Individual Bayesian Information Matrix for Predicting Estimation Error and Shrinkage of Individual Parameters Accounting for Data Below the Limit of Quantification

Abstract

Purpose: In mixed models, the relative standard errors (RSE) and shrinkage of individual parameters can be predicted from the individual Bayesian information matrix (MBF).We proposed an approach accounting for data below the limit of quantification (LOQ) in MBF.Methods: MBF is the sum of the expectation of the individual Fisher information (MIF) which can be evaluated by First-Order linearization and the inverse of random effect variance.We expressed the individual information as a weighted sum of predicted MIF for every possible design composing of measurements above and/or below LOQ. When evaluating MIF, we derived the likelihood expressed as the product of the likelihood of observed data and the 2 probability for data to be below LOQ. The relevance of RSE and shrinkage predicted by MBF in absence or presence of data below LOQ were evaluated by simulations, using a pharmacokinetic/viral kinetic model defined by differential equations. Results:Simulations showed good agreement between predicted and observed RSE and shrinkage in absence or presence of data below LOQ. We found that RSE and shrinkage increased with sparser designs and with data below LOQ. Conclusions:The proposed method based on MBF adequately predicted individual RSE and shrinkage, allowing for evaluation of a large number of scenarios without extensive simulations. KEYWORDS: Bayesian Fisher information matrix; Data below the limit of quantification;Nonlinear mixed effect models; Optimal design; Shrinkage. ABBREVIATIONS: FO

Dosage optimization of voriconazole in children with haematological malignancies based on population pharmacokinetics

“…This is probably a result of the small sample size with sparse sampling design in the current study. 43,44 Fourth, the lack of MIC values made it impossible for us to do pharmacokinetic/pharmacodynamic analysis like Xu et al 45 Finally, this study was a single-centre study with a sparse sampling design, and the recommended protocol is only a reference, necessitating further research in the future.…”

PFIM 4.0, an extended R program for design evaluation and optimization in nonlinear mixed-effect models

“…We plan to include other new developments and extensions in future versions of PFIM: allowing non-diagonal variance-covariance matrix of the random effects for correlation between these random effects [52], accounting for fixed effects for the influence of continuous covariates, including the possibility for the user to specify cost functions in the Fedorov-Wynn algorithm to penalize less feasible designs [53], taking into account the probability of data being observed or below the limit of quantification at each sampling time [4,54]. It would also be interesting to include in PFIM different optimality criteria and to explore alternative algorithms for finding optimum designs such as multiplicative algorithms [55] or discrete particle swarm optimization [56].…”

Determination of Bayesian Cramér–Rao Bounds for Estimating Uncertainties in the Bio-Optical Properties of the Water Column, the Seabed Depth and Composition in a Coastal Environment

“…where d i,j denotes the (i, j) element of Σ −1 . It has been shown that J MAP can be approximated by [80,84,85]:…”