different Types of Stool-based Sensitivity The results of gFOBT-base screening program based on meta-analysis show that the pooled test sensitivity was 52.5% (95CI: 51.2%-53.8%), and the episode sensitivity was 49.6% (95CI: 48.3%-50.9%). The program sensitivity was 33.3% (32.0%-34.5%). The results of FIT-based screening program based on meta-analysis show that the test sensitivity was 80.9% (95CI: 80.0%-81.7%) and episode sensitivity was 65.7% (95CI: 64.6%-67.3%). The program sensitivity of FIT-based screening program was reduced to 34.7% (95 CI: 33.7%-35.7%). Part II: Interval cancer in FIT-based screening program The baselined incidence rate and progress rate were estimated as 0.00151 (95% CI: 0.00147-0.00155) and 0.36 (95%CI:0.34-0.38), respectively. The sensitivity of screening test was estimated as 75.47% (95%CI:72.99%-77.80%). The effect size of male sex and old age on the occurrence of CRC was estimated as 1.75 (95%CI:1.68-1.82) and 1.79 (95%CI:1.73-1.86), respectively. The effect size of male sex and old age on the progression of CRC was estimated as 0.82 (95%CI:0.77-0.87) and 0.91 (95%CI:0.86-0.97), respectively. FIT interval CRCs from biennial regime mainly resulted from newly developed CRCs (68.8%). The corresponding figures were reduced to 61.1% for annual program but increased to 74.7% for triennial program. The estimated results based on the generalized non-linear joint measurement error model gave the sensitivity of 71.9% (95% CI: 71.0-72.8%) and 93.6% (95% CI: 91.9-94.9%) for FIT and colonoscopy, respectively. Based on these estimates, 61% and 39% of ICs resulted from the path of newly develop and false negative, respectively for FIT screen. The corresponding figure for colonoscopy is 88% for newly developed, and 12% for false negative. Part III: Survival by detection model After adjusting for age, sex, location, and treatment, compared with that for CRCs in screening nonparticipants, aHR of was 0.56(0.50-0.64) for CRC in colonoscopy noncompliers, 0.57 (0.52-0.64) for FIT IC, 0.42 (0.32–0.54) for colonoscopy IC, 0.30 (0.27-0.33) for prevalent screen-detected CRC, and 0.22 (0.19-0.26) for subsequent screen-detected CRC. After adjustment for lead-time and truncation, the aHR was 0.25 (0.22-0.30) for subsequent screen-detected CRC, and 0.63 (0.58–0.69) for prevalent screen-detected CRC. Based on the estimated results of generalized non-linear joint measurement error model and survival for each detection modes, the biennial program compared with triennial one resulted in the life-year gained by 2337 person-years, among which 2314 person-years and 23 person-years resulted from FIT test and colonoscopy, respectively. Conclusions. A new generalized non-linear measurement error regression models was developed to model contributory causes of FIT and Colonoscopy ICs to estimate the impact of inter-screening interval on the reduction of deaths from CRC attributed to each type of ICs making use of the lead-time and truncation-adjusted survival model. |
URI: | http://tdr.lib.ntu.edu.tw/handle/123456789/707 |
DOI: | 10.6342/NTU201902518 |
全文授權: | 同意授權(全球公開) |
顯示於系所單位: | 流行病學與預防醫學研究所 |
文件中的檔案:
檔案 | 大小 | 格式 | |
---|---|---|---|
ntu-108-1.pdf | 2.46 MB | Adobe PDF | 檢視/開啟 |
系統中的文件,除了特別指名其著作權條款之外,均受到著作權保護,並且保留所有的權利。