Krouwer J S, Schlain B
Ciba Corning Diagnostics Corp., Medfield, MA 02052.
Clin Chem. 1993 Aug;39(8):1689-93.
We present a statistical method to quantify deviations from linearity for assays that veer from linear assay responses. Our procedure handles the common case of unequally spaced analyte levels and nonconstant variance and provides a least-squares estimate with a confidence interval for the amount of deviation from assay linearity at a specified analyte concentration. This estimate of assay bias due to nonlinearity goes beyond the NCCLS EP6 lack-of-fit test, which tests for only the presence of nonlinearity. Knowing that nonlinearity is present is insufficient; users need to know the magnitude of the bias caused by nonlinearity. Our method can also be used with multifactor designs that estimate other systematic assay effects such as drift and carryover, thus obviating the need for a separate protocol to assess linearity. The procedure is carried out by adding extra columns to the design matrix corresponding to the concentration level(s) of interest. The extra columns, which replace the quadratic column, are orthogonal to all other columns. We describe a general method of constructing the new columns, and illustrate the procedure with a manual ammonia assay example dataset from EP6.
我们提出了一种统计方法，用于量化线性检测响应出现偏差的检测方法与线性的偏离程度。我们的程序处理了分析物水平不等距和方差非恒定的常见情况，并提供了一个最小二乘估计值以及在指定分析物浓度下与检测线性度偏离量的置信区间。这种由于非线性导致的检测偏差估计超出了NCCLS EP6拟合不足检验，后者仅测试非线性的存在。仅仅知道存在非线性是不够的；用户需要知道由非线性引起的偏差大小。我们的方法还可用于估计其他系统检测效应（如漂移和残留）的多因素设计，从而无需单独的方案来评估线性度。该程序是通过在设计矩阵中添加与感兴趣的浓度水平相对应的额外列来执行的。这些取代二次列的额外列与所有其他列正交。我们描述了构建新列的一般方法，并用来自EP6的手动氨检测示例数据集说明了该程序。
