糞便潛血梯度為基礎評估大腸直腸癌篩服務計畫之統計模型

Abstract

背景：儘管利用化學法糞便潛血試驗進行大腸直腸癌篩檢在降低死亡率的效益已經經由隨機對照試驗證明，但這並不代表大規模服務性篩檢也能得到相同的益處。再者，使用更精確的免疫化學糞便檢查於大規模服務性計畫之效益評估除了有其科學實證上的探索價值之外，更能提供有效醫療資源分配之參考。 目的：我的實習報告主要目的在發展以糞便潛血梯度為基礎之大腸直腸癌篩檢服務計畫評估工具，架構於三階段馬可夫模型進一步開發評估臺灣大腸癌篩檢效益之統計模型，並將建立的模型應用於其他國家。最後，透過結合臺灣和其他國家大腸直腸癌服務性篩檢計畫的資料建立貝氏卜瓦松回歸模型，考慮不同篩檢參與率下調整選擇性偏差後的臨床偵測大腸直腸癌個案之相對風險。 材料與方法：本報告分析由實習單位—臺灣篩檢評估中心所提供2004年至2014年臺灣全國大腸直腸癌篩檢資料用於統計模型開發，以三階段馬可夫模型為基礎，探究糞便潛血濃度在疾病發生（啟動因子）和進展（促進因子）之梯度作用大小，並運用貝氏卜瓦松回歸模型整合來自臺灣和荷蘭篩檢計劃的資料計算調整選擇性偏差後的篩檢效益。 結果：糞便潛血濃度劑量反應關係不僅存在臨床症前期的發生率，也存在臨床症前期到臨床期的轉移速率，且前者較後者梯度關係更為顯著。使用免疫法糞便潛血試驗在篩檢率為60%及切點值為20g/g時大腸直腸癌臨床偵測個案風險下降了24.4%(95% CI 18.3%-30.1%; RR=0.756 (95% CI: 0.699-0.817))。 將本模式應用於荷蘭資料，結果發現以他們開始進行全國性篩檢的情境之下（出席率為71.3％，切點值為15g/g）預期大腸直腸癌臨床偵測個案風險可下降42.7%(95% CI: 38.3-46.7%)，若荷蘭在計畫執行半年後的切點值調整(提高為50g/g)狀態，預期大腸直腸癌臨床偵測個案風險下降至37% (95% CI: 32.4%-41.3%)。以臺灣情境(篩檢率為60％，切點值為20g/g)之下進行貝氐卜瓦松迴歸分析進行推估，則意向治療在大腸直腸癌臨床偵測個案風險之相對危險性是0.30 (95% CI: 0.30-0.31)。 結論：本研究針對使用免疫法糞便潛血試驗之篩檢計畫，發展一個考慮了糞便潛血濃度梯度效應成功地開發評估大腸直腸癌篩檢的統計模型，此已應用於其他國家，如荷蘭，並進行效益推估，並可用以做為以統合分析證明免疫法糞便潛血試驗篩檢在大腸直腸癌臨床偵測個案風險之降低效益。

Background Despite the efficacy of colorectal cancer screening in terms of mortality reduction using stool-based tool of gFOBT have been proved by using randomized controlled trial, it does not imply the same benefits that result from population-based screening programs. Further, the incremental efficacy brought by the introduction of new tool such as immunochemical fecal test (FIT) is of great interest, not only for the scientific reasons but also for the efficient allocation of medical resources. Objective My practicum aims to develop a f-Hb-gradient-guided statistical model underpinning with the Markov process for evaluation of service colorectal cancer screening based on Taiwan National Colorectal Cancer Screening Data, and apply the proposed model to other countries, and finally to provide a self-selection bias adjusted relative risk of FIT screening with different attendance rates by using Bayesian Poisson regression model incorporating data from Taiwanese and Netherlands program. Methods The data on screening registry of Taiwan Nationwide Colorectal Cancer Screening between 2004 and 2014 provide by the Centre of Taiwan Screening Programme Evaluation are used for the development of the statistical models for the evaluation of service screening. We developed a f-Hb-gradient-guided three-state Markov model for CRC natural history, of which the gradient role of fecal hemoglobin on the incidence (initiators) and progression (promoters) was quantified. The effectiveness from screening adjusted with self-selection bias as assessed using Poisson regression model with Bayesian underpinning to incorporate evidence from Taiwan and Netherlands screening programs. Results There are distinct dose-response relationships for FIT levels not only noted for the incidence of entering PCDP but also the transition from PCDP to CP. The gradient relationship for the incidence of PCDP was more remarkable than that for transition for PCDP to CP. Given 60% screening rate and the cutoff at 20g/g, the reduction in clinical CRC as a result of FIT test was 24.4% (95% CI 18.3%-30.1%; RR=0.756 (95% CI: 0.699-0.817)). The expected clinical CRC reduction was 42.7% (95% CI: 38.3-46.7%) given Netherlands scenario in the initiation of their nationwide screening (attendance rate of 71.3% with a cut-off at 15 g/g), and reduced to 37% (95% CI: 32.4%-41.3%) with cut-off elevated to 50 g/g in the second half of their program. The intention-to-treat RRs is 30% (95% CI: 30-31%) with 60% attendance rate given cut-off level of 20 g/g. Conclusion In this study, we successfully develop a statistical model for the evaluation of CRC screening program using FIT as screening tool with consideration of f-Hb-gradient. The proposed f-Hb-gradient-guided model has been applied to other data like the Netherland to yield the results of meta-analysis on the results of effectiveness in reducing CRC in clinical phase (CP).