Population PK Modeling of Dapivirine released from Vaginal Rings
Theo Reijmers (1), Stefan Zeiser (1), Marieke van den Dobbelsteen (1), Annalene Nel (2), Jeremy Nuttall (2), Neliëtte van Niekerk (2), Jeroen Elassaiss-Schaap (1,3)
(1) Kinesis Pharma BV; (2) International Partnership for Microbicides, Silver Spring, MD, USA; (3) PD-value BV
Objectives: Dapivirine is a non-nucleoside reverse transcriptase inhibitor with potent antiviral activity against HIV-1. International Partnership for Microbicides has developed a vaginal ring containing dapivirine (25 mg) to reduce the risk of HIV infection through sexual intercourse for women. These rings are placed in the upper third of the vagina resulting in a sustained release of dapivirine for more than a month. A population PK model was built to describe concentrations of dapivirine in vaginal fluids and plasma. The objective was to provide a robust aggregate for exposure from vaginal rings across Phase 1 and Phase 3 trials and to report change in exposure under different conditions or special populations.
Methods: Non-linear mixed effects PK modeling using NONMEM was performed based on vaginal fluid and plasma dapivirine concentrations. Two models were developed, a model specific for the fluid data, and a model describing the fluid and plasma data together. The impact of covariates on the fluid model was tested using the stepwise covariate modeling (SCM) method.
Results: A one-compartmental model with first-order absorption, parameterized in elimination rate and volume of distribution (V), was found to describe all fluid data optimally. After performing SCM and screening for clinical significance, only the relationships of pH of vaginal fluids and site ID with V were retained. The parameters of the fluid model were well estimated with standard errors (<25%) close to the bootstrap results. Simulation properties as assessed using a VPC were considered adequate with most predictions matching the observations at medians and the extremes of the distribution. The fluid-plasma link model was developed by assuming a steady-state concentration ratio between plasma and fluid. Uncertainty of parameter estimates of the model (<33%) were similar to those obtained with a bootstrap. The performance in the VPC of the fluid-plasma link model was not completely adequate (underprediction of median and overprediction of 95th percentile plasma concentrations).
Conclusions: The final fluid model was adequate in terms of parameter estimates, bootstrap performance and simulation properties. The pH of vaginal fluid was included as a covariate and resulted in a 22% reduction in exposure at the 90th percentile of the population. Reduced exposure at 2 South-African study sites (29% and 66%) might be explained by different adherence patterns of ring use that may vary according to local culture.
