Evaluating In Vivo-In Vitro Correlation Using a Bayesian Approach

Bayesian estimation in NONMEM

Bayesian estimation is a powerful but underutilized tool for answering drug development questions. In this tutorial, the principles of Bayesian model development, assessment, and prior selection will be outlined. An example pharmacokinetic (PK) model will be used to demonstrate the implementation of Bayesian modeling using the nonlinear mixed-effects modeling software NONMEM.

Bayesian knowledge integration for an in vitro-in vivo correlation model

The primary goal of "in vitro-in vivo correlation" (IVIVC) is the reliable prediction of the in vivo serum concentration-time course, based on the in vitro drug dissolution or release profiles. IVIVC methods are particularly appropriate for formulations that are released over an extended period of time or with a lag in absorption and may support approving a change in formulation of a drug without additional bioequivalence trials in human subjects. Most of the current IVIVC models are assessed using frequentist methods, such as linear regression, based on averaged data and entail complex and potentially unstable mathematical deconvolution. The proposed IVIVC approach includes (a) a nonlinear-mixed effects model for the in vitro release data; (b) a population pharmacokinetic (PK) compartment model for the in vivo immediate release (IR) data; and (c) a system of ordinal differential equations (ODEs), containing the submodels (a) and (b), which approximates and predicts the in vivo controlled release (CR) data. The innovation in this paper consists of splitting the parameter space between submodels (a) and (b) versus (c). Subsequently, the uncertainty on these parameters is accounted for using a Bayesian framework, that is estimates from the first two submodels serve as priors for the Bayesian hierarchical third submodel. As such, the Bayesian method explained ensures a natural integration and transfer of knowledge between various sources of information, balancing possible differences in sample size and parameter uncertainty of in vitro and in vivo studies. Consequently, it is a very flexible approach yielding results for a broad range of data situations. The application of the method is demonstrated for a transdermal patch (TD).

Defining level A IVIVC dissolution specifications based on individual in vitro dissolution profiles of a controlled release formulation

Regulatory guidelines recommend that, when a level A IVIVC is established, dissolution specification should be established using averaged data and the maximum difference between AUC and C_max between the reference and test formulations cannot be greater than 20%. However, averaging data assumes a loss of information and may reflect a bias in the results. The objective of the current work is to present a new approach to establish dissolution specifications using a new methodology (individual approach) instead of average data (classical approach). Different scenarios were established based on the relationship between in vitro-in vivo dissolution rate coefficient using a level A IVIVC of a controlled release formulation. Then, in order to compare this new approach with the classical one, six additional batches were simulated. For each batch, 1000 simulations of a dissolution assay were run. C_max ratios between the reference formulation and each batch were calculated showing that the individual approach was more sensitive and able to detect differences between the reference and the batch formulation compared to the classical approach. Additionally, the new methodology displays wider dissolution specification limits than the classical approach, ensuring that any tablet from the new batch would generate in vivo profiles which its AUC or C_max ratio will be out of the 0.8-1.25 range, taking into account the in vitro and in vivo variability of the new batches developed.

A Simple Approach for Comparing the In Vitro Dissolution Profiles of Highly Variable Drug Products: a Proposal

When in vitro dissolution profile variability prohibits the use of the F2 metric, there currently is no satisfactory alternative available. Published reports evaluating alternative approaches such as Multivariate Statistical Distance and use of a bootstrap F2 identify sources of bias that can limit the utility of these alternatives. Within veterinary medicine, an additional complication is the potential magnitude of interlot variability associated with dosage forms containing "natural" ingredients. In situations when both interlot and intralot variability need to be factored in the test and reference profile comparison, we designed a method that integrates such concepts as F2, USP S1 and S2 criteria and statistical tolerance limits. Unlike F2, this alternative approach integrates a statistical confidence into the determination through the use of tolerance limits about the reference product profile. Moreover, while differences in product variability, along with differences in mean profiles, will influence the comparability assessment, this method does not impose the need to confirm homogeneity of variances: there is not direct statistical comparison of test versus reference dissolution data. For more typical situations when interlot variability is not a concern, the F2 component can be omitted from the profile comparison. Lastly, by being a model-independent approach, we avoid the potential for introducing error into the comparability determination due either to model misspecification or problems associated with a lack of collinearity. This manuscript details this alternative approach and the results of performance characterization efforts to illustrate its behavior under a range of potential situations.