Search

This article highlights some numerical challenges when implementing the bounded integer model for composite score modeling and suggests an improved implementation. The improvement is based on an approximation of the logarithm of the error function. After presenting the derivation of the improved implementation, the article compares the performance of the algorithm to a naive implementation of the log-likelihood using both simulations and a real data example. In the simulation setting, the improved algorithm yielded more precise and less biased parameter estimates when the within-subject variability was small and estimation was performed using the Laplace algorithm. The estimation results did not differ between implementations when the SAEM algorithm was used. For the real data example, bootstrap results differed between implementations with the improved implementation producing identical or better objective function values. Based on the findings in this article, the improved implementation is suggested as the new default log-likelihood implementation for the bounded integer model. Electronic supplementary material The online version of this article (10.1007/s10928-020-09727-8) contains supplementary material, which is available to authorized users.

Graphic abstract Lack of data is an obvious limitation to what can be modelled. However, aggregate data in the form of means and possibly (co)variances, as well as previously published pharmacometric models, are often available. Being able to use all available data is desirable, and therefore this paper will outline several methods for using aggregate data as the basis of parameter estimation. The presented methods can be used for estimation of parameters from aggregate data, and as a computationally efficient alternative for the stochastic simulation and estimation procedure. They also allow for population PK/PD optimal design in the case when the data-generating model is different from the data-analytic model, a scenario for which no solutions have previously been available. Mathematical analysis and computational results confirm that the aggregate-data FO algorithm converges to the same estimates as the individual-data FO and yields near-identical standard errors when used in optimal design. The aggregate-data MC algorithm will asymptotically converge to the exactly correct parameter estimates if the data-generating model is the same as the data-analytic model. The performance of the aggregate-data methods were also compared to stochastic simulations and estimations (SSEs) when the data-generating model is different from the data-analytic model. The aggregate-data FO optimal design correctly predicted the sampling distributions of 200 models fitted to simulated datasets with the individual-data FO method. Supplementary Information The online version contains supplementary material available at 10.1007/s10928-021-09760-1.

Physiologically based pharmacokinetic (PBPK) models are an important type of systems model used commonly in drug development before commencement of first-in-human studies. Due to structural complexity, these models are not easily utilised for future data-driven population pharmacokinetic (PK) analyses that require simpler models. In the current study we aimed to explore and automate methods of simplifying PBPK models using a proper lumping technique. A linear 17-state PBPK model for fentanyl was identified from the literature. Four methods were developed to search the optimal lumped model, including full enumeration (the reference method), non-adaptive random search (NARS), scree plot plus NARS, and simulated annealing (SA). For exploratory purposes, it was required that the total area under the fentanyl arterial concentration–time curve (AUC) between the lumped and original models differ by 0.002% at maximum. In full enumeration, a 4-state lumped model satisfying the exploratory criterion was found. In NARS, a lumped model with the same number of lumped states was found, requiring a large number of random samples. The scree plot provided a starting lumped model to NARS and the search completed within a short time. In SA, a 4-state lumped model was consistently delivered. In simplify an existing linear fentanyl PBPK model, SA was found to be robust and the most efficient and may be suitable for general application to other larger-scale linear systems. Ultimately, simplified PBPK systems with fundamental mechanisms may be readily used for data-driven PK analyses. Electronic supplementary material The online version of this article (10.1007/s10928-019-09644-5) contains supplementary material, which is available to authorized users.

Bridging fundamental approaches to model optimization for pharmacometricians, systems pharmacologists and statisticians is a critical issue. These fields rely primarily on Maximum Likelihood and Extended Least Squares metrics with iterative estimation of parameters. Our research combines adaptive chaos synchronization and grid search to estimate physiological and pharmacological systems with emergent properties by exploring deterministic methods that are more appropriate and have potentially superior performance than classical numerical approaches, which minimize the sum of squares or maximize the likelihood. We illustrate these issues with an established model of cortisol in human with nonlinear dynamics. The model describes cortisol kinetics over time, including its chaotic oscillations, by a delay differential equation. We demonstrate that chaos synchronization helps to avoid the tendency of the gradient-based optimization algorithms to end up in a local minimum. The subsequent analysis illustrates that the hybrid adaptive chaos synchronization for estimation of linear parameters with coarse-to-fine grid search for optimal values of non-linear parameters can be applied iteratively to accurately estimate parameters and effectively track trajectories for a wide class of noisy chaotic systems. Electronic supplementary material The online version of this article (10.1007/s10928-019-09629-4) contains supplementary material, which is available to authorized users.

Input estimation is employed in cases where it is desirable to recover the form of an input function which cannot be directly observed and for which there is no model for the generating process. In pharmacokinetic and pharmacodynamic modelling, input estimation in linear systems (deconvolution) is well established, while the nonlinear case is largely unexplored. In this paper, a rigorous definition of the input-estimation problem is given, and the choices involved in terms of modelling assumptions and estimation algorithms are discussed. In particular, the paper covers Maximum a Posteriori estimates using techniques from optimal control theory, and full Bayesian estimation using Markov Chain Monte Carlo (MCMC) approaches. These techniques are implemented using the optimisation software CasADi, and applied to two example problems: one where the oral absorption rate and bioavailability of the drug eflornithine are estimated using pharmacokinetic data from rats, and one where energy intake is estimated from body-mass measurements of mice exposed to monoclonal antibodies targeting the fibroblast growth factor receptor (FGFR) 1c. The results from the analysis are used to highlight the strengths and weaknesses of the methods used when applied to sparsely sampled data. The presented methods for optimal control are fast and robust, and can be recommended for use in drug discovery. The MCMC-based methods can have long running times and require more expertise from the user. The rigorous definition together with the illustrative examples and suggestions for software serve as a highly promising starting point for application of input-estimation methods to problems in drug discovery. Electronic supplementary material The online version of this article (doi:10.1007/s10928-016-9467-z) contains supplementary material, which is available to authorized users.

The purpose of this study was to assess the pharmacokinetics of dexmedetomidine in the ICU settings during the prolonged infusion and to compare it with the existing literature data using the Bayesian population modeling with literature-based informative priors. Thirty-eight patients were included in the analysis with concentration measurements obtained at two occasions: first from 0 to 24 h after infusion initiation and second from 0 to 8 h after infusion end. Data analysis was conducted using WinBUGS software. The prior information on dexmedetomidine pharmacokinetics was elicited from the literature study pooling results from a relatively large group of 95 children. A two compartment PK model, with allometrically scaled parameters, maturation of clearance and t-student residual distribution on a log-scale was used to describe the data. The incorporation of time-dependent (different between two occasions) PK parameters improved the model. It was observed that volume of distribution is 1.5-fold higher during the second occasion. There was also an evidence of increased (1.3-fold) clearance for the second occasion with posterior probability equal to 62 %. This work demonstrated the usefulness of Bayesian modeling with informative priors in analyzing pharmacokinetic data and comparing it with existing literature knowledge. Electronic supplementary material The online version of this article (doi:10.1007/s10928-016-9474-0) contains supplementary material, which is available to authorized users.

Cardiotoxicity is among the top drug safety concerns, and is of specific interest in tuberculosis, where this is a known or potential adverse event of current and emerging treatment regimens. As there is a need for a tool, beyond the QT interval, to quantify cardiotoxicity early in drug development, an empirical decision tree based classifier was developed to predict the risk of Torsades de pointes (TdP). The cardiac risk algorithm was developed using pseudo-electrocardiogram (ECG) outputs derived from cardiac myocyte electromechanical model simulations of increasing concentrations of 96 reference compounds which represented a range of clinical TdP risk. The algorithm correctly classified 89% of reference compounds with moderate sensitivity and high specificity (71 and 96%, respectively) as well as 10 out of 12 external validation compounds and the anti-TB drugs moxifloxacin and bedaquiline. The cardiac risk algorithm is suitable to help inform early drug development decisions in TB and will evolve with the addition of emerging data. Electronic supplementary material The online version of this article (10.1007/s10928-018-9580-2) contains supplementary material, which is available to authorized users.

Estimating the power for a non-linear mixed-effects model-based analysis is challenging due to the lack of a closed form analytic expression. Often, computationally intensive Monte Carlo studies need to be employed to evaluate the power of a planned experiment. This is especially time consuming if full power versus sample size curves are to be obtained. A novel parametric power estimation (PPE) algorithm utilizing the theoretical distribution of the alternative hypothesis is presented in this work. The PPE algorithm estimates the unknown non-centrality parameter in the theoretical distribution from a limited number of Monte Carlo simulation and estimations. The estimated parameter linearly scales with study size allowing a quick generation of the full power versus study size curve. A comparison of the PPE with the classical, purely Monte Carlo-based power estimation (MCPE) algorithm for five diverse pharmacometric models showed an excellent agreement between both algorithms, with a low bias of less than 1.2 % and higher precision for the PPE. The power extrapolated from a specific study size was in a very good agreement with power curves obtained with the MCPE algorithm. PPE represents a promising approach to accelerate the power calculation for non-linear mixed effect models.

A limitation in traditional stepwise population pharmacokinetic model building is the difficulty in handling interactions between model components. To address this issue, a method was previously introduced which couples NONMEM parameter estimation and model fitness evaluation to a single-objective, hybrid genetic algorithm for global optimization of the model structure. In this study, the generalizability of this approach for pharmacokinetic model building is evaluated by comparing (1) correct and spurious covariate relationships in a simulated dataset resulting from automated stepwise covariate modeling, Lasso methods, and single-objective hybrid genetic algorithm approaches to covariate identification and (2) information criteria values, model structures, convergence, and model parameter values resulting from manual stepwise versus single-objective, hybrid genetic algorithm approaches to model building for seven compounds. Both manual stepwise and single-objective, hybrid genetic algorithm approaches to model building were applied, blinded to the results of the other approach, for selection of the compartment structure as well as inclusion and model form of inter-individual and inter-occasion variability, residual error, and covariates from a common set of model options. For the simulated dataset, stepwise covariate modeling identified three of four true covariates and two spurious covariates; Lasso identified two of four true and 0 spurious covariates; and the single-objective, hybrid genetic algorithm identified three of four true covariates and one spurious covariate. For the clinical datasets, the Akaike information criterion was a median of 22.3 points lower (range of 470.5 point decrease to 0.1 point decrease) for the best single-objective hybrid genetic-algorithm candidate model versus the final manual stepwise model: the Akaike information criterion was lower by greater than 10 points for four compounds and differed by less than 10 points for three compounds. The root mean squared error and absolute mean prediction error of the best single-objective hybrid genetic algorithm candidates were a median of 0.2 points higher (range of 38.9 point decrease to 27.3 point increase) and 0.02 points lower (range of 0.98 point decrease to 0.74 point increase), respectively, than that of the final stepwise models. In addition, the best single-objective, hybrid genetic algorithm candidate models had successful convergence and covariance steps for each compound, used the same compartment structure as the manual stepwise approach for 6 of 7 (86 %) compounds, and identified 54 % (7 of 13) of covariates included by the manual stepwise approach and 16 covariate relationships not included by manual stepwise models. The model parameter values between the final manual stepwise and best single-objective, hybrid genetic algorithm models differed by a median of 26.7 % (q1 = 4.9 % and q3 = 57.1 %). Finally, the single-objective, hybrid genetic algorithm approach was able to identify models capable of estimating absorption rate parameters for four compounds that the manual stepwise approach did not identify. The single-objective, hybrid genetic algorithm represents a general pharmacokinetic model building methodology whose ability to rapidly search the feasible solution space leads to nearly equivalent or superior model fits to pharmacokinetic data. Electronic supplementary material The online version of this article (doi:10.1007/s10928-012-9258-0) contains supplementary material, which is available to authorized users.

The citalopram for Alzheimer’s disease trial evaluated citalopram for the management for agitation in Alzheimer’s disease patients. Sparse data was available from this elderly patient population. A nonlinear mixed effects population pharmacokinetic modeling approach was used to describe the pharmacokinetics of R- and S-citalopram and their primary metabolite (desmethylcitalopram). A structural model with 4 compartments (one compartment/compound) with linear oral absorption and elimination described the data adequately. Overall, the model showed that clearance of the R-enantiomer was slower than the clearance of the S-enantiomer. Without accounting for any patient-specific covariates, the population estimate of the metabolic clearance of citalopram was 8.6 (R-citalopram) and 14 L/h (S-citalopram). The population estimate of the clearance of desmethylcitalopram was 23.8 (R-Dcit) and 38.5 L/h (S-Dcit). Several patient-specific covariates were found to have a significant effect on the pharmacokinetics of R,S-citalopram and desmethylcitalopram. A significant difference in the metabolic clearance of R-citalopram between males and females (13 vs 9.05 L/h) was identified in this analysis. Both R- and S-citalopram metabolic clearance decreased with age. Additionally, consistent with literature reports S-citalopram metabolic clearance increased with increasing body weight and was significantly influenced by CYPC19 genotype, with a difference of 5.8 L/h between extensive/rapid and intermediate/poor metabolizers. R,S-desmethylcitalopram clearance increased with increasing body weight. This model may allow for the opportunity to delineate the effect of R- and S-citalopram on pharmacodynamics outcomes related to the management of agitation in Alzheimer’s disease. Electronic supplementary material The online version of this article (doi:10.1007/s10928-015-9457-6) contains supplementary material, which is available to authorized users.

Population pharmacokinetic (PK) modeling methods can be statistically classified as either parametric or nonparametric (NP). Each classification can be divided into maximum likelihood (ML) or Bayesian (B) approazches. In this paper we discuss the nonparametric case using both maximum likelihood and Bayesian approaches. We present two nonparametric methods for estimating the unknown joint population distribution of model parameter values in a pharmacokinetic/pharmacodynamic (PK/PD) dataset. The first method is the NP Adaptive Grid (NPAG). The second is the NP Bayesian (NPB) algorithm with a stick-breaking process to construct a Dirichlet prior. Our objective is to compare the performance of these two methods using a simulated PK/PD dataset. Our results showed excellent performance of NPAG and NPB in a realistically simulated PK study. This simulation allowed us to have benchmarks in the form of the true population parameters to compare with the estimates produced by the two methods, while incorporating challenges like unbalanced sample times and sample numbers as well as the ability to include the covariate of patient weight. We conclude that both NPML and NPB can be used in realistic PK/PD population analysis problems. The advantages of one versus the other are discussed in the paper. NPAG and NPB are implemented in R and freely available for download within the Pmetrics package from www.lapk.org.

Few scientific contributions have made significant impact unless there was a champion who had the vision to see the potential for its use in seemingly disparate areas-and who then drove active implementation. In this paper, we present a historical summary of the development of non-linear mixed effects (NLME) modeling up to the more recent extensions of this statistical methodology. The paper places strong emphasis on the pivotal role played by Lewis B. Sheiner (1940-2004), who used this statistical methodology to elucidate solutions to real problems identified in clinical practice and in medical research and on how he drove implementation of the proposed solutions. A succinct overview of the evolution of the NLME modeling methodology is presented as well as ideas on how its expansion helped to provide guidance for a more scientific view of (model-based) drug development that reduces empiricism in favor of critical quantitative thinking and decision making.

International audience; The aim of this paper is to provide an overview of pharmacometric models that involve some latent process with Markovian dynamics. Such models include hidden Markov models which may be useful for describing the dynamics of a disease state that jumps from one state to another at discrete times. On the contrary, diffusion models are continuous-time and continuous-state Markov models that are relevant for modelling non observed phenomena that fluctuate continuously and randomly over time. We show that an extension of these models to mixed effects models is straightforward in a population context. We then show how the Forward-Backward algorithm used for inference in hidden Markov models and the extended Kalman filter used for inference in diffusion models can be combined with standard inference algorithms in mixed effects models for estimating the parameters of the model. The use of these models is illustrated with two applications: a hidden Markov model for describing the epileptic activity of a large number of patients and a stochastic differential equation based model for describing the pharmacokinetics of theophyllin.

An innovative probabilistic rule is proposed to predict the clinical significance or clinical insignificance of DDI. This rule is coupled with a hierarchical Bayesian model approach to summarize substrate/inhibitor's PK models from multiple published resources. This approach incorporates between-subject and between-study variances into DDI prediction. Hence, it can predict both population-average and subject-specific AUCR. The clinically significant DDI, weak DDI, and clinically insignificant inhibitions are decided by the probabilities of predicted AUCR falling into three intervals, (-infinity, 1.25), (1.25, 2), and (2, infinity). The main advantage of this probabilistic rule to predict clinical significance of DDI over the deterministic rule is that the probabilistic rule considers the sample variability, and the decision is independent of sampling variation; while deterministic rule based decision will vary from sample to sample. The probabilistic rule proposed in this paper is best suited for the situation when in vivo PK studies and models are available for both the inhibitor and substrate. An early decision on clinically significant or clinically insignificant inhibition can avoid additional DDI studies. Ketoconazole and midazolam are used as an interaction pair to illustrate our idea. AUCR predictions incorporating between-subject variability always have greater variances than population-average AUCR predictions. A clinically insignificant AUCR at population-average level is not necessarily true when considering between-subject variability. Additional simulation studies suggest that predicted AUCRs highly depend on the interaction constant K(i) and dose combinations.

The non-linear mixed-effects model based on stochastic differential equations (SDEs) provides an attractive residual error model, that is able to handle serially correlated residuals typically arising from structural mis-specification of the true underlying model. The use of SDEs also opens up for new tools for model development and easily allows for tracking of unknown inputs and parameters over time. An algorithm for maximum likelihood estimation of the model has earlier been proposed, and the present paper presents the first general implementation of this algorithm. The implementation is done in Matlab and also demonstrates the use of parallel computing for improved estimation times. The use of the implementation is illustrated by two examples of application which focus on the ability of the model to estimate unknown inputs facilitated by the extension to SDEs. The first application is a deconvolution-type estimation of the insulin secretion rate based on a linear two-compartment model for C-peptide measurements. In the second application the model is extended to also give an estimate of the time varying liver extraction based on both C-peptide and insulin measurements.

The objective of this paper is to determine optimal blood sampling time windows for the estimation of pharmacokinetic (PK) parameters by a population approach within the clinical constraints. A population PK model was developed to describe a reference phase II PK dataset. Using this model and the parameter estimates, D-optimal sampling times were determined by optimising the determinant of the population Fisher information matrix (PFIM) using PFIM_ _M 1.2 and the modified Fedorov exchange algorithm. Optimal sampling time windows were then determined by allowing the D-optimal windows design to result in a specified level of efficiency when compared to the fixed-times D-optimal design. The best results were obtained when K(a) and IIV on K(a) were fixed. Windows were determined using this approach assuming 90% level of efficiency and uniform sample distribution. Four optimal sampling time windows were determined as follow: at trough between 22 h and new drug administration; between 2 and 4 h after dose for all patients; and for 1/3 of the patients only 2 sampling time windows between 4 and 10 h after dose, equal to [4 h-5 h 05] and [9 h 10-10 h]. This work permitted the determination of an optimal design, with suitable sampling time windows which was then evaluated by simulations. The sampling time windows will be used to define the sampling schedule in a prospective phase II study.

In PK/PD data analysis, statistical models involving random variables are developed. At times an analysis can be carried out by conditioning on certain random events involving these variables. This paper attempts to clarify issues regarding conditioning. In particular, conditioning is examined as it relates to a number of disparate practical matters: missing covariate values, dose titration, BQL data, "no change from baseline" data, and the use of a truncated intraindividual probability distribution for PK observations.

The question of whether or not a mixed effects model is required when modelling binary data with inter-subject variability and within subject correlation was reported in this journal by Yano et al. (J. Pharmacokin. Pharmacodyn. 28:389-412 [2001]). That report used simulation experiments to demonstrate that, under certain circumstances, the use of a fixed effects model produced more accurate estimates of the fixed effect parameters than those produced by a mixed effects model. The Laplace approximation to the likelihood was used when fitting the mixed effects model. This paper repeats one of those simulation experiments, with two binary observations recorded for every subject, and uses both the Laplace and the adaptive Gaussian quadrature approximations to the likelihood when fitting the mixed effects model. The results show that the estimates produced using the Laplace approximation include a small number of extreme outliers. This was not the case when using the adaptive Gaussian quadrature approximation. Further examination of these outliers shows that they arise in situations in which the Laplace approximation seriously overestimates the likelihood in an extreme region of the parameter space. It is also demonstrated that when the number of observations per subject is increased from two to three, the estimates based on the Laplace approximation no longer include any extreme outliers. The root mean squared error is a combination of the bias and the variability of the estimates. Increasing the sample size is known to reduce the variability of an estimator with a consequent reduction in its root mean squared error. The estimates based on the fixed effects model are inherently biased and this bias acts as a lower bound for the root mean squared error of these estimates. Consequently, it might be expected that for data sets with a greater number of subjects the estimates based on the mixed effects model would be more accurate than those based on the fixed effects model. This is borne out by the results of a further simulation experiment with an increased number of subjects in each set of data. The difference in the interpretation of the parameters of the fixed and mixed effects models is discussed. It is demonstrated that the mixed effects model and parameter estimates can be used to estimate the parameters of the fixed effects model but not vice versa.

In this paper, the two non-linear mixed-effects programs NONMEM and NLME were compared for their use in population pharmacokinetic/pharmacodynamic (PK/PD) modelling. We have described the first-order conditional estimation (FOCE) method as implemented in NONMEM and the alternating algorithm in NLME proposed by Lindstrom and Bates. The two programs were tested using clinical PK/PD data of a new gonadotropin-releasing hormone (GnRH) antagonist degarelix currently being developed for prostate cancer treatment. The pharmacokinetics of intravenous administered degarelix was analysed using a three compartment model while the pharmacodynamics was analysed using a turnover model with a pool compartment. The results indicated that the two algorithms produce consistent parameter estimates. The bias and precision of the two algorithms were further investigated using a parametric bootstrap procedure which showed that NONMEM produced more accurate results than NLME together with the nlmeODE package for this specific study.

The paper compares performance of Nonmem estimation methods--first order conditional estimation with interaction (FOCEI), iterative two stage (ITS), Monte Carlo importance sampling (IMP), importance sampling assisted by mode a posteriori (IMPMAP), stochastic approximation expectation-maximization (SAEM), and Markov chain Monte Carlo Bayesian (BAYES), on the simulated examples of a monoclonal antibody with target-mediated drug disposition (TMDD), demonstrates how optimization of the estimation options improves performance, and compares standard errors of Nonmem parameter estimates with those predicted by PFIM 3.2 optimal design software. In the examples of the one- and two-target quasi-steady-state TMDD models with rich sampling, the parameter estimates and standard errors of the new Nonmem 7.2.0 ITS, IMP, IMPMAP, SAEM and BAYES estimation methods were similar to the FOCEI method, although larger deviation from the true parameter values (those used to simulate the data) was observed using the BAYES method for poorly identifiable parameters. Standard errors of the parameter estimates were in general agreement with the PFIM 3.2 predictions. The ITS, IMP, and IMPMAP methods with the convergence tester were the fastest methods, reducing the computation time by about ten times relative to the FOCEI method. Use of lower computational precision requirements for the FOCEI method reduced the estimation time by 3-5 times without compromising the quality of the parameter estimates, and equaled or exceeded the speed of the SAEM and BAYES methods. Use of parallel computations with 4-12 processors running on the same computer improved the speed proportionally to the number of processors with the efficiency (for 12 processor run) in the range of 85-95% for all methods except BAYES, which had parallelization efficiency of about 70%.

Computer models of biological systems grow more complex as computing power increase. Often these models are defined as differential equations and no analytical solutions exist. Numerical integration is used to approximate the solution; this can be computationally intensive, time consuming and be a large proportion of the total computer runtime. The performance of different integration methods depend on the mathematical properties of the differential equations system at hand. In this paper we investigate the possibility of runtime gains by calculating parts of or the whole differential equations system at given time intervals, outside of the differential equations solver. This approach was tested on nine models defined as differential equations with the goal to reduce runtime while maintaining model fit, based on the objective function value. The software used was NONMEM. In four models the computational runtime was successfully reduced (by 59-96%). The differences in parameter estimates, compared to using only the differential equations solver were less than 12% for all fixed effects parameters. For the variance parameters, estimates were within 10% for the majority of the parameters. Population and individual predictions were similar and the differences in OFV were between 1 and -14 units. When computational runtime seriously affects the usefulness of a model we suggest evaluating this approach for repetitive elements of model building and evaluation such as covariate inclusions or bootstraps.

This paper presents an example of how optimal design methodology was used to help design a phase II clinical study. The planned analysis would relate the clinical endpoint to exposure (measured via the area under the curve (AUC)), rather than dose. Optimal design methodology was used to compare a number of candidate phase II designs, and an algorithm for finding optimal designs was employed. The sigmoidal E(max) with baseline (E₀) model was used to relate the clinical endpoint to individual subject AUCs, and the primary metrics were D optimality and the standard error (SE) of the AUC required to yield a clinically relevant change in the clinical endpoint. The performance of the candidate designs were compared across four different 'true' exposure response relationships (determined from the analysis of an earlier proof of concept (PoC) study). The results suggested the total sample size should be increased from the planned 540 individuals, and that the optimal design with 700 individuals would be equivalent to 812 individuals with the reference design (a 16% gain). The performance with this design was considered acceptable, although all designs performed poorly if the true exposure response relationship was very flat. This work allowed a prospective assessment of the likely performance and precision from the exposure response modelling prior to the start of the phase II study, and hence allowed the design to be revised to ensure the subsequent analysis would be of most value.

This paper describes an approach for optimising sampling windows for population pharmacokinetic experiments. Sampling windows designs are more practical in late phase drug development where patients are enrolled in many centres and in out-patient clinic settings. Collection of samples under the uncontrolled environment at these centres at fixed times may be problematic and can result in uninformative data. Population pharmacokinetic sampling windows design provides an opportunity to control when samples are collected by allowing some flexibility and yet provide satisfactory parameter estimation. This approach uses information obtained from previous experiments about the model and parameter estimates to optimise sampling windows for population pharmacokinetic experiments within a space of admissible sampling windows sequences. The optimisation is based on a continuous design and in addition to sampling windows the structure of the population design in terms of the proportion of subjects in elementary designs, number of elementary designs in the population design and number of sampling windows per elementary design is also optimised. The results obtained showed that optimal sampling windows designs obtained using this approach are very efficient for estimating population PK parameters and provide greater flexibility in terms of when samples are collected. The results obtained also showed that the generalized equivalence theorem holds for this approach.

Population pharmacokinetic (PK) (or pharmacodynamic (PD)) modelling aims to analyse the variability of drug kinetics (or dynamics) between numerous subjects belonging to a population. Such variability includes inter- and intra-individual sources leading to important differences between the variation ranges, the relative concentrations and the global shapes of PK profiles. These various sources of variability suggest that the distance metrics between the subjects can be examined under different aspects. Some subjects are so distant from the majority that they tend to be atypical or outliers. This paper presents three multivariate statistical methods to diagnose the outliers within a full population PK dataset, prior to any modelling step. Each method combined all the concentration-time variables to analyse the differences between patients by referring to a distance criterion: (a) Correspondence analysis (CA) used the chi-square distance to highlight the most atypical profiles in terms of relative concentrations; (b) Mahalanobis distance was calculated to extract PK profiles showing atypical shapes due to atypical variations in concentration; (c) Andrews method combined all the concentration variables into a Fourier transformation to give sine-cosine curves showing the clustering behaviours of subjects under the Euclidean distance criterion. After identification of outlier subjects, these methods can also be used to extract the concentration values which cause the atypical states of the patients. Therefore, the outliers will incorporate different variability sources of the PK dataset according to each method and independently of any PK modelling. Finally, a significant positive trend was found between the number of times outlier concentrations were detected (by one, two or three diagnostics) and the NPDE metrics of these concentrations (after a PK modelling): NPDE were highest when the corresponding concentration was detected by more diagnostics a priori. The application of a priori outlier diagnostics is illustrated here on two PK datasets: stimulated cortisol by synacthen and capecitabine administrated orally.

Long term therapy with antiretroviral agents in HIV-infected patients often result in failure to suppress the virus load. Imperfect adherence to prescribed antiviral drugs is an important factor explaining the resurgence of virus. A better understanding of the factors responsible for the virological failure is important for the development of new treatment strategies. Many complex non-linear models have been developed to describe and simulate the dynamics of HIV-1 virus. Those complicated viral dynamic models have not been used in clinical trials to estimate HIV dynamics parameters, due to their complexity, until the recent development of simplification and approximation techniques. The estimation of the parameters associated with the dynamics from real data has been mostly limited to linearized models that can only explain the decay (suppression) of the virus following antiviral treatment. Moreover, no complete characterization of typical clinical data in terms of inter-subject variability and identification of important covariates effecting HIV-1 dynamics has been attempted. The objective of our paper was to develop a hierarchical non-linear mixed effect model characterizing inter-subject variability in the long-term response to treatment of HIV-1 RNA, and show how the model can be used to quantify the effect of important covariates, such as physiological variables, adherence to treatment or previous exposure to treatment, on the dynamics of HIV-1 RNA. As an example we report the analysis of AIDS clinical trial data from AACTG 398, which shows that patients with previous exposure to treatment show faster death rates for HIV-1, and that higher adherence to treatment is associated with lower reproductive ratio.

Effective long-term treatment of Type 2 Diabetes Mellitus (T2DM) implies modification of the disease processes that cause this progressive disorder. This paper proposes a mechanism-based approach to disease progression modeling of T2DM that aims to provide the ability to describe and quantify the effects of treatment on the time-course of the progressive loss of beta-cell function and insulin-sensitivity underlying T2DM. It develops a population pharmacodynamic model that incorporates mechanism-based representations of the homeostatic feedback relationships between fasting levels of plasma glucose (FPG) and fasting serum insulin (FSI), and the physiological feed-forward relationship between FPG and glycosylated hemoglobin A1c (HbA1c). This model was developed on data from two parallel one-year studies comparing the effects of pioglitazone relative to metformin or sulfonylurea treatment in 2,408 treatment-naïve T2DM patients. It was found that the model provided accurate descriptions of the time-courses of FPG and HbA1c for different treatment arms. It allowed the identification of the long-term effects of different treatments on loss of beta-cell function and insulin-sensitivity, independently from their immediate anti-hyperglycemic effects modeled at their specific sites of action. Hence it avoided the confounding of these effects that is inherent in point estimates of beta-cell function and insulin-sensitivity such as the widely used HOMA-%B and HOMA-%S. It was also found that metformin therapy did not result in a reduction in FSI levels in conjunction with reduced FPG levels, as expected for an insulin-sensitizer, whereas pioglitazone therapy did. It is concluded that, although its current implementation leaves room for further improvement, the mechanism-based approach presented here constitutes a promising conceptual advance in the study of T2DM disease progression and disease modification.

The mathematical model structure selected to describe system behavior is at least partially dependent on the proposed use of the model. In this paper, a pharmacokinetic(PK)/pharmacodynamic (PD) model for use in drug delivery algorithm synthesis is developed. The antitumor agent 9-nitrocamptothecin (9NC) was administered orally to severe combined immunodeficient (SCID) mice bearing subcutaneously implanted HT29 human colon xenografts, and the effect of 9NC on those xenografts was characterized. Different PK model structures were considered in characterizing the dynamics of the drug concentration in the plasma. Akaike's Information Criterion (AIC) was used to select the model structure maximizing fit accuracy while simultaneously minimizing the number of model parameters. The resulting PK model was a set of coupled linear ordinary differential equations able to describe the nonlinear dynamic behavior (e.g. plateauing, etc.) of the drug concentrations observed in the plasma. Pharmacodynamics were modeled by characterizing tumor growth in both the untreated and drug-treated animals. The resulting PK/PD model related drug administration to effect, and this model has a structure that facilitates future control algorithm synthesis. Control algorithms in this context would directly utilize PK/PD model predictions. These predictions would be used to determine the amount and frequency of drug administration in order to reduce the tumor burden without violating clinically relevant constraints. This methodology could then be used to aid the clinician in selecting dose levels and schedules, and extension to patient tailored treatment may eventually be feasible with this approach.