Automated concentration-QT data preparation, model selection and reporting in R

Introduction:Since the publication of the ICH E14 guidance in 2015, QT interval prolongation as-sessment can be carried out with a concentration-QTc modeling approach as part of single- or mul-tiple- dose escalation studies, instead of conducting a thorough QT/QTc study [ 1]. The scientific white paper by Garnett et al. [ 2] provides technical details for how to perform and report concen-tration-QTc modeling. Their recommendations include a pre-specified linear mixed-effects model and diagnostic plots to include in the report. This pre-specification provides the opportunity to automate the workflow. In this work, we pre-sent an R package which can perform a concentration-QTc analysis in an automated way from data preparation to full report generation.

Methods: The automated workflow is implemented using R (for data preparation), MonolixSuite (for parameter estimation, accessed via R using the lixoftConnectors), and Quarto (for reporting). In addition to the linear concentration-QTc relationship, other non-linear relationships are implemented, including loglinear, Emax, Emax with sigmoidicity, and delayed-effect models based on an effect compartment. The workflow includes three main steps corresponding to three separate R functions:

Data preparation: depending on the columns already present in the input dataset, the function computes QTc, QTc baseline, baseline-corrected QTc (ΔQTc), baseline- and placebo-corrected QTc (ΔΔQTc), and averages QTc over triplicates. The resulting formatted dataset also contains columns required for model implementation, such as time as a factor (categorical covariate), centered baseline as a continuous covariate, and drug concentration as an independent variable. Concentration-QTc analysis: using the formatted dataset and concentration-QTc models (linear, loglinear, Emax, Emax with sigmoidicity, delayed effect, or user-defined), Monolix projects are created and run to estimate the model parameters.

Reporting: the report generation is based on a Quarto template which includes the exploratory data analysis plots, a comparison of the different tested models, model diagnostic plots, and the 90% prediction interval for ΔΔQTc. The Quarto template can be modified by the user to customize or include additional elements. The proposed workflow and R functions are applied to several real and simulated datasets: the crossover datasets analyzed in Johannesen et al. [ 3– 4], the simulated dataset from the white paper supplement which displays a non-linear concentration-QTc relationship [ 2], and Vanoxerine with placebo and two different dose groups [ 5– 6].

Results: The R functions are able to handle the diverse designs and situations present in the tested real and simulated datasets. Only a few lines of R code are needed to create the report from the input dataset.

Conclusions: These findings demonstrate that the concentration-QTc R package provides a viable automated solution for QTc liability assessment. The associated time savings gives scientists the opportunity to focus on the interpretation of results rather than on the details of analysis implementation. The R functions can be freely downloaded from [ 7].