Performance of different population pharmacokinetic algorithms
| Autor: | Julie Grenier, Philippe Colucci, Jacques Turgeon, Corinne Seng Yue, Murray P. Ducharme |
|---|---|
| Rok vydání: | 2011 |
| Předmět: |
Pharmacology
education.field_of_study Maximum likelihood expectation maximization Population Reproducibility of Results Variance (accounting) Residual Models Biological NONMEM Reduction (complexity) Pharmacokinetics Bias Data Interpretation Statistical Expectation–maximization algorithm Humans Pharmacology (medical) Computer Simulation Drug Monitoring education Algorithm Algorithms Mathematics |
| Zdroj: | Therapeutic drug monitoring. 33(5) |
| ISSN: | 1536-3694 |
| Popis: | Background There has been an increased focus on population pharmacokinetics (PK) to improve the drug development process since the "Critical Path paper" by the Food and Drug Administration. This increased interest has given rise to additional algorithms. Objectives The purpose of this exercise was to compare the new algorithms iterative-2-stage (ITS) and maximum likelihood expectation maximization (MLEM) available in ADAPT 5 with other methods. Methods A total of 29 clinical trials with different study designs were simulated. Different algorithms were used to fit the simulated data, and the estimated parameters were compared with the true values. The algorithms ITS and MLEM were compared with the standard-2-stage, Iterative-2-Stage (IT2S) method in the IT2S package and the first-order conditional estimate (FOCE) method in NONMEM version VI. Imprecision and bias for the population PK parameters, variances, and individual PK parameters were used to compare the methods. Results Population PK parameters were well estimated and bias low for all nonlinear mixed effect modeling approaches. These approaches were superior to the standard-2-stage analyses. The algorithm MLEM was better than IT2S and ITS in predicting the PK and variability parameters. Residual variability was better estimated using MLEM and FOCE. A difference in the estimation of the variance exists between FOCE and the other methods. Variances estimated with FOCE often had shrinkage issues, whereas MLEM in ADAPT 5 had practically no shrinkage problems. Using MLEM, a reduction from 3000 to 1000 samples in the expectation maximization step had no impact on the results. Conclusions The new algorithm MLEM in ADAPT 5 was consistently better than IT2S and ITS in its prediction of PK parameters, variances, and the residual variability. It was comparable with the FOCE method with significantly fewer shrinkage issues in the estimation of variance. The number of samples used in the expectation maximization step with MLEM did not influence the results. |
| Databáze: | OpenAIRE |
| Externí odkaz: |
|