Lorenzo Cifelli 1, Jeroen Elassaiss-schaap 1
1 PD-value (Utrecht, Netherlands)
Introduction/Objective:
IMPRESM is a modeling tool for PK profile estimation and simulation, based on semi-parametric smoothing using P-splines and an impulse-response framework. Due to its flexible (‘one size fits all’) nature, it is especially well-suited to describe not only (linear) PK profiles, but also complicated absorption profiles, such as those exhibiting a ‘double-peak’. The framework allows one to fit a smooth curve to PK data, but also to extrapolate it to different dosing regimes encompassing different dose levels, dosing intervals and administration routes. It requires limited input from the investigator as the fitting and smoothness optimization is automated. A detailed introduction to the framework for the single-subject case can be found in [1].
This work extends the IMPRESM framework to ‘popPK’ modeling and covariate analysis, including the evaluation of the covariate effect and their extrapolation in simulations. First a population or reference PK profile is chosen. This may be the fitted PK profile for one of the subjects or a combined fit to all subjects using IMPRESM. Second, the reference profile is written as the convolution of an absorption and an elimination component, with the former mostly influencing the initial part of the PK curve, whereas the latter the terminal part. In certain situations the investigator may choose to decompose further the contribution from the elimination, analogously to using multiple compartments in traditional popPK modeling. Finally, with the contribution of absorption and elimination separated, they may be scaled, both in terms of their amplitude (‘vertical scaling’) and in time (‘horizontal scaling’), thus enabling the estimation of individual profiles as well as the inter-individual variance-covariance matrix, within a random-effects framework.
Methods
Three key steps are taken to support the work. First, the Markov chain Monte Carlo algorithm is used for the estimation of the population model and covariate effect. Second, the framework is evaluated on synthetic data generated from a traditional two-compartmental model incorporating inter-individual variability and body weight as covariate.. Third, the results from IMPRESM are compared to those using nlmixr.
Results
This proof-of-concept analysis highlighted links between the allometric scaling of clearance, volume of distribution and absorption rate constant. The results indicated that scaling the absorption and elimination components vertically is akin to scaling the volume of distribution, whereas horizontal scaling is akin to scaling rate parameters (e.g. clearance over volume). Similarly to traditional allometric scaling, the framework enables simulations under different covariate values. Furthermore, the framework can be used to estimate the covariate effects within the individual effects, rather than constraining them to fixed values. Finally, the method was evaluated on synthetic data for comparison of the estimated parameters with their ‘true’ values, and with nlmixr, showing comparable results.
Conclusion
This proof of concept study extends the single-subject framework of [1] to population modeling and covariate analysis. Due to its flexibility, the framework requires little input from the investigator and, therefore, a swift characterisation of the inter-individual variability and a covariate analysis is enabled. The framework was tested on synthetic data, recovering some of the essence of allometric scaling and parameter interpretability. Finally, the IMPRESM framework was found to be comparable to nlmixr.
References:
[1] L. Cifelli, P. Eilers, J. Elassiass-Schaap. Impulse-response modeling of PK curves utilizing P-splines. In preparation, 2026
Reference: PAGE 34 (2026) Abstr 12252 [www.page-meeting.org/?abstract=12252]
Poster: Methodology - New Modelling Approaches