(M1030-06-37) Predicting Drug Release from EVA-Based Long-Acting Implants: Mechanisms, Modeling, and Simulation of Drug Release

Purpose: Nexplanon^® is an ethylene vinyl acetate (EVA) based implant prevents pregnancy for three years. The long acting effect is achieved through its controlled release of etonogestrel. The implant is a rod with a core/shell structure (4 cm long with 2 mm diameter), consisting of a core layer composed of EVA, etonogestrel, and barium sulfate, surrounded by a rate-controlling EVA “skin” layer. The two ends of the implant are not covered by the skin and etonogestrel is released from both the two ends and the skin. Despite its extensive use, drug release mechanisms of Nexplanon^® are not thoroughly investigated. Moreover, characterization of etonogestrel release over the entire three years duration is time-consuming and labor-intensive. To address this challenge, drug release mechanisms were characterized and described using mechanistic models. Then the parameters of these models were calibrated using short term experimental release data. The final model was then used to simulate the long term drug release.

Methods: Drug release was tested in 0.00075 % polysorbate 80 aqueous solution at 37 ºC at 150 rpm. To test skin-only release, implant ends were sealed with etonogestrel-impermeable Loctite^® 4011 glue. For ends-only release, the skin was sealed. Film permeation experiments in diffusion cells was used to determine etonogestrel solubility and diffusivity in EVA [1]. Ends release and skin release were modeled separately and combined to calculate the total release (Figure 1A). Because respective effective release areas of skin release and ends release reduced overtime due to depletion zones build up, model predicted release rates were adjusted using a depletion zone factor (Figure 1B). Ends release data were first fitted using a diffusion-based semi-empirical equation (t^-0.5) [2]. Because ends release also conformed to the homogeneous Higuchi model [3], effective transport parameters in the ends depletion zones were determined by comparing the coefficients of t^-0.5 in the empirical and mechanistic equations. These parameters were applied to skin depletion zone assuming structural similarity in ends and skin depletion zones (Figure 1B). Skin burst release was modeled using first order kinetic [4] [5]. The subsequent sustained skin release was modeled using a membrane matrix model (Paul model) [6]. Boundary layer effect was factored by comparing the experimental rate to the theoretical rate predicted by Fick’s law. This modeling approach is showed in Figure 1C. Drug release was assumed to be controlled by diffusion only. For Paul model, the cylindrical core/shell structure was approximated using a plane geometry. Curve fitting was performed using GraphPad Prism. Model errors were quantified using root mean square error (RMSE). To validate the model, the predicted release profile was compared to published three years in vitro release data.

Results: Ends release rates were linear to t^-0.5 over three months following 31.0 × t^-0.5 + 3.0 (Figure 2A). This indicated a diffusion controlled kinetic and a zero order kinetic of ends release [2] . Skin burst release completed in 20 days following first order kinetic (Figure 2B). The subsequent release from the skin at day 21 was reduced by 40% due to the presence of an un-stirred boundary layer (F_HDBL = 0.6, see Figure 2C). The predicted skin release (post day 20) closely matched the three months experimental data (Figure 2C). The final model accurately predicted the total implant release over three months (Figure 2D, RMSE = 1.0). Then the full release profile simulated using the final model closely followed the trend of published in vitro data for up to 3.5 years (Figure 3A), with overall 9.6 % under-prediction (Figure 3B). This deviation is attributed to different testing conditions between data used for model development and reference data, which were tested in a larger media volume and at a higher agitation speed. Overall, the model demonstrated good predictivity and effectively captured key characteristics of long-term release behavior.

Conclusion: The semi-empirical model developed using three months in vitro release data successfully predicted post three months release profile for up to 3.5 years. This was achieved by three aspects. First, the semi-empirical approach reflects implant properties and drug release mechanisms by incorporating experimental data and physical phenomena during drug release, including depletion zones formation overtime, drug supersaturation, and influence of un-stirred boundary layer. Second, the total drug release from implant was deconvoluted into ends-only release and skin-only release, which were effectively modeled separately. These two models were then combined together by accounting for reduction in effective release area due to the depletion zones formation to calculate total release. Third, an empirical-to-mechanistic mapping approach was applied to extract transport parameters within the depletion zones. The findings from this study offer valuable insights into the drug release mechanisms of reservoir type EVA-based long-acting implants with a core/shell structure and a rate-controlling membrane. On going studies to fine tune the model include (1) accounting for drug dissolution in model and (2) numerically solve Paul model in a cylindrical geometry using simulation software (e.g., COMSOL).

References: 1. Lee, E.K., et al., Transport of steroids in poly (etherurethane) and poly (ethylene vinyl acetate) membranes. Journal of membrane science, 1985. 24(2): p. 125-143.
2. Ritger, P.L. and N.A. Peppas, A simple equation for description of solute release I. Fickian and non-fickian release from non-swellable devices in the form of slabs, spheres, cylinders or discs. Journal of controlled release, 1987. 5(1): p. 23-36.
3. Higuchi, T., Mechanism of sustained‐action medication. Theoretical analysis of rate of release of solid drugs dispersed in solid matrices. Journal of Pharmaceutical Sciences, 1963. 52(12): p. 1145-1149.
4. Ren, A., et al., Manufacture, characterization, and elucidation of drug release mechanisms of etonogestrel implants based on ethylene vinyl acetate. Journal of Pharmaceutical Sciences, 2024.
5. Siepmann, J. and F. Siepmann, Modeling of diffusion controlled drug delivery. Journal of controlled release, 2012. 161(2): p. 351-362.
6. Paul, D., Solute release from membrane—matrix composites. Journal of membrane science, 1984. 21(2): p. 203-207.

Acknowledgements: This work was supported by the Broad Agency Announcement (BAA) Contract # 75F40122C00019 from the U.S. Food and Drug Administration (FDA). The content reflects the views of the authors and should not be construed to present FDA’s views or policies.


Figure 1. (A) Schematic illustration of paths of drug release: from skin (dashed arrow) and from ends (solid arrow). (B) Schematic illustration of reduction of effective release area caused by buildup of depletion zones overtime. Schemes are cross sectional view and not to scale. (C) Step by step roadmap of semi-empirical model development.


Figure 2. Semi-empirical model development: (A) Measured three months ends release, empirical curve fitting using diffusion-based model, and fitting of effective transport parameters using homogeneous Higuchi model. (B) Measured skin burst release (first 20 days) and its curve fitting using first order kinetic model. (C) Measured skin sustained release (post day 20), quantification of boundary layer effect, and predicted skin release using modified Paul model. (D) Measured three months Nexplanon® release and predicted release using the semi-empirical mode. Experimental data points were reported as mean ± SD (N = 3).


Figure 3. Semi-empirical model validation: (A) Published 3.5 years in vitro and in vivo drug release data and predicted drug release profile using the semi-empirical model developed in this study. (B) Drug release rate (mcg/day) prediction errors of semi-empirical model comparing to published in vitro release rate at day 20, day 40, ends of year 1, 2, and 3. Error bars in (A) were adapted from data source.