以糞便血色素濃度預測大腸直腸癌死亡之替代終點研究

Abstract

研究背景及目的 欲了解危險因子是否與感興趣的結果，例如疾病之死亡有相關，通常需要長時間的追蹤及觀察，而替代終點為驗證此因果關係之取代方法，但需經嚴謹的統計方法予以定義。本研究欲應用Prentice與Freedman所提出檢視替代終點之方法，利用社區為基礎之大腸直腸癌篩檢世代資料來評估對大腸直腸癌死亡風險具劑量效應之糞便血紅素濃度因子，是否可利用大腸直腸癌癌症病理分期及篩檢偵測模式作為大腸直腸癌死亡之替代終點進行檢視及驗證。 研究方法 本研究利用三個以社區為基礎之大腸直腸癌篩檢族群，研究樣本共204,052位參與糞便免疫法篩檢至少一次以上之40歲以上民眾，平均經4.14年追蹤時間之資料進行分析。基於Prentice’s替代終點檢視準則之於存活分析方法，以及Freedman’s替代終點準則之於廣義線性模式方法，進一步來驗證糞便血紅素濃度在大腸直腸癌死亡具劑量效應下，驗證癌症病理分期和篩檢偵測模式可否做為大腸直腸癌死亡之替代終點，利用增速事件時間分析模式和羅吉斯迴歸模式配合社區實證資料以檢定Prentice及Freedman所提出之替代終點驗證準則。 研究結果 首次糞便血紅素濃度測量值在篩檢偵測和臨床偵測癌症之表現不同，篩檢偵測癌症潛血值較低並有顯著差異(357.44±730.26與426.74±767.22, P<0.001)。在不同的腫瘤病理分期間，其首次糞便血紅素濃度亦有不同。這些發現皆符合Prentice’s替代終點驗證準則統計上若p則q，若非q則非p之定義。在調整年齡、性別和腫瘤病理分期後，就沒有逐漸增加的糞便血紅素濃度效果，迴歸係數則呈現反向的變化，而腫瘤病理分期在大腸直腸癌死亡風險表現上，迴歸係數仍維持相同方向性。這樣的結果說明了在腫瘤期別下，糞便血紅素濃度對大腸直腸癌死亡之條件獨立性，符合Prentice’s及Freedman’s替代終點驗證準則，這樣的結果在二元迴歸模式中也有相類似的發現。 由於調整和未調整替代終點後的迴歸係數間有相反的效果，Freedman's的解釋比例在所有的分析幾乎是100%，但僅調整腫瘤期別的二元迴歸模式下為96.8% (95%信賴區間為96.4%-97.2%)。藉由用首次糞便血紅素濃度分別對侵襲性大腸直腸癌及大腸直腸癌死亡做預測，對侵襲性大腸直腸癌的作業接受曲線下面積為74.6 %，十分很接近對大腸直腸癌死亡之72.2%。而利用糞便血紅素濃度對大腸直腸癌死亡做預測(72.2%)也與利用腫瘤期別之面積相當接近(73.8%)。 結論 本研究呈現如何應用替代終點的統計定義去評估某些歸因因子或危險因子對長期追蹤結果的效應，這樣的優點可免除長期追蹤的時間並同時獲得統計效益。本研究成功的應用Prentice’s及Freedman’s替代終點驗證統計準則並結合迴歸分析於探討糞便血紅素濃度與大腸直腸癌死亡之關係，驗證當長期追蹤結果為大腸直腸癌死亡時，腫瘤病理分期或篩檢偵測模式可做為大腸直腸癌死亡之替代終點。

Background Surrogate endpoint under the context of strict statistical definition is a good idea for dispensing with long-term follow-up when the effect of certain risk factor on long-term sequence is required to be validated. The recent postulate on the effect of fecal hemoglobin (f-Hb) concentration on colorectal cancer (CRC) mortality is a good example. Aim We aimed to assess whether the effect of an incremental increase in f-Hb on the risk of death from CRC can be validated by using tumor stage and detection mode as surrogate endpoint under strict statistical definition of Prentice’s and Freedman’s criteria based on data on community-based CRC screening. Material and Methods A total of 205,042 participates with average of 4.14 years of follow-up, comprising three cohorts aged over 40 years who underwent CRC screening with fecal immunochemical test, were enrolled in our analysis. Prentice’s criteria under the context of survival domain and Freedman’s criteria under the context of generalized linear model were applied to validate the definition of the surrogate endpoint such as tumor stage and detection mode for the effect of f-Hb on CRC mortality. Accelerated failure time (AFT) model and the logistic regression model were operated to test these two conceptual criteria. Result The average baseline f-Hb concentration was different between screen-detected CRC and clinical-detected CRC. The difference of average baseline f-Hb across tumor stages was also noted. An incremental increase in both of baseline and updated f-Hb on the risk of advanced CRC cancer were found. These findings met statistical definition for surrogate endpoint using Prentice’s criteria with full equivalent logic argument (p⇔q). There was lacking of incremental effect of f-Hb on the risk for CRC death after adjustment for age, gender and tumor stages, showing there is a dramatic change of regression coefficients regarding the incremental effect of f-Hb. The effect of tumor stage on CRC death remained identical. These finding upheld Prentice’s and Freedman’s criteria on the auxillary argument about conditional independent effect of the f-Hb group on CRC death given tumor stage information has already known. Due to the opposite sign of regression coefficients between the unadjusted and adjusted for surrogate endpoints, Freedman's Proportion Explained (FPE) in all analysis were 100% but only 96.8% (95% CI: 96.4%-97.2%) of PE was found in binary model making allowance of tumor stage only. The 74.6 % of area under curve (AUC) for advanced CRC was close to that of 72.2% of AUC for CRC death by using baseline f-Hb as predictor. The predictive validity for f-Hb (72.2%) was also close to that for advanced CRC (73.8%). Conclusion We demonstrate how to apply statistical definition of surrogate endpoint to evaluate the effect of certain attribute or risk factor on long-term outcome in order to dispense with long-term follow-up and also to gain statistical efficiency. Prentice’s and Freedman’s criteria in combination with regression models were successfully applied to evaluating the effect of f-Hb on CRC mortality by using tumor stage and possibly detection mode as surrogate endpoint.