A latent variable approach to account for correlated inputs in global sensitivity analysis

Area of Science:

• Pharmacometrics and Computational Toxicology
• Mathematical Modeling in Pharmacology
• Drug Development and Regulatory Science

Background:

• Model-based methods, including physiologically-based pharmacokinetics (PBPK), are crucial for drug development decision-making.
• Global sensitivity analysis (GSA) is increasingly used for quality assessment of model-informed inference.
• Interpreting correlated factors within GSA presents a significant challenge.

Purpose of the Study:

• To develop and evaluate a novel latent variable approach for addressing correlated factors in GSA.
• To improve the interpretability and practical application of GSA in the presence of input parameter correlations.
• To provide a robust method for sensitivity analysis in complex biological and pharmacokinetic models.

Main Methods:

• Developed a latent variable approach modeling input correlations via a latent variable and unique variances.
• Applied the latent variable approach to algebraic models and a PBPK case study.
• Compared the latent variable approach against standard GSA methods (Sobol's without/with grouping, Kucherenko).

Main Results:

• The latent variable approach provides unique and interpretable sensitivity indices while preserving factor correlations.
• Compared methods either assume independence, group variables, or suffer from interpretation difficulties.
• The proposed method offers a practical solution for GSA with correlated inputs, overcoming limitations of existing techniques.

Conclusions:

• The latent variable approach offers a practical, interpretable, and easily implementable method for GSA with correlated inputs.
• This approach does not violate the independence assumption often inherent in GSA methods.
• It enhances the utility of GSA for supporting model-informed decision-making in drug development.