201507fda行业指南：分析方法验证(中英文)(下).doc

201507 FDA行业指南：分析方法验证（中英文）（下）

VII. STATISTICAL ANALYSIS AND MODELS 统计学分析和模型 A. Statistics 统计学 Statistical analysis of

validation data can be used to evaluate validation characteristics against predetermined acceptance criteria.

All statistical procedures and parameters used in the analysis of the data should be based on sound principles

and appropriate for the intended evaluation. Several statistical methods are useful for assessing validation characteristics, for example, an analysis of variance (ANOVA) to assess regression analysis R (correlation coefficient) and R squared (coefficient of determination) or linear regression to measure linearity. Many statistical methods used for assessing validation characteristics rely

on population normality, and it is important to determine whether or not to reject this assumption. There are many techniques, such as histograms, normality tests, and probability plots that can be used to evaluate the observed distribution. It may be appropriate to transform the data to better fit the normal distribution or apply distribution-free (nonparametric) approaches when the observed data are

not normally distributed. Appropriate literature or text

should be consulted for information on statistical procedures to use when developing new test methods, evaluating existing test methods or evaluating measurement system performance, as well as other

general information on the interpretation and treatment of analytical data[18].The data analysis should be assured either by using appropriately validated software or independent verification for correctness.验证数据的统计学分析可以用于评估验证的属性是否符合预定的可接受标准。

所有用于数据分析的统计学程序和参数均应是基于合理的原

则，并适合于既定评估。有几个统计学方法用于评估验证

属性颇为有用，例如，变量分析（ ANOVA ）用于评估相关性分析 R（相关因子）和 R 平方（判定系数或拟合优度）或线性回归用于测量线性。许多用于评估验证属性的统计学方法依赖于样本的正态性，决定是否拒绝该假设很重要。有许多技术，如柱状图、正态分布和概率图，可以用于评估所观察到的分布情况。如果观察到的数据是非正态分布的，则将数据转换成为更为正态分布或应用非正态分布（无参数）方法会更为恰当。在研发新的分析方法、评估现有分析方法、或评估测量系统性能时，应参考适当的文献或文件来获取关于统计学程序的信息，以及关于分析数据诠释和处理的其它

通用信息。数据分析应采用经过适当验证的软件，否则应单

独确认其正确性。 B. Models模型

Some analytical methods might use chemometric and/or multivariate models. When developing these models, the number of samples to provide adequate statistical power

and range for model development and validation should be considered. Suitable software should be used for data analysis. Model parameters should be deliberately varied to test model robustness. 有些分析方法可能会使用化学计

量学和 /或多变量模型。如果研发的这些模型、样品数据可以提供足够的统计功效和范围用于建模，则应考虑进行验证。

可以使用适当的软件进行数据分析。应该设计变化模型参数

来测试模型的耐用性。VIII. LIFE CYCLE MANAGEMENT OF ANALYTICAL PROCEDURES分析方法的生命周期管理 Once an analytical procedure (including compendial methods) is successfully validated (or verified) and implemented, the procedure should be followed during the

life cycle of the product to continually assure that it remains

fit for its intended purpose. Trend analysis on method performance should be performed at regular intervals to evaluate the need to optimize the analytical procedure or to revalidate all or a part of the analytical procedure. If an