谷浓度采样设计和算法对群体和个体药动学参数估算的影响

Effects of multiple-trough sampling design and algorithm on the estimation of population and individual pharmacokinetic parameters

本研究以符合一室和两室模型特征的奥卡西平和他克莫司为典型药物，采用蒙特卡洛法模拟服药后的稳态谷浓度，且均用一室模型拟合，考察不同多谷浓度采样方案（每名受试者采集1～4个谷浓度）和NONMEM软件中的不同算法对群体和个体药动学参数估算的影响。结果表明：表观清除率（CL／F）估算的准确度和精密度优于其他参数。对于奥卡西平，增加每名受试者的谷浓度采样次数，所有参数估算的准确度和精密度均有改善。对于他克莫司，随着采样增加，CL／F与其个体间变异的估算更稳定，残差变异估算也更可靠，但前者的估算准确度无改善。算法中蒙特卡洛重要抽样法（IMP）和基于后验估计的IMP法（IMPMAP）较稳定可靠。个体参数的估算仅与算法中是否含个体间变异和残差变异的交互作用有关。随着变异水平减小，群体或个体参数估算的准确度和精密度均有改善，且残差变异的影响大于个体间变异。上述结果可为基于谷浓度的群体药动学研究提供参考。

The purpose of this study is to investigate the effects of multiple-trough sampling design and nonlinear mixed effect modeling （NONMEM） algorithm on the estimation of population and individual pharmacokinetic parameters. Oxcarbazepine and tacrolimus were used as one-compartment and two-compartment model drugs, respectively. Seven sampling designs were investigated using various number of trough concentrations per individual ranging from 1-4. Monte Carlo simulations were performed to produce state- steady trough concentrations. One-compartment model was used to fit simulated data from oxcarbazepine and tacrolimus. The accuracy and precision of the estimated parameters were evaluated using the median prediction error （PE）, the median absolute PE and boxplot. The results indicated that trough concentrations could yield reliable estimates of apparent clearance （CL/F）. For oxcarbazepine, as the number of trough concentrations per subject increased, the accuracy and precision of CL/F, between-subject variability （BSV） of CL/F and residual variability （RUV） tended to be improved. For tacrolimus, however, although no improvement were observed in the accuracy of CL/F and BSV of CL/F, the PE distribution ranges were significantly narrowed and the RUV estimates were less bias and imprecise. In terms of algorithm, Monte Carlo importance sampling （IMP）and IMP assisted by mode a posteriori estimation （IMPMAP） were consistently better than other methods. Additionally, the sampling design had no significant effects on the individual parameter estimates, which were only depended on the interaction between BSV and RUV in various algorithms. Decreased in BSV and RUV levels can improve the accuracy and precision of the estimation for both population and individual pharmacokinetic parameter estimates.