Evard Hanno, Kruve Anneli, Leito Ivo
University of Tartu, Institute of Chemistry, Ravila 14a, Tartu, 50411, Estonia.
Anal Chim Acta. 2016 Oct 26;942:23-39. doi: 10.1016/j.aca.2016.08.043. Epub 2016 Sep 13.
A large body of literature exists on the limit of detection (LOD), but there is still a lot of confusion about this important validation parameter. This confusion mainly stems from its statistically complex background. The goal of this two-part tutorial is to discuss and clarify the topic of LOD for practitioners. The two main conclusions of this tutorial are: (1) the choice of how to estimate LOD should be based on the purpose of the analytical method that is being validated (e.g. considerable effort should not be made to estimate LOD for a method that is not used for detecting traces in the vicinity of LOD), and (2) LOD estimates are strongly dependent on different assumptions and the approach used, and therefore caution must be exercised when using the estimate or when comparing different estimates. Part I of the tutorial contains a theoretical discussion (without excessively sophisticated statistics) and part II contains examples based on experimental data. In Part I, LOD and other definitions related to it are reviewed, and their estimation and use are discussed. The assumptions and practicality of different approaches to estimate LOD are compared. Different aspects of the analytical method that strongly influence LOD estimates (e.g. linearity, scedasticity and day-to-day variability of LOD) together with experimental design considerations when estimating LOD are discussed. In part II, LOD is estimated for the LC-MS/MS analysis method to detect pesticides on separate days. The performance of different tests on the data necessary for LOD estimation are discussed and the results of different approaches under review in this tutorial are compared. In conclusion, a decision tree is proposed for estimating and monitoring LOD. A detailed working procedure for estimating LOD is presented. Several topics are pointed out in which further research and discussion is needed.
关于检测限（LOD）已有大量文献，但对于这个重要的验证参数仍存在很多混淆之处。这种混淆主要源于其统计学上的复杂背景。这个分为两部分的教程的目的是为从业者讨论并阐明检测限这一主题。本教程的两个主要结论是：（1）如何估计检测限的选择应基于正在验证的分析方法的目的（例如，对于不用于检测接近检测限的痕量的方法，不应花费大量精力来估计检测限），以及（2）检测限估计强烈依赖于不同的假设和所使用的方法，因此在使用该估计值或比较不同估计值时必须谨慎。教程的第一部分包含理论讨论（不过多涉及复杂的统计学），第二部分包含基于实验数据的示例。在第一部分中，回顾了检测限及其相关的其他定义，并讨论了它们的估计和使用。比较了估计检测限的不同方法的假设和实用性。讨论了对检测限估计有强烈影响的分析方法的不同方面（例如线性、方差齐性和检测限的日常变异性）以及估计检测限时的实验设计考虑因素。在第二部分中，对用于在不同日期检测农药的液相色谱 - 串联质谱（LC-MS/MS）分析方法进行了检测限估计。讨论了对检测限估计所需数据进行的不同测试的性能，并比较了本教程中所探讨的不同方法的结果。总之，提出了一个用于估计和监测检测限的决策树。给出了估计检测限的详细工作程序。指出了几个需要进一步研究和讨论的主题。
