Empirical Comparison of Five Parametric Curves in Fifty Dose-Response Studies

One of the most common parametric curves used for dose response modeling and target dose estimation is the Emax model. Previous meta-analyses have shown that the Emax provides an adequate description of the dose-response relationship for many studies, but direct comparisons with other model families were not reported. We addressed two questions in this paper. First, does any one model family obtain superior fits to the data across many dose response studies? Second, which model families should generally be included in the candidate model set? We performed an exploratory analysis to quantitatively compare goodness-of-fit between the Emax and competing two- and three-parameter model families. No single model family obtained the best penalized fit in most studies. While the Log-linear model often yielded the lowest information criterion, a cluster analysis revealed that the Emax model achieved the most robust fit across studies.

用于剂量反应建模与靶剂量估计的最常用参数曲线之一为Emax模型（Emax model）。既往荟萃分析表明，Emax模型可对诸多剂量反应研究中的剂量-响应关系做出合理描述，但尚未见其与其他模型族的直接比较报道。本文针对两个问题展开研究：其一，是否存在某一模型族可在多数剂量反应研究中获得更优的数据拟合效果；其二，候选模型集通常应包含哪些模型族。我们开展了探索性分析，对Emax模型与其他竞争的二参数、三参数模型族的拟合优度进行定量比较。结果显示，并无单一模型族能在多数研究中获得最佳惩罚拟合效果。尽管对数线性模型往往能取得最低的信息准则值，但聚类分析结果表明，Emax模型在各研究间的拟合稳健性最优。