|摘要: ||風險評估的研究中，劑量反應模型可以用來對資料作配適，並加以預測在某特定劑量下 反應的機率，也可經由模型所產生的參數估計進一步陳述資料 (summarize data)。大 多數情況，反應比例為100p%時所需劑量是值得我們去探討的。此時的劑量定義為100p% 有效劑量(effective dose)，並註記為ED100p 。一般而言，我們可以根據模式以建立ED100p 的區間估計。ED100p 信賴區間的估計方法已經在許多單一毒素或化學物的劑量反應模式被 建構完成。相較之下，並沒有太多的研究去探討累積風險評估下有效劑量之區間估計。 陳等人(2001, 2003, 2005)利用劑量加法 (dose-addition) 的概念，提出了一個新劑量模 式以估計混和物聯合暴露所產生的累積風險 (cumulative risk)，但是卻未特別針對本 主題說明與討論。本計畫目的在於累積風險評估下，提出計算混和物有效劑量的信賴區 間之方法。本計畫將對二分反應變項，延伸陳等的模式，並應用delta, likelihood, and parametric bootstrap approaches 等估計方法用以發展出不同的區間估計式。本計畫除了提 出估計過程必要的程式或演算法，期望以實務資料驗證提出的區間估計方法，並運用電 腦資料模擬做相互比較，以評估不同估計式間的差異。
In the filed of risk assessment, when a dose-response model is fitted to data, the model can be used not only to predict the probability of response at a given dose but also to summarize the data through the associated parameter estimates. In many cases, it is desirable to estimate the dose associated with 100p% response. Such dose is usually defined as the p% effective dose and denoted by ED100p. Generally, one can construct, under the model, an interval estimation of the ED100p. Various methods for calculating confidence intervals of ED100p have been well established for a single toxin or chemical. Comparatively, less attention has been focused on interval estimation within the context of cumulative risk assessment. Chen et al. (2001, 2003, 2005) have proposed an useful approach applying dose-addition for estimating cumulative risk from exposure to multiple chemicals, but have not addressed this problem in their work. The objective of this proposal is to present procedures calculating confidence intervals for mixture effective dose of multiple chemicals. Different interval estimators based on delta, likelihood, and parametric bootstrap approaches will be derived and constructed under an extended dose-addition model for quantal response which contains the model of Chen et al. (2001) as a special case. Necessary program or algorithm for calculating the desired estimates and the confidence intervals will be provided. We will demonstrate the use of the derived interval estimators with some real data sets (from published books or articles) or simulated data, and their performance will be also investigated via computer simulation.