Geraldine Celliere 1, Stephanie Kollmann 1, Jonathan Chauvin 1
1 Simulations Plus (Paris, France)
Introduction: The assessment of proarrhythmic risk via the analysis of QT interval prolongation is an integral part of drug development. Since the publication of the ICH E14 Q&A in 2015, QT assessment can be carried out with a concentration-QTc modeling approach [1]. The white paper by Garnett et al. [2] provides technical details for how to perform concentration-QTc modeling, based on a pre-specified linear mixed-effects model.
One of the assumptions to be verified before applying the pre-specified model is the absence of time delay between the drug concentration and its effect on ΔQTc. When a delay is detected and cannot be explained by the presence of a metabolite, the white paper suggests using a PK/PD model with an effect compartment. How this model is then used to derive the ΔΔQTc prediction interval versus concentration plot, which serves as a basis for excluding a 10-ms QTc prolongation effect, has not been addressed in detail in the literature. In this work, we present a step-by-step case study on Vanoxerine [3], showing how to (1) detect a delay, (2) implement a PK/PD model with an effect compartment and (3) derive the ΔΔQTc prediction interval versus concentration plot in case of a delay. It builds on top of our previous work on conc-QTc analysis automation [4, 5].
Methods: The study was a double-blind, placebo-controlled, dose-escalating trial assessing the safety, tolerance, and pharmacokinetics of Vanoxerine in cocaine-experienced volunteers [3].
The delay detection is performed using the MonolixSuite-based conc-QTc R package [4]. This package automatically generates a standard report containing the exploratory data analysis, assessment of model assumptions for the linear model, modeling results, and the model-derived ΔΔQTc prediction at concentrations of interest. For the delay detection, two diagnostics are proposed: a phase plot of the mean ΔQTc versus concentration connected in temporal order, and a comparison of the goodness of fit (BICc) of the pre-specified linear model versus a model with effect compartment.
When the model with effect compartment provides a better fit, this model cannot be used directly to derive the ΔΔQTc prediction interval vs concentration plot. Indeed, a given concentration does not translate directly into a corresponding ΔQTc or ΔΔQTc effect. The ΔQTc evolution depends on the entire past concentration evolution via the ODE for the effect compartment.
To be able to simulate a concentration evolution and its effect on ΔQTc or ΔΔQTc, it is necessary to develop a PK/PD model. The model is the same as above, except that the drug concentration is modeled too and it is the model-predicted concentration which influences ΔQTc. Several alternative models for the conc-ΔQTc relationship are also investigated: Emax, Hill and log-linear.
The PK/PD model is then used to simulate a typical individual (without random effects) at several dose levels. The uncertainty of the parameters is accounted for by sampling from the Fisher Information Matrix. For each dose level, the maximum concentration and maximum ΔΔQTc are recorded and used to generate the ΔΔQTc versus concentration plot. The confidence interval is computed as percentiles over the uncertainty replicates.
Results: For Vanoxerine, the ΔQTc vs concentration phase plot shows a counterclockwise pattern suggesting the presence of hysteresis. In addition, the model with effect compartment has a 17 points better BICc with an estimated delay of 2.3h. We thus conclude that the assumptions to apply the pre-specified model are not fulfilled and proceed with a PK/PD model. The PK data is well captured by a 2-compartment model with first-order absorption. For the PD part, we start from the pre-specified model. The addition of the effect compartment improves the BICc by 14 points, and the Emax relationship improves it by 8 points. The model with both modifications captures the data well and is used to perform simulations to compute the 90% confidence interval for the model-predicted ΔΔQTc. The upper bound of the confidence interval crosses the 10-ms threshold for a dose of 15 mg, corresponding to an average Cmax of 2.5 ng/mL.
Conclusion: This example demonstrates a robust conc-QTc analysis workflow in the case where a delay has been detected and the pre-specified model cannot be applied. The publicly available R scripts can serve as a template for the analysis of other drugs.
References:
[1] ICH E14 Guideline (2015) The clinical evaluation of QT/QTc interval prolongation and proarrhythmic potential for non-antiarrhythmic drugs. Questions & answers (R3).
[2] Garnett C, Bonate PL, Dang Q et al (2018) Scientific white paper on concentration-QTc modeling. J Pharmacokinet Pharmacodyn 45:383–397
[3] Preti A. Vanoxerine National Institute on Drug Abuse. Curr Opin Investig Drugs. 2000 Oct;1(2):241-51.
[4] Simulations Plus, Inc. R package for concentration-QT analysis. https://monolixsuite.slp-software.com/r-functions/2024R1/package-for-conc-qtc-analysis
[5] Celliere G, Krause A, Bonnefois G, Chavin J (2025). Beyond the linear model in concentration-QT analysis. J Pharmacokinet Pharmacodyn 52(3):31
Reference: PAGE 34 (2026) Abstr 12221 [www.page-meeting.org/?abstract=12221]
Poster: Drug/Disease Modelling - Safety