Lever M, Munster D J
Clin Biochem. 1977 Apr;10(2):56-64. doi: 10.1016/s0009-9120(77)90754-8.
The random errors in an analytical method are additive and can be classified into analytical response-dependent and -independent terms. Non-random errors, caused by systematic faults in the analytical procedure, are not always distriguishable from the random errors, but some cases of non-linear assay response and unsuitable standardisation can be studied usefully in models without random error. Interlaboratory quality control programs cannot distinguish systematic and random error until the pattern of results on a number of specimens, or pairs of specimens, can be studied. In this case linear regression analysis is a powerful method for distinguishing different forms of error especially when response-dependent random errors do not predominate. The range of concentrations used for regression whould be as wide as that in which quantitative distinctions are used in clincal diagnosis and treatment. Preliminary reports, of the results on which the regression analysis is based, are most suitably presented on Youden diagrams with paired specimens.
分析方法中的随机误差具有累加性，可分为与分析响应相关和无关的项。由分析过程中的系统故障引起的非随机误差，并不总是能与随机误差区分开来，但在没有随机误差的模型中，可以有效地研究一些非线性分析响应和不合适标准化的情况。实验室间质量控制程序在能够研究多个标本或标本对的结果模式之前，无法区分系统误差和随机误差。在这种情况下，线性回归分析是区分不同形式误差的有力方法，尤其是当与响应相关的随机误差不占主导时。用于回归的浓度范围应与临床诊断和治疗中进行定量区分的范围一样宽。回归分析所基于的结果的初步报告，最适合用配对标本在尤登图上呈现。
