Haven M C, Orsulak P J, Arnold L L, Crowley G
Clin Chem. 1987 Jul;33(7):1207-10.
In an attempt to optimize curve fitting for immunoradiometric assays, we investigated eight data-reduction methods with two commercially available assays of thyrotropin. In four of these methods linear data-reduction models are used: logit-log programs of Iso-Data, Micromedic, and Hewlitt-Packard, and probit-log of Hewlitt-Packard. The other four were nonlinear data-reduction models: Iso-Data's "French curve" (modified spline), four-parameter logistic function, and point-to-point methods, as well as a nonlinear least squares method. In using the eight data-reduction methods on data from analyses of 78 patients' samples, we found clinically relevant differences between models. In fact, differences found by changing data-reduction models were greater than the difference between the two commercial kits.
为了优化免疫放射分析的曲线拟合，我们用两种市售促甲状腺激素检测方法研究了八种数据简化方法。在这些方法中，有四种使用线性数据简化模型：Iso-Data、Micromedic和惠普公司的对数-对数程序，以及惠普公司的概率-对数程序。另外四种是非线性数据简化模型：Iso-Data的“法国曲线”（修正样条）、四参数逻辑函数和逐点法，以及非线性最小二乘法。在对78例患者样本分析所得数据使用这八种数据简化方法时，我们发现各模型之间存在临床相关差异。事实上，因改变数据简化模型而发现的差异大于两种市售试剂盒之间的差异。
