從臺灣肝癌發現高比例帶有馬兜鈴酸特有的基因突變特徵研究的新聞事件：以流行病學觀點評述

Misclassification and Bias Publicized in Research Results of Aristolochic Acid Mutational Signature in Taiwan HCCs

2017年10月國際期刊《科學轉譯醫學》（Science Translational Medicine）報導從臺灣肝癌組織檢體中發現帶有馬兜鈴酸特有的基因突變特徵（aristolochic acid mutational signature）的比例並沒有因為2003年法令禁止使用含馬兜鈴酸植物而下降的研究，這個研究結果引起科學家們懷疑在政府法令禁止後是否民眾仍然持續地使用含有馬兜鈴酸的植物。依據Rozen團隊主張2003年之前與2003年之後馬兜鈴酸的暴露比例並沒有因法令禁止而下降，本文將進行對研究論文「Aristolochic acids and their derivatives are widely implicated in liver cancers in Taiwan and throughout Asia」以2003年法令禁止使用含馬兜鈴酸植物為切點，依據其研究目的、研究方法、研究結果及結論以流行病學的觀點評述。如果用2003年法令禁止使用含馬兜鈴酸植物的當年做切點來檢視臺灣肝癌組織檢體中發現帶有馬兜鈴酸特有的基因突變特徵，的確比例並沒有因為禁止使用而下降。然而經過敏感度分析（sensitivity analysis），發現當用2006至2012年個別的年份當切點來檢視時，帶有馬兜鈴酸特有的基因突變特徵比例逐年明顯的下降，比較於2003年以前跟2003年以後帶有馬兜鈴酸特有的基因突變特徵比例從上升4.6%至2012年之前跟2012年之後下降19.7%，差距有24.3個百分點。同時，僅用北臺灣兩個醫學中心的肝癌檢體推論代表臺灣的整體肝癌現況，產生樣本不具代表性（representative）及在缺乏暴露資料的揭露及校正不同國家人口數的情況下，做跨國的比較，亦產生選擇性偏差（selection bias），導致有過度推論之嫌。評讀一篇科學性的論文，應該時時秉持猜測與否證的態度，這一篇刊登在《科學轉譯醫學》的論文，由於錯誤分類（misclassification）產生錯誤的推論，以及用北臺灣有限的小樣本推論臺灣現況，帶入選擇性偏差的干擾因素（selection bias）。本篇以流行病學觀點評述的論文之重要性在於使醫療從業人員學習如何客觀的判斷一個新的研究成果。 Science Translational Medicine published article “Aristolochic acids and their derivatives are widely implicated in liver cancers in Taiwan and throughout Asia”, has made statements for no difference in the prevalence or in the numbers or proportion of aristolochic acid mutational signatures in Taiwan hepatocellular carcinomas (HCC) before and after 2003, when Taiwan government legally banned herbal remedies containing aristolochic acid. This upraises concerns if public is consistency using aristolochic acid contained herbals. The aim of this article is to comment and illustrate underlying bias in this published research results. Authors did not take time trend analysis into consideration in the final analysis and reported no percentage difference for the prevalence of aristolochic acid mutational signature by using the cut-point 2003, before and after. However, when two additional alternative cut-points were selected, a decreasing trend from 71.4% in 2006 to 63.6% in 2012 for the attributable proportion of detected AA mutational signature was found. And, cross country comparisons areharshly evaluated by using sample collected from just two north medical centers are bizarre and overestimate its inferences, with poor disclosure of relative information and conspicuously disregard of denominator for each country other than Taiwan. It is suggested to go through the process of conjecture and refutation to scientific articles back and forth, in order to efficiently prevent misclassification and biases.