A latent variable approach to account for correlated inputs in global sensitivity analysis with cases from pharmacological systems modelling

Abstract: In pharmaceutical research and development decision-making related to drug candidate selection, efficacy and safety is commonly supported through modelling and simulation (M&S). Among others, physiologically-based pharmacokinetic models are used to describe drug absorption, distribution and metabolism in human. Global sensitivity analysis (GSA) is gaining interest in the pharmacological M&S community as an important element for quality assessment of model-based inference. Physiological models often present inter-correlated parameters. The inclusion of correlated factors in GSA and the sensitivity indices interpretation has proven an issue for these models. Here we devise and evaluate a latent variable approach for dealing with correlated factors in GSA. This approach describes the correlation between two model inputs through the causal relationship of three independent factors: the latent variable and the unique variances of the two correlated parameters. Then, GSA is performed with the classical variance-based method. We applied the latent variable approach to a set of algebraic models and a case from physiologically-based pharmacokinetics. Then, we compared our approach to Sobol's GSA assuming no correlations, Sobol's GSA with groups and the Kucherenko approach. The relative ease of implementation and interpretation makes this a simple approach for carrying out GSA for models with correlated input factors.