Bayesian design criteria: computation, comparison, and application to a pharmacokinetic and a pharmacodynamic model

The Use of Microdosing for In vivo Phenotyping of Cytochrome P450 Enzymes: Where Do We Stand? A Narrative Review

Cytochrome P450 (CYP) enzymes play a central role in the elimination of approximately 80% of all clinically used drugs. Differences in CYP enzyme activity between individuals can contribute to interindividual variability in exposure and, therefore, treatment outcome. In vivo CYP enzyme activity could be determined with phenotyping. Currently, (sub)therapeutic doses are used for in vivo phenotyping, which can lead to side effects. The use of microdoses (100 µg) for in vivo phenotyping for CYP enzymes could overcome the limitations associated with the use of (sub)therapeutic doses of substrates. The aim of this review is to provide a critical overview of the application of microdosing for in vivo phenotyping of CYP enzymes. A literature search was performed to find drug-drug interaction studies of CYP enzyme substrates that used microdoses of the respective substrates. A substrate was deemed sensitive to changes in CYP enzyme activity when the pharmacokinetics of the substrate significantly changed during inhibition and induction of the enzyme. On the basis of the currently available evidence, the use of microdosing for in vivo phenotyping for subtypes CYP1A2, CYP2C9, CYP2D6, and CYP2E1 is not recommended. Microdosing can be used for the in vivo phenotyping of CYP2C19 and CYP3A. The recommended microdose phenotyping test for CYP2C19 is measuring the omeprazole area-under-the-concentration-time curve over 24 h (AUC0-24) after administration of a single 100 µg dose. CYP3A activity could be best determined with a 0.1-75 µg dose of midazolam, and subsequently measuring AUC extrapolated to infinity (AUC∞) or clearance. Moreover, there are two metrics available for midazolam using a limited sampling strategy: AUC over 10 h (AUC0-10) and AUC from 2 to 4 h (AUC2-4).

"De-Shrinking" EBEs: The Solution for Bayesian Therapeutic Drug Monitoring

BACKGROUND

Therapeutic drug monitoring (TDM) aims at individualising a dosage regimen and is increasingly being performed by estimating individual pharmacokinetic parameters via empirical Bayes estimates (EBEs). However, EBEs suffer from shrinkage that makes them biased. This bias is a weakness for TDM and probably a barrier to the acceptance of drug dosage adjustments by prescribers.

OBJECTIVE

The aim of this article is to propose a methodology that allows a correction of EBE shrinkage and an improvement in their precision.

METHODS

As EBEs are defined, they can be seen as a special case of ridge estimators depending on a parameter usually denoted λ. After a bias correction depending on λ, we chose λ so that the individual pharmacokinetic estimations have minimal imprecision. Our estimate is by construction always better than EBE with respect to bias (i.e. shrinkage) and precision.

RESULTS

We illustrate the performance of this approach with two different drugs: iohexol and isavuconazole. Depending on the patient's actual pharmacokinetic parameter values, the improvement given by our approach ranged from 0 to 100%.

CONCLUSION

This innovative methodology is promising since, to the best of our knowledge, no other individual shrinkage correction has been proposed.

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

BACKGROUND AND OBJECTIVE

Nonlinear mixed-effect models (NLMEMs) are increasingly used for the analysis of longitudinal studies during drug development. When designing these studies, the expected Fisher information matrix (FIM) can be used instead of performing time-consuming clinical trial simulations. The function PFIM is the first tool for design evaluation and optimization that has been developed in R. In this article, we present an extended version, PFIM 4.0, which includes several new features.

METHODS

Compared with version 3.0, PFIM 4.0 includes a more complete pharmacokinetic/pharmacodynamic library of models and accommodates models including additional random effects for inter-occasion variability as well as discrete covariates. A new input method has been added to specify user-defined models through an R function. Optimization can be performed assuming some fixed parameters or some fixed sampling times. New outputs have been added regarding the FIM such as eigenvalues, conditional numbers, and the option of saving the matrix obtained after evaluation or optimization. Previously obtained results, which are summarized in a FIM, can be taken into account in evaluation or optimization of one-group protocols. This feature enables the use of PFIM for adaptive designs. The Bayesian individual FIM has been implemented, taking into account a priori distribution of random effects. Designs for maximum a posteriori Bayesian estimation of individual parameters can now be evaluated or optimized and the predicted shrinkage is also reported. It is also possible to visualize the graphs of the model and the sensitivity functions without performing evaluation or optimization.

RESULTS

The usefulness of these approaches and the simplicity of use of PFIM 4.0 are illustrated by two examples: (i) an example of designing a population pharmacokinetic study accounting for previous results, which highlights the advantage of adaptive designs; (ii) an example of Bayesian individual design optimization for a pharmacodynamic study, showing that the Bayesian individual FIM can be a useful tool in therapeutic drug monitoring, allowing efficient prediction of estimation precision and shrinkage for individual parameters.

CONCLUSION

PFIM 4.0 is a useful tool for design evaluation and optimization of longitudinal studies in pharmacometrics and is freely available at http://www.pfim.biostat.fr.

Experiment design for nonparametric models based on minimizing Bayes Risk: application to voriconazole¹

An experimental design approach is presented for individualized therapy in the special case where the prior information is specified by a nonparametric (NP) population model. Here, a NP model refers to a discrete probability model characterized by a finite set of support points and their associated weights. An important question arises as to how to best design experiments for this type of model. Many experimental design methods are based on Fisher information or other approaches originally developed for parametric models. While such approaches have been used with some success across various applications, it is interesting to note that they largely fail to address the fundamentally discrete nature of the NP model. Specifically, the problem of identifying an individual from a NP prior is more naturally treated as a problem of classification, i.e., to find a support point that best matches the patient's behavior. This paper studies the discrete nature of the NP experiment design problem from a classification point of view. Several new insights are provided including the use of Bayes Risk as an information measure, and new alternative methods for experiment design. One particular method, denoted as MMopt (multiple-model optimal), will be examined in detail and shown to require minimal computation while having distinct advantages compared to existing approaches. Several simulated examples, including a case study involving oral voriconazole in children, are given to demonstrate the usefulness of MMopt in pharmacokinetics applications.

Inter occasion variability in individual optimal design

Inter occasion variability (IOV) is of importance to consider in the development of a design where individual pharmacokinetic or pharmacodynamic parameters are of interest. IOV may adversely affect the precision of maximum a posteriori (MAP) estimated individual parameters, yet the influence of inclusion of IOV in optimal design for estimation of individual parameters has not been investigated. In this work two methods of including IOV in the maximum a posteriori Fisher information matrix (FIM_MAP) are evaluated: (i) MAP_occ—the IOV is included as a fixed effect deviation per occasion and individual, and (ii) POP_occ—the IOV is included as an occasion random effect. Sparse sampling schedules were designed for two test models and compared to a scenario where IOV is ignored, either by omitting known IOV (Omit) or by mimicking a situation where unknown IOV has inflated the IIV (Inflate). Accounting for IOV in the FIM_MAP markedly affected the designs compared to ignoring IOV and, as evaluated by stochastic simulation and estimation, resulted in superior precision in the individual parameters. In addition MAP_occ and POP_occ accurately predicted precision and shrinkage. For the investigated designs, the MAP_occ method was on average slightly superior to POP_occ and was less computationally intensive.

Powers of the likelihood ratio test and the correlation test using empirical bayes estimates for various shrinkages in population pharmacokinetics

We compared the powers of the likelihood ratio test (LRT) and the Pearson correlation test (CT) from empirical Bayes estimates (EBEs) for various designs and shrinkages in the context of nonlinear mixed-effect modeling. Clinical trial simulation was performed with a simple pharmacokinetic model with various weight (WT) effects on volume (V). Data sets were analyzed with NONMEM 7.2 using first-order conditional estimation with interaction and stochastic approximation expectation maximization algorithms. The powers of LRT and CT in detecting the link between individual WT and V or clearance were computed to explore hidden or induced correlations, respectively. Although the different designs and variabilities could be related to the large shrinkage of the EBEs, type 1 errors and powers were similar in LRT and CT in all cases. Power was mostly influenced by covariate effect size and, to a lesser extent, by the informativeness of the design. Further studies with more models are needed.

Influence of a priori Information, Designs, and Undetectable Data on Individual Parameters Estimation and Prediction of Hepatitis C Treatment Outcome

Hepatitis C viral kinetic analysis based on nonlinear mixed effect models can be used to individualize treatment. For that purpose, it is necessary to obtain precise estimation of individual parameters. Here, we evaluated by simulation the influence on Bayesian individual parameter estimation and outcome prediction of a priori information on population parameters, viral load sampling designs, and methods for handling data below detection limit (BDL). We found that a precise estimation of both individual parameters and treatment outcome could be obtained using as few as six measurements in the first month of therapy. This result remained valid even when incorrect a priori information on population parameters was set as long as the parameters were identifiable and BDL data were properly handled. However, setting wrong values for a priori population parameters could lead to severe estimation/prediction errors if BDL data were ignored and not properly accounted in the likelihood function.

Prediction of shrinkage of individual parameters using the bayesian information matrix in non-linear mixed effect models with evaluation in pharmacokinetics

PURPOSE

When information is sparse, individual parameters derived from a non-linear mixed effects model analysis can shrink to the mean. The objective of this work was to predict individual parameter shrinkage from the Bayesian information matrix (M BF ). We 1) Propose and evaluate an approximation of M BF by First-Order linearization (FO), 2) Explore by simulations the relationship between shrinkage and precision of estimates and 3) Evaluate prediction of shrinkage and individual parameter precision.

METHODS

We approximated M BF using FO. From the shrinkage formula in linear mixed effects models, we derived the predicted shrinkage from M BF . Shrinkage values were generated for parameters of two pharmacokinetic models by varying the structure and the magnitude of the random effect and residual error models as well as the design. We then evaluated the approximation of M BF FO and compared it to Monte-Carlo (MC) simulations. We finally compared expected and observed shrinkage as well as the predicted and estimated Standard Errors (SE) of individual parameters.

RESULTS

M BF FO was similar to M BF MC. Predicted and observed shrinkages were close . Predicted and estimated SE were similar.

CONCLUSIONS

M BF FO enables prediction of shrinkage and SE of individual parameters. It can be used for design optimization.

Application of the optimal design approach to improve a pretransplant drug dose finding design for ciclosporin

A time and sampling intensive pretransplant test dose design was to be reduced, but at the same time optimized so that there was no loss in the precision of predicting the individual pharmacokinetic (PK) estimates of posttransplant dosing. The following variables were optimized simultaneously: sampling times, ciclosporin dose, time of second dose, infusion duration, and administration order, using a published ciclosporin population PK model as prior information. The original design was reduced from 22 samples to 6 samples/patient and both doses (intravenous oral) were administered within 8 hours. Compared with the prior information given by the published ciclosporin population PK model, the expected standard deviations (SDs) of the individual parameters for clearance and bioavailability could be reduced by, on average, 40% under the optimized sparse designs. The gain of performing the original rich design compared with the optimal reduced design, considering the standard errors of the parameter estimates, was found to be minimal. This application demonstrates, in a practical clinical scenario, how optimal design techniques may be used to improve diagnostic procedures given available software and methods.

Application of optimal design methodologies in clinical pharmacology experiments

Pharmacokinetics and pharmacodynamics data are often analysed by mixed-effects modelling techniques (also known as population analysis), which has become a standard tool in the pharmaceutical industries for drug development. The last 10 years has witnessed considerable interest in the application of experimental design theories to population pharmacokinetic and pharmacodynamic experiments. Design of population pharmacokinetic experiments involves selection and a careful balance of a number of design factors. Optimal design theory uses prior information about the model and parameter estimates to optimize a function of the Fisher information matrix to obtain the best combination of the design factors. This paper provides a review of the different approaches that have been described in the literature for optimal design of population pharmacokinetic and pharmacodynamic experiments. It describes options that are available and highlights some of the issues that could be of concern as regards practical application. It also discusses areas of application of optimal design theories in clinical pharmacology experiments. It is expected that as the awareness about the benefits of this approach increases, more people will embrace it and ultimately will lead to more efficient population pharmacokinetic and pharmacodynamic experiments and can also help to reduce both cost and time during drug development.

Robust population pharmacokinetic experiment design

The population approach to estimating mixed effects model parameters of interest in pharmacokinetic (PK) studies has been demonstrated to be an effective method in quantifying relevant population drug properties. The information available for each individual is usually sparse. As such, care should be taken to ensure that the information gained from each population experiment is as efficient as possible by designing the experiment optimally, according to some criterion. The classic approach to this problem is to design "good" sampling schedules, usually addressed by the D-optimality criterion. This method has the drawback of requiring exact advanced knowledge (expected values) of the parameters of interest. Often, this information is not available. Additionally, if such prior knowledge about the parameters is misspecified, this approach yields designs that may not be robust for parameter estimation. In order to incorporate uncertainty in the prior parameter specification, a number of criteria have been suggested. We focus on ED-optimality. This criterion leads to a difficult numerical problem, which is made tractable here by a novel approximation of the expectation integral usually solved by stochastic integration techniques. We present two case studies as evidence of the robustness of ED-optimal designs in the face of misspecified prior information. Estimates from replicate simulated population data show that such misspecified ED-optimal designs recover parameter estimates that are better than similarly misspecified D-optimal designs, and approach estimates gained from D-optimal designs where the parameters are correctly specified.

A program for the optimum design of pharmacokinetic, pharmacodynamic, drug metabolism and drug-drug interaction models

Planning any experiment includes issues such as how many samples are to be taken and their location given some predictor variable. Often a model is used to explain these data; hence including this formally in the design will be beneficial for any subsequent parameter estimation and modelling. A number of criteria for model oriented experiments, which maximise the information content of the collected data are available. In this paper we present a program, Optdes, to investigate the optimal design of pharmacokinetic, pharmacodynamic, drug metabolism and drug-drug interaction models. Using the developed software the location of either a predetermined number of design points (exact designs) or together with the proportion of samples at each point (continuous designs) can be determined. Local as well as Bayesian designs can be optimised by either D- or A-optimality criteria. Although often the optimal design cannot be applied for practical reasons, alternative designs can be readily evaluated.

Limited and optimal sampling strategies for etoposide and etoposide catechol in children with leukemia

Etoposide is used to treat childhood malignancies, and its plasma pharmacokinetics have been related to pharmacodynamic endpoints. Limiting the number of samples should facilitate the assessment of etoposide pharmacokinetics in children. We compared limited sampling strategies using multiple linear regression of plasma concentrations and clearance with Bayesian methods of estimating clearance using compartmental pharmacokinetic models. Optimal sampling times were estimated in the multiple linear regression method by determining the combination of two samples which maximized the correlation coefficient, and in the Bayesian estimation approach by minimizing the variance in estimates of clearance. Clearance estimates were compared to the actual clearances from Monte Carlo-simulated data and predicted clearances estimated using all available plasma concentrations in clinical data from children with acute lymphoblastic leukemia. Multiple linear regression poorly predicted clearance (mean bias 8.3%, precision 17.5%), but improved if plasma concentrations were logarithmically transformed (mean bias 1.4%, precision 12.5%). Bayesian estimation methods with optimal samples gave the best overall prediction (mean bias 2.5%, precision 6.8%) and also performed better than regression methods for abnormally high or low clearances. We conclude that Bayesian estimation with limited sampling gives the best estimates of etoposide clearance.

Potential benefit of Bayesian forecasting for therapeutic drug monitoring in neonates

Therapeutic drug monitoring in neonate has been hampered by invasiveness of blood samplings raising ethical problems. A methodologic approach has been developped in adults and in children that is still unsufficiently developped in neonates, the Bayesian forecasting of drug plasma concentration. This method is particularly attractive in neonates using a few blood samples from an individual patient and more informations from a prior patient sample representative of the population the individual patient belongs to. The present article aims at reviewing the different procedures and methods to minimize invasiveness during therapeutic drug monitoring in neonate and at reviewing the methods for improving the quality of different dose adjustments using a Bayesian approach.

Applications of population approaches in toxicology

Many experimental or observational studies in toxicology are best analysed in a population framework. Recent examples include investigations of the extent and origin of intra-individual variability in toxicity studies, incorporation of genotypic information to address intra-individual variability, optimal design of experiments, and extension of toxicokinetic modelling to the analysis of biomarker studies. Bayesian statistics provide powerful numerical methods for fitting population models, particularly when complex mechanistic models are involved. Challenges and limitations to the use of population models, in terms of basic structure, computational burden, ease of implementation and data accessibility, are identified and discussed.

The iterative two-stage population approach to IVGTT minimal modeling: improved precision with reduced sampling. Intravenous glucose tolerance test

The minimal model method is widely used to estimate glucose effectiveness (S(G)) and insulin sensitivity (S(I)) from intravenous glucose tolerance test (IVGTT) data. In the standard protocol (sIVGTT, 0.33 g/kg glucose bolus given at time 0), which allows the simultaneous assessment of beta-cell function, the precision of the individualized estimates often degrades and particularly so in the presence of reduced sampling schedules. Here, we investigated the use of a population approach, the iterative two-stage (ITS) approach, to analyze 16 sIVGTTs in healthy subjects and to obtain refined estimates of S(G) and S(I) in the population and in the individual subjects. The ITS is based on calculation of the population mean and standard deviation of the parameters at each iteration and then use of them as prior information for the individual analyses. Theoretically, the use of a prior in the ITS should improve the precision of the individual estimates. The customary approach (standard two stage, STS), where modeling is performed separately for each individual subject, does not take the population knowledge into account. We used both frequent (FSS, 30 samples) and (quasi-optimally) reduced (RSS, 14 samples) sampling schedules. For the FSS, STS gave estimates (mean +/- SD) for S(G) = 2.66 +/- 1.09 x 10(-2). min(-1) and S(I) = 6.46 +/- 6.99 10(-4). min(-1). microU(-1). ml, with an average precision of 51 (range 5-176) and 33% (3-91), respectively. RSS radically worsened the precision of both S(G) and S(I). However, RSS and ITS gave S(G) = 2.59 +/- 0.73 and S(I) = 6.06 +/- 7.28, with an average precision of 23 (12-42) and 27% (), respectively. In conclusion, population minimal modeling of sIVGTT data improves the precision of individual estimates of glucose effectiveness and insulin sensitivity, as the theory predicts, and, even with reduced sampling, the improvement is substantial.

Optimal sampling times for Bayesian estimation of the pharmacokinetic parameters of nortriptyline during therapeutic drug monitoring

Sampling times for Bayesian estimation of the pharmacokinetic parameters of an antidepressant drug, nortriptyline, during its therapeutic drug monitoring were optimized. Our attention was focused on designs including a limited number of measurements: one, two, and three sample designs in which sampling times had to be chosen between 0 and 24 hr after the last intake of a test-dose study. The optimization was conducted in four groups of patients defined by their gender and the administration or not of concomitant drugs inhibiting the metabolism of nortriptyline. The Bayesian design criterion was defined as the expected information provided by an experiment. A stochastic approximation algorithm, the Kiefer-Wolfowitz algorithm, was used for the criterion maximization under experimental constraints. Results showed that optimal Bayesian sampling times differ between patients in monotherapy and polytherapy. For one-sample designs the measurements have to be performed either at the lower (0 hr) or at the upper (24 hr) bound of the admissible interval. Replications were often found for 2- and 3-point designs. Other sampling designs can lead to criterion close to the optimum and can therefore be performed without great loss of information. In contrast, we found that several designs lead to low values of the information criterion, which justifies the approach.

Optimal statistical design for toxicokinetic studies

Optimal statistical design strategies are applied to toxicokinetic experiments, for determining proper allocations of subjects and/or spacings of sampling times under a variety of nonlinear concentration-time models. The strategies include: (i) optimal allocations of subjects assuming the placement of time points is fixed, (ii) optimal spacing of design time points while assuming an equal allocation of subjects per time points and (iii) allocations/time-point spacings optimized jointly. Emphasis is placed on the first case, where a variance-minimization method is illustrated for optimizing the allocations when estimating specific toxicokinetic parameters. Appeals to forms of D-optimality are also considered, for cases when no specific toxicokinetic parameter is of specialized interest.

Stochastic optimization algorithms of a Bayesian design criterion for Bayesian parameter estimation of nonlinear regression models: application in pharmacokinetics

This article proposes three stochastic algorithms to optimize a Bayesian design criterion for Bayesian estimation of the parameters of nonlinear regression models; this criterion is the information expected from an experiment. The first algorithm is based on a stochastic version of the simplex with an adaptive sampling procedure. The others are stochastic approximation algorithms: the Kiefer-Wolfowitz and the pseudogradient algorithms. We first present the information criterion and the optimization algorithms. The efficiency of each algorithm for optimizing this Bayesian design criterion is then assessed by a simulation study for a nonlinear model assuming a discrete prior distribution. An application for designing an experiment to estimate the kinetics of radioiodine thyroid uptake is then proposed.

Comparison of ED, EID, and API criteria for the robust optimization of sampling times in pharmacokinetics

Optimization of the sampling schedule can be used in pharmacokinetic (PK) experiments to increase the accuracy and the precision of parameter estimation or to reduce the number of samples required. Several optimization criteria that formally incorporate prior parameter uncertainty have been proposed earlier. These criteria consist in finding the sampling schedule that maximizes the expectation (over a given parameter distribution) of det F (ED-optimality) or Log(det F) (API-optimality), or minimizes the expectation of 1/det F (EID-optimality), where F is the Fisher information matrix. The precision and the accuracy of parameter estimation after having fitted a PK model to a small number of optimal data points (determined according to D, ED, EID, and API criteria) or to a naive sampling schedule were compared in a Monte Carlo simulation study. A one-compartment model with first-order absorption rate (3 parameters) and a two-compartment model with zero-order infusion rate (4 parameters) were considered. Data were simulated for 300 subjects with both structural models, combined with several residual error models (homoscedastic, heteroscedastic with constant or variable coefficient of variation). Interindividual variabilities in PK parameters ranged from 25-66%. ED-, EID-, and API-optimal sampling times were calculated using the software OSP-Fit. Three or five samples were allowed for parameter estimation by extended least-squares. Performances of each design criterion were evaluated in terms of mean prediction error, root mean squared error, and number of acceptable estimates (i.e., with a SE less than 30%). Compared to the D-optimal design, the EID and API designs reduced the bias and the imprecision of the estimation of the parameters having a large interindividual variability. Moreover, the API design resulted in some cases in a higher number of acceptable estimates.