2019 - Stockholm - Sweden

PAGE 2019: Methodology - New Modelling Approaches
Vangelis Karalis

An In Vitro – In Vivo Simulation Methodology for Predicting the Outcome of Bioequivalence Studies

Eleni Karatza, Vangelis Karalis

Department of Pharmacy, School of Health Sciences, National and Kapodistrian University of Athens, Greece

Introduction: The role of modeling and simulation methodologies in drug development is emerging.

Objectives: To develop an in vitro - in vivo simulation (IVIVS) approach aiming at predicting the in vivo pharmacokinetic (PK) performance and the probability of success of a bioequivalence (BE) study. Relying only on in vitro dissolution data, the final aim of this methodology is to assist the R&D department selecting the appropriate test product lot (if many), clinical design, and sample size (N). This IVIVS approach was applied twice to two different irbesartan/hydrochlorothiazide BE studies.

Methods: The IVIVS methodology can conceptually be split into three steps: in vitro, in vivo, and IVIVS (1). A) In vitro: Mathematical models (e.g., Weibull, first-order) are used to describe the individual dissolution profiles at the three pH values. In a subsequent step, these models are used to simulate a single dissolution profile. Several scenarios are used to imitate drug transit through the gastrointestinal tract in terms of residence time at each pH condition (i.e., stomach, small intestine). B) In vivo: Using literature or actual in vivo concentration-time data for the drug under study, a PK model is developed using non-liner mixed effects (NLME) modeling approaches. C) IVIVS: This step applies a joint in vitro – in vivo model incorporating the models developed in ‘A’ and ‘B’. Using the mean and variability estimates for each PK parameter, Monte Carlo simulations are performed and virtual subjects are generated assuming several levels of between- and within-subject variabilities and PK scenarios. The typical BE measures (Cmax, AUCt, and AUCinf) are calculated for each simulated individual using non-compartmental approaches. In turn, the estimated BE measures are set into a certain clinical design (e.g., 2x2, 3x3, 2x4). For each BE study, the typical statistical assessment is applied as imposed by the regulatory authorities (2,3). Each virtual BE study is repeated for thousands/millions of times and the statistical power is estimated. The same procedure can be followed several times taking into consideration the sample size, clinical design, sampling scheme, and anticipated difference in the PK parameters between the two formulations. Thus, an overall table with the statistical power of each scenario is constructed where the most probable case can be selected. In order to validate the reliability of the proposed methodology, the whole procedure is applied twice, firstly using literature PK parameters and secondly using estimated PK parameters from in vivo data through NLME approaches. The two BE studies utilized in this analysis refer to 2x2, single dose irbesartan/hydrochlorothiazide trials [study 1: 300 mg /25 mg, N=32 (2011), study 2: 150 mg / 12.5 mg, N=46 (2008)]. The computational work is performed in MATLAB®which is finally implemented in the form of a graphical user interface (GUI), while Monolix®2018R2 was used for the in vivo fittings of step ‘B’.

Results: In study 1, the actual in vitro data were utilized, while the PK information for irbesartan/hydrochlorothiazide came from the literature. Based on the IVIVS methodology, the overall predicted statistical power for the several scenarios tested was 81%. The actual BE study, as illustrated in the relevant report, resulted in a power of 86%. For study 2, the new dissolution data were utilized, while the PK estimates came from NLME fitting to the C-t data of study 1. For irbesartan, a two-compartment model was derived assuming first order absorption and elimination from the central compartment and combined residual error model (ka=0.665 h-1, Cl/F= 1.34*104l/h, V1/F= 3.64*104l, Q/F= 8.92*103l/h, V2/F= 6.61*104l/h). In case of hydrochlorothiazide, a similar PK model was found with the following population estimates: Tlag=0.404 h, ka=0.773 h-1, V1/F=1.37*105ml, V2/F=1.46*105ml, Q/F=2.54*104(ml/h), and CL/F=3.45*104l/h. Using the same scenarios, as in case of the first IVIVS for study 1, the overall predicted probability of success for 46 subjects was 91%, whereas the actual power of the study reported in the protocol was 89%.

Conclusions: A new in vitro – in vivo simulation methodology is presented aiming at predicting the in vivo outcome (drug plasma concentration and probability of success of a BE study) based on in vitro dissolution studies. The proposed methodology was applied successfully to two irbesartan/hydrochlorothiazide BE studies. 

[1] Karatza E and Karalis V. A Semi-Physiological Modeling & Simulation approach for guided decision making in R&D. AAPS Annual meeting and exposition (Washington DC, November 4-7, 2018, T1130-05-036).
[2] EMA Guideline on the investigation of bioequivalence. Doc. Ref.: CPMP/EWP/QWP/1401/98 Rev. 1/ Corr ** (2010).
[3] FDA Guidance for Industry. Bioavailability and Bioequivalence Studies for Orally Administered Drug Products — General Considerations (2003).

Reference: PAGE 28 (2019) Abstr 8860 [www.page-meeting.org/?abstract=8860]
Poster: Methodology - New Modelling Approaches