2009 - St. Petersburg - Russia

PAGE 2009: Methodology- Other topics
Paul Baverel

Two PsN Bootstrapping Routines for Obtaining Uncertainty Measurement around the Nonparametric Distribution Obtained in NONMEM VI

Paul G. Baverel, Radojka M. Savic, Mats O. Karlsson

Department of Pharmaceutical Biosciences, Uppsala University, Sweden

Objectives: To assess the performance of two different resampling-based methods aiming at quantifying uncertainty around nonparametric distribution (NPD).

Methods: An IV-bolus PK model was used to simulate three data sets of 200 individuals following a rich sampling design and for which the random effects conformed to a (i) normal, (ii) bimodal and (iii) heavy-tailed underlying distributional shapes. After subsequent estimation with FOCE-NONP in NONMEM VI, uncertainty measurement was derived for each parameter marginal density by means of two different bootstrapping procedures. The first, full, method described in a previous work [1] relies on N bootstrap samples of the original data and a reanalysis of both the preceding parametric as well as the nonparametric step. The second, simplified, method relies on bootstrap sampling of the vectors of individual probabilities associated with each unique support point of the NPD. From the bootstrap samples, confidence intervals (CIs) around the original NPD are calculated. This simplified method is far less computer intensive than the first one (ca 2 min vs. 4 hr for this example). In order to compare the outcomes of the full and simplified methods, the 95% CI of the true marginal density was derived by standard bootstrapping procedure [2] from the true individual parameters and this was used as reference to compute mean errors (MEs) of the 95% CI width for each method.

Results: Overall, both versions performed similarly with respect to the trend of the true uncertainty around the NPD for each distributional case. MEs of the distance across 95% CI boundaries for CL were equal to: (i) -0.024 and -0.012, (ii) -0.021 and -0.006, (iii) -0.009 and 0 for the full and simplified, respectively.

Conclusions: Two bootstrapping methods suitable for estimating uncertainty around NPD are currently available. Little bias was induced by these techniques when comparing the 95% CI width regardless of the distributional shape in case of rich sampling design.

References:
[1] Savic RM, Baverel PG, Karlsson MO. A novel bootstrap method for obtaining uncertainty measurement around the nonparametric distribution. PAGE 17 (2008) Abstr 1390 [www.page-meeting.org/?abstract=1390].
[2] Efron B. Bootstrap methods: another look at the jackknife. Ann Stat 1979; 7:1-26.




Reference: PAGE 18 (2009) Abstr 1520 [www.page-meeting.org/?abstract=1520]
Poster: Methodology- Other topics
Click to open PDF poster/presentation (click to open)
Top