PKPD modeling for rapid dose selection with IMPRES-M, a novel smoothing approach.
Lorenzo Cifelli1, Jeroen Elassaiss-Schaap1
1PD-value D.V.
Introduction Selecting optimal doses for repeated administration in clinical trials is a critical task in drug development. Traditional pharmacokinetic (PK) and pharmacodynamic (PD) modeling approaches rely on mechanistic compartmental models. While these models provide valuable insights into drug behavior, they can be time-consuming, especially in early-stage trials where limited information on drug disposition is available. IMPRES-M [1] provides an alternative approach by directly estimating PK profiles from observed concentration data over time using a smoothing technique, bypassing the need for model development using e.g. compartmental equations. This work evaluates the feasibility of using IMPRES-M to extrapolate PK profiles and guide dose selection for repeated dosing, aiming to maximize PD impact while ensuring efficient decision-making. Specifically, we consider a scenario where PKPD data are available after a single oral dose, and the goal is to identify the dose that achieves, on average, 75% of the PD maximum effect during the final day of a two-week daily dosing regimen. Objective To identify the optimal dose for repeated administration based on PD response by leveraging IMPRES-M PK estimation and extrapolation, facilitating data-driven dose selection in drug development. Methods A simulated dataset mimicking a real PK-PD study was used. PK data were generated from a two-compartment model with extravascular administration and parameters: Ka = 1.5 h-1, CL = 0.3 L/h, Vc= 1.0 L, Q = 0.9 L/h, Vp = 2.1 L. Inter-individual variability (SD = 0.4) and residual error (SD = 0.2) were incorporated. The PD responses were simulated from an inhibitory indirect turnover model defined as follows dR/dt = kin(1- C/(C+IC50)) – kout*R, (1) with IC50=6 mg/L, kout=0.6 h-1, R(0)=R0=20 mg/L and kin=kout*R0. Inter-individual variability (SD = 0.1) and additive residual error were included. The drug concentration, C(t), and the PD response were sampled at times 0.1, 0.5, 1.0, 2.5, 6, 12, 18, 24, 36 , and 48 hours after a single oral dose. Three-fold increasing dosing levels of 50 mg, 150 mg, and 450 mg were considered, and for each dose level, data were simulated for three subjects per dose. PK profiles were estimated individually using IMPRES-M, which computes PK profiles from observed concentration measurements over time using a smoothing regression technique. This approach makes no assumptions about the underlying PK process, allowing the resulting function to be entirely data-driven. Specifically, a penalized B-spline approach [2] was employed, where the PK function C(t) is estimated as a linear combination of basis functions Bj(t) such that C(t)= SBj (t)ßj., where ßj are coefficients estimated from the observed data. A penalization is added on the ßjs coefficients to ensure a smooth PK function. An inhibitory indirect turnover PD model was then individually fitted to each subject’s PD observations using the smoothed PK profiles. Specifically, the PD model defined as in (1) was estimated using a standard optimization routine in R, applying the explicit formula derived from the ODE equation (1) as follows R(t) = exp(-kout t) {R0 + kin ?(1-C(t)/(C(t)+I50))exp(kout t)dt}, (2) with the integral evaluated numerically. To determine the dose achieving at least 75% of the maximal PD effect on average during the final day of a two-week daily dosing regimen, PK curves were extrapolated across a dose range of 10 mg to 1500 mg. The corresponding PD responses were computed, identifying the optimal dose. Results IMPRES-M successfully estimated PK profiles for all nine subjects across the three dose levels and accurately extrapolated exposure for the new dosing regimens. The estimated PD parameters where within 10% of the simulation model. By integrating PK extrapolations with the PD model, the analysis determined that a 400 mg daily dose achieved at least 75% of the maximum PD effect during the final day of treatment for all individuals. Conclusion This study highlights the potential of IMPRES-M in accelerating PKPD modeling for dose selection in drug development. By enabling rapid PK extrapolations without requiring explicit model development, this approach streamlines dose optimization, facilitating more efficient decision-making.
[1] Elassaiss-Schaap, J., Cifelli, L., & Eilers P.H.. PAGE 32 (2024) Abstr 11200. “Construction of IMPRES-M, a non-parametric impulse-response modeling method, in the context of varying pharmacokinetic profiles”. [https://www.page-meeting.org/default.asp?abstract=11200] [2] Eilers, P. H., & Marx, B. D. (2021). Practical smoothing: The joys of P-splines. Cambridge University Press.