(T1530-10-56) Development of In Vitro/In Vivo Correlation for Long-Acting Suspensions via Bootstrapping

Purpose: In vitro/in vivo correlation (IVIVC) is a mathematical model to predict the in vivo performance (e.g., plasma concentration-time profiles) of a drug product via its in vitro characteristics (e.g., dissolution profiles). This tool is of great importance to expedite the development life-cycle for pharmaceutical industry and reduce the burden for regulatory assessment, especially for long-acting injectable (LAI) drug products. However, owing to the inherent complexity and relative novelty, no regulatory guidance is available for IVIVC development of LAIs. Moreover, per reported pharmacokinetic analyses, LAI dosage forms are prone to exhibit much higher inter-patient variability compared to conventional formulation types [1, 2]. Currently, relevant publications and an industry survey indicate most of the researchers tend to use mean (both in vitro and in vivo) data to establish IVIVCs, which may discount the inherent variability of LAIs, and hence result in stringent IVIVCs with poor predictive performance. Therefore, it is critical to use a statistical tool to represent highly variable in vivo profiles and more accurately evaluate IVIVC. In this study, an innovative bootstrapping strategy is introduced for the first time for IVIVC development and evaluation to capture the inherent high inter-subject variabilities observed with LAI aqueous suspensions.

Methods: Briefly, three LAI suspensions including single agent product 1, combination product 1, and combination product 2 were used for model development. In vitro dissolution and rat plasma concentration – time (Cp – t) profiles of these formulations were extracted from a published study [3]. The Cp – t profile of each individual rat was deconvoluted to obtain in vivo absorption profile via Loo-Riegelman method using the IVIVC Module in GastroPlus software (Version 9.8.1003, SimulationsPlus, Lancaster, CA). Random selection of internal formulations was conducted to create different formulation combinations for IVIVC development and validation. Then, for each combination, bootstrapping was performed using in vitro release profiles (n = 4 for each formulation) and deconvoluted in vivo absorption profiles (n = 3 or 4 for different formulations) from the selected internal formulations. Correlations were established between two datasets (in vitro release profiles, in vivo absorption profiles) using a random sampling with replacement technique known as bootstrapping, which resulted in multiple linear correlations, each characterized by parameters specific to a given in vivo-in vitro pairing. The mean and 95% confidence intervals of the resulting parameter distribution were used to establish the IVIVC. Then, the established IVIVCs were used to convert the in vitro release profiles to predicted in vivo absorption profiles. The model was evaluated by comparison between predicted and observed in vivo absorption profiles, as well as predicted and observed Cp – t profiles. The critical PK parameters (AUC and Cmax) for both internal and external formulations were calculated and compared.

Results: The developed IVIVC models were used to predict the in vivo absorption profiles, then compared with the observed in vivo profiles (Figure 1). In general, IVIVCs established using different formulation combinations exhibited similar results, indicating good comparability and robustness of the established IVIVCs. Moreover, it is noted that the mean predicted absorption profiles overlapped well with the mean observed absorption profiles for combination drug products 1 and 2. Reasonably good overlapping ( < 10% difference) of predicted and observed mean profiles was observed for single agent drug product 1. More importantly, it should be noted that the prediction ranges (blue shade) all fell into the observed data ranges (red shade), and very close to the observed mean profiles for all formulations. In addition, the predicted in vivo absorption profiles were convoluted into predicted Cp-t profiles and compared with the observed Cp-t profiles to further understand the predictability of the developed IVIVCs (Figure 2). Overall, results indicated good predictability (lower and upper end of the predicted 95 CI mostly within the observed data range). These results corresponded well with the predicted versus observed data shown in previous section. These models were further validated by comparing predicted and observed PK parameters (Cmax and AUC). Conventionally, mean values have been compared based on the percent prediction error, based on the criteria of the FDA guidance. Considering the high inter-subject variability, 95% CI ranges were compared between predicted and observed data sets in this work. As shown in Table 1, predicted range of both Cmax and AUC were within the observed data range. This further confirms the capability of bootstrapped IVIVC to incorporate the inter-subject variability for complex LAI dosage forms.

Conclusion: This work demonstrates the feasibility of establishing IVIVCs via bootstrapping technique for complex LAI dosage forms. More importantly, it implies the capability of bootstrapped IVIVCs to capture the variability of observed in vivo data within a reasonable prediction range. The work provides an innovative modeling strategy and may pave the way for future IVIVC development and assessment.

References: 1. Landovitz, R.J., et al., Safety, tolerability, and pharmacokinetics of long-acting injectable cabotegravir in low-risk HIV-uninfected individuals: HPTN 077, a phase 2a randomized controlled trial. PLoS medicine, 2018. 15(11): p. e1002690.
2. Johnson, A.R., et al., A retrospective analysis of preclinical and clinical pharmacokinetics from administration of long-acting aqueous suspensions. Pharmaceutical Research, 2023. 40(7): p. 1641-1656.
3. Jain, K.M.H., et al., Accelerated in vitro release testing using USP apparatus 2 to predict rat in vivo performance of single agent and combination long-acting injectable suspension products. Journal of Pharmaceutical Sciences, 2024.

Figure 1. Comparison of predicted versus observed in vivo absorption profiles of three suspension drug products following intramuscular injection.

Figure 2. Comparison of predicted versus observed plasma concentration – time profiles of different suspension drug products (combination drug product 1 and combination drug product 2 as the internal formulations).

Table 1. Validation of the IVIVC developed using three suspension formulations.