評估大腸直腸癌族群篩檢過度偵測數量模型

Quantitative Models for Modelling Overdetection in Population-based Screening for Colorectal Cancer

廖翎均(Ling-Chun Liao)

指導教授 ： 陳秀熙

國立臺灣大學/公共衛生學院/流行病學與預防醫學研究所/碩士(2018年)

https://doi.org/10.6342/NTU201802429

摘要

背景:對於目標癌症藉由不同的工具達到早期發現進而降低其死亡率為族群篩檢的主要目的，但此一過程中所衍生的過度偵測問題亦有許多的爭論。隨著族群篩檢對於不同癌症之廣泛運用，如何對篩檢中過度偵測進行量化評估對於篩檢之適切推行有迫切之重要性。目前在族群乳癌與攝護腺癌的篩檢已知存有過度偵測的情形，此兩種主要癌症之族群篩檢計畫之過度偵測亦在許多文獻中有所討論。而大腸直腸癌篩檢運用包含化學法糞便潛血檢查與免疫法糞便潛血檢查之族群篩檢計畫中的過度偵測則甚少被提及。
目的:本論文旨在發展一系列的方法對於族群篩檢中的過度偵測進行評估與量化。目前大腸直腸癌篩檢中的過度偵測在文獻上並未有許多探討，本研究將運用所發展的方法於以化學法以及免疫法為工具之大腸直腸癌族群篩檢。藉由此多種不同方式的評估方法結合大腸直腸癌疾病進展達到量化並評估大腸直腸癌篩檢中的過度偵測機轉。
資料來源:對於大腸直腸癌篩檢本研究運用兩個隨機分派試驗研究(化學法糞便潛血檢查)與台灣族群篩檢資料(免疫法糞便潛血檢查)對於其中之過度偵測進行量性評估與分析。
研究方法:本研究發展三種對於族群篩檢中的過度偵測之量性評估方法。第一種方法為累積發生率曲線圖示法，透過比較隨機分派研究中篩檢組與未篩檢組之晚期癌症與非晚期癌症的累計發生率達到對於過度偵測之量性評估。第二類方法則運用數據模擬，藉由三階段隨機過程描述癌症之病程發展(無疾病期→無症狀可偵測期(臨床症狀前期)→臨床症狀期)達到估計疾病自然進展各階段之進展速度，同時對於篩檢工具敏感度加以考量，運用此實證評估結果推估得到篩檢邀請組中的預期個案數並且與篩檢組實際得到之觀察個案數相比得到過度偵測量性評估結果(標準過度偵測比率)。
本研究運用前述之方法於大腸直腸癌隨機分派研究資料，對於在篩檢計畫中之過度偵測進行量性評估。對於大腸直腸癌癌隨機分派研究則可藉由對照組作為比較基準評估過度偵測之情形；而台灣服務性族群篩檢之過度偵測評估，本研究則以篩檢前期之發生率資料作為比較基準，推估過度偵測比例(SOR，標準個案比之倒數-1)*100%。在前述對於族群篩檢中的過度偵測之了解下，本研究發展基於五階段大直直腸癌疾病進展模式之過度偵測評估方法，結合前述之標準化過度偵測率(SOR)達到量化過度偵測以及釐清其發生之機轉。
結果:運用累積發生率曲線圖示法於兩個大腸直腸癌隨機分派研究顯示英國與丹麥之研究中的過度偵測分別為31.1%與24.9%。但此一運用累積發生率之比較估計得到之數值由於並未將大腸直腸腺腫以及認知提高等因素納入考量，因此可能為具有偏誤之結果。
運用本研究提出之三階段馬可夫模型評估大腸直腸癌篩檢過度偵測之結果顯示，在兩個運用化學法糞便潛血檢查的歐洲隨機分派研究中過度偵測之情形接近9%；而在台灣大型篩檢服務(運用免疫法糞便潛血檢查)則為7%。進一步運用五階段大腸直腸癌疾病進展模式，英國與丹麥研究之過度偵測比率分別為6.1%及9.2%。運用本研究發展之標準化過度偵測率(SOR)對於兩個隨機分派研究評估之結果顯示，在排除過度偵測後，英國與丹麥之研究中SOR分別由1.29與1.16下降至1.16與0.97。此一結果顯示，造成過度偵測之機轉主要源於大腸直腸癌疾病進展中，由早期臨床症前期轉移至晚期臨床症前期之疾病進展路徑所致。
結論:本論文系統性的發展出一系列量化估計過度偵測的統計方法，運用於評估大腸直腸癌族群篩檢中的過度偵測。藉由本研究發展之標準化過度偵測率(SOR)除了可達到對於過度偵測之量性評估外，亦進一步運用臨床症前期進展至臨床期之癌症期別變化達到對於大腸直腸癌篩檢中造成度偵測之機轉有所了解之目的。

關鍵字

大腸直腸癌 ； 族群篩檢 ； 過度偵測 ； 隨機分派

並列摘要

Background
Overdetection resulting from population-based screening for common cancers has been debatable over the past decade, but the formal methodology for quantitative estimation of overdetection is lacking. Contextually, mass screening for breast cancer and prostate cancer has been well documented, but colorectal cancer with faecal occult blood (FOB) test and faecal immunological test (FIT) on this thorny issue has been scarcely highlighted.
Objective
This thesis aims to develop various quantitative methods for overdetection and to apply these methods to colorectal cancer with either guaiac FOB test (gFOBT) or FIT to understand the mechanism of overdetection concerning the natural course of colorectal cancer.
Data sources We used data on two randomised controlled trials on gFOBT and Taiwan population-based service screening using FIT to assess the proposed methods to quantify overdetection in population-based cancer screening.
Method
The graphic method was first proposed by comparing cumulative incidence curves of advanced and non-advanced cancers. There are two modelling approaches used for quantitative methods. First, we developed the standardized overdiagnosis ratio (SOR), and the expected (numerator) is simulated by a three-state stochastic process merely based on data from the screened group against the observed (denominator) either from the control group in the randomised controlled trial or the comparator from the pre-screening period in the service screening. The three state were, free of CRC, pre-clinical detectable phase, and the clinical phase in conjunction with the sensitivity of FOB test. Such a design was applied to two randomised controlled trials and one service screening program to estimate the proportion of overdetection of colorectal cancer resulting from FOB test and FIT by calculating (SOR-1)*100%.
The second proposed modelling approach is the development of a five-state Markov model in conjunction with SOR to quantify overdetection. Besides, to evaluate overdetection, the mechanism of overdetection through the pathway of progression to advanced pre-clinical detectable phase (PCDP) or the pathway to non-advanced clinical phase (CP) based on SOR was assessed in the light of the five-state Markov model.
Results
Using the graphic method, the average of over-detection was 31.1% and 24.9% for the UK trial and the Denmark trial, respectively. However, the estimated proportion of over-diagnosis using the graphic method may be biased because the natural history of adenoma and high awareness in the routine clinical practice have not considered.
According to the proposed three-state disease progression model the 9% overdetection of CRC is expected in population-based screening for CRC with FOBT test based on the two randomised control trials in Europe and 7% overdetection of CRC were noted for FIT while using in a nationwide screening program in Taiwan.
The estimated results derived from the five-state Markov model were 6.1% and 9.2% for the UK and Denmark trial, respectively. Indicated by SOR derived from the five-state Markov model, the important pathway for overdetection in population-based CRC screening is mainly the pathway from non-advanced PCDP to advanced PCDP. The pathway is supported by the evidence that SOR is deflated from the estimated results of 1.29 to 1.16 and 1.16 to 0.97 for the UK and Denmark trial, respectively.
Conclusions
This thesis systematically developed a series of statistical methods including the graphic method and the disease natural progression model for quantitative assessment of over-detection in colorectal cancer. The index of SOR was proposed to both assess the extent of over-detection and also to elucidate the mechanism of how overdetection affects the progression of colorectal cancer from PCDP to CP in the light of information on cancer stage.

並列關鍵字

Colorectal cancer ； population-based screening ； overdetection ； randomised controlled

參考文獻

Auvinen, A., S. M. Moss, T. L. Tammela, K. Taari, M. J. Roobol, F. H. Schroder, C. H. Bangma, S. Carlsson, G. Aus, M. Zappa, D. Puliti, L. J. Denis, V. Nelen, M. Kwiatkowski, M. Randazzo, A. Paez, M. Lujan and J. Hugosson (2016). "Absolute Effect of Prostate Cancer Screening: Balance of Benefits and Harms by Center within the European Randomized Study of Prostate Cancer Screening." Clin Cancer Res 22(1): 243-249.

Chamberlain, J., S. M. Moss, A. E. Kirkpatrick, M. Michell and L. Johns (1993). "National Health Service breast screening programme results for 1991-2." Bmj 307(6900): 353-356.

Chen, T. H., A. M. Yen, J. C. Fann, P. Gordon, S. L. Chen, S. Y. Chiu, C. Y. Hsu, K. J. Chang, W. C. Lee, K. G. Yeoh, H. Saito, S. Promthet, C. Hamashima, A. Maidin, F. Robinson and L. Z. Zhao (2017). "Clarifying the debate on population-based screening for breast cancer with mammography: A systematic review of randomized controlled trials on mammography with Bayesian meta-analysis and causal model." Medicine (Baltimore) 96(3): e5684.

Chen, T. H., H. S. Kuo, M. F. Yen, M. S. Lai, L. Tabar and S. W. Duffy (2000). "Estimation of sojourn time in chronic disease screening without data on interval cases." Biometrics 56(1): 167-172.

Chiu, H. M., S. L. Chen, A. M. Yen, S. Y. Chiu, J. C. Fann, Y. C. Lee, S. L. Pan, M. S. Wu, C. S. Liao, H. H. Chen, S. L. Koong and S. T. Chiou (2015). "Effectiveness of fecal immunochemical testing in reducing colorectal cancer mortality from the One Million Taiwanese Screening Program." Cancer 121(18): 3221-3229.