(T-033) Practical PK/PD Modeling of Lymphocytes and CD4+ T Cells in HIV Patients Treated with the Nucleoside Reverse Transcriptase Translocation Inhibitor (NRTTI) Islatravir
Tuesday, October 21, 2025
7:00 AM - 1:45 PM MDT
Location: Colorado A
Bill Poland, PhD – Certara; Irene Bae, PharmD – Certara; Michelle Pham – Merck and Co., Inc.; Brian Maas – Merck and Co., Inc.
VP & Lead Scientist Certara Portola Valley, California, United States
Disclosure(s):
Bill Poland: No financial relationships to disclose
Objectives: Islatravir (ISL) is a nucleoside reverse transcriptase translocation inhibitor being studied for treatment of HIV-1. The objective was to compare a fast, approximate method to a standard sequential method for estimating the long-term relationships of ISL and of active control to total lymphocyte and CD4+ T-cell counts.
Methods: Separate lymphocyte and CD4 pharmacodynamic models were developed with the same basic structure: turnover with estimated constant zero-order cell generation and exposure-dependent first-order cell elimination. A depot compartment was included in the CD4 model to better capture CD4 time-profiles. The elimination rate was assumed to increase with the intracellular concentration of the active metabolite, islatravir triphosphate (ISL-TP), in peripheral blood mononuclear cells (PBMCs), less a minimum or threshold ISL-TP level, raised to a power. Estimated parameters included this threshold and power, baseline cell counts for treatment-naïve and virologically suppressed participants separately, and turnover model parameters.
A separate population pharmacokinetic (PK) model was fitted to observed islatravir plasma and ISL-TP concentrations. In the standard sequential PK/PD method, individual post hoc (empirical Bayes) parameter estimates from the PK model were used to regenerate continuous-time plasma ISL and ISL-TP for the cell count modeling. However, ISL-TP, and especially the cell counts, change on a much slower time scale than plasma ISL, resulting in multi-day estimation runs (despite parallelization) using NONMEM’s FOCE-I method, due to a very large dataset of almost 70,000 observations. Therefore, in the approximate method, ISL-TP was precalculated and added to the dataset as a time-varying covariate, allowing removal of the dosing and PK calculations in the estimation. Weekly-average ISL-TP for each subject and week, approximated from simulations, was selected as a compromise between speed and accuracy. Daily, weekly, and monthly ISL dosing and dose changes in the dataset were accounted for, as was NONMEM’s “next-observation-carried-back” treatment of time-varying covariates. The final models were rerun with the full sequential method to check accuracy.
Results: The approximate method reduced NONMEM estimation time by 2 orders of magnitude to a little less than 1 hour. Estimates were comparable to the standard sequential method, e.g., parameters for the slope of the elimination rate effect from the approximate and standard methods were 0.0174 vs. 0.0168/µM of ISL-TP for lymphocytes, and 0.00737 vs. 0.00738/µM of ISL-TP for CD4 cells.
Conclusions: Reduction of the PK in PK/PD modeling to a time-varying covariate was sufficiently accurate for this problem and allowed model exploration that otherwise would have been prohibitively time consuming. This approach may be useful whenever computation time is limiting sequential PK-PD modeling due to very different PK and PD time scales and a large dataset.
