Elham Haem (1,2), Kajsa Harling (2), Seyyed Mohammad Taghi Ayatollahi (1) Najaf Zare(1) and Mats O. Karlsson (2)
(1) Department of Biostatistics, Shiraz University of Medical Sciences school of Medicine, Shiraz, Iran,(2) Department of Pharmaceutical Biosciences, Uppsala University, Uppsala, Sweden.
Objectives: This study aimed to propose and implement a new version of lasso, Adjusted Adaptive Lasso [1], for non-linear mixed effects models. The new method takes into account the standard errors of the maximum likelihood coefficient estimators for covariate estimation and selection in population pharmacokinetics when there is multicollinearity. We also compared mean absolute prediction error and error of estimated coefficient of covariates with Adaptive Lasso [2] and simple Lasso [3].
Methods: Data sets were simulated with 20 40, 80 or 120 individuals, in which each subject had three PK observations, from a one compartment i.v.bolus model. Ten covariates were created by sampling from a multivariate normal distribution with no, low (0.2), moderate (0.5) and high (0.7) correlation among them. Four true covariates with different magnitudes influenced clearance. Adjusted Adaptive Lasso, Adaptive Lasso and Lasso were implemented using PsN and NONMEM 7.3. The methods were compared to each other in terms of mean absolute prediction error and error of estimated coefficient of covariates in 16 scenarios with different dataset sizes and distinct values of correlation among covariates. All simulation scenarios were replicated 100 times.
Results: Simulation results showed that Adjusted Adaptive Lasso had advantage in terms of both mean absolute prediction error and error of estimated coefficient of covariates comparing with other methods when data sets were small and in particular for no, small and moderate correlation among covariates. However, the benefit was negligible for large data sets.
Conclusions: Adjusted Adaptive Lasso outperformed Adaptive Lasso and simple Lasso in obtaining a predictive covariate model on small data sets.
References:
[1] Algamal Z. Y & Lee M. H . Adjusted adaptive lasso in high-dimensional poisson regression model. Modern Applied Science (2015), 9(4):170–177.
[2] Zou H. The adaptive lasso and its oracle properties. Journal of the American Statistical Association (2006), 101:1418–1429.
[3] Ribbing J, Nyberg J, Caster O & Jonsson E.N. The lasso-a novel method for predictive covariate model building in nonlinear mixed effects models. J Pharmacokin Pharmacodyn (2007). 34 (4):485–517 .
Reference: PAGE 25 (2016) Abstr 5821 [www.page-meeting.org/?abstract=5821]
Poster: Methodology - Covariate/Variability Models