Felicien Le Louedec 1, Ari Brekkan 1, Niclas E. Jonsson
1 Pharmetheus AB (Uppsala, )
Introduction
Exposure-response (ER) analyses of repeatedly administered drugs often rely on the use of an exposure metric at steady-state: the area under the concentration-time curve at steady-state (AUCtau,ss) or the maximum concentration at steady-state (Cmaxss). The choice of one or another often relies on empirical considerations or conventional wisdom such as “AUC for efficacy, Cmax for toxicity”. The objectives of this work were to identify scenarios where AUCtau,ss and Cmaxss are decorrelated and to determine the statistical power to detect a drug response using either when decorrelated.
Methods
(1) PK simulations: A global sensitivity analysis across four PK structural models with 2-compartment kinetics: “standard”, “slow absorption”, “slow clearance”, and “non-linear” (Michaelis-Menten) elimination.
(2) Parameter grid: 10,000 unique parameter vectors were simulated per model (100 subjects per vector), varying inter-individual variability (IIV), typical typical clearance (TVCL), volume (TVV), dosing intervals, and IIV correlation matrices. Three dosing intervals were also simulated.
(3) Machine learning: Random Forest variable importance (VIP) was utilized to identify the primary drivers of exposure metric decorrelation (defined as Pearson r < 0.4).
(4) PK/PD Power Analysis: A highly decorrelated parameter set from the standard model was used to simulate repeated daily dosing (1000 mg) in 40 subjects over 100 trial iterations.
(5) Response Modeling: PD was simulated using an indirect response model (inhibition of Kin) driven by an Emax function and the continuous function of concentration vs time. Statistical power (p < 0.05 via linear regression) of AUCtau,ss vs. Cmaxss was evaluated against Peak and Mean PD responses across permutations of PD half-life (1, 6, and 24 hours), Hill coefficients (1 to 8), and IC50 values.
Results
(1) High (r > 0.6) correlations between AUCtau,ss and Cmaxss were observed for the majority of scenarios (~90% of scenarios featuring slow absorption, slow clearance and non-linear and ~60% of scenarios featuring standard PK)
(2) Low (r < 0.4) correlations between AUCtau,ss and Cmaxss were observed in <5% of scenarios featuring slow absorption, slow clearance and non-linear and <20% of scenarios featuring standard PK.
(3) Longer dosing intervals tend to result in more decorrelations in the standard PK model, as expected. The most important factor driving decorrelation in the standard PK scenario was the correlation in IIV between CL and V.
(4) When considering multi-dimensional factors, low correlations were observed for scenarios in the standard model with low typical half-life (log(2)*TVV/TVCL) and low IIV of CL and high IIV of V and low correlation between CL and V.
(5) AUCtau,ss largely outperformed or equaled Cmaxss on any non-steep ER relationship scenarios (HILL <= 2). Cmaxss tended to outperform AUCtau,ss with a very steep ER relationship (HILL >= 4).
Discussion
Routine empirical exploration of Cmaxss alongside AUCtau,ss in standard ER analyses offers limited value due to their inherently high correlation. It could be meaningfully decorrelated if different regimens (e.g. bid and qd),or mixed formulations/routes of administrations were studied. Even in scenarios where the metrics decouple (limited), AUCtau,ss remains the more robust predictor for mean and peak PD responses in standard turnover model. Statistical power was only higher with Cmaxss in systems with highly reactive, steep-sigmoidal target mechanisms. Multiple testing of exposure metrics could lead to misinterpretation of the relationship between exposure and response. Exploration of Cmaxss in addition to AUCtau,ss should be avoided if these results are proven to be robust.
Reference: PAGE 34 (2026) Abstr 12278 [www.page-meeting.org/?abstract=12278]
Poster: Methodology - Other topics