Refereed conference paper presented and published in conference proceedings
香港中文大学研究人员 ( 现职)
| 林泰宁教授 (药剂学院) |
全文
| 没有全文档案提供 |
其它资讯
摘要Purpose
To evaluate subject classification performance on fitting a bimodal mixture model with NONMEM?, and to develop an algorithm that quantitatively predicts classification accuracy, through a simulation study.
Methods
Pharmacokinetic datasets (n=2048) were simulated, each of which assumes a bimodal distribution of drug clearance, with variations in typical value of clearance (TCVL, of the subpopulation with slower clearance), the ratio of clearances of the two subpopulations (rCL), inter-individual variability in clearance (CVCL), the mixing proportion (MIXP, which represents the proportion of subjects with slower clearance) and other factors such as sample size, number of samplings per subject and typical values of volume of distribution and absorption rate constant. Each simulated dataset was then fitted to a mixture model specifying two subpopulations with different typical values of clearance, using NONMEM? version 7.2 (Icon PLC). The one-compartment model with first-order absorption was applied, with the use of a proportional residual error model. First-order conditional estimation with interaction was used for parameter estimation. Each dataset was also fitted with single-population model in order to obtain the change in objective function value (dOFV) when the mixture model is removed. The fitting outputs were extracted and the accuracy of subject classification was expressed by ACCLn,OR=ln[c/(n+1)]–ln[1–c/(n+1)], where c is the number of subjects correctly classified and n is the sample size, for each dataset. All datasets with any abnormality in minimization and/or covariance step were excluded from subsequent analyses. Quantitative relationship between ACCLn,OR and various fitting outputs, including parameter estimates, relative standard errors and shrinkages of random effects were investigated using curve estimation in SPSS version 23.0.0.0 (IBM?). Then, NONMEM? was used to build the
prediction model for ACCLn,OR. Results of backward elimination and prediction and residual plots were inspected for diagnosis of the final model. Bootstrapping and testing of the model on 200 separately simulated datasets were done to validate the predictive performance of the final model for ACCLn,OR.
Results
Optimization of 1,306 datasets terminated successfully and were included in subsequent analyses. The residual error model was adjusted for observed heteroscedasticity with respect to dOFV. In the final model, dOFV was found to be the most predictive factor for classification accuracy. Other factors that were found to have significant association with ACCLn,OR included sample size, TVCL, rCL and MIXP. Backward elimination and diagnostic plots produced satisfactory results. Bootstrapping means were found to be very close to the final model’s parameter estimates (See Table 1 and Figure 1). Among the 200 datasets in the validation set, 163 of them did not produce any fitting error, among which the observed ACCLn,OR in 152 (93.3%) datasets fell into the 95% confidence interval of the prediction by our final model.
Conclusion
We have successfully developed a predictive model for the accuracy in subject classification in fitting a bimodal mixture model using NONMEM? to pharmacokinetic data with subpopulations with different typical values of drug clearance. This provides modelers with a novel estimator for assessing the appropriateness and accuracy of their developed mixture models.
出版社接受日期22.07.2016
着者HUI Ka Ho, LAM Tai Ning
会议名称2016 AAPS Annual Meeting and Exposition
会议开始日13.11.2016
会议完结日17.11.2016
会议地点Colorado Convention Center, Denver, Colorado
会议国家美国
出版年份2016
语言英式英语
关键词Systems Pharmacology, Pharmacokinetics, Pharmacometrics, Modeling and simulation, Mixture model
