2021 - Online - In the cloud

PAGE 2021: Methodology - Other topics
Simone Zannoni

Investigating the identifiability of one-stage IVIVC population models for extended-release dosage forms

Simone Zannoni (1), Ari Brekkan (2), Giovanni Smania (2)

(1) University of Padova, Italy, (2) Pharmetheus AB, Sweden

Objectives: In-vitro/in-vivo correlation (IVIVC) models can aid in the development of modified-release dosage forms such as extended-release (ER) products [1]. Besides allowing for non-linear disposition kinetics, one-stage (or convolution-based) IVIVC population models can account for inter-individual variability (IIV) in drug dissolution and absorption [2]. These models are often developed based on Phase I cross-over pharmacokinetic (PK) studies in healthy subjects, where the sample size is relatively low (10-20 subjects). The objective of this analysis was to evaluate the identifiability of one-stage IVIVC population models developed from small Phase I PK studies.

Methods: A published model for methylphenidate hydrochloride (MPH) ER capsules was used as a case-study [1], in which the in-vitro dissolution was described by a double-Weibull function. Two linear time scaling models were evaluated: (i) tvitro = tvivo (no time scaling) and (ii) tvitro = 0.5·tvivo (in-vivo absorption twice as slow as in-vitro dissolution). The Phase I study was designed as a cross-over study where each healthy subject received 3 ER formulations (slow, medium and fast release), with the individual disposition parameters - quantifying the unit impulse response - assumed to be known. First, optimal design techniques were used to derive adequate PK sampling schedules for the Phase I study, assuming a sample size of 12 and 24 subjects. Then, 100 stochastic simulations and re-estimations (SSE) were performed to assess precision and accuracy of the IVIVC model parameters. Finally, FDA internal predictability criteria for Level A IVIVC were evaluated [3]. Two competing models were investigated: a model with a single IIV term on the absorption rate (Model A) vs. a model with 5 IIV terms on each of the Weibull parameters (Model B). Design optimization was done using the R package popED [4], PsN [5] was used to carry out the SSE step.

Results: Optimal design suggested that the expected precision in structural IVIVC parameters was adequate for all scenarios, as shown by predicted relative standard errors (RSEs) < 10%. The predicted RSE for IIV parameters was consistently < 30% for Model A, while in Model B some IIV parameters were associated with an RSE > 30%. SSE confirmed the adequate precision and indicated satisfactory accuracy in structural IVIVC parameters for both models. Poorer accuracy and precision in IIV parameters were obtained for Model B (|relative bias| < 63%, RMSE < 70%) compared to Model A (|relative bias| < 8%, RMSE < 24%). Importantly, SSE showed that the percentage of study replicates that satisfied the internal predictability criteria for a Level A IVIVC with tvitro = 0.5·tvivo was 83% (n=12) and 95% (n=24) for Model A compared to 58% (n=12) and 77% (n=24) for Model B.

Conclusions: In light of the small size of the clinical studies employed in the development of IVIVC models, IIV parameters should be employed with parsimony as overparameterization could result in a loss of power to detect a Level A IVIVC. A prospective investigation of the model identifiability given the study design can help mitigating the risk of IVIVC failure.

[1] Gomeni R, Fang LL, Bressolle-Gomeni F, Spencer TJ, Faraone SV, Babiskin A. A General Framework for Assessing In vitro/In vivo Correlation as a Tool for Maximizing the Benefit-Risk Ratio of a Treatment Using a Convolution-Based Modeling Approach. CPT Pharmacometrics Syst Pharmacol. 2019 Feb;8(2):97-106.
[2] https://www.ema.europa.eu/en/documents/scientific-guideline/guideline-pharmacokinetic-clinical-evaluation-modified-release-dosage-forms_en.pdf
[3] https://www.fda.gov/media/70939/download
[4] Nyberg J, Ueckert S, Stroemberg EA, Hennig S, Karlsson MO, Hooker AC (2012). “PopED: An extended, parallelized, nonlinear mixed effects models optimal design tool.” Computer Methods and Programs in Biomedicine, 108.
[5] Lindbom L, Ribbing J and Jonsson EN. Perl-speaks-NONMEM (PsN) - a Perl module for NONMEM related programming. Comp Meth Prog Biomed 75 (2), 2004.

Reference: PAGE 29 (2021) Abstr 9691 [www.page-meeting.org/?abstract=9691]
Poster: Methodology - Other topics
Click to open PDF poster/presentation (click to open)