Imbimbo B P, Martinelli P, Rocchetti M, Ferrari G, Bassotti G, Imbimbo E
Mediolamum Farmaceutici, Milan, Italy.
Biopharm Drug Dispos. 1991 Mar;12(2):139-47. doi: 10.1002/bdd.2510120207.
Several statistical and empirical approaches have been proposed to select the multiexponential equation that best describes the time course of the plasma concentration of a drug. Recently, a new criterion (Ip) has been proposed according to which the model that best interprets a set of experimental data points is the one with the smallest area between the approximate confidence limits of estimated plasma concentration. We used large Montecarlo simulations to compare the ability of different selection criteria to select the correct model from data generated with an independent, normally distributed random error. The new criterion (Ip), Akaike's information criterion, the Schwartz test, and the F ratio test were studied. In this situation, the correct model was known and the performances of different selecting methods were assessed by examining their sensitivity to the number of exponential terms, the number of data points, and the size of the exponents in the true model. Mono-, bi-, and triexponential equations were studied. Overall mean percentages of right identification were 98.1 per cent for the new index, 82.8 per cent for Akaike's information criterion, 89.5 per cent for the Schwartz test, and 97.7 per cent for the F ratio test. The Akaike and Schwartz tests were not as efficient as the other tests with few (8-10) data points. The Ip and the F test raise the percentages of right identification of the model when the hybrid elimination rate macroconstants differ by at least a factor of four.(ABSTRACT TRUNCATED AT 250 WORDS)
已经提出了几种统计和实证方法来选择最能描述药物血浆浓度随时间变化过程的多指数方程。最近，有人提出了一种新的标准（Ip），根据该标准，最能解释一组实验数据点的模型是估计血浆浓度的近似置信限之间面积最小的模型。我们使用大型蒙特卡罗模拟来比较不同选择标准从具有独立正态分布随机误差生成的数据中选择正确模型的能力。研究了新的标准（Ip）、赤池信息准则、施瓦茨检验和F比率检验。在这种情况下，正确的模型是已知的，通过检查它们对指数项数量、数据点数量和真实模型中指数大小的敏感性来评估不同选择方法的性能。研究了单指数、双指数和三指数方程。新指数的正确识别总体平均百分比为98.1%，赤池信息准则为82.8%，施瓦茨检验为89.5%，F比率检验为97.7%。当数据点较少（8 - 10个）时，赤池检验和施瓦茨检验不如其他检验有效。当混合消除速率宏观常数相差至少四倍时，Ip和F检验提高了模型正确识别的百分比。（摘要截断于250字）
