Hamada C, Yoshino K, Matsumoto K, Nomura M, Yoshimura I
Department of Pharmacoepidemiology, Faculty of Medicine, University of Tokyo, Japan.
J Toxicol Sci. 1998 Aug;23(3):173-81. doi: 10.2131/jts.23.3_173.
An appropriate statistical methodology in toxicity studies has been discussed over the last two decades and many statistical methods have already been proposed. Many practical problems, however, still remain unresolved and most pharmaceutical industries have been using a tree-type algorithm routinely to analyze repeated-dose toxicity study data. In considering routine use of statistical analysis in toxicological studies, standardization of statistical methodology is necessary and the decision tree has an important role. In this article, the problems, relating to tree-type algorithms are summarized. Then we propose a new tree-type algorithm, which targets quantitative data in repeated-dose studies in rodents, usually sample size per group between 10 to 20, based on the following two important principles: "using a parametric method" and "suitable for intuition of toxicologists". An example of its application to actual toxicity study data is demonstrated. The performance of this new method is also evaluated using historical data. However, it should be noted that the intention of this paper is not to make a definite solution of the decision tree. Several other alternatives can be considered. Since there is no single theoretically correct solution of tree-type algorithms, too formal a use of the decision tree is not recommended. We must not forget the exploratory nature of evaluating repeated toxicity data.
在过去二十年里，人们一直在讨论毒性研究中合适的统计方法，并且已经提出了许多统计方法。然而，许多实际问题仍未得到解决，大多数制药行业一直在常规使用树形算法来分析重复剂量毒性研究数据。在考虑毒理学研究中统计分析的常规应用时，统计方法的标准化是必要的，而决策树具有重要作用。在本文中，总结了与树形算法相关的问题。然后，我们基于以下两个重要原则提出了一种新的树形算法，该算法针对啮齿动物重复剂量研究中的定量数据，通常每组样本量在10到20之间：“使用参数方法”和“适合毒理学家的直觉”。展示了其应用于实际毒性研究数据的一个例子。还使用历史数据评估了这种新方法的性能。然而，应该注意的是，本文的目的不是对决策树做出确定的解决方案。可以考虑其他几种替代方法。由于树形算法没有单一的理论上正确的解决方案，不建议过度形式化地使用决策树。我们绝不能忘记评估重复毒性数据的探索性本质。
